diff --git "a/split_einsum_v2/compiled/Unet.mlmodelc/model.mil" "b/split_einsum_v2/compiled/Unet.mlmodelc/model.mil" new file mode 100644--- /dev/null +++ "b/split_einsum_v2/compiled/Unet.mlmodelc/model.mil" @@ -0,0 +1,17620 @@ +program(1.0) +[buildInfo = dict, tensor>({{"coremlc-component-MIL", "5.30.0"}, {"coremlc-version", "1839.0.0"}})] +{ + func main(tensor encoder_hidden_states, tensor sample, tensor timestep) { + tensor var_25 = const()[name = tensor("op_25"), val = tensor(-1)]; + tensor var_42_axes_0 = const()[name = tensor("op_42_axes_0"), val = tensor([1])]; + tensor var_42_cast = expand_dims(axes = var_42_axes_0, x = timestep)[name = tensor("op_42_cast")]; + tensor var_44_to_fp16 = const()[name = tensor("op_44_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(64)))]; + tensor emb_3_cast = mul(x = var_42_cast, y = var_44_to_fp16)[name = tensor("emb_3_cast")]; + tensor var_49_cast = sin(x = emb_3_cast)[name = tensor("op_49_cast")]; + tensor var_50_cast = cos(x = emb_3_cast)[name = tensor("op_50_cast")]; + tensor emb_interleave_0 = const()[name = tensor("emb_interleave_0"), val = tensor(false)]; + tensor emb_cast = concat(axis = var_25, interleave = emb_interleave_0, values = (var_49_cast, var_50_cast))[name = tensor("emb_cast")]; + tensor var_54_begin_0 = const()[name = tensor("op_54_begin_0"), val = tensor([0, 160])]; + tensor var_54_end_0 = const()[name = tensor("op_54_end_0"), val = tensor([2, 320])]; + tensor var_54_end_mask_0 = const()[name = tensor("op_54_end_mask_0"), val = tensor([true, true])]; + tensor var_54_cast = slice_by_index(begin = var_54_begin_0, end = var_54_end_0, end_mask = var_54_end_mask_0, x = emb_cast)[name = tensor("op_54_cast")]; + tensor var_56_begin_0 = const()[name = tensor("op_56_begin_0"), val = tensor([0, 0])]; + tensor var_56_end_0 = const()[name = tensor("op_56_end_0"), val = tensor([2, 160])]; + tensor var_56_end_mask_0 = const()[name = tensor("op_56_end_mask_0"), val = tensor([true, false])]; + tensor var_56_cast = slice_by_index(begin = var_56_begin_0, end = var_56_end_0, end_mask = var_56_end_mask_0, x = emb_cast)[name = tensor("op_56_cast")]; + tensor sample_interleave_0 = const()[name = tensor("sample_interleave_0"), val = tensor(false)]; + tensor sample_cast = concat(axis = var_25, interleave = sample_interleave_0, values = (var_54_cast, var_56_cast))[name = tensor("sample_cast")]; + tensor var_59 = const()[name = tensor("op_59"), val = tensor(1)]; + tensor var_66_axes_0 = const()[name = tensor("op_66_axes_0"), val = tensor([-1])]; + tensor var_66_cast = expand_dims(axes = var_66_axes_0, x = sample_cast)[name = tensor("op_66_cast")]; + tensor input_1_axes_0 = const()[name = tensor("input_1_axes_0"), val = tensor([-1])]; + tensor input_1_cast = expand_dims(axes = input_1_axes_0, x = var_66_cast)[name = tensor("input_1_cast")]; + tensor var_70 = const()[name = tensor("op_70"), val = tensor([1, 1])]; + tensor var_72 = const()[name = tensor("op_72"), val = tensor([1, 1])]; + tensor input_3_pad_type_0 = const()[name = tensor("input_3_pad_type_0"), val = tensor("custom")]; + tensor input_3_pad_0 = const()[name = tensor("input_3_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor time_embedding_linear_1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(448))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(307712))), name = tensor("time_embedding_linear_1_weight_to_fp16_palettized"), shape = tensor([1280, 320, 1, 1])]; + tensor time_embedding_linear_1_bias_to_fp16 = const()[name = tensor("time_embedding_linear_1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(307904)))]; + tensor input_3_cast = conv(bias = time_embedding_linear_1_bias_to_fp16, dilations = var_72, groups = var_59, pad = input_3_pad_0, pad_type = input_3_pad_type_0, strides = var_70, weight = time_embedding_linear_1_weight_to_fp16_palettized, x = input_1_cast)[name = tensor("input_3_cast")]; + tensor input_5_cast = silu(x = input_3_cast)[name = tensor("input_5_cast")]; + tensor var_78 = const()[name = tensor("op_78"), val = tensor([1, 1])]; + tensor var_80 = const()[name = tensor("op_80"), val = tensor([1, 1])]; + tensor input_13_pad_type_0 = const()[name = tensor("input_13_pad_type_0"), val = tensor("custom")]; + tensor input_13_pad_0 = const()[name = tensor("input_13_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor time_embedding_linear_2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(310528))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1539392))), name = tensor("time_embedding_linear_2_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor time_embedding_linear_2_bias_to_fp16 = const()[name = tensor("time_embedding_linear_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1539584)))]; + tensor input_13_cast = conv(bias = time_embedding_linear_2_bias_to_fp16, dilations = var_80, groups = var_59, pad = input_13_pad_0, pad_type = input_13_pad_type_0, strides = var_78, weight = time_embedding_linear_2_weight_to_fp16_palettized, x = input_5_cast)[name = tensor("input_13_cast")]; + tensor var_86 = const()[name = tensor("op_86"), val = tensor(1)]; + tensor var_89 = const()[name = tensor("op_89"), val = tensor([1, 1])]; + tensor var_91 = const()[name = tensor("op_91"), val = tensor([1, 1])]; + tensor input_7_pad_type_0 = const()[name = tensor("input_7_pad_type_0"), val = tensor("custom")]; + tensor input_7_pad_0 = const()[name = tensor("input_7_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor conv_in_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1542208))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1550912))), name = tensor("conv_in_weight_to_fp16_palettized"), shape = tensor([320, 4, 3, 3])]; + tensor conv_in_bias_to_fp16 = const()[name = tensor("conv_in_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1551104)))]; + tensor input_7_cast = conv(bias = conv_in_bias_to_fp16, dilations = var_91, groups = var_86, pad = input_7_pad_0, pad_type = input_7_pad_type_0, strides = var_89, weight = conv_in_weight_to_fp16_palettized, x = sample)[name = tensor("input_7_cast")]; + tensor var_95 = const()[name = tensor("op_95"), val = tensor(3)]; + tensor var_115 = const()[name = tensor("op_115"), val = tensor(true)]; + tensor var_120 = const()[name = tensor("op_120"), val = tensor(1)]; + tensor reshape_0_shape_0 = const()[name = tensor("reshape_0_shape_0"), val = tensor([2, 32, 10, 64, 64])]; + tensor reshape_0_cast = reshape(shape = reshape_0_shape_0, x = input_7_cast)[name = tensor("reshape_0_cast")]; + tensor reduce_mean_0_axes_0 = const()[name = tensor("reduce_mean_0_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_0_keep_dims_0 = const()[name = tensor("reduce_mean_0_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_0_cast = reduce_mean(axes = reduce_mean_0_axes_0, keep_dims = reduce_mean_0_keep_dims_0, x = reshape_0_cast)[name = tensor("reduce_mean_0_cast")]; + tensor sub_0_cast = sub(x = reshape_0_cast, y = reduce_mean_0_cast)[name = tensor("sub_0_cast")]; + tensor square_0_cast = square(x = sub_0_cast)[name = tensor("square_0_cast")]; + tensor reduce_mean_2_axes_0 = const()[name = tensor("reduce_mean_2_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_2_keep_dims_0 = const()[name = tensor("reduce_mean_2_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_2_cast = reduce_mean(axes = reduce_mean_2_axes_0, keep_dims = reduce_mean_2_keep_dims_0, x = square_0_cast)[name = tensor("reduce_mean_2_cast")]; + tensor add_0_y_0_to_fp16 = const()[name = tensor("add_0_y_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_0_cast = add(x = reduce_mean_2_cast, y = add_0_y_0_to_fp16)[name = tensor("add_0_cast")]; + tensor sqrt_0_cast = sqrt(x = add_0_cast)[name = tensor("sqrt_0_cast")]; + tensor real_div_0_cast = real_div(x = sub_0_cast, y = sqrt_0_cast)[name = tensor("real_div_0_cast")]; + tensor reshape_1_shape_0 = const()[name = tensor("reshape_1_shape_0"), val = tensor([2, 320, 64, 64])]; + tensor reshape_1_cast = reshape(shape = reshape_1_shape_0, x = real_div_0_cast)[name = tensor("reshape_1_cast")]; + tensor add_1_mean_0_to_fp16 = const()[name = tensor("add_1_mean_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1551808)))]; + tensor add_1_variance_0_to_fp16 = const()[name = tensor("add_1_variance_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1552512)))]; + tensor add_1_gamma_0_to_fp16 = const()[name = tensor("add_1_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1553216)))]; + tensor add_1_beta_0_to_fp16 = const()[name = tensor("add_1_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1553920)))]; + tensor add_1_epsilon_0_to_fp16 = const()[name = tensor("add_1_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_1_cast = batch_norm(beta = add_1_beta_0_to_fp16, epsilon = add_1_epsilon_0_to_fp16, gamma = add_1_gamma_0_to_fp16, mean = add_1_mean_0_to_fp16, variance = add_1_variance_0_to_fp16, x = reshape_1_cast)[name = tensor("add_1_cast")]; + tensor input_11_cast = silu(x = add_1_cast)[name = tensor("input_11_cast")]; + tensor var_142 = const()[name = tensor("op_142"), val = tensor([1, 1])]; + tensor var_144 = const()[name = tensor("op_144"), val = tensor([1, 1])]; + tensor hidden_states_1_pad_type_0 = const()[name = tensor("hidden_states_1_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_1_pad_0 = const()[name = tensor("hidden_states_1_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor down_blocks_0_resnets_0_conv1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1554624))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2245888))), name = tensor("down_blocks_0_resnets_0_conv1_weight_to_fp16_palettized"), shape = tensor([320, 320, 3, 3])]; + tensor down_blocks_0_resnets_0_conv1_bias_to_fp16 = const()[name = tensor("down_blocks_0_resnets_0_conv1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2246080)))]; + tensor hidden_states_1_cast = conv(bias = down_blocks_0_resnets_0_conv1_bias_to_fp16, dilations = var_144, groups = var_120, pad = hidden_states_1_pad_0, pad_type = hidden_states_1_pad_type_0, strides = var_142, weight = down_blocks_0_resnets_0_conv1_weight_to_fp16_palettized, x = input_11_cast)[name = tensor("hidden_states_1_cast")]; + tensor input_15_cast = silu(x = input_13_cast)[name = tensor("input_15_cast")]; + tensor var_150 = const()[name = tensor("op_150"), val = tensor([1, 1])]; + tensor var_152 = const()[name = tensor("op_152"), val = tensor([1, 1])]; + tensor temb_1_pad_type_0 = const()[name = tensor("temb_1_pad_type_0"), val = tensor("custom")]; + tensor temb_1_pad_0 = const()[name = tensor("temb_1_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_0_resnets_0_time_emb_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2246784))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2554048))), name = tensor("down_blocks_0_resnets_0_time_emb_proj_weight_to_fp16_palettized"), shape = tensor([320, 1280, 1, 1])]; + tensor down_blocks_0_resnets_0_time_emb_proj_bias_to_fp16 = const()[name = tensor("down_blocks_0_resnets_0_time_emb_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2554240)))]; + tensor temb_1_cast = conv(bias = down_blocks_0_resnets_0_time_emb_proj_bias_to_fp16, dilations = var_152, groups = var_120, pad = temb_1_pad_0, pad_type = temb_1_pad_type_0, strides = var_150, weight = down_blocks_0_resnets_0_time_emb_proj_weight_to_fp16_palettized, x = input_15_cast)[name = tensor("temb_1_cast")]; + tensor input_17_cast = add(x = hidden_states_1_cast, y = temb_1_cast)[name = tensor("input_17_cast")]; + tensor reshape_4_shape_0 = const()[name = tensor("reshape_4_shape_0"), val = tensor([2, 32, 10, 64, 64])]; + tensor reshape_4_cast = reshape(shape = reshape_4_shape_0, x = input_17_cast)[name = tensor("reshape_4_cast")]; + tensor reduce_mean_3_axes_0 = const()[name = tensor("reduce_mean_3_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_3_keep_dims_0 = const()[name = tensor("reduce_mean_3_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_3_cast = reduce_mean(axes = reduce_mean_3_axes_0, keep_dims = reduce_mean_3_keep_dims_0, x = reshape_4_cast)[name = tensor("reduce_mean_3_cast")]; + tensor sub_2_cast = sub(x = reshape_4_cast, y = reduce_mean_3_cast)[name = tensor("sub_2_cast")]; + tensor square_1_cast = square(x = sub_2_cast)[name = tensor("square_1_cast")]; + tensor reduce_mean_5_axes_0 = const()[name = tensor("reduce_mean_5_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_5_keep_dims_0 = const()[name = tensor("reduce_mean_5_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_5_cast = reduce_mean(axes = reduce_mean_5_axes_0, keep_dims = reduce_mean_5_keep_dims_0, x = square_1_cast)[name = tensor("reduce_mean_5_cast")]; + tensor add_2_y_0_to_fp16 = const()[name = tensor("add_2_y_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_2_cast = add(x = reduce_mean_5_cast, y = add_2_y_0_to_fp16)[name = tensor("add_2_cast")]; + tensor sqrt_1_cast = sqrt(x = add_2_cast)[name = tensor("sqrt_1_cast")]; + tensor real_div_1_cast = real_div(x = sub_2_cast, y = sqrt_1_cast)[name = tensor("real_div_1_cast")]; + tensor reshape_5_shape_0 = const()[name = tensor("reshape_5_shape_0"), val = tensor([2, 320, 64, 64])]; + tensor reshape_5_cast = reshape(shape = reshape_5_shape_0, x = real_div_1_cast)[name = tensor("reshape_5_cast")]; + tensor add_3_gamma_0_to_fp16 = const()[name = tensor("add_3_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2554944)))]; + tensor add_3_beta_0_to_fp16 = const()[name = tensor("add_3_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2555648)))]; + tensor add_3_epsilon_0_to_fp16 = const()[name = tensor("add_3_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_3_cast = batch_norm(beta = add_3_beta_0_to_fp16, epsilon = add_3_epsilon_0_to_fp16, gamma = add_3_gamma_0_to_fp16, mean = add_1_mean_0_to_fp16, variance = add_1_variance_0_to_fp16, x = reshape_5_cast)[name = tensor("add_3_cast")]; + tensor input_21_cast = silu(x = add_3_cast)[name = tensor("input_21_cast")]; + tensor var_162 = const()[name = tensor("op_162"), val = tensor([1, 1])]; + tensor var_164 = const()[name = tensor("op_164"), val = tensor([1, 1])]; + tensor hidden_states_3_pad_type_0 = const()[name = tensor("hidden_states_3_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_3_pad_0 = const()[name = tensor("hidden_states_3_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor down_blocks_0_resnets_0_conv2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2556352))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3247616))), name = tensor("down_blocks_0_resnets_0_conv2_weight_to_fp16_palettized"), shape = tensor([320, 320, 3, 3])]; + tensor down_blocks_0_resnets_0_conv2_bias_to_fp16 = const()[name = tensor("down_blocks_0_resnets_0_conv2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3247808)))]; + tensor hidden_states_3_cast = conv(bias = down_blocks_0_resnets_0_conv2_bias_to_fp16, dilations = var_164, groups = var_120, pad = hidden_states_3_pad_0, pad_type = hidden_states_3_pad_type_0, strides = var_162, weight = down_blocks_0_resnets_0_conv2_weight_to_fp16_palettized, x = input_21_cast)[name = tensor("hidden_states_3_cast")]; + tensor hidden_states_5_cast = add(x = input_7_cast, y = hidden_states_3_cast)[name = tensor("hidden_states_5_cast")]; + tensor reshape_8_shape_0 = const()[name = tensor("reshape_8_shape_0"), val = tensor([2, 32, 10, 64, 64])]; + tensor reshape_8_cast = reshape(shape = reshape_8_shape_0, x = hidden_states_5_cast)[name = tensor("reshape_8_cast")]; + tensor reduce_mean_6_axes_0 = const()[name = tensor("reduce_mean_6_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_6_keep_dims_0 = const()[name = tensor("reduce_mean_6_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_6_cast = reduce_mean(axes = reduce_mean_6_axes_0, keep_dims = reduce_mean_6_keep_dims_0, x = reshape_8_cast)[name = tensor("reduce_mean_6_cast")]; + tensor sub_4_cast = sub(x = reshape_8_cast, y = reduce_mean_6_cast)[name = tensor("sub_4_cast")]; + tensor square_2_cast = square(x = sub_4_cast)[name = tensor("square_2_cast")]; + tensor reduce_mean_8_axes_0 = const()[name = tensor("reduce_mean_8_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_8_keep_dims_0 = const()[name = tensor("reduce_mean_8_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_8_cast = reduce_mean(axes = reduce_mean_8_axes_0, keep_dims = reduce_mean_8_keep_dims_0, x = square_2_cast)[name = tensor("reduce_mean_8_cast")]; + tensor add_4_y_0_to_fp16 = const()[name = tensor("add_4_y_0_to_fp16"), val = tensor(0x1.1p-20)]; + tensor add_4_cast = add(x = reduce_mean_8_cast, y = add_4_y_0_to_fp16)[name = tensor("add_4_cast")]; + tensor sqrt_2_cast = sqrt(x = add_4_cast)[name = tensor("sqrt_2_cast")]; + tensor real_div_2_cast = real_div(x = sub_4_cast, y = sqrt_2_cast)[name = tensor("real_div_2_cast")]; + tensor reshape_9_shape_0 = const()[name = tensor("reshape_9_shape_0"), val = tensor([2, 320, 64, 64])]; + tensor reshape_9_cast = reshape(shape = reshape_9_shape_0, x = real_div_2_cast)[name = tensor("reshape_9_cast")]; + tensor add_5_gamma_0_to_fp16 = const()[name = tensor("add_5_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3248512)))]; + tensor add_5_beta_0_to_fp16 = const()[name = tensor("add_5_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3249216)))]; + tensor add_5_epsilon_0_to_fp16 = const()[name = tensor("add_5_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_5_cast = batch_norm(beta = add_5_beta_0_to_fp16, epsilon = add_5_epsilon_0_to_fp16, gamma = add_5_gamma_0_to_fp16, mean = add_1_mean_0_to_fp16, variance = add_1_variance_0_to_fp16, x = reshape_9_cast)[name = tensor("add_5_cast")]; + tensor var_184 = const()[name = tensor("op_184"), val = tensor([1, 1])]; + tensor var_186 = const()[name = tensor("op_186"), val = tensor([1, 1])]; + tensor hidden_states_7_pad_type_0 = const()[name = tensor("hidden_states_7_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_7_pad_0 = const()[name = tensor("hidden_states_7_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_0_attentions_0_proj_in_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3249920))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3326784))), name = tensor("down_blocks_0_attentions_0_proj_in_weight_to_fp16_palettized"), shape = tensor([320, 320, 1, 1])]; + tensor down_blocks_0_attentions_0_proj_in_bias_to_fp16 = const()[name = tensor("down_blocks_0_attentions_0_proj_in_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3326976)))]; + tensor hidden_states_7_cast = conv(bias = down_blocks_0_attentions_0_proj_in_bias_to_fp16, dilations = var_186, groups = var_120, pad = hidden_states_7_pad_0, pad_type = hidden_states_7_pad_type_0, strides = var_184, weight = down_blocks_0_attentions_0_proj_in_weight_to_fp16_palettized, x = add_5_cast)[name = tensor("hidden_states_7_cast")]; + tensor var_191 = const()[name = tensor("op_191"), val = tensor([2, 320, 1, 4096])]; + tensor inputs_1_cast = reshape(shape = var_191, x = hidden_states_7_cast)[name = tensor("inputs_1_cast")]; + tensor var_201 = const()[name = tensor("op_201"), val = tensor([1])]; + tensor channels_mean_1_cast = reduce_mean(axes = var_201, keep_dims = var_115, x = inputs_1_cast)[name = tensor("channels_mean_1_cast")]; + tensor zero_mean_1_cast = sub(x = inputs_1_cast, y = channels_mean_1_cast)[name = tensor("zero_mean_1_cast")]; + tensor zero_mean_sq_1_cast = mul(x = zero_mean_1_cast, y = zero_mean_1_cast)[name = tensor("zero_mean_sq_1_cast")]; + tensor var_205 = const()[name = tensor("op_205"), val = tensor([1])]; + tensor var_206_cast = reduce_mean(axes = var_205, keep_dims = var_115, x = zero_mean_sq_1_cast)[name = tensor("op_206_cast")]; + tensor var_207_to_fp16 = const()[name = tensor("op_207_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_208_cast = add(x = var_206_cast, y = var_207_to_fp16)[name = tensor("op_208_cast")]; + tensor denom_1_epsilon_0_to_fp16 = const()[name = tensor("denom_1_epsilon_0_to_fp16"), val = tensor(0x1p-24)]; + tensor denom_1_cast = rsqrt(epsilon = denom_1_epsilon_0_to_fp16, x = var_208_cast)[name = tensor("denom_1_cast")]; + tensor out_1_cast = mul(x = zero_mean_1_cast, y = denom_1_cast)[name = tensor("out_1_cast")]; + tensor var_212_to_fp16 = const()[name = tensor("op_212_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3327680)))]; + tensor var_213_cast = add(x = out_1_cast, y = var_212_to_fp16)[name = tensor("op_213_cast")]; + tensor var_215_to_fp16 = const()[name = tensor("op_215_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3328384)))]; + tensor hidden_states_9_cast = mul(x = var_213_cast, y = var_215_to_fp16)[name = tensor("hidden_states_9_cast")]; + tensor var_222 = const()[name = tensor("op_222"), val = tensor([1, 1])]; + tensor var_224 = const()[name = tensor("op_224"), val = tensor([1, 1])]; + tensor q_1_pad_type_0 = const()[name = tensor("q_1_pad_type_0"), val = tensor("custom")]; + tensor q_1_pad_0 = const()[name = tensor("q_1_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_0_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3329088))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3405952))), name = tensor("down_blocks_0_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16_palettized"), shape = tensor([320, 320, 1, 1])]; + tensor q_1_cast = conv(dilations = var_224, groups = var_120, pad = q_1_pad_0, pad_type = q_1_pad_type_0, strides = var_222, weight = down_blocks_0_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16_palettized, x = hidden_states_9_cast)[name = tensor("q_1_cast")]; + tensor var_228 = const()[name = tensor("op_228"), val = tensor([1, 1])]; + tensor var_230 = const()[name = tensor("op_230"), val = tensor([1, 1])]; + tensor k_1_pad_type_0 = const()[name = tensor("k_1_pad_type_0"), val = tensor("custom")]; + tensor k_1_pad_0 = const()[name = tensor("k_1_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_0_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3406144))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3483008))), name = tensor("down_blocks_0_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16_palettized"), shape = tensor([320, 320, 1, 1])]; + tensor k_1_cast = conv(dilations = var_230, groups = var_120, pad = k_1_pad_0, pad_type = k_1_pad_type_0, strides = var_228, weight = down_blocks_0_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16_palettized, x = hidden_states_9_cast)[name = tensor("k_1_cast")]; + tensor var_234 = const()[name = tensor("op_234"), val = tensor([1, 1])]; + tensor var_236 = const()[name = tensor("op_236"), val = tensor([1, 1])]; + tensor v_1_pad_type_0 = const()[name = tensor("v_1_pad_type_0"), val = tensor("custom")]; + tensor v_1_pad_0 = const()[name = tensor("v_1_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_0_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3483200))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3560064))), name = tensor("down_blocks_0_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16_palettized"), shape = tensor([320, 320, 1, 1])]; + tensor v_1_cast = conv(dilations = var_236, groups = var_120, pad = v_1_pad_0, pad_type = v_1_pad_type_0, strides = var_234, weight = down_blocks_0_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16_palettized, x = hidden_states_9_cast)[name = tensor("v_1_cast")]; + tensor var_240_begin_0 = const()[name = tensor("op_240_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_240_end_0 = const()[name = tensor("op_240_end_0"), val = tensor([2, 64, 1, 4096])]; + tensor var_240_end_mask_0 = const()[name = tensor("op_240_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_240_cast = slice_by_index(begin = var_240_begin_0, end = var_240_end_0, end_mask = var_240_end_mask_0, x = q_1_cast)[name = tensor("op_240_cast")]; + tensor var_244_begin_0 = const()[name = tensor("op_244_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_244_end_0 = const()[name = tensor("op_244_end_0"), val = tensor([2, 128, 1, 4096])]; + tensor var_244_end_mask_0 = const()[name = tensor("op_244_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_244_cast = slice_by_index(begin = var_244_begin_0, end = var_244_end_0, end_mask = var_244_end_mask_0, x = q_1_cast)[name = tensor("op_244_cast")]; + tensor var_248_begin_0 = const()[name = tensor("op_248_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_248_end_0 = const()[name = tensor("op_248_end_0"), val = tensor([2, 192, 1, 4096])]; + tensor var_248_end_mask_0 = const()[name = tensor("op_248_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_248_cast = slice_by_index(begin = var_248_begin_0, end = var_248_end_0, end_mask = var_248_end_mask_0, x = q_1_cast)[name = tensor("op_248_cast")]; + tensor var_252_begin_0 = const()[name = tensor("op_252_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_252_end_0 = const()[name = tensor("op_252_end_0"), val = tensor([2, 256, 1, 4096])]; + tensor var_252_end_mask_0 = const()[name = tensor("op_252_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_252_cast = slice_by_index(begin = var_252_begin_0, end = var_252_end_0, end_mask = var_252_end_mask_0, x = q_1_cast)[name = tensor("op_252_cast")]; + tensor var_256_begin_0 = const()[name = tensor("op_256_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_256_end_0 = const()[name = tensor("op_256_end_0"), val = tensor([2, 320, 1, 4096])]; + tensor var_256_end_mask_0 = const()[name = tensor("op_256_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_256_cast = slice_by_index(begin = var_256_begin_0, end = var_256_end_0, end_mask = var_256_end_mask_0, x = q_1_cast)[name = tensor("op_256_cast")]; + tensor var_259_begin_0 = const()[name = tensor("op_259_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_259_end_0 = const()[name = tensor("op_259_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_259_end_mask_0 = const()[name = tensor("op_259_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_259_cast = slice_by_index(begin = var_259_begin_0, end = var_259_end_0, end_mask = var_259_end_mask_0, x = var_240_cast)[name = tensor("op_259_cast")]; + tensor var_260_begin_0 = const()[name = tensor("op_260_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_260_end_0 = const()[name = tensor("op_260_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_260_end_mask_0 = const()[name = tensor("op_260_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_260_cast = slice_by_index(begin = var_260_begin_0, end = var_260_end_0, end_mask = var_260_end_mask_0, x = var_240_cast)[name = tensor("op_260_cast")]; + tensor var_261_begin_0 = const()[name = tensor("op_261_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_261_end_0 = const()[name = tensor("op_261_end_0"), val = tensor([2, 64, 1, 1536])]; + tensor var_261_end_mask_0 = const()[name = tensor("op_261_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_261_cast = slice_by_index(begin = var_261_begin_0, end = var_261_end_0, end_mask = var_261_end_mask_0, x = var_240_cast)[name = tensor("op_261_cast")]; + tensor var_262_begin_0 = const()[name = tensor("op_262_begin_0"), val = tensor([0, 0, 0, 1536])]; + tensor var_262_end_0 = const()[name = tensor("op_262_end_0"), val = tensor([2, 64, 1, 2048])]; + tensor var_262_end_mask_0 = const()[name = tensor("op_262_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_262_cast = slice_by_index(begin = var_262_begin_0, end = var_262_end_0, end_mask = var_262_end_mask_0, x = var_240_cast)[name = tensor("op_262_cast")]; + tensor var_263_begin_0 = const()[name = tensor("op_263_begin_0"), val = tensor([0, 0, 0, 2048])]; + tensor var_263_end_0 = const()[name = tensor("op_263_end_0"), val = tensor([2, 64, 1, 2560])]; + tensor var_263_end_mask_0 = const()[name = tensor("op_263_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_263_cast = slice_by_index(begin = var_263_begin_0, end = var_263_end_0, end_mask = var_263_end_mask_0, x = var_240_cast)[name = tensor("op_263_cast")]; + tensor var_264_begin_0 = const()[name = tensor("op_264_begin_0"), val = tensor([0, 0, 0, 2560])]; + tensor var_264_end_0 = const()[name = tensor("op_264_end_0"), val = tensor([2, 64, 1, 3072])]; + tensor var_264_end_mask_0 = const()[name = tensor("op_264_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_264_cast = slice_by_index(begin = var_264_begin_0, end = var_264_end_0, end_mask = var_264_end_mask_0, x = var_240_cast)[name = tensor("op_264_cast")]; + tensor var_265_begin_0 = const()[name = tensor("op_265_begin_0"), val = tensor([0, 0, 0, 3072])]; + tensor var_265_end_0 = const()[name = tensor("op_265_end_0"), val = tensor([2, 64, 1, 3584])]; + tensor var_265_end_mask_0 = const()[name = tensor("op_265_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_265_cast = slice_by_index(begin = var_265_begin_0, end = var_265_end_0, end_mask = var_265_end_mask_0, x = var_240_cast)[name = tensor("op_265_cast")]; + tensor var_266_begin_0 = const()[name = tensor("op_266_begin_0"), val = tensor([0, 0, 0, 3584])]; + tensor var_266_end_0 = const()[name = tensor("op_266_end_0"), val = tensor([2, 64, 1, 4096])]; + tensor var_266_end_mask_0 = const()[name = tensor("op_266_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_266_cast = slice_by_index(begin = var_266_begin_0, end = var_266_end_0, end_mask = var_266_end_mask_0, x = var_240_cast)[name = tensor("op_266_cast")]; + tensor var_267_begin_0 = const()[name = tensor("op_267_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_267_end_0 = const()[name = tensor("op_267_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_267_end_mask_0 = const()[name = tensor("op_267_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_267_cast = slice_by_index(begin = var_267_begin_0, end = var_267_end_0, end_mask = var_267_end_mask_0, x = var_244_cast)[name = tensor("op_267_cast")]; + tensor var_268_begin_0 = const()[name = tensor("op_268_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_268_end_0 = const()[name = tensor("op_268_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_268_end_mask_0 = const()[name = tensor("op_268_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_268_cast = slice_by_index(begin = var_268_begin_0, end = var_268_end_0, end_mask = var_268_end_mask_0, x = var_244_cast)[name = tensor("op_268_cast")]; + tensor var_269_begin_0 = const()[name = tensor("op_269_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_269_end_0 = const()[name = tensor("op_269_end_0"), val = tensor([2, 64, 1, 1536])]; + tensor var_269_end_mask_0 = const()[name = tensor("op_269_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_269_cast = slice_by_index(begin = var_269_begin_0, end = var_269_end_0, end_mask = var_269_end_mask_0, x = var_244_cast)[name = tensor("op_269_cast")]; + tensor var_270_begin_0 = const()[name = tensor("op_270_begin_0"), val = tensor([0, 0, 0, 1536])]; + tensor var_270_end_0 = const()[name = tensor("op_270_end_0"), val = tensor([2, 64, 1, 2048])]; + tensor var_270_end_mask_0 = const()[name = tensor("op_270_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_270_cast = slice_by_index(begin = var_270_begin_0, end = var_270_end_0, end_mask = var_270_end_mask_0, x = var_244_cast)[name = tensor("op_270_cast")]; + tensor var_271_begin_0 = const()[name = tensor("op_271_begin_0"), val = tensor([0, 0, 0, 2048])]; + tensor var_271_end_0 = const()[name = tensor("op_271_end_0"), val = tensor([2, 64, 1, 2560])]; + tensor var_271_end_mask_0 = const()[name = tensor("op_271_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_271_cast = slice_by_index(begin = var_271_begin_0, end = var_271_end_0, end_mask = var_271_end_mask_0, x = var_244_cast)[name = tensor("op_271_cast")]; + tensor var_272_begin_0 = const()[name = tensor("op_272_begin_0"), val = tensor([0, 0, 0, 2560])]; + tensor var_272_end_0 = const()[name = tensor("op_272_end_0"), val = tensor([2, 64, 1, 3072])]; + tensor var_272_end_mask_0 = const()[name = tensor("op_272_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_272_cast = slice_by_index(begin = var_272_begin_0, end = var_272_end_0, end_mask = var_272_end_mask_0, x = var_244_cast)[name = tensor("op_272_cast")]; + tensor var_273_begin_0 = const()[name = tensor("op_273_begin_0"), val = tensor([0, 0, 0, 3072])]; + tensor var_273_end_0 = const()[name = tensor("op_273_end_0"), val = tensor([2, 64, 1, 3584])]; + tensor var_273_end_mask_0 = const()[name = tensor("op_273_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_273_cast = slice_by_index(begin = var_273_begin_0, end = var_273_end_0, end_mask = var_273_end_mask_0, x = var_244_cast)[name = tensor("op_273_cast")]; + tensor var_274_begin_0 = const()[name = tensor("op_274_begin_0"), val = tensor([0, 0, 0, 3584])]; + tensor var_274_end_0 = const()[name = tensor("op_274_end_0"), val = tensor([2, 64, 1, 4096])]; + tensor var_274_end_mask_0 = const()[name = tensor("op_274_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_274_cast = slice_by_index(begin = var_274_begin_0, end = var_274_end_0, end_mask = var_274_end_mask_0, x = var_244_cast)[name = tensor("op_274_cast")]; + tensor var_275_begin_0 = const()[name = tensor("op_275_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_275_end_0 = const()[name = tensor("op_275_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_275_end_mask_0 = const()[name = tensor("op_275_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_275_cast = slice_by_index(begin = var_275_begin_0, end = var_275_end_0, end_mask = var_275_end_mask_0, x = var_248_cast)[name = tensor("op_275_cast")]; + tensor var_276_begin_0 = const()[name = tensor("op_276_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_276_end_0 = const()[name = tensor("op_276_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_276_end_mask_0 = const()[name = tensor("op_276_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_276_cast = slice_by_index(begin = var_276_begin_0, end = var_276_end_0, end_mask = var_276_end_mask_0, x = var_248_cast)[name = tensor("op_276_cast")]; + tensor var_277_begin_0 = const()[name = tensor("op_277_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_277_end_0 = const()[name = tensor("op_277_end_0"), val = tensor([2, 64, 1, 1536])]; + tensor var_277_end_mask_0 = const()[name = tensor("op_277_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_277_cast = slice_by_index(begin = var_277_begin_0, end = var_277_end_0, end_mask = var_277_end_mask_0, x = var_248_cast)[name = tensor("op_277_cast")]; + tensor var_278_begin_0 = const()[name = tensor("op_278_begin_0"), val = tensor([0, 0, 0, 1536])]; + tensor var_278_end_0 = const()[name = tensor("op_278_end_0"), val = tensor([2, 64, 1, 2048])]; + tensor var_278_end_mask_0 = const()[name = tensor("op_278_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_278_cast = slice_by_index(begin = var_278_begin_0, end = var_278_end_0, end_mask = var_278_end_mask_0, x = var_248_cast)[name = tensor("op_278_cast")]; + tensor var_279_begin_0 = const()[name = tensor("op_279_begin_0"), val = tensor([0, 0, 0, 2048])]; + tensor var_279_end_0 = const()[name = tensor("op_279_end_0"), val = tensor([2, 64, 1, 2560])]; + tensor var_279_end_mask_0 = const()[name = tensor("op_279_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_279_cast = slice_by_index(begin = var_279_begin_0, end = var_279_end_0, end_mask = var_279_end_mask_0, x = var_248_cast)[name = tensor("op_279_cast")]; + tensor var_280_begin_0 = const()[name = tensor("op_280_begin_0"), val = tensor([0, 0, 0, 2560])]; + tensor var_280_end_0 = const()[name = tensor("op_280_end_0"), val = tensor([2, 64, 1, 3072])]; + tensor var_280_end_mask_0 = const()[name = tensor("op_280_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_280_cast = slice_by_index(begin = var_280_begin_0, end = var_280_end_0, end_mask = var_280_end_mask_0, x = var_248_cast)[name = tensor("op_280_cast")]; + tensor var_281_begin_0 = const()[name = tensor("op_281_begin_0"), val = tensor([0, 0, 0, 3072])]; + tensor var_281_end_0 = const()[name = tensor("op_281_end_0"), val = tensor([2, 64, 1, 3584])]; + tensor var_281_end_mask_0 = const()[name = tensor("op_281_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_281_cast = slice_by_index(begin = var_281_begin_0, end = var_281_end_0, end_mask = var_281_end_mask_0, x = var_248_cast)[name = tensor("op_281_cast")]; + tensor var_282_begin_0 = const()[name = tensor("op_282_begin_0"), val = tensor([0, 0, 0, 3584])]; + tensor var_282_end_0 = const()[name = tensor("op_282_end_0"), val = tensor([2, 64, 1, 4096])]; + tensor var_282_end_mask_0 = const()[name = tensor("op_282_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_282_cast = slice_by_index(begin = var_282_begin_0, end = var_282_end_0, end_mask = var_282_end_mask_0, x = var_248_cast)[name = tensor("op_282_cast")]; + tensor var_283_begin_0 = const()[name = tensor("op_283_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_283_end_0 = const()[name = tensor("op_283_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_283_end_mask_0 = const()[name = tensor("op_283_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_283_cast = slice_by_index(begin = var_283_begin_0, end = var_283_end_0, end_mask = var_283_end_mask_0, x = var_252_cast)[name = tensor("op_283_cast")]; + tensor var_284_begin_0 = const()[name = tensor("op_284_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_284_end_0 = const()[name = tensor("op_284_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_284_end_mask_0 = const()[name = tensor("op_284_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_284_cast = slice_by_index(begin = var_284_begin_0, end = var_284_end_0, end_mask = var_284_end_mask_0, x = var_252_cast)[name = tensor("op_284_cast")]; + tensor var_285_begin_0 = const()[name = tensor("op_285_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_285_end_0 = const()[name = tensor("op_285_end_0"), val = tensor([2, 64, 1, 1536])]; + tensor var_285_end_mask_0 = const()[name = tensor("op_285_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_285_cast = slice_by_index(begin = var_285_begin_0, end = var_285_end_0, end_mask = var_285_end_mask_0, x = var_252_cast)[name = tensor("op_285_cast")]; + tensor var_286_begin_0 = const()[name = tensor("op_286_begin_0"), val = tensor([0, 0, 0, 1536])]; + tensor var_286_end_0 = const()[name = tensor("op_286_end_0"), val = tensor([2, 64, 1, 2048])]; + tensor var_286_end_mask_0 = const()[name = tensor("op_286_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_286_cast = slice_by_index(begin = var_286_begin_0, end = var_286_end_0, end_mask = var_286_end_mask_0, x = var_252_cast)[name = tensor("op_286_cast")]; + tensor var_287_begin_0 = const()[name = tensor("op_287_begin_0"), val = tensor([0, 0, 0, 2048])]; + tensor var_287_end_0 = const()[name = tensor("op_287_end_0"), val = tensor([2, 64, 1, 2560])]; + tensor var_287_end_mask_0 = const()[name = tensor("op_287_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_287_cast = slice_by_index(begin = var_287_begin_0, end = var_287_end_0, end_mask = var_287_end_mask_0, x = var_252_cast)[name = tensor("op_287_cast")]; + tensor var_288_begin_0 = const()[name = tensor("op_288_begin_0"), val = tensor([0, 0, 0, 2560])]; + tensor var_288_end_0 = const()[name = tensor("op_288_end_0"), val = tensor([2, 64, 1, 3072])]; + tensor var_288_end_mask_0 = const()[name = tensor("op_288_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_288_cast = slice_by_index(begin = var_288_begin_0, end = var_288_end_0, end_mask = var_288_end_mask_0, x = var_252_cast)[name = tensor("op_288_cast")]; + tensor var_289_begin_0 = const()[name = tensor("op_289_begin_0"), val = tensor([0, 0, 0, 3072])]; + tensor var_289_end_0 = const()[name = tensor("op_289_end_0"), val = tensor([2, 64, 1, 3584])]; + tensor var_289_end_mask_0 = const()[name = tensor("op_289_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_289_cast = slice_by_index(begin = var_289_begin_0, end = var_289_end_0, end_mask = var_289_end_mask_0, x = var_252_cast)[name = tensor("op_289_cast")]; + tensor var_290_begin_0 = const()[name = tensor("op_290_begin_0"), val = tensor([0, 0, 0, 3584])]; + tensor var_290_end_0 = const()[name = tensor("op_290_end_0"), val = tensor([2, 64, 1, 4096])]; + tensor var_290_end_mask_0 = const()[name = tensor("op_290_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_290_cast = slice_by_index(begin = var_290_begin_0, end = var_290_end_0, end_mask = var_290_end_mask_0, x = var_252_cast)[name = tensor("op_290_cast")]; + tensor var_291_begin_0 = const()[name = tensor("op_291_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_291_end_0 = const()[name = tensor("op_291_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_291_end_mask_0 = const()[name = tensor("op_291_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_291_cast = slice_by_index(begin = var_291_begin_0, end = var_291_end_0, end_mask = var_291_end_mask_0, x = var_256_cast)[name = tensor("op_291_cast")]; + tensor var_292_begin_0 = const()[name = tensor("op_292_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_292_end_0 = const()[name = tensor("op_292_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_292_end_mask_0 = const()[name = tensor("op_292_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_292_cast = slice_by_index(begin = var_292_begin_0, end = var_292_end_0, end_mask = var_292_end_mask_0, x = var_256_cast)[name = tensor("op_292_cast")]; + tensor var_293_begin_0 = const()[name = tensor("op_293_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_293_end_0 = const()[name = tensor("op_293_end_0"), val = tensor([2, 64, 1, 1536])]; + tensor var_293_end_mask_0 = const()[name = tensor("op_293_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_293_cast = slice_by_index(begin = var_293_begin_0, end = var_293_end_0, end_mask = var_293_end_mask_0, x = var_256_cast)[name = tensor("op_293_cast")]; + tensor var_294_begin_0 = const()[name = tensor("op_294_begin_0"), val = tensor([0, 0, 0, 1536])]; + tensor var_294_end_0 = const()[name = tensor("op_294_end_0"), val = tensor([2, 64, 1, 2048])]; + tensor var_294_end_mask_0 = const()[name = tensor("op_294_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_294_cast = slice_by_index(begin = var_294_begin_0, end = var_294_end_0, end_mask = var_294_end_mask_0, x = var_256_cast)[name = tensor("op_294_cast")]; + tensor var_295_begin_0 = const()[name = tensor("op_295_begin_0"), val = tensor([0, 0, 0, 2048])]; + tensor var_295_end_0 = const()[name = tensor("op_295_end_0"), val = tensor([2, 64, 1, 2560])]; + tensor var_295_end_mask_0 = const()[name = tensor("op_295_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_295_cast = slice_by_index(begin = var_295_begin_0, end = var_295_end_0, end_mask = var_295_end_mask_0, x = var_256_cast)[name = tensor("op_295_cast")]; + tensor var_296_begin_0 = const()[name = tensor("op_296_begin_0"), val = tensor([0, 0, 0, 2560])]; + tensor var_296_end_0 = const()[name = tensor("op_296_end_0"), val = tensor([2, 64, 1, 3072])]; + tensor var_296_end_mask_0 = const()[name = tensor("op_296_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_296_cast = slice_by_index(begin = var_296_begin_0, end = var_296_end_0, end_mask = var_296_end_mask_0, x = var_256_cast)[name = tensor("op_296_cast")]; + tensor var_297_begin_0 = const()[name = tensor("op_297_begin_0"), val = tensor([0, 0, 0, 3072])]; + tensor var_297_end_0 = const()[name = tensor("op_297_end_0"), val = tensor([2, 64, 1, 3584])]; + tensor var_297_end_mask_0 = const()[name = tensor("op_297_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_297_cast = slice_by_index(begin = var_297_begin_0, end = var_297_end_0, end_mask = var_297_end_mask_0, x = var_256_cast)[name = tensor("op_297_cast")]; + tensor var_298_begin_0 = const()[name = tensor("op_298_begin_0"), val = tensor([0, 0, 0, 3584])]; + tensor var_298_end_0 = const()[name = tensor("op_298_end_0"), val = tensor([2, 64, 1, 4096])]; + tensor var_298_end_mask_0 = const()[name = tensor("op_298_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_298_cast = slice_by_index(begin = var_298_begin_0, end = var_298_end_0, end_mask = var_298_end_mask_0, x = var_256_cast)[name = tensor("op_298_cast")]; + tensor k_3_perm_0 = const()[name = tensor("k_3_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_303_begin_0 = const()[name = tensor("op_303_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_303_end_0 = const()[name = tensor("op_303_end_0"), val = tensor([2, 4096, 1, 64])]; + tensor var_303_end_mask_0 = const()[name = tensor("op_303_end_mask_0"), val = tensor([true, true, true, false])]; + tensor transpose_31 = transpose(perm = k_3_perm_0, x = k_1_cast)[name = tensor("transpose_31")]; + tensor var_303_cast = slice_by_index(begin = var_303_begin_0, end = var_303_end_0, end_mask = var_303_end_mask_0, x = transpose_31)[name = tensor("op_303_cast")]; + tensor var_307_begin_0 = const()[name = tensor("op_307_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_307_end_0 = const()[name = tensor("op_307_end_0"), val = tensor([2, 4096, 1, 128])]; + tensor var_307_end_mask_0 = const()[name = tensor("op_307_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_307_cast = slice_by_index(begin = var_307_begin_0, end = var_307_end_0, end_mask = var_307_end_mask_0, x = transpose_31)[name = tensor("op_307_cast")]; + tensor var_311_begin_0 = const()[name = tensor("op_311_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_311_end_0 = const()[name = tensor("op_311_end_0"), val = tensor([2, 4096, 1, 192])]; + tensor var_311_end_mask_0 = const()[name = tensor("op_311_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_311_cast = slice_by_index(begin = var_311_begin_0, end = var_311_end_0, end_mask = var_311_end_mask_0, x = transpose_31)[name = tensor("op_311_cast")]; + tensor var_315_begin_0 = const()[name = tensor("op_315_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_315_end_0 = const()[name = tensor("op_315_end_0"), val = tensor([2, 4096, 1, 256])]; + tensor var_315_end_mask_0 = const()[name = tensor("op_315_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_315_cast = slice_by_index(begin = var_315_begin_0, end = var_315_end_0, end_mask = var_315_end_mask_0, x = transpose_31)[name = tensor("op_315_cast")]; + tensor var_319_begin_0 = const()[name = tensor("op_319_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_319_end_0 = const()[name = tensor("op_319_end_0"), val = tensor([2, 4096, 1, 320])]; + tensor var_319_end_mask_0 = const()[name = tensor("op_319_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_319_cast = slice_by_index(begin = var_319_begin_0, end = var_319_end_0, end_mask = var_319_end_mask_0, x = transpose_31)[name = tensor("op_319_cast")]; + tensor var_321_begin_0 = const()[name = tensor("op_321_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_321_end_0 = const()[name = tensor("op_321_end_0"), val = tensor([2, 64, 1, 4096])]; + tensor var_321_end_mask_0 = const()[name = tensor("op_321_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_321_cast = slice_by_index(begin = var_321_begin_0, end = var_321_end_0, end_mask = var_321_end_mask_0, x = v_1_cast)[name = tensor("op_321_cast")]; + tensor var_325_begin_0 = const()[name = tensor("op_325_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_325_end_0 = const()[name = tensor("op_325_end_0"), val = tensor([2, 128, 1, 4096])]; + tensor var_325_end_mask_0 = const()[name = tensor("op_325_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_325_cast = slice_by_index(begin = var_325_begin_0, end = var_325_end_0, end_mask = var_325_end_mask_0, x = v_1_cast)[name = tensor("op_325_cast")]; + tensor var_329_begin_0 = const()[name = tensor("op_329_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_329_end_0 = const()[name = tensor("op_329_end_0"), val = tensor([2, 192, 1, 4096])]; + tensor var_329_end_mask_0 = const()[name = tensor("op_329_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_329_cast = slice_by_index(begin = var_329_begin_0, end = var_329_end_0, end_mask = var_329_end_mask_0, x = v_1_cast)[name = tensor("op_329_cast")]; + tensor var_333_begin_0 = const()[name = tensor("op_333_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_333_end_0 = const()[name = tensor("op_333_end_0"), val = tensor([2, 256, 1, 4096])]; + tensor var_333_end_mask_0 = const()[name = tensor("op_333_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_333_cast = slice_by_index(begin = var_333_begin_0, end = var_333_end_0, end_mask = var_333_end_mask_0, x = v_1_cast)[name = tensor("op_333_cast")]; + tensor var_337_begin_0 = const()[name = tensor("op_337_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_337_end_0 = const()[name = tensor("op_337_end_0"), val = tensor([2, 320, 1, 4096])]; + tensor var_337_end_mask_0 = const()[name = tensor("op_337_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_337_cast = slice_by_index(begin = var_337_begin_0, end = var_337_end_0, end_mask = var_337_end_mask_0, x = v_1_cast)[name = tensor("op_337_cast")]; + tensor var_341_equation_0 = const()[name = tensor("op_341_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_341_cast = einsum(equation = var_341_equation_0, values = (var_303_cast, var_259_cast))[name = tensor("op_341_cast")]; + tensor var_342_to_fp16 = const()[name = tensor("op_342_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1_cast = mul(x = var_341_cast, y = var_342_to_fp16)[name = tensor("aw_chunk_1_cast")]; + tensor var_345_equation_0 = const()[name = tensor("op_345_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_345_cast = einsum(equation = var_345_equation_0, values = (var_303_cast, var_260_cast))[name = tensor("op_345_cast")]; + tensor var_346_to_fp16 = const()[name = tensor("op_346_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3_cast = mul(x = var_345_cast, y = var_346_to_fp16)[name = tensor("aw_chunk_3_cast")]; + tensor var_349_equation_0 = const()[name = tensor("op_349_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_349_cast = einsum(equation = var_349_equation_0, values = (var_303_cast, var_261_cast))[name = tensor("op_349_cast")]; + tensor var_350_to_fp16 = const()[name = tensor("op_350_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_5_cast = mul(x = var_349_cast, y = var_350_to_fp16)[name = tensor("aw_chunk_5_cast")]; + tensor var_353_equation_0 = const()[name = tensor("op_353_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_353_cast = einsum(equation = var_353_equation_0, values = (var_303_cast, var_262_cast))[name = tensor("op_353_cast")]; + tensor var_354_to_fp16 = const()[name = tensor("op_354_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_7_cast = mul(x = var_353_cast, y = var_354_to_fp16)[name = tensor("aw_chunk_7_cast")]; + tensor var_357_equation_0 = const()[name = tensor("op_357_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_357_cast = einsum(equation = var_357_equation_0, values = (var_303_cast, var_263_cast))[name = tensor("op_357_cast")]; + tensor var_358_to_fp16 = const()[name = tensor("op_358_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_9_cast = mul(x = var_357_cast, y = var_358_to_fp16)[name = tensor("aw_chunk_9_cast")]; + tensor var_361_equation_0 = const()[name = tensor("op_361_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_361_cast = einsum(equation = var_361_equation_0, values = (var_303_cast, var_264_cast))[name = tensor("op_361_cast")]; + tensor var_362_to_fp16 = const()[name = tensor("op_362_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_11_cast = mul(x = var_361_cast, y = var_362_to_fp16)[name = tensor("aw_chunk_11_cast")]; + tensor var_365_equation_0 = const()[name = tensor("op_365_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_365_cast = einsum(equation = var_365_equation_0, values = (var_303_cast, var_265_cast))[name = tensor("op_365_cast")]; + tensor var_366_to_fp16 = const()[name = tensor("op_366_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_13_cast = mul(x = var_365_cast, y = var_366_to_fp16)[name = tensor("aw_chunk_13_cast")]; + tensor var_369_equation_0 = const()[name = tensor("op_369_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_369_cast = einsum(equation = var_369_equation_0, values = (var_303_cast, var_266_cast))[name = tensor("op_369_cast")]; + tensor var_370_to_fp16 = const()[name = tensor("op_370_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_15_cast = mul(x = var_369_cast, y = var_370_to_fp16)[name = tensor("aw_chunk_15_cast")]; + tensor var_373_equation_0 = const()[name = tensor("op_373_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_373_cast = einsum(equation = var_373_equation_0, values = (var_307_cast, var_267_cast))[name = tensor("op_373_cast")]; + tensor var_374_to_fp16 = const()[name = tensor("op_374_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_17_cast = mul(x = var_373_cast, y = var_374_to_fp16)[name = tensor("aw_chunk_17_cast")]; + tensor var_377_equation_0 = const()[name = tensor("op_377_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_377_cast = einsum(equation = var_377_equation_0, values = (var_307_cast, var_268_cast))[name = tensor("op_377_cast")]; + tensor var_378_to_fp16 = const()[name = tensor("op_378_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_19_cast = mul(x = var_377_cast, y = var_378_to_fp16)[name = tensor("aw_chunk_19_cast")]; + tensor var_381_equation_0 = const()[name = tensor("op_381_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_381_cast = einsum(equation = var_381_equation_0, values = (var_307_cast, var_269_cast))[name = tensor("op_381_cast")]; + tensor var_382_to_fp16 = const()[name = tensor("op_382_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_21_cast = mul(x = var_381_cast, y = var_382_to_fp16)[name = tensor("aw_chunk_21_cast")]; + tensor var_385_equation_0 = const()[name = tensor("op_385_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_385_cast = einsum(equation = var_385_equation_0, values = (var_307_cast, var_270_cast))[name = tensor("op_385_cast")]; + tensor var_386_to_fp16 = const()[name = tensor("op_386_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_23_cast = mul(x = var_385_cast, y = var_386_to_fp16)[name = tensor("aw_chunk_23_cast")]; + tensor var_389_equation_0 = const()[name = tensor("op_389_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_389_cast = einsum(equation = var_389_equation_0, values = (var_307_cast, var_271_cast))[name = tensor("op_389_cast")]; + tensor var_390_to_fp16 = const()[name = tensor("op_390_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_25_cast = mul(x = var_389_cast, y = var_390_to_fp16)[name = tensor("aw_chunk_25_cast")]; + tensor var_393_equation_0 = const()[name = tensor("op_393_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_393_cast = einsum(equation = var_393_equation_0, values = (var_307_cast, var_272_cast))[name = tensor("op_393_cast")]; + tensor var_394_to_fp16 = const()[name = tensor("op_394_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_27_cast = mul(x = var_393_cast, y = var_394_to_fp16)[name = tensor("aw_chunk_27_cast")]; + tensor var_397_equation_0 = const()[name = tensor("op_397_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_397_cast = einsum(equation = var_397_equation_0, values = (var_307_cast, var_273_cast))[name = tensor("op_397_cast")]; + tensor var_398_to_fp16 = const()[name = tensor("op_398_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_29_cast = mul(x = var_397_cast, y = var_398_to_fp16)[name = tensor("aw_chunk_29_cast")]; + tensor var_401_equation_0 = const()[name = tensor("op_401_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_401_cast = einsum(equation = var_401_equation_0, values = (var_307_cast, var_274_cast))[name = tensor("op_401_cast")]; + tensor var_402_to_fp16 = const()[name = tensor("op_402_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_31_cast = mul(x = var_401_cast, y = var_402_to_fp16)[name = tensor("aw_chunk_31_cast")]; + tensor var_405_equation_0 = const()[name = tensor("op_405_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_405_cast = einsum(equation = var_405_equation_0, values = (var_311_cast, var_275_cast))[name = tensor("op_405_cast")]; + tensor var_406_to_fp16 = const()[name = tensor("op_406_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_33_cast = mul(x = var_405_cast, y = var_406_to_fp16)[name = tensor("aw_chunk_33_cast")]; + tensor var_409_equation_0 = const()[name = tensor("op_409_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_409_cast = einsum(equation = var_409_equation_0, values = (var_311_cast, var_276_cast))[name = tensor("op_409_cast")]; + tensor var_410_to_fp16 = const()[name = tensor("op_410_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_35_cast = mul(x = var_409_cast, y = var_410_to_fp16)[name = tensor("aw_chunk_35_cast")]; + tensor var_413_equation_0 = const()[name = tensor("op_413_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_413_cast = einsum(equation = var_413_equation_0, values = (var_311_cast, var_277_cast))[name = tensor("op_413_cast")]; + tensor var_414_to_fp16 = const()[name = tensor("op_414_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_37_cast = mul(x = var_413_cast, y = var_414_to_fp16)[name = tensor("aw_chunk_37_cast")]; + tensor var_417_equation_0 = const()[name = tensor("op_417_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_417_cast = einsum(equation = var_417_equation_0, values = (var_311_cast, var_278_cast))[name = tensor("op_417_cast")]; + tensor var_418_to_fp16 = const()[name = tensor("op_418_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_39_cast = mul(x = var_417_cast, y = var_418_to_fp16)[name = tensor("aw_chunk_39_cast")]; + tensor var_421_equation_0 = const()[name = tensor("op_421_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_421_cast = einsum(equation = var_421_equation_0, values = (var_311_cast, var_279_cast))[name = tensor("op_421_cast")]; + tensor var_422_to_fp16 = const()[name = tensor("op_422_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_41_cast = mul(x = var_421_cast, y = var_422_to_fp16)[name = tensor("aw_chunk_41_cast")]; + tensor var_425_equation_0 = const()[name = tensor("op_425_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_425_cast = einsum(equation = var_425_equation_0, values = (var_311_cast, var_280_cast))[name = tensor("op_425_cast")]; + tensor var_426_to_fp16 = const()[name = tensor("op_426_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_43_cast = mul(x = var_425_cast, y = var_426_to_fp16)[name = tensor("aw_chunk_43_cast")]; + tensor var_429_equation_0 = const()[name = tensor("op_429_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_429_cast = einsum(equation = var_429_equation_0, values = (var_311_cast, var_281_cast))[name = tensor("op_429_cast")]; + tensor var_430_to_fp16 = const()[name = tensor("op_430_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_45_cast = mul(x = var_429_cast, y = var_430_to_fp16)[name = tensor("aw_chunk_45_cast")]; + tensor var_433_equation_0 = const()[name = tensor("op_433_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_433_cast = einsum(equation = var_433_equation_0, values = (var_311_cast, var_282_cast))[name = tensor("op_433_cast")]; + tensor var_434_to_fp16 = const()[name = tensor("op_434_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_47_cast = mul(x = var_433_cast, y = var_434_to_fp16)[name = tensor("aw_chunk_47_cast")]; + tensor var_437_equation_0 = const()[name = tensor("op_437_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_437_cast = einsum(equation = var_437_equation_0, values = (var_315_cast, var_283_cast))[name = tensor("op_437_cast")]; + tensor var_438_to_fp16 = const()[name = tensor("op_438_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_49_cast = mul(x = var_437_cast, y = var_438_to_fp16)[name = tensor("aw_chunk_49_cast")]; + tensor var_441_equation_0 = const()[name = tensor("op_441_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_441_cast = einsum(equation = var_441_equation_0, values = (var_315_cast, var_284_cast))[name = tensor("op_441_cast")]; + tensor var_442_to_fp16 = const()[name = tensor("op_442_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_51_cast = mul(x = var_441_cast, y = var_442_to_fp16)[name = tensor("aw_chunk_51_cast")]; + tensor var_445_equation_0 = const()[name = tensor("op_445_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_445_cast = einsum(equation = var_445_equation_0, values = (var_315_cast, var_285_cast))[name = tensor("op_445_cast")]; + tensor var_446_to_fp16 = const()[name = tensor("op_446_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_53_cast = mul(x = var_445_cast, y = var_446_to_fp16)[name = tensor("aw_chunk_53_cast")]; + tensor var_449_equation_0 = const()[name = tensor("op_449_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_449_cast = einsum(equation = var_449_equation_0, values = (var_315_cast, var_286_cast))[name = tensor("op_449_cast")]; + tensor var_450_to_fp16 = const()[name = tensor("op_450_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_55_cast = mul(x = var_449_cast, y = var_450_to_fp16)[name = tensor("aw_chunk_55_cast")]; + tensor var_453_equation_0 = const()[name = tensor("op_453_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_453_cast = einsum(equation = var_453_equation_0, values = (var_315_cast, var_287_cast))[name = tensor("op_453_cast")]; + tensor var_454_to_fp16 = const()[name = tensor("op_454_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_57_cast = mul(x = var_453_cast, y = var_454_to_fp16)[name = tensor("aw_chunk_57_cast")]; + tensor var_457_equation_0 = const()[name = tensor("op_457_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_457_cast = einsum(equation = var_457_equation_0, values = (var_315_cast, var_288_cast))[name = tensor("op_457_cast")]; + tensor var_458_to_fp16 = const()[name = tensor("op_458_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_59_cast = mul(x = var_457_cast, y = var_458_to_fp16)[name = tensor("aw_chunk_59_cast")]; + tensor var_461_equation_0 = const()[name = tensor("op_461_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_461_cast = einsum(equation = var_461_equation_0, values = (var_315_cast, var_289_cast))[name = tensor("op_461_cast")]; + tensor var_462_to_fp16 = const()[name = tensor("op_462_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_61_cast = mul(x = var_461_cast, y = var_462_to_fp16)[name = tensor("aw_chunk_61_cast")]; + tensor var_465_equation_0 = const()[name = tensor("op_465_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_465_cast = einsum(equation = var_465_equation_0, values = (var_315_cast, var_290_cast))[name = tensor("op_465_cast")]; + tensor var_466_to_fp16 = const()[name = tensor("op_466_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_63_cast = mul(x = var_465_cast, y = var_466_to_fp16)[name = tensor("aw_chunk_63_cast")]; + tensor var_469_equation_0 = const()[name = tensor("op_469_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_469_cast = einsum(equation = var_469_equation_0, values = (var_319_cast, var_291_cast))[name = tensor("op_469_cast")]; + tensor var_470_to_fp16 = const()[name = tensor("op_470_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_65_cast = mul(x = var_469_cast, y = var_470_to_fp16)[name = tensor("aw_chunk_65_cast")]; + tensor var_473_equation_0 = const()[name = tensor("op_473_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_473_cast = einsum(equation = var_473_equation_0, values = (var_319_cast, var_292_cast))[name = tensor("op_473_cast")]; + tensor var_474_to_fp16 = const()[name = tensor("op_474_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_67_cast = mul(x = var_473_cast, y = var_474_to_fp16)[name = tensor("aw_chunk_67_cast")]; + tensor var_477_equation_0 = const()[name = tensor("op_477_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_477_cast = einsum(equation = var_477_equation_0, values = (var_319_cast, var_293_cast))[name = tensor("op_477_cast")]; + tensor var_478_to_fp16 = const()[name = tensor("op_478_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_69_cast = mul(x = var_477_cast, y = var_478_to_fp16)[name = tensor("aw_chunk_69_cast")]; + tensor var_481_equation_0 = const()[name = tensor("op_481_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_481_cast = einsum(equation = var_481_equation_0, values = (var_319_cast, var_294_cast))[name = tensor("op_481_cast")]; + tensor var_482_to_fp16 = const()[name = tensor("op_482_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_71_cast = mul(x = var_481_cast, y = var_482_to_fp16)[name = tensor("aw_chunk_71_cast")]; + tensor var_485_equation_0 = const()[name = tensor("op_485_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_485_cast = einsum(equation = var_485_equation_0, values = (var_319_cast, var_295_cast))[name = tensor("op_485_cast")]; + tensor var_486_to_fp16 = const()[name = tensor("op_486_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_73_cast = mul(x = var_485_cast, y = var_486_to_fp16)[name = tensor("aw_chunk_73_cast")]; + tensor var_489_equation_0 = const()[name = tensor("op_489_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_489_cast = einsum(equation = var_489_equation_0, values = (var_319_cast, var_296_cast))[name = tensor("op_489_cast")]; + tensor var_490_to_fp16 = const()[name = tensor("op_490_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_75_cast = mul(x = var_489_cast, y = var_490_to_fp16)[name = tensor("aw_chunk_75_cast")]; + tensor var_493_equation_0 = const()[name = tensor("op_493_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_493_cast = einsum(equation = var_493_equation_0, values = (var_319_cast, var_297_cast))[name = tensor("op_493_cast")]; + tensor var_494_to_fp16 = const()[name = tensor("op_494_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_77_cast = mul(x = var_493_cast, y = var_494_to_fp16)[name = tensor("aw_chunk_77_cast")]; + tensor var_497_equation_0 = const()[name = tensor("op_497_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_497_cast = einsum(equation = var_497_equation_0, values = (var_319_cast, var_298_cast))[name = tensor("op_497_cast")]; + tensor var_498_to_fp16 = const()[name = tensor("op_498_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_79_cast = mul(x = var_497_cast, y = var_498_to_fp16)[name = tensor("aw_chunk_79_cast")]; + tensor var_500_cast = softmax(axis = var_120, x = aw_chunk_1_cast)[name = tensor("op_500_cast")]; + tensor var_501_cast = softmax(axis = var_120, x = aw_chunk_3_cast)[name = tensor("op_501_cast")]; + tensor var_502_cast = softmax(axis = var_120, x = aw_chunk_5_cast)[name = tensor("op_502_cast")]; + tensor var_503_cast = softmax(axis = var_120, x = aw_chunk_7_cast)[name = tensor("op_503_cast")]; + tensor var_504_cast = softmax(axis = var_120, x = aw_chunk_9_cast)[name = tensor("op_504_cast")]; + tensor var_505_cast = softmax(axis = var_120, x = aw_chunk_11_cast)[name = tensor("op_505_cast")]; + tensor var_506_cast = softmax(axis = var_120, x = aw_chunk_13_cast)[name = tensor("op_506_cast")]; + tensor var_507_cast = softmax(axis = var_120, x = aw_chunk_15_cast)[name = tensor("op_507_cast")]; + tensor var_508_cast = softmax(axis = var_120, x = aw_chunk_17_cast)[name = tensor("op_508_cast")]; + tensor var_509_cast = softmax(axis = var_120, x = aw_chunk_19_cast)[name = tensor("op_509_cast")]; + tensor var_510_cast = softmax(axis = var_120, x = aw_chunk_21_cast)[name = tensor("op_510_cast")]; + tensor var_511_cast = softmax(axis = var_120, x = aw_chunk_23_cast)[name = tensor("op_511_cast")]; + tensor var_512_cast = softmax(axis = var_120, x = aw_chunk_25_cast)[name = tensor("op_512_cast")]; + tensor var_513_cast = softmax(axis = var_120, x = aw_chunk_27_cast)[name = tensor("op_513_cast")]; + tensor var_514_cast = softmax(axis = var_120, x = aw_chunk_29_cast)[name = tensor("op_514_cast")]; + tensor var_515_cast = softmax(axis = var_120, x = aw_chunk_31_cast)[name = tensor("op_515_cast")]; + tensor var_516_cast = softmax(axis = var_120, x = aw_chunk_33_cast)[name = tensor("op_516_cast")]; + tensor var_517_cast = softmax(axis = var_120, x = aw_chunk_35_cast)[name = tensor("op_517_cast")]; + tensor var_518_cast = softmax(axis = var_120, x = aw_chunk_37_cast)[name = tensor("op_518_cast")]; + tensor var_519_cast = softmax(axis = var_120, x = aw_chunk_39_cast)[name = tensor("op_519_cast")]; + tensor var_520_cast = softmax(axis = var_120, x = aw_chunk_41_cast)[name = tensor("op_520_cast")]; + tensor var_521_cast = softmax(axis = var_120, x = aw_chunk_43_cast)[name = tensor("op_521_cast")]; + tensor var_522_cast = softmax(axis = var_120, x = aw_chunk_45_cast)[name = tensor("op_522_cast")]; + tensor var_523_cast = softmax(axis = var_120, x = aw_chunk_47_cast)[name = tensor("op_523_cast")]; + tensor var_524_cast = softmax(axis = var_120, x = aw_chunk_49_cast)[name = tensor("op_524_cast")]; + tensor var_525_cast = softmax(axis = var_120, x = aw_chunk_51_cast)[name = tensor("op_525_cast")]; + tensor var_526_cast = softmax(axis = var_120, x = aw_chunk_53_cast)[name = tensor("op_526_cast")]; + tensor var_527_cast = softmax(axis = var_120, x = aw_chunk_55_cast)[name = tensor("op_527_cast")]; + tensor var_528_cast = softmax(axis = var_120, x = aw_chunk_57_cast)[name = tensor("op_528_cast")]; + tensor var_529_cast = softmax(axis = var_120, x = aw_chunk_59_cast)[name = tensor("op_529_cast")]; + tensor var_530_cast = softmax(axis = var_120, x = aw_chunk_61_cast)[name = tensor("op_530_cast")]; + tensor var_531_cast = softmax(axis = var_120, x = aw_chunk_63_cast)[name = tensor("op_531_cast")]; + tensor var_532_cast = softmax(axis = var_120, x = aw_chunk_65_cast)[name = tensor("op_532_cast")]; + tensor var_533_cast = softmax(axis = var_120, x = aw_chunk_67_cast)[name = tensor("op_533_cast")]; + tensor var_534_cast = softmax(axis = var_120, x = aw_chunk_69_cast)[name = tensor("op_534_cast")]; + tensor var_535_cast = softmax(axis = var_120, x = aw_chunk_71_cast)[name = tensor("op_535_cast")]; + tensor var_536_cast = softmax(axis = var_120, x = aw_chunk_73_cast)[name = tensor("op_536_cast")]; + tensor var_537_cast = softmax(axis = var_120, x = aw_chunk_75_cast)[name = tensor("op_537_cast")]; + tensor var_538_cast = softmax(axis = var_120, x = aw_chunk_77_cast)[name = tensor("op_538_cast")]; + tensor var_539_cast = softmax(axis = var_120, x = aw_chunk_79_cast)[name = tensor("op_539_cast")]; + tensor var_541_equation_0 = const()[name = tensor("op_541_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_541_cast = einsum(equation = var_541_equation_0, values = (var_321_cast, var_500_cast))[name = tensor("op_541_cast")]; + tensor var_543_equation_0 = const()[name = tensor("op_543_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_543_cast = einsum(equation = var_543_equation_0, values = (var_321_cast, var_501_cast))[name = tensor("op_543_cast")]; + tensor var_545_equation_0 = const()[name = tensor("op_545_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_545_cast = einsum(equation = var_545_equation_0, values = (var_321_cast, var_502_cast))[name = tensor("op_545_cast")]; + tensor var_547_equation_0 = const()[name = tensor("op_547_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_547_cast = einsum(equation = var_547_equation_0, values = (var_321_cast, var_503_cast))[name = tensor("op_547_cast")]; + tensor var_549_equation_0 = const()[name = tensor("op_549_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_549_cast = einsum(equation = var_549_equation_0, values = (var_321_cast, var_504_cast))[name = tensor("op_549_cast")]; + tensor var_551_equation_0 = const()[name = tensor("op_551_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_551_cast = einsum(equation = var_551_equation_0, values = (var_321_cast, var_505_cast))[name = tensor("op_551_cast")]; + tensor var_553_equation_0 = const()[name = tensor("op_553_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_553_cast = einsum(equation = var_553_equation_0, values = (var_321_cast, var_506_cast))[name = tensor("op_553_cast")]; + tensor var_555_equation_0 = const()[name = tensor("op_555_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_555_cast = einsum(equation = var_555_equation_0, values = (var_321_cast, var_507_cast))[name = tensor("op_555_cast")]; + tensor var_557_equation_0 = const()[name = tensor("op_557_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_557_cast = einsum(equation = var_557_equation_0, values = (var_325_cast, var_508_cast))[name = tensor("op_557_cast")]; + tensor var_559_equation_0 = const()[name = tensor("op_559_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_559_cast = einsum(equation = var_559_equation_0, values = (var_325_cast, var_509_cast))[name = tensor("op_559_cast")]; + tensor var_561_equation_0 = const()[name = tensor("op_561_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_561_cast = einsum(equation = var_561_equation_0, values = (var_325_cast, var_510_cast))[name = tensor("op_561_cast")]; + tensor var_563_equation_0 = const()[name = tensor("op_563_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_563_cast = einsum(equation = var_563_equation_0, values = (var_325_cast, var_511_cast))[name = tensor("op_563_cast")]; + tensor var_565_equation_0 = const()[name = tensor("op_565_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_565_cast = einsum(equation = var_565_equation_0, values = (var_325_cast, var_512_cast))[name = tensor("op_565_cast")]; + tensor var_567_equation_0 = const()[name = tensor("op_567_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_567_cast = einsum(equation = var_567_equation_0, values = (var_325_cast, var_513_cast))[name = tensor("op_567_cast")]; + tensor var_569_equation_0 = const()[name = tensor("op_569_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_569_cast = einsum(equation = var_569_equation_0, values = (var_325_cast, var_514_cast))[name = tensor("op_569_cast")]; + tensor var_571_equation_0 = const()[name = tensor("op_571_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_571_cast = einsum(equation = var_571_equation_0, values = (var_325_cast, var_515_cast))[name = tensor("op_571_cast")]; + tensor var_573_equation_0 = const()[name = tensor("op_573_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_573_cast = einsum(equation = var_573_equation_0, values = (var_329_cast, var_516_cast))[name = tensor("op_573_cast")]; + tensor var_575_equation_0 = const()[name = tensor("op_575_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_575_cast = einsum(equation = var_575_equation_0, values = (var_329_cast, var_517_cast))[name = tensor("op_575_cast")]; + tensor var_577_equation_0 = const()[name = tensor("op_577_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_577_cast = einsum(equation = var_577_equation_0, values = (var_329_cast, var_518_cast))[name = tensor("op_577_cast")]; + tensor var_579_equation_0 = const()[name = tensor("op_579_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_579_cast = einsum(equation = var_579_equation_0, values = (var_329_cast, var_519_cast))[name = tensor("op_579_cast")]; + tensor var_581_equation_0 = const()[name = tensor("op_581_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_581_cast = einsum(equation = var_581_equation_0, values = (var_329_cast, var_520_cast))[name = tensor("op_581_cast")]; + tensor var_583_equation_0 = const()[name = tensor("op_583_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_583_cast = einsum(equation = var_583_equation_0, values = (var_329_cast, var_521_cast))[name = tensor("op_583_cast")]; + tensor var_585_equation_0 = const()[name = tensor("op_585_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_585_cast = einsum(equation = var_585_equation_0, values = (var_329_cast, var_522_cast))[name = tensor("op_585_cast")]; + tensor var_587_equation_0 = const()[name = tensor("op_587_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_587_cast = einsum(equation = var_587_equation_0, values = (var_329_cast, var_523_cast))[name = tensor("op_587_cast")]; + tensor var_589_equation_0 = const()[name = tensor("op_589_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_589_cast = einsum(equation = var_589_equation_0, values = (var_333_cast, var_524_cast))[name = tensor("op_589_cast")]; + tensor var_591_equation_0 = const()[name = tensor("op_591_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_591_cast = einsum(equation = var_591_equation_0, values = (var_333_cast, var_525_cast))[name = tensor("op_591_cast")]; + tensor var_593_equation_0 = const()[name = tensor("op_593_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_593_cast = einsum(equation = var_593_equation_0, values = (var_333_cast, var_526_cast))[name = tensor("op_593_cast")]; + tensor var_595_equation_0 = const()[name = tensor("op_595_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_595_cast = einsum(equation = var_595_equation_0, values = (var_333_cast, var_527_cast))[name = tensor("op_595_cast")]; + tensor var_597_equation_0 = const()[name = tensor("op_597_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_597_cast = einsum(equation = var_597_equation_0, values = (var_333_cast, var_528_cast))[name = tensor("op_597_cast")]; + tensor var_599_equation_0 = const()[name = tensor("op_599_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_599_cast = einsum(equation = var_599_equation_0, values = (var_333_cast, var_529_cast))[name = tensor("op_599_cast")]; + tensor var_601_equation_0 = const()[name = tensor("op_601_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_601_cast = einsum(equation = var_601_equation_0, values = (var_333_cast, var_530_cast))[name = tensor("op_601_cast")]; + tensor var_603_equation_0 = const()[name = tensor("op_603_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_603_cast = einsum(equation = var_603_equation_0, values = (var_333_cast, var_531_cast))[name = tensor("op_603_cast")]; + tensor var_605_equation_0 = const()[name = tensor("op_605_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_605_cast = einsum(equation = var_605_equation_0, values = (var_337_cast, var_532_cast))[name = tensor("op_605_cast")]; + tensor var_607_equation_0 = const()[name = tensor("op_607_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_607_cast = einsum(equation = var_607_equation_0, values = (var_337_cast, var_533_cast))[name = tensor("op_607_cast")]; + tensor var_609_equation_0 = const()[name = tensor("op_609_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_609_cast = einsum(equation = var_609_equation_0, values = (var_337_cast, var_534_cast))[name = tensor("op_609_cast")]; + tensor var_611_equation_0 = const()[name = tensor("op_611_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_611_cast = einsum(equation = var_611_equation_0, values = (var_337_cast, var_535_cast))[name = tensor("op_611_cast")]; + tensor var_613_equation_0 = const()[name = tensor("op_613_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_613_cast = einsum(equation = var_613_equation_0, values = (var_337_cast, var_536_cast))[name = tensor("op_613_cast")]; + tensor var_615_equation_0 = const()[name = tensor("op_615_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_615_cast = einsum(equation = var_615_equation_0, values = (var_337_cast, var_537_cast))[name = tensor("op_615_cast")]; + tensor var_617_equation_0 = const()[name = tensor("op_617_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_617_cast = einsum(equation = var_617_equation_0, values = (var_337_cast, var_538_cast))[name = tensor("op_617_cast")]; + tensor var_619_equation_0 = const()[name = tensor("op_619_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_619_cast = einsum(equation = var_619_equation_0, values = (var_337_cast, var_539_cast))[name = tensor("op_619_cast")]; + tensor var_621_interleave_0 = const()[name = tensor("op_621_interleave_0"), val = tensor(false)]; + tensor var_621_cast = concat(axis = var_95, interleave = var_621_interleave_0, values = (var_541_cast, var_543_cast, var_545_cast, var_547_cast, var_549_cast, var_551_cast, var_553_cast, var_555_cast))[name = tensor("op_621_cast")]; + tensor var_623_interleave_0 = const()[name = tensor("op_623_interleave_0"), val = tensor(false)]; + tensor var_623_cast = concat(axis = var_95, interleave = var_623_interleave_0, values = (var_557_cast, var_559_cast, var_561_cast, var_563_cast, var_565_cast, var_567_cast, var_569_cast, var_571_cast))[name = tensor("op_623_cast")]; + tensor var_625_interleave_0 = const()[name = tensor("op_625_interleave_0"), val = tensor(false)]; + tensor var_625_cast = concat(axis = var_95, interleave = var_625_interleave_0, values = (var_573_cast, var_575_cast, var_577_cast, var_579_cast, var_581_cast, var_583_cast, var_585_cast, var_587_cast))[name = tensor("op_625_cast")]; + tensor var_627_interleave_0 = const()[name = tensor("op_627_interleave_0"), val = tensor(false)]; + tensor var_627_cast = concat(axis = var_95, interleave = var_627_interleave_0, values = (var_589_cast, var_591_cast, var_593_cast, var_595_cast, var_597_cast, var_599_cast, var_601_cast, var_603_cast))[name = tensor("op_627_cast")]; + tensor var_629_interleave_0 = const()[name = tensor("op_629_interleave_0"), val = tensor(false)]; + tensor var_629_cast = concat(axis = var_95, interleave = var_629_interleave_0, values = (var_605_cast, var_607_cast, var_609_cast, var_611_cast, var_613_cast, var_615_cast, var_617_cast, var_619_cast))[name = tensor("op_629_cast")]; + tensor input_25_interleave_0 = const()[name = tensor("input_25_interleave_0"), val = tensor(false)]; + tensor input_25_cast = concat(axis = var_120, interleave = input_25_interleave_0, values = (var_621_cast, var_623_cast, var_625_cast, var_627_cast, var_629_cast))[name = tensor("input_25_cast")]; + tensor var_635 = const()[name = tensor("op_635"), val = tensor([1, 1])]; + tensor var_637 = const()[name = tensor("op_637"), val = tensor([1, 1])]; + tensor var_639_pad_type_0 = const()[name = tensor("op_639_pad_type_0"), val = tensor("custom")]; + tensor var_639_pad_0 = const()[name = tensor("op_639_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_0_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3560256))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3637120))), name = tensor("down_blocks_0_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16_palettized"), shape = tensor([320, 320, 1, 1])]; + tensor down_blocks_0_attentions_0_transformer_blocks_0_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("down_blocks_0_attentions_0_transformer_blocks_0_attn1_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3637312)))]; + tensor var_639_cast = conv(bias = down_blocks_0_attentions_0_transformer_blocks_0_attn1_to_out_0_bias_to_fp16, dilations = var_637, groups = var_120, pad = var_639_pad_0, pad_type = var_639_pad_type_0, strides = var_635, weight = down_blocks_0_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16_palettized, x = input_25_cast)[name = tensor("op_639_cast")]; + tensor inputs_3_cast = add(x = var_639_cast, y = inputs_1_cast)[name = tensor("inputs_3_cast")]; + tensor var_643 = const()[name = tensor("op_643"), val = tensor([1])]; + tensor channels_mean_3_cast = reduce_mean(axes = var_643, keep_dims = var_115, x = inputs_3_cast)[name = tensor("channels_mean_3_cast")]; + tensor zero_mean_3_cast = sub(x = inputs_3_cast, y = channels_mean_3_cast)[name = tensor("zero_mean_3_cast")]; + tensor zero_mean_sq_3_cast = mul(x = zero_mean_3_cast, y = zero_mean_3_cast)[name = tensor("zero_mean_sq_3_cast")]; + tensor var_647 = const()[name = tensor("op_647"), val = tensor([1])]; + tensor var_648_cast = reduce_mean(axes = var_647, keep_dims = var_115, x = zero_mean_sq_3_cast)[name = tensor("op_648_cast")]; + tensor var_649_to_fp16 = const()[name = tensor("op_649_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_650_cast = add(x = var_648_cast, y = var_649_to_fp16)[name = tensor("op_650_cast")]; + tensor denom_3_epsilon_0_to_fp16 = const()[name = tensor("denom_3_epsilon_0_to_fp16"), val = tensor(0x1p-24)]; + tensor denom_3_cast = rsqrt(epsilon = denom_3_epsilon_0_to_fp16, x = var_650_cast)[name = tensor("denom_3_cast")]; + tensor out_3_cast = mul(x = zero_mean_3_cast, y = denom_3_cast)[name = tensor("out_3_cast")]; + tensor var_654_to_fp16 = const()[name = tensor("op_654_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3638016)))]; + tensor var_655_cast = add(x = out_3_cast, y = var_654_to_fp16)[name = tensor("op_655_cast")]; + tensor var_657_to_fp16 = const()[name = tensor("op_657_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3638720)))]; + tensor hidden_states_11_cast = mul(x = var_655_cast, y = var_657_to_fp16)[name = tensor("hidden_states_11_cast")]; + tensor var_664 = const()[name = tensor("op_664"), val = tensor([1, 1])]; + tensor var_666 = const()[name = tensor("op_666"), val = tensor([1, 1])]; + tensor q_3_pad_type_0 = const()[name = tensor("q_3_pad_type_0"), val = tensor("custom")]; + tensor q_3_pad_0 = const()[name = tensor("q_3_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_0_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3639424))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3716288))), name = tensor("down_blocks_0_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16_palettized"), shape = tensor([320, 320, 1, 1])]; + tensor q_3_cast = conv(dilations = var_666, groups = var_120, pad = q_3_pad_0, pad_type = q_3_pad_type_0, strides = var_664, weight = down_blocks_0_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16_palettized, x = hidden_states_11_cast)[name = tensor("q_3_cast")]; + tensor var_670 = const()[name = tensor("op_670"), val = tensor([1, 1])]; + tensor var_672 = const()[name = tensor("op_672"), val = tensor([1, 1])]; + tensor k_5_pad_type_0 = const()[name = tensor("k_5_pad_type_0"), val = tensor("custom")]; + tensor k_5_pad_0 = const()[name = tensor("k_5_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_0_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3716480))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3962304))), name = tensor("down_blocks_0_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized"), shape = tensor([320, 1024, 1, 1])]; + tensor k_5_cast = conv(dilations = var_672, groups = var_120, pad = k_5_pad_0, pad_type = k_5_pad_type_0, strides = var_670, weight = down_blocks_0_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor("k_5_cast")]; + tensor var_676 = const()[name = tensor("op_676"), val = tensor([1, 1])]; + tensor var_678 = const()[name = tensor("op_678"), val = tensor([1, 1])]; + tensor v_3_pad_type_0 = const()[name = tensor("v_3_pad_type_0"), val = tensor("custom")]; + tensor v_3_pad_0 = const()[name = tensor("v_3_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_0_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3962496))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4208320))), name = tensor("down_blocks_0_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized"), shape = tensor([320, 1024, 1, 1])]; + tensor v_3_cast = conv(dilations = var_678, groups = var_120, pad = v_3_pad_0, pad_type = v_3_pad_type_0, strides = var_676, weight = down_blocks_0_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor("v_3_cast")]; + tensor var_682_begin_0 = const()[name = tensor("op_682_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_682_end_0 = const()[name = tensor("op_682_end_0"), val = tensor([2, 64, 1, 4096])]; + tensor var_682_end_mask_0 = const()[name = tensor("op_682_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_682_cast = slice_by_index(begin = var_682_begin_0, end = var_682_end_0, end_mask = var_682_end_mask_0, x = q_3_cast)[name = tensor("op_682_cast")]; + tensor var_686_begin_0 = const()[name = tensor("op_686_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_686_end_0 = const()[name = tensor("op_686_end_0"), val = tensor([2, 128, 1, 4096])]; + tensor var_686_end_mask_0 = const()[name = tensor("op_686_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_686_cast = slice_by_index(begin = var_686_begin_0, end = var_686_end_0, end_mask = var_686_end_mask_0, x = q_3_cast)[name = tensor("op_686_cast")]; + tensor var_690_begin_0 = const()[name = tensor("op_690_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_690_end_0 = const()[name = tensor("op_690_end_0"), val = tensor([2, 192, 1, 4096])]; + tensor var_690_end_mask_0 = const()[name = tensor("op_690_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_690_cast = slice_by_index(begin = var_690_begin_0, end = var_690_end_0, end_mask = var_690_end_mask_0, x = q_3_cast)[name = tensor("op_690_cast")]; + tensor var_694_begin_0 = const()[name = tensor("op_694_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_694_end_0 = const()[name = tensor("op_694_end_0"), val = tensor([2, 256, 1, 4096])]; + tensor var_694_end_mask_0 = const()[name = tensor("op_694_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_694_cast = slice_by_index(begin = var_694_begin_0, end = var_694_end_0, end_mask = var_694_end_mask_0, x = q_3_cast)[name = tensor("op_694_cast")]; + tensor var_698_begin_0 = const()[name = tensor("op_698_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_698_end_0 = const()[name = tensor("op_698_end_0"), val = tensor([2, 320, 1, 4096])]; + tensor var_698_end_mask_0 = const()[name = tensor("op_698_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_698_cast = slice_by_index(begin = var_698_begin_0, end = var_698_end_0, end_mask = var_698_end_mask_0, x = q_3_cast)[name = tensor("op_698_cast")]; + tensor var_701_begin_0 = const()[name = tensor("op_701_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_701_end_0 = const()[name = tensor("op_701_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_701_end_mask_0 = const()[name = tensor("op_701_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_701_cast = slice_by_index(begin = var_701_begin_0, end = var_701_end_0, end_mask = var_701_end_mask_0, x = var_682_cast)[name = tensor("op_701_cast")]; + tensor var_702_begin_0 = const()[name = tensor("op_702_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_702_end_0 = const()[name = tensor("op_702_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_702_end_mask_0 = const()[name = tensor("op_702_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_702_cast = slice_by_index(begin = var_702_begin_0, end = var_702_end_0, end_mask = var_702_end_mask_0, x = var_682_cast)[name = tensor("op_702_cast")]; + tensor var_703_begin_0 = const()[name = tensor("op_703_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_703_end_0 = const()[name = tensor("op_703_end_0"), val = tensor([2, 64, 1, 1536])]; + tensor var_703_end_mask_0 = const()[name = tensor("op_703_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_703_cast = slice_by_index(begin = var_703_begin_0, end = var_703_end_0, end_mask = var_703_end_mask_0, x = var_682_cast)[name = tensor("op_703_cast")]; + tensor var_704_begin_0 = const()[name = tensor("op_704_begin_0"), val = tensor([0, 0, 0, 1536])]; + tensor var_704_end_0 = const()[name = tensor("op_704_end_0"), val = tensor([2, 64, 1, 2048])]; + tensor var_704_end_mask_0 = const()[name = tensor("op_704_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_704_cast = slice_by_index(begin = var_704_begin_0, end = var_704_end_0, end_mask = var_704_end_mask_0, x = var_682_cast)[name = tensor("op_704_cast")]; + tensor var_705_begin_0 = const()[name = tensor("op_705_begin_0"), val = tensor([0, 0, 0, 2048])]; + tensor var_705_end_0 = const()[name = tensor("op_705_end_0"), val = tensor([2, 64, 1, 2560])]; + tensor var_705_end_mask_0 = const()[name = tensor("op_705_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_705_cast = slice_by_index(begin = var_705_begin_0, end = var_705_end_0, end_mask = var_705_end_mask_0, x = var_682_cast)[name = tensor("op_705_cast")]; + tensor var_706_begin_0 = const()[name = tensor("op_706_begin_0"), val = tensor([0, 0, 0, 2560])]; + tensor var_706_end_0 = const()[name = tensor("op_706_end_0"), val = tensor([2, 64, 1, 3072])]; + tensor var_706_end_mask_0 = const()[name = tensor("op_706_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_706_cast = slice_by_index(begin = var_706_begin_0, end = var_706_end_0, end_mask = var_706_end_mask_0, x = var_682_cast)[name = tensor("op_706_cast")]; + tensor var_707_begin_0 = const()[name = tensor("op_707_begin_0"), val = tensor([0, 0, 0, 3072])]; + tensor var_707_end_0 = const()[name = tensor("op_707_end_0"), val = tensor([2, 64, 1, 3584])]; + tensor var_707_end_mask_0 = const()[name = tensor("op_707_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_707_cast = slice_by_index(begin = var_707_begin_0, end = var_707_end_0, end_mask = var_707_end_mask_0, x = var_682_cast)[name = tensor("op_707_cast")]; + tensor var_708_begin_0 = const()[name = tensor("op_708_begin_0"), val = tensor([0, 0, 0, 3584])]; + tensor var_708_end_0 = const()[name = tensor("op_708_end_0"), val = tensor([2, 64, 1, 4096])]; + tensor var_708_end_mask_0 = const()[name = tensor("op_708_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_708_cast = slice_by_index(begin = var_708_begin_0, end = var_708_end_0, end_mask = var_708_end_mask_0, x = var_682_cast)[name = tensor("op_708_cast")]; + tensor var_709_begin_0 = const()[name = tensor("op_709_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_709_end_0 = const()[name = tensor("op_709_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_709_end_mask_0 = const()[name = tensor("op_709_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_709_cast = slice_by_index(begin = var_709_begin_0, end = var_709_end_0, end_mask = var_709_end_mask_0, x = var_686_cast)[name = tensor("op_709_cast")]; + tensor var_710_begin_0 = const()[name = tensor("op_710_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_710_end_0 = const()[name = tensor("op_710_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_710_end_mask_0 = const()[name = tensor("op_710_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_710_cast = slice_by_index(begin = var_710_begin_0, end = var_710_end_0, end_mask = var_710_end_mask_0, x = var_686_cast)[name = tensor("op_710_cast")]; + tensor var_711_begin_0 = const()[name = tensor("op_711_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_711_end_0 = const()[name = tensor("op_711_end_0"), val = tensor([2, 64, 1, 1536])]; + tensor var_711_end_mask_0 = const()[name = tensor("op_711_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_711_cast = slice_by_index(begin = var_711_begin_0, end = var_711_end_0, end_mask = var_711_end_mask_0, x = var_686_cast)[name = tensor("op_711_cast")]; + tensor var_712_begin_0 = const()[name = tensor("op_712_begin_0"), val = tensor([0, 0, 0, 1536])]; + tensor var_712_end_0 = const()[name = tensor("op_712_end_0"), val = tensor([2, 64, 1, 2048])]; + tensor var_712_end_mask_0 = const()[name = tensor("op_712_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_712_cast = slice_by_index(begin = var_712_begin_0, end = var_712_end_0, end_mask = var_712_end_mask_0, x = var_686_cast)[name = tensor("op_712_cast")]; + tensor var_713_begin_0 = const()[name = tensor("op_713_begin_0"), val = tensor([0, 0, 0, 2048])]; + tensor var_713_end_0 = const()[name = tensor("op_713_end_0"), val = tensor([2, 64, 1, 2560])]; + tensor var_713_end_mask_0 = const()[name = tensor("op_713_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_713_cast = slice_by_index(begin = var_713_begin_0, end = var_713_end_0, end_mask = var_713_end_mask_0, x = var_686_cast)[name = tensor("op_713_cast")]; + tensor var_714_begin_0 = const()[name = tensor("op_714_begin_0"), val = tensor([0, 0, 0, 2560])]; + tensor var_714_end_0 = const()[name = tensor("op_714_end_0"), val = tensor([2, 64, 1, 3072])]; + tensor var_714_end_mask_0 = const()[name = tensor("op_714_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_714_cast = slice_by_index(begin = var_714_begin_0, end = var_714_end_0, end_mask = var_714_end_mask_0, x = var_686_cast)[name = tensor("op_714_cast")]; + tensor var_715_begin_0 = const()[name = tensor("op_715_begin_0"), val = tensor([0, 0, 0, 3072])]; + tensor var_715_end_0 = const()[name = tensor("op_715_end_0"), val = tensor([2, 64, 1, 3584])]; + tensor var_715_end_mask_0 = const()[name = tensor("op_715_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_715_cast = slice_by_index(begin = var_715_begin_0, end = var_715_end_0, end_mask = var_715_end_mask_0, x = var_686_cast)[name = tensor("op_715_cast")]; + tensor var_716_begin_0 = const()[name = tensor("op_716_begin_0"), val = tensor([0, 0, 0, 3584])]; + tensor var_716_end_0 = const()[name = tensor("op_716_end_0"), val = tensor([2, 64, 1, 4096])]; + tensor var_716_end_mask_0 = const()[name = tensor("op_716_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_716_cast = slice_by_index(begin = var_716_begin_0, end = var_716_end_0, end_mask = var_716_end_mask_0, x = var_686_cast)[name = tensor("op_716_cast")]; + tensor var_717_begin_0 = const()[name = tensor("op_717_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_717_end_0 = const()[name = tensor("op_717_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_717_end_mask_0 = const()[name = tensor("op_717_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_717_cast = slice_by_index(begin = var_717_begin_0, end = var_717_end_0, end_mask = var_717_end_mask_0, x = var_690_cast)[name = tensor("op_717_cast")]; + tensor var_718_begin_0 = const()[name = tensor("op_718_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_718_end_0 = const()[name = tensor("op_718_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_718_end_mask_0 = const()[name = tensor("op_718_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_718_cast = slice_by_index(begin = var_718_begin_0, end = var_718_end_0, end_mask = var_718_end_mask_0, x = var_690_cast)[name = tensor("op_718_cast")]; + tensor var_719_begin_0 = const()[name = tensor("op_719_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_719_end_0 = const()[name = tensor("op_719_end_0"), val = tensor([2, 64, 1, 1536])]; + tensor var_719_end_mask_0 = const()[name = tensor("op_719_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_719_cast = slice_by_index(begin = var_719_begin_0, end = var_719_end_0, end_mask = var_719_end_mask_0, x = var_690_cast)[name = tensor("op_719_cast")]; + tensor var_720_begin_0 = const()[name = tensor("op_720_begin_0"), val = tensor([0, 0, 0, 1536])]; + tensor var_720_end_0 = const()[name = tensor("op_720_end_0"), val = tensor([2, 64, 1, 2048])]; + tensor var_720_end_mask_0 = const()[name = tensor("op_720_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_720_cast = slice_by_index(begin = var_720_begin_0, end = var_720_end_0, end_mask = var_720_end_mask_0, x = var_690_cast)[name = tensor("op_720_cast")]; + tensor var_721_begin_0 = const()[name = tensor("op_721_begin_0"), val = tensor([0, 0, 0, 2048])]; + tensor var_721_end_0 = const()[name = tensor("op_721_end_0"), val = tensor([2, 64, 1, 2560])]; + tensor var_721_end_mask_0 = const()[name = tensor("op_721_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_721_cast = slice_by_index(begin = var_721_begin_0, end = var_721_end_0, end_mask = var_721_end_mask_0, x = var_690_cast)[name = tensor("op_721_cast")]; + tensor var_722_begin_0 = const()[name = tensor("op_722_begin_0"), val = tensor([0, 0, 0, 2560])]; + tensor var_722_end_0 = const()[name = tensor("op_722_end_0"), val = tensor([2, 64, 1, 3072])]; + tensor var_722_end_mask_0 = const()[name = tensor("op_722_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_722_cast = slice_by_index(begin = var_722_begin_0, end = var_722_end_0, end_mask = var_722_end_mask_0, x = var_690_cast)[name = tensor("op_722_cast")]; + tensor var_723_begin_0 = const()[name = tensor("op_723_begin_0"), val = tensor([0, 0, 0, 3072])]; + tensor var_723_end_0 = const()[name = tensor("op_723_end_0"), val = tensor([2, 64, 1, 3584])]; + tensor var_723_end_mask_0 = const()[name = tensor("op_723_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_723_cast = slice_by_index(begin = var_723_begin_0, end = var_723_end_0, end_mask = var_723_end_mask_0, x = var_690_cast)[name = tensor("op_723_cast")]; + tensor var_724_begin_0 = const()[name = tensor("op_724_begin_0"), val = tensor([0, 0, 0, 3584])]; + tensor var_724_end_0 = const()[name = tensor("op_724_end_0"), val = tensor([2, 64, 1, 4096])]; + tensor var_724_end_mask_0 = const()[name = tensor("op_724_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_724_cast = slice_by_index(begin = var_724_begin_0, end = var_724_end_0, end_mask = var_724_end_mask_0, x = var_690_cast)[name = tensor("op_724_cast")]; + tensor var_725_begin_0 = const()[name = tensor("op_725_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_725_end_0 = const()[name = tensor("op_725_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_725_end_mask_0 = const()[name = tensor("op_725_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_725_cast = slice_by_index(begin = var_725_begin_0, end = var_725_end_0, end_mask = var_725_end_mask_0, x = var_694_cast)[name = tensor("op_725_cast")]; + tensor var_726_begin_0 = const()[name = tensor("op_726_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_726_end_0 = const()[name = tensor("op_726_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_726_end_mask_0 = const()[name = tensor("op_726_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_726_cast = slice_by_index(begin = var_726_begin_0, end = var_726_end_0, end_mask = var_726_end_mask_0, x = var_694_cast)[name = tensor("op_726_cast")]; + tensor var_727_begin_0 = const()[name = tensor("op_727_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_727_end_0 = const()[name = tensor("op_727_end_0"), val = tensor([2, 64, 1, 1536])]; + tensor var_727_end_mask_0 = const()[name = tensor("op_727_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_727_cast = slice_by_index(begin = var_727_begin_0, end = var_727_end_0, end_mask = var_727_end_mask_0, x = var_694_cast)[name = tensor("op_727_cast")]; + tensor var_728_begin_0 = const()[name = tensor("op_728_begin_0"), val = tensor([0, 0, 0, 1536])]; + tensor var_728_end_0 = const()[name = tensor("op_728_end_0"), val = tensor([2, 64, 1, 2048])]; + tensor var_728_end_mask_0 = const()[name = tensor("op_728_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_728_cast = slice_by_index(begin = var_728_begin_0, end = var_728_end_0, end_mask = var_728_end_mask_0, x = var_694_cast)[name = tensor("op_728_cast")]; + tensor var_729_begin_0 = const()[name = tensor("op_729_begin_0"), val = tensor([0, 0, 0, 2048])]; + tensor var_729_end_0 = const()[name = tensor("op_729_end_0"), val = tensor([2, 64, 1, 2560])]; + tensor var_729_end_mask_0 = const()[name = tensor("op_729_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_729_cast = slice_by_index(begin = var_729_begin_0, end = var_729_end_0, end_mask = var_729_end_mask_0, x = var_694_cast)[name = tensor("op_729_cast")]; + tensor var_730_begin_0 = const()[name = tensor("op_730_begin_0"), val = tensor([0, 0, 0, 2560])]; + tensor var_730_end_0 = const()[name = tensor("op_730_end_0"), val = tensor([2, 64, 1, 3072])]; + tensor var_730_end_mask_0 = const()[name = tensor("op_730_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_730_cast = slice_by_index(begin = var_730_begin_0, end = var_730_end_0, end_mask = var_730_end_mask_0, x = var_694_cast)[name = tensor("op_730_cast")]; + tensor var_731_begin_0 = const()[name = tensor("op_731_begin_0"), val = tensor([0, 0, 0, 3072])]; + tensor var_731_end_0 = const()[name = tensor("op_731_end_0"), val = tensor([2, 64, 1, 3584])]; + tensor var_731_end_mask_0 = const()[name = tensor("op_731_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_731_cast = slice_by_index(begin = var_731_begin_0, end = var_731_end_0, end_mask = var_731_end_mask_0, x = var_694_cast)[name = tensor("op_731_cast")]; + tensor var_732_begin_0 = const()[name = tensor("op_732_begin_0"), val = tensor([0, 0, 0, 3584])]; + tensor var_732_end_0 = const()[name = tensor("op_732_end_0"), val = tensor([2, 64, 1, 4096])]; + tensor var_732_end_mask_0 = const()[name = tensor("op_732_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_732_cast = slice_by_index(begin = var_732_begin_0, end = var_732_end_0, end_mask = var_732_end_mask_0, x = var_694_cast)[name = tensor("op_732_cast")]; + tensor var_733_begin_0 = const()[name = tensor("op_733_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_733_end_0 = const()[name = tensor("op_733_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_733_end_mask_0 = const()[name = tensor("op_733_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_733_cast = slice_by_index(begin = var_733_begin_0, end = var_733_end_0, end_mask = var_733_end_mask_0, x = var_698_cast)[name = tensor("op_733_cast")]; + tensor var_734_begin_0 = const()[name = tensor("op_734_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_734_end_0 = const()[name = tensor("op_734_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_734_end_mask_0 = const()[name = tensor("op_734_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_734_cast = slice_by_index(begin = var_734_begin_0, end = var_734_end_0, end_mask = var_734_end_mask_0, x = var_698_cast)[name = tensor("op_734_cast")]; + tensor var_735_begin_0 = const()[name = tensor("op_735_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_735_end_0 = const()[name = tensor("op_735_end_0"), val = tensor([2, 64, 1, 1536])]; + tensor var_735_end_mask_0 = const()[name = tensor("op_735_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_735_cast = slice_by_index(begin = var_735_begin_0, end = var_735_end_0, end_mask = var_735_end_mask_0, x = var_698_cast)[name = tensor("op_735_cast")]; + tensor var_736_begin_0 = const()[name = tensor("op_736_begin_0"), val = tensor([0, 0, 0, 1536])]; + tensor var_736_end_0 = const()[name = tensor("op_736_end_0"), val = tensor([2, 64, 1, 2048])]; + tensor var_736_end_mask_0 = const()[name = tensor("op_736_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_736_cast = slice_by_index(begin = var_736_begin_0, end = var_736_end_0, end_mask = var_736_end_mask_0, x = var_698_cast)[name = tensor("op_736_cast")]; + tensor var_737_begin_0 = const()[name = tensor("op_737_begin_0"), val = tensor([0, 0, 0, 2048])]; + tensor var_737_end_0 = const()[name = tensor("op_737_end_0"), val = tensor([2, 64, 1, 2560])]; + tensor var_737_end_mask_0 = const()[name = tensor("op_737_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_737_cast = slice_by_index(begin = var_737_begin_0, end = var_737_end_0, end_mask = var_737_end_mask_0, x = var_698_cast)[name = tensor("op_737_cast")]; + tensor var_738_begin_0 = const()[name = tensor("op_738_begin_0"), val = tensor([0, 0, 0, 2560])]; + tensor var_738_end_0 = const()[name = tensor("op_738_end_0"), val = tensor([2, 64, 1, 3072])]; + tensor var_738_end_mask_0 = const()[name = tensor("op_738_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_738_cast = slice_by_index(begin = var_738_begin_0, end = var_738_end_0, end_mask = var_738_end_mask_0, x = var_698_cast)[name = tensor("op_738_cast")]; + tensor var_739_begin_0 = const()[name = tensor("op_739_begin_0"), val = tensor([0, 0, 0, 3072])]; + tensor var_739_end_0 = const()[name = tensor("op_739_end_0"), val = tensor([2, 64, 1, 3584])]; + tensor var_739_end_mask_0 = const()[name = tensor("op_739_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_739_cast = slice_by_index(begin = var_739_begin_0, end = var_739_end_0, end_mask = var_739_end_mask_0, x = var_698_cast)[name = tensor("op_739_cast")]; + tensor var_740_begin_0 = const()[name = tensor("op_740_begin_0"), val = tensor([0, 0, 0, 3584])]; + tensor var_740_end_0 = const()[name = tensor("op_740_end_0"), val = tensor([2, 64, 1, 4096])]; + tensor var_740_end_mask_0 = const()[name = tensor("op_740_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_740_cast = slice_by_index(begin = var_740_begin_0, end = var_740_end_0, end_mask = var_740_end_mask_0, x = var_698_cast)[name = tensor("op_740_cast")]; + tensor k_7_perm_0 = const()[name = tensor("k_7_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_745_begin_0 = const()[name = tensor("op_745_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_745_end_0 = const()[name = tensor("op_745_end_0"), val = tensor([2, 77, 1, 64])]; + tensor var_745_end_mask_0 = const()[name = tensor("op_745_end_mask_0"), val = tensor([true, true, true, false])]; + tensor transpose_30 = transpose(perm = k_7_perm_0, x = k_5_cast)[name = tensor("transpose_30")]; + tensor var_745_cast = slice_by_index(begin = var_745_begin_0, end = var_745_end_0, end_mask = var_745_end_mask_0, x = transpose_30)[name = tensor("op_745_cast")]; + tensor var_749_begin_0 = const()[name = tensor("op_749_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_749_end_0 = const()[name = tensor("op_749_end_0"), val = tensor([2, 77, 1, 128])]; + tensor var_749_end_mask_0 = const()[name = tensor("op_749_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_749_cast = slice_by_index(begin = var_749_begin_0, end = var_749_end_0, end_mask = var_749_end_mask_0, x = transpose_30)[name = tensor("op_749_cast")]; + tensor var_753_begin_0 = const()[name = tensor("op_753_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_753_end_0 = const()[name = tensor("op_753_end_0"), val = tensor([2, 77, 1, 192])]; + tensor var_753_end_mask_0 = const()[name = tensor("op_753_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_753_cast = slice_by_index(begin = var_753_begin_0, end = var_753_end_0, end_mask = var_753_end_mask_0, x = transpose_30)[name = tensor("op_753_cast")]; + tensor var_757_begin_0 = const()[name = tensor("op_757_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_757_end_0 = const()[name = tensor("op_757_end_0"), val = tensor([2, 77, 1, 256])]; + tensor var_757_end_mask_0 = const()[name = tensor("op_757_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_757_cast = slice_by_index(begin = var_757_begin_0, end = var_757_end_0, end_mask = var_757_end_mask_0, x = transpose_30)[name = tensor("op_757_cast")]; + tensor var_761_begin_0 = const()[name = tensor("op_761_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_761_end_0 = const()[name = tensor("op_761_end_0"), val = tensor([2, 77, 1, 320])]; + tensor var_761_end_mask_0 = const()[name = tensor("op_761_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_761_cast = slice_by_index(begin = var_761_begin_0, end = var_761_end_0, end_mask = var_761_end_mask_0, x = transpose_30)[name = tensor("op_761_cast")]; + tensor var_763_begin_0 = const()[name = tensor("op_763_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_763_end_0 = const()[name = tensor("op_763_end_0"), val = tensor([2, 64, 1, 77])]; + tensor var_763_end_mask_0 = const()[name = tensor("op_763_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_763_cast = slice_by_index(begin = var_763_begin_0, end = var_763_end_0, end_mask = var_763_end_mask_0, x = v_3_cast)[name = tensor("op_763_cast")]; + tensor var_767_begin_0 = const()[name = tensor("op_767_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_767_end_0 = const()[name = tensor("op_767_end_0"), val = tensor([2, 128, 1, 77])]; + tensor var_767_end_mask_0 = const()[name = tensor("op_767_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_767_cast = slice_by_index(begin = var_767_begin_0, end = var_767_end_0, end_mask = var_767_end_mask_0, x = v_3_cast)[name = tensor("op_767_cast")]; + tensor var_771_begin_0 = const()[name = tensor("op_771_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_771_end_0 = const()[name = tensor("op_771_end_0"), val = tensor([2, 192, 1, 77])]; + tensor var_771_end_mask_0 = const()[name = tensor("op_771_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_771_cast = slice_by_index(begin = var_771_begin_0, end = var_771_end_0, end_mask = var_771_end_mask_0, x = v_3_cast)[name = tensor("op_771_cast")]; + tensor var_775_begin_0 = const()[name = tensor("op_775_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_775_end_0 = const()[name = tensor("op_775_end_0"), val = tensor([2, 256, 1, 77])]; + tensor var_775_end_mask_0 = const()[name = tensor("op_775_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_775_cast = slice_by_index(begin = var_775_begin_0, end = var_775_end_0, end_mask = var_775_end_mask_0, x = v_3_cast)[name = tensor("op_775_cast")]; + tensor var_779_begin_0 = const()[name = tensor("op_779_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_779_end_0 = const()[name = tensor("op_779_end_0"), val = tensor([2, 320, 1, 77])]; + tensor var_779_end_mask_0 = const()[name = tensor("op_779_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_779_cast = slice_by_index(begin = var_779_begin_0, end = var_779_end_0, end_mask = var_779_end_mask_0, x = v_3_cast)[name = tensor("op_779_cast")]; + tensor var_783_equation_0 = const()[name = tensor("op_783_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_783_cast = einsum(equation = var_783_equation_0, values = (var_745_cast, var_701_cast))[name = tensor("op_783_cast")]; + tensor var_784_to_fp16 = const()[name = tensor("op_784_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_81_cast = mul(x = var_783_cast, y = var_784_to_fp16)[name = tensor("aw_chunk_81_cast")]; + tensor var_787_equation_0 = const()[name = tensor("op_787_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_787_cast = einsum(equation = var_787_equation_0, values = (var_745_cast, var_702_cast))[name = tensor("op_787_cast")]; + tensor var_788_to_fp16 = const()[name = tensor("op_788_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_83_cast = mul(x = var_787_cast, y = var_788_to_fp16)[name = tensor("aw_chunk_83_cast")]; + tensor var_791_equation_0 = const()[name = tensor("op_791_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_791_cast = einsum(equation = var_791_equation_0, values = (var_745_cast, var_703_cast))[name = tensor("op_791_cast")]; + tensor var_792_to_fp16 = const()[name = tensor("op_792_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_85_cast = mul(x = var_791_cast, y = var_792_to_fp16)[name = tensor("aw_chunk_85_cast")]; + tensor var_795_equation_0 = const()[name = tensor("op_795_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_795_cast = einsum(equation = var_795_equation_0, values = (var_745_cast, var_704_cast))[name = tensor("op_795_cast")]; + tensor var_796_to_fp16 = const()[name = tensor("op_796_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_87_cast = mul(x = var_795_cast, y = var_796_to_fp16)[name = tensor("aw_chunk_87_cast")]; + tensor var_799_equation_0 = const()[name = tensor("op_799_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_799_cast = einsum(equation = var_799_equation_0, values = (var_745_cast, var_705_cast))[name = tensor("op_799_cast")]; + tensor var_800_to_fp16 = const()[name = tensor("op_800_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_89_cast = mul(x = var_799_cast, y = var_800_to_fp16)[name = tensor("aw_chunk_89_cast")]; + tensor var_803_equation_0 = const()[name = tensor("op_803_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_803_cast = einsum(equation = var_803_equation_0, values = (var_745_cast, var_706_cast))[name = tensor("op_803_cast")]; + tensor var_804_to_fp16 = const()[name = tensor("op_804_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_91_cast = mul(x = var_803_cast, y = var_804_to_fp16)[name = tensor("aw_chunk_91_cast")]; + tensor var_807_equation_0 = const()[name = tensor("op_807_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_807_cast = einsum(equation = var_807_equation_0, values = (var_745_cast, var_707_cast))[name = tensor("op_807_cast")]; + tensor var_808_to_fp16 = const()[name = tensor("op_808_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_93_cast = mul(x = var_807_cast, y = var_808_to_fp16)[name = tensor("aw_chunk_93_cast")]; + tensor var_811_equation_0 = const()[name = tensor("op_811_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_811_cast = einsum(equation = var_811_equation_0, values = (var_745_cast, var_708_cast))[name = tensor("op_811_cast")]; + tensor var_812_to_fp16 = const()[name = tensor("op_812_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_95_cast = mul(x = var_811_cast, y = var_812_to_fp16)[name = tensor("aw_chunk_95_cast")]; + tensor var_815_equation_0 = const()[name = tensor("op_815_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_815_cast = einsum(equation = var_815_equation_0, values = (var_749_cast, var_709_cast))[name = tensor("op_815_cast")]; + tensor var_816_to_fp16 = const()[name = tensor("op_816_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_97_cast = mul(x = var_815_cast, y = var_816_to_fp16)[name = tensor("aw_chunk_97_cast")]; + tensor var_819_equation_0 = const()[name = tensor("op_819_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_819_cast = einsum(equation = var_819_equation_0, values = (var_749_cast, var_710_cast))[name = tensor("op_819_cast")]; + tensor var_820_to_fp16 = const()[name = tensor("op_820_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_99_cast = mul(x = var_819_cast, y = var_820_to_fp16)[name = tensor("aw_chunk_99_cast")]; + tensor var_823_equation_0 = const()[name = tensor("op_823_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_823_cast = einsum(equation = var_823_equation_0, values = (var_749_cast, var_711_cast))[name = tensor("op_823_cast")]; + tensor var_824_to_fp16 = const()[name = tensor("op_824_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_101_cast = mul(x = var_823_cast, y = var_824_to_fp16)[name = tensor("aw_chunk_101_cast")]; + tensor var_827_equation_0 = const()[name = tensor("op_827_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_827_cast = einsum(equation = var_827_equation_0, values = (var_749_cast, var_712_cast))[name = tensor("op_827_cast")]; + tensor var_828_to_fp16 = const()[name = tensor("op_828_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_103_cast = mul(x = var_827_cast, y = var_828_to_fp16)[name = tensor("aw_chunk_103_cast")]; + tensor var_831_equation_0 = const()[name = tensor("op_831_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_831_cast = einsum(equation = var_831_equation_0, values = (var_749_cast, var_713_cast))[name = tensor("op_831_cast")]; + tensor var_832_to_fp16 = const()[name = tensor("op_832_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_105_cast = mul(x = var_831_cast, y = var_832_to_fp16)[name = tensor("aw_chunk_105_cast")]; + tensor var_835_equation_0 = const()[name = tensor("op_835_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_835_cast = einsum(equation = var_835_equation_0, values = (var_749_cast, var_714_cast))[name = tensor("op_835_cast")]; + tensor var_836_to_fp16 = const()[name = tensor("op_836_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_107_cast = mul(x = var_835_cast, y = var_836_to_fp16)[name = tensor("aw_chunk_107_cast")]; + tensor var_839_equation_0 = const()[name = tensor("op_839_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_839_cast = einsum(equation = var_839_equation_0, values = (var_749_cast, var_715_cast))[name = tensor("op_839_cast")]; + tensor var_840_to_fp16 = const()[name = tensor("op_840_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_109_cast = mul(x = var_839_cast, y = var_840_to_fp16)[name = tensor("aw_chunk_109_cast")]; + tensor var_843_equation_0 = const()[name = tensor("op_843_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_843_cast = einsum(equation = var_843_equation_0, values = (var_749_cast, var_716_cast))[name = tensor("op_843_cast")]; + tensor var_844_to_fp16 = const()[name = tensor("op_844_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_111_cast = mul(x = var_843_cast, y = var_844_to_fp16)[name = tensor("aw_chunk_111_cast")]; + tensor var_847_equation_0 = const()[name = tensor("op_847_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_847_cast = einsum(equation = var_847_equation_0, values = (var_753_cast, var_717_cast))[name = tensor("op_847_cast")]; + tensor var_848_to_fp16 = const()[name = tensor("op_848_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_113_cast = mul(x = var_847_cast, y = var_848_to_fp16)[name = tensor("aw_chunk_113_cast")]; + tensor var_851_equation_0 = const()[name = tensor("op_851_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_851_cast = einsum(equation = var_851_equation_0, values = (var_753_cast, var_718_cast))[name = tensor("op_851_cast")]; + tensor var_852_to_fp16 = const()[name = tensor("op_852_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_115_cast = mul(x = var_851_cast, y = var_852_to_fp16)[name = tensor("aw_chunk_115_cast")]; + tensor var_855_equation_0 = const()[name = tensor("op_855_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_855_cast = einsum(equation = var_855_equation_0, values = (var_753_cast, var_719_cast))[name = tensor("op_855_cast")]; + tensor var_856_to_fp16 = const()[name = tensor("op_856_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_117_cast = mul(x = var_855_cast, y = var_856_to_fp16)[name = tensor("aw_chunk_117_cast")]; + tensor var_859_equation_0 = const()[name = tensor("op_859_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_859_cast = einsum(equation = var_859_equation_0, values = (var_753_cast, var_720_cast))[name = tensor("op_859_cast")]; + tensor var_860_to_fp16 = const()[name = tensor("op_860_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_119_cast = mul(x = var_859_cast, y = var_860_to_fp16)[name = tensor("aw_chunk_119_cast")]; + tensor var_863_equation_0 = const()[name = tensor("op_863_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_863_cast = einsum(equation = var_863_equation_0, values = (var_753_cast, var_721_cast))[name = tensor("op_863_cast")]; + tensor var_864_to_fp16 = const()[name = tensor("op_864_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_121_cast = mul(x = var_863_cast, y = var_864_to_fp16)[name = tensor("aw_chunk_121_cast")]; + tensor var_867_equation_0 = const()[name = tensor("op_867_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_867_cast = einsum(equation = var_867_equation_0, values = (var_753_cast, var_722_cast))[name = tensor("op_867_cast")]; + tensor var_868_to_fp16 = const()[name = tensor("op_868_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_123_cast = mul(x = var_867_cast, y = var_868_to_fp16)[name = tensor("aw_chunk_123_cast")]; + tensor var_871_equation_0 = const()[name = tensor("op_871_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_871_cast = einsum(equation = var_871_equation_0, values = (var_753_cast, var_723_cast))[name = tensor("op_871_cast")]; + tensor var_872_to_fp16 = const()[name = tensor("op_872_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_125_cast = mul(x = var_871_cast, y = var_872_to_fp16)[name = tensor("aw_chunk_125_cast")]; + tensor var_875_equation_0 = const()[name = tensor("op_875_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_875_cast = einsum(equation = var_875_equation_0, values = (var_753_cast, var_724_cast))[name = tensor("op_875_cast")]; + tensor var_876_to_fp16 = const()[name = tensor("op_876_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_127_cast = mul(x = var_875_cast, y = var_876_to_fp16)[name = tensor("aw_chunk_127_cast")]; + tensor var_879_equation_0 = const()[name = tensor("op_879_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_879_cast = einsum(equation = var_879_equation_0, values = (var_757_cast, var_725_cast))[name = tensor("op_879_cast")]; + tensor var_880_to_fp16 = const()[name = tensor("op_880_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_129_cast = mul(x = var_879_cast, y = var_880_to_fp16)[name = tensor("aw_chunk_129_cast")]; + tensor var_883_equation_0 = const()[name = tensor("op_883_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_883_cast = einsum(equation = var_883_equation_0, values = (var_757_cast, var_726_cast))[name = tensor("op_883_cast")]; + tensor var_884_to_fp16 = const()[name = tensor("op_884_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_131_cast = mul(x = var_883_cast, y = var_884_to_fp16)[name = tensor("aw_chunk_131_cast")]; + tensor var_887_equation_0 = const()[name = tensor("op_887_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_887_cast = einsum(equation = var_887_equation_0, values = (var_757_cast, var_727_cast))[name = tensor("op_887_cast")]; + tensor var_888_to_fp16 = const()[name = tensor("op_888_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_133_cast = mul(x = var_887_cast, y = var_888_to_fp16)[name = tensor("aw_chunk_133_cast")]; + tensor var_891_equation_0 = const()[name = tensor("op_891_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_891_cast = einsum(equation = var_891_equation_0, values = (var_757_cast, var_728_cast))[name = tensor("op_891_cast")]; + tensor var_892_to_fp16 = const()[name = tensor("op_892_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_135_cast = mul(x = var_891_cast, y = var_892_to_fp16)[name = tensor("aw_chunk_135_cast")]; + tensor var_895_equation_0 = const()[name = tensor("op_895_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_895_cast = einsum(equation = var_895_equation_0, values = (var_757_cast, var_729_cast))[name = tensor("op_895_cast")]; + tensor var_896_to_fp16 = const()[name = tensor("op_896_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_137_cast = mul(x = var_895_cast, y = var_896_to_fp16)[name = tensor("aw_chunk_137_cast")]; + tensor var_899_equation_0 = const()[name = tensor("op_899_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_899_cast = einsum(equation = var_899_equation_0, values = (var_757_cast, var_730_cast))[name = tensor("op_899_cast")]; + tensor var_900_to_fp16 = const()[name = tensor("op_900_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_139_cast = mul(x = var_899_cast, y = var_900_to_fp16)[name = tensor("aw_chunk_139_cast")]; + tensor var_903_equation_0 = const()[name = tensor("op_903_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_903_cast = einsum(equation = var_903_equation_0, values = (var_757_cast, var_731_cast))[name = tensor("op_903_cast")]; + tensor var_904_to_fp16 = const()[name = tensor("op_904_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_141_cast = mul(x = var_903_cast, y = var_904_to_fp16)[name = tensor("aw_chunk_141_cast")]; + tensor var_907_equation_0 = const()[name = tensor("op_907_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_907_cast = einsum(equation = var_907_equation_0, values = (var_757_cast, var_732_cast))[name = tensor("op_907_cast")]; + tensor var_908_to_fp16 = const()[name = tensor("op_908_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_143_cast = mul(x = var_907_cast, y = var_908_to_fp16)[name = tensor("aw_chunk_143_cast")]; + tensor var_911_equation_0 = const()[name = tensor("op_911_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_911_cast = einsum(equation = var_911_equation_0, values = (var_761_cast, var_733_cast))[name = tensor("op_911_cast")]; + tensor var_912_to_fp16 = const()[name = tensor("op_912_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_145_cast = mul(x = var_911_cast, y = var_912_to_fp16)[name = tensor("aw_chunk_145_cast")]; + tensor var_915_equation_0 = const()[name = tensor("op_915_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_915_cast = einsum(equation = var_915_equation_0, values = (var_761_cast, var_734_cast))[name = tensor("op_915_cast")]; + tensor var_916_to_fp16 = const()[name = tensor("op_916_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_147_cast = mul(x = var_915_cast, y = var_916_to_fp16)[name = tensor("aw_chunk_147_cast")]; + tensor var_919_equation_0 = const()[name = tensor("op_919_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_919_cast = einsum(equation = var_919_equation_0, values = (var_761_cast, var_735_cast))[name = tensor("op_919_cast")]; + tensor var_920_to_fp16 = const()[name = tensor("op_920_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_149_cast = mul(x = var_919_cast, y = var_920_to_fp16)[name = tensor("aw_chunk_149_cast")]; + tensor var_923_equation_0 = const()[name = tensor("op_923_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_923_cast = einsum(equation = var_923_equation_0, values = (var_761_cast, var_736_cast))[name = tensor("op_923_cast")]; + tensor var_924_to_fp16 = const()[name = tensor("op_924_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_151_cast = mul(x = var_923_cast, y = var_924_to_fp16)[name = tensor("aw_chunk_151_cast")]; + tensor var_927_equation_0 = const()[name = tensor("op_927_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_927_cast = einsum(equation = var_927_equation_0, values = (var_761_cast, var_737_cast))[name = tensor("op_927_cast")]; + tensor var_928_to_fp16 = const()[name = tensor("op_928_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_153_cast = mul(x = var_927_cast, y = var_928_to_fp16)[name = tensor("aw_chunk_153_cast")]; + tensor var_931_equation_0 = const()[name = tensor("op_931_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_931_cast = einsum(equation = var_931_equation_0, values = (var_761_cast, var_738_cast))[name = tensor("op_931_cast")]; + tensor var_932_to_fp16 = const()[name = tensor("op_932_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_155_cast = mul(x = var_931_cast, y = var_932_to_fp16)[name = tensor("aw_chunk_155_cast")]; + tensor var_935_equation_0 = const()[name = tensor("op_935_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_935_cast = einsum(equation = var_935_equation_0, values = (var_761_cast, var_739_cast))[name = tensor("op_935_cast")]; + tensor var_936_to_fp16 = const()[name = tensor("op_936_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_157_cast = mul(x = var_935_cast, y = var_936_to_fp16)[name = tensor("aw_chunk_157_cast")]; + tensor var_939_equation_0 = const()[name = tensor("op_939_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_939_cast = einsum(equation = var_939_equation_0, values = (var_761_cast, var_740_cast))[name = tensor("op_939_cast")]; + tensor var_940_to_fp16 = const()[name = tensor("op_940_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_159_cast = mul(x = var_939_cast, y = var_940_to_fp16)[name = tensor("aw_chunk_159_cast")]; + tensor var_942_cast = softmax(axis = var_120, x = aw_chunk_81_cast)[name = tensor("op_942_cast")]; + tensor var_943_cast = softmax(axis = var_120, x = aw_chunk_83_cast)[name = tensor("op_943_cast")]; + tensor var_944_cast = softmax(axis = var_120, x = aw_chunk_85_cast)[name = tensor("op_944_cast")]; + tensor var_945_cast = softmax(axis = var_120, x = aw_chunk_87_cast)[name = tensor("op_945_cast")]; + tensor var_946_cast = softmax(axis = var_120, x = aw_chunk_89_cast)[name = tensor("op_946_cast")]; + tensor var_947_cast = softmax(axis = var_120, x = aw_chunk_91_cast)[name = tensor("op_947_cast")]; + tensor var_948_cast = softmax(axis = var_120, x = aw_chunk_93_cast)[name = tensor("op_948_cast")]; + tensor var_949_cast = softmax(axis = var_120, x = aw_chunk_95_cast)[name = tensor("op_949_cast")]; + tensor var_950_cast = softmax(axis = var_120, x = aw_chunk_97_cast)[name = tensor("op_950_cast")]; + tensor var_951_cast = softmax(axis = var_120, x = aw_chunk_99_cast)[name = tensor("op_951_cast")]; + tensor var_952_cast = softmax(axis = var_120, x = aw_chunk_101_cast)[name = tensor("op_952_cast")]; + tensor var_953_cast = softmax(axis = var_120, x = aw_chunk_103_cast)[name = tensor("op_953_cast")]; + tensor var_954_cast = softmax(axis = var_120, x = aw_chunk_105_cast)[name = tensor("op_954_cast")]; + tensor var_955_cast = softmax(axis = var_120, x = aw_chunk_107_cast)[name = tensor("op_955_cast")]; + tensor var_956_cast = softmax(axis = var_120, x = aw_chunk_109_cast)[name = tensor("op_956_cast")]; + tensor var_957_cast = softmax(axis = var_120, x = aw_chunk_111_cast)[name = tensor("op_957_cast")]; + tensor var_958_cast = softmax(axis = var_120, x = aw_chunk_113_cast)[name = tensor("op_958_cast")]; + tensor var_959_cast = softmax(axis = var_120, x = aw_chunk_115_cast)[name = tensor("op_959_cast")]; + tensor var_960_cast = softmax(axis = var_120, x = aw_chunk_117_cast)[name = tensor("op_960_cast")]; + tensor var_961_cast = softmax(axis = var_120, x = aw_chunk_119_cast)[name = tensor("op_961_cast")]; + tensor var_962_cast = softmax(axis = var_120, x = aw_chunk_121_cast)[name = tensor("op_962_cast")]; + tensor var_963_cast = softmax(axis = var_120, x = aw_chunk_123_cast)[name = tensor("op_963_cast")]; + tensor var_964_cast = softmax(axis = var_120, x = aw_chunk_125_cast)[name = tensor("op_964_cast")]; + tensor var_965_cast = softmax(axis = var_120, x = aw_chunk_127_cast)[name = tensor("op_965_cast")]; + tensor var_966_cast = softmax(axis = var_120, x = aw_chunk_129_cast)[name = tensor("op_966_cast")]; + tensor var_967_cast = softmax(axis = var_120, x = aw_chunk_131_cast)[name = tensor("op_967_cast")]; + tensor var_968_cast = softmax(axis = var_120, x = aw_chunk_133_cast)[name = tensor("op_968_cast")]; + tensor var_969_cast = softmax(axis = var_120, x = aw_chunk_135_cast)[name = tensor("op_969_cast")]; + tensor var_970_cast = softmax(axis = var_120, x = aw_chunk_137_cast)[name = tensor("op_970_cast")]; + tensor var_971_cast = softmax(axis = var_120, x = aw_chunk_139_cast)[name = tensor("op_971_cast")]; + tensor var_972_cast = softmax(axis = var_120, x = aw_chunk_141_cast)[name = tensor("op_972_cast")]; + tensor var_973_cast = softmax(axis = var_120, x = aw_chunk_143_cast)[name = tensor("op_973_cast")]; + tensor var_974_cast = softmax(axis = var_120, x = aw_chunk_145_cast)[name = tensor("op_974_cast")]; + tensor var_975_cast = softmax(axis = var_120, x = aw_chunk_147_cast)[name = tensor("op_975_cast")]; + tensor var_976_cast = softmax(axis = var_120, x = aw_chunk_149_cast)[name = tensor("op_976_cast")]; + tensor var_977_cast = softmax(axis = var_120, x = aw_chunk_151_cast)[name = tensor("op_977_cast")]; + tensor var_978_cast = softmax(axis = var_120, x = aw_chunk_153_cast)[name = tensor("op_978_cast")]; + tensor var_979_cast = softmax(axis = var_120, x = aw_chunk_155_cast)[name = tensor("op_979_cast")]; + tensor var_980_cast = softmax(axis = var_120, x = aw_chunk_157_cast)[name = tensor("op_980_cast")]; + tensor var_981_cast = softmax(axis = var_120, x = aw_chunk_159_cast)[name = tensor("op_981_cast")]; + tensor var_983_equation_0 = const()[name = tensor("op_983_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_983_cast = einsum(equation = var_983_equation_0, values = (var_763_cast, var_942_cast))[name = tensor("op_983_cast")]; + tensor var_985_equation_0 = const()[name = tensor("op_985_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_985_cast = einsum(equation = var_985_equation_0, values = (var_763_cast, var_943_cast))[name = tensor("op_985_cast")]; + tensor var_987_equation_0 = const()[name = tensor("op_987_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_987_cast = einsum(equation = var_987_equation_0, values = (var_763_cast, var_944_cast))[name = tensor("op_987_cast")]; + tensor var_989_equation_0 = const()[name = tensor("op_989_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_989_cast = einsum(equation = var_989_equation_0, values = (var_763_cast, var_945_cast))[name = tensor("op_989_cast")]; + tensor var_991_equation_0 = const()[name = tensor("op_991_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_991_cast = einsum(equation = var_991_equation_0, values = (var_763_cast, var_946_cast))[name = tensor("op_991_cast")]; + tensor var_993_equation_0 = const()[name = tensor("op_993_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_993_cast = einsum(equation = var_993_equation_0, values = (var_763_cast, var_947_cast))[name = tensor("op_993_cast")]; + tensor var_995_equation_0 = const()[name = tensor("op_995_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_995_cast = einsum(equation = var_995_equation_0, values = (var_763_cast, var_948_cast))[name = tensor("op_995_cast")]; + tensor var_997_equation_0 = const()[name = tensor("op_997_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_997_cast = einsum(equation = var_997_equation_0, values = (var_763_cast, var_949_cast))[name = tensor("op_997_cast")]; + tensor var_999_equation_0 = const()[name = tensor("op_999_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_999_cast = einsum(equation = var_999_equation_0, values = (var_767_cast, var_950_cast))[name = tensor("op_999_cast")]; + tensor var_1001_equation_0 = const()[name = tensor("op_1001_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1001_cast = einsum(equation = var_1001_equation_0, values = (var_767_cast, var_951_cast))[name = tensor("op_1001_cast")]; + tensor var_1003_equation_0 = const()[name = tensor("op_1003_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1003_cast = einsum(equation = var_1003_equation_0, values = (var_767_cast, var_952_cast))[name = tensor("op_1003_cast")]; + tensor var_1005_equation_0 = const()[name = tensor("op_1005_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1005_cast = einsum(equation = var_1005_equation_0, values = (var_767_cast, var_953_cast))[name = tensor("op_1005_cast")]; + tensor var_1007_equation_0 = const()[name = tensor("op_1007_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1007_cast = einsum(equation = var_1007_equation_0, values = (var_767_cast, var_954_cast))[name = tensor("op_1007_cast")]; + tensor var_1009_equation_0 = const()[name = tensor("op_1009_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1009_cast = einsum(equation = var_1009_equation_0, values = (var_767_cast, var_955_cast))[name = tensor("op_1009_cast")]; + tensor var_1011_equation_0 = const()[name = tensor("op_1011_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1011_cast = einsum(equation = var_1011_equation_0, values = (var_767_cast, var_956_cast))[name = tensor("op_1011_cast")]; + tensor var_1013_equation_0 = const()[name = tensor("op_1013_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1013_cast = einsum(equation = var_1013_equation_0, values = (var_767_cast, var_957_cast))[name = tensor("op_1013_cast")]; + tensor var_1015_equation_0 = const()[name = tensor("op_1015_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1015_cast = einsum(equation = var_1015_equation_0, values = (var_771_cast, var_958_cast))[name = tensor("op_1015_cast")]; + tensor var_1017_equation_0 = const()[name = tensor("op_1017_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1017_cast = einsum(equation = var_1017_equation_0, values = (var_771_cast, var_959_cast))[name = tensor("op_1017_cast")]; + tensor var_1019_equation_0 = const()[name = tensor("op_1019_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1019_cast = einsum(equation = var_1019_equation_0, values = (var_771_cast, var_960_cast))[name = tensor("op_1019_cast")]; + tensor var_1021_equation_0 = const()[name = tensor("op_1021_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1021_cast = einsum(equation = var_1021_equation_0, values = (var_771_cast, var_961_cast))[name = tensor("op_1021_cast")]; + tensor var_1023_equation_0 = const()[name = tensor("op_1023_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1023_cast = einsum(equation = var_1023_equation_0, values = (var_771_cast, var_962_cast))[name = tensor("op_1023_cast")]; + tensor var_1025_equation_0 = const()[name = tensor("op_1025_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1025_cast = einsum(equation = var_1025_equation_0, values = (var_771_cast, var_963_cast))[name = tensor("op_1025_cast")]; + tensor var_1027_equation_0 = const()[name = tensor("op_1027_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1027_cast = einsum(equation = var_1027_equation_0, values = (var_771_cast, var_964_cast))[name = tensor("op_1027_cast")]; + tensor var_1029_equation_0 = const()[name = tensor("op_1029_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1029_cast = einsum(equation = var_1029_equation_0, values = (var_771_cast, var_965_cast))[name = tensor("op_1029_cast")]; + tensor var_1031_equation_0 = const()[name = tensor("op_1031_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1031_cast = einsum(equation = var_1031_equation_0, values = (var_775_cast, var_966_cast))[name = tensor("op_1031_cast")]; + tensor var_1033_equation_0 = const()[name = tensor("op_1033_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1033_cast = einsum(equation = var_1033_equation_0, values = (var_775_cast, var_967_cast))[name = tensor("op_1033_cast")]; + tensor var_1035_equation_0 = const()[name = tensor("op_1035_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1035_cast = einsum(equation = var_1035_equation_0, values = (var_775_cast, var_968_cast))[name = tensor("op_1035_cast")]; + tensor var_1037_equation_0 = const()[name = tensor("op_1037_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1037_cast = einsum(equation = var_1037_equation_0, values = (var_775_cast, var_969_cast))[name = tensor("op_1037_cast")]; + tensor var_1039_equation_0 = const()[name = tensor("op_1039_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1039_cast = einsum(equation = var_1039_equation_0, values = (var_775_cast, var_970_cast))[name = tensor("op_1039_cast")]; + tensor var_1041_equation_0 = const()[name = tensor("op_1041_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1041_cast = einsum(equation = var_1041_equation_0, values = (var_775_cast, var_971_cast))[name = tensor("op_1041_cast")]; + tensor var_1043_equation_0 = const()[name = tensor("op_1043_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1043_cast = einsum(equation = var_1043_equation_0, values = (var_775_cast, var_972_cast))[name = tensor("op_1043_cast")]; + tensor var_1045_equation_0 = const()[name = tensor("op_1045_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1045_cast = einsum(equation = var_1045_equation_0, values = (var_775_cast, var_973_cast))[name = tensor("op_1045_cast")]; + tensor var_1047_equation_0 = const()[name = tensor("op_1047_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1047_cast = einsum(equation = var_1047_equation_0, values = (var_779_cast, var_974_cast))[name = tensor("op_1047_cast")]; + tensor var_1049_equation_0 = const()[name = tensor("op_1049_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1049_cast = einsum(equation = var_1049_equation_0, values = (var_779_cast, var_975_cast))[name = tensor("op_1049_cast")]; + tensor var_1051_equation_0 = const()[name = tensor("op_1051_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1051_cast = einsum(equation = var_1051_equation_0, values = (var_779_cast, var_976_cast))[name = tensor("op_1051_cast")]; + tensor var_1053_equation_0 = const()[name = tensor("op_1053_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1053_cast = einsum(equation = var_1053_equation_0, values = (var_779_cast, var_977_cast))[name = tensor("op_1053_cast")]; + tensor var_1055_equation_0 = const()[name = tensor("op_1055_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1055_cast = einsum(equation = var_1055_equation_0, values = (var_779_cast, var_978_cast))[name = tensor("op_1055_cast")]; + tensor var_1057_equation_0 = const()[name = tensor("op_1057_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1057_cast = einsum(equation = var_1057_equation_0, values = (var_779_cast, var_979_cast))[name = tensor("op_1057_cast")]; + tensor var_1059_equation_0 = const()[name = tensor("op_1059_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1059_cast = einsum(equation = var_1059_equation_0, values = (var_779_cast, var_980_cast))[name = tensor("op_1059_cast")]; + tensor var_1061_equation_0 = const()[name = tensor("op_1061_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1061_cast = einsum(equation = var_1061_equation_0, values = (var_779_cast, var_981_cast))[name = tensor("op_1061_cast")]; + tensor var_1063_interleave_0 = const()[name = tensor("op_1063_interleave_0"), val = tensor(false)]; + tensor var_1063_cast = concat(axis = var_95, interleave = var_1063_interleave_0, values = (var_983_cast, var_985_cast, var_987_cast, var_989_cast, var_991_cast, var_993_cast, var_995_cast, var_997_cast))[name = tensor("op_1063_cast")]; + tensor var_1065_interleave_0 = const()[name = tensor("op_1065_interleave_0"), val = tensor(false)]; + tensor var_1065_cast = concat(axis = var_95, interleave = var_1065_interleave_0, values = (var_999_cast, var_1001_cast, var_1003_cast, var_1005_cast, var_1007_cast, var_1009_cast, var_1011_cast, var_1013_cast))[name = tensor("op_1065_cast")]; + tensor var_1067_interleave_0 = const()[name = tensor("op_1067_interleave_0"), val = tensor(false)]; + tensor var_1067_cast = concat(axis = var_95, interleave = var_1067_interleave_0, values = (var_1015_cast, var_1017_cast, var_1019_cast, var_1021_cast, var_1023_cast, var_1025_cast, var_1027_cast, var_1029_cast))[name = tensor("op_1067_cast")]; + tensor var_1069_interleave_0 = const()[name = tensor("op_1069_interleave_0"), val = tensor(false)]; + tensor var_1069_cast = concat(axis = var_95, interleave = var_1069_interleave_0, values = (var_1031_cast, var_1033_cast, var_1035_cast, var_1037_cast, var_1039_cast, var_1041_cast, var_1043_cast, var_1045_cast))[name = tensor("op_1069_cast")]; + tensor var_1071_interleave_0 = const()[name = tensor("op_1071_interleave_0"), val = tensor(false)]; + tensor var_1071_cast = concat(axis = var_95, interleave = var_1071_interleave_0, values = (var_1047_cast, var_1049_cast, var_1051_cast, var_1053_cast, var_1055_cast, var_1057_cast, var_1059_cast, var_1061_cast))[name = tensor("op_1071_cast")]; + tensor input_27_interleave_0 = const()[name = tensor("input_27_interleave_0"), val = tensor(false)]; + tensor input_27_cast = concat(axis = var_120, interleave = input_27_interleave_0, values = (var_1063_cast, var_1065_cast, var_1067_cast, var_1069_cast, var_1071_cast))[name = tensor("input_27_cast")]; + tensor var_1077 = const()[name = tensor("op_1077"), val = tensor([1, 1])]; + tensor var_1079 = const()[name = tensor("op_1079"), val = tensor([1, 1])]; + tensor var_1081_pad_type_0 = const()[name = tensor("op_1081_pad_type_0"), val = tensor("custom")]; + tensor var_1081_pad_0 = const()[name = tensor("op_1081_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_0_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4208512))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4285376))), name = tensor("down_blocks_0_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16_palettized"), shape = tensor([320, 320, 1, 1])]; + tensor down_blocks_0_attentions_0_transformer_blocks_0_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("down_blocks_0_attentions_0_transformer_blocks_0_attn2_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4285568)))]; + tensor var_1081_cast = conv(bias = down_blocks_0_attentions_0_transformer_blocks_0_attn2_to_out_0_bias_to_fp16, dilations = var_1079, groups = var_120, pad = var_1081_pad_0, pad_type = var_1081_pad_type_0, strides = var_1077, weight = down_blocks_0_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16_palettized, x = input_27_cast)[name = tensor("op_1081_cast")]; + tensor inputs_5_cast = add(x = var_1081_cast, y = inputs_3_cast)[name = tensor("inputs_5_cast")]; + tensor var_1085 = const()[name = tensor("op_1085"), val = tensor([1])]; + tensor channels_mean_5_cast = reduce_mean(axes = var_1085, keep_dims = var_115, x = inputs_5_cast)[name = tensor("channels_mean_5_cast")]; + tensor zero_mean_5_cast = sub(x = inputs_5_cast, y = channels_mean_5_cast)[name = tensor("zero_mean_5_cast")]; + tensor zero_mean_sq_5_cast = mul(x = zero_mean_5_cast, y = zero_mean_5_cast)[name = tensor("zero_mean_sq_5_cast")]; + tensor var_1089 = const()[name = tensor("op_1089"), val = tensor([1])]; + tensor var_1090_cast = reduce_mean(axes = var_1089, keep_dims = var_115, x = zero_mean_sq_5_cast)[name = tensor("op_1090_cast")]; + tensor var_1091_to_fp16 = const()[name = tensor("op_1091_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_1092_cast = add(x = var_1090_cast, y = var_1091_to_fp16)[name = tensor("op_1092_cast")]; + tensor denom_5_epsilon_0_to_fp16 = const()[name = tensor("denom_5_epsilon_0_to_fp16"), val = tensor(0x1p-24)]; + tensor denom_5_cast = rsqrt(epsilon = denom_5_epsilon_0_to_fp16, x = var_1092_cast)[name = tensor("denom_5_cast")]; + tensor out_5_cast = mul(x = zero_mean_5_cast, y = denom_5_cast)[name = tensor("out_5_cast")]; + tensor var_1096_to_fp16 = const()[name = tensor("op_1096_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4286272)))]; + tensor var_1097_cast = add(x = out_5_cast, y = var_1096_to_fp16)[name = tensor("op_1097_cast")]; + tensor var_1099_to_fp16 = const()[name = tensor("op_1099_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4286976)))]; + tensor input_29_cast = mul(x = var_1097_cast, y = var_1099_to_fp16)[name = tensor("input_29_cast")]; + tensor var_1107 = const()[name = tensor("op_1107"), val = tensor([1, 1])]; + tensor var_1109 = const()[name = tensor("op_1109"), val = tensor([1, 1])]; + tensor var_1111_pad_type_0 = const()[name = tensor("op_1111_pad_type_0"), val = tensor("custom")]; + tensor var_1111_pad_0 = const()[name = tensor("op_1111_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_0_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4287680))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4902144))), name = tensor("down_blocks_0_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16_palettized"), shape = tensor([2560, 320, 1, 1])]; + tensor down_blocks_0_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4902336))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4904320))), name = tensor("down_blocks_0_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16_palettized"), shape = tensor([2560])]; + tensor var_1111_cast = conv(bias = down_blocks_0_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16_palettized, dilations = var_1109, groups = var_120, pad = var_1111_pad_0, pad_type = var_1111_pad_type_0, strides = var_1107, weight = down_blocks_0_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16_palettized, x = input_29_cast)[name = tensor("op_1111_cast")]; + tensor var_1112_split_sizes_0 = const()[name = tensor("op_1112_split_sizes_0"), val = tensor([1280, 1280])]; + tensor var_1112_axis_0 = const()[name = tensor("op_1112_axis_0"), val = tensor(1)]; + tensor var_1112_cast_0, tensor var_1112_cast_1 = split(axis = var_1112_axis_0, split_sizes = var_1112_split_sizes_0, x = var_1111_cast)[name = tensor("op_1112_cast")]; + tensor var_1114_mode_0 = const()[name = tensor("op_1114_mode_0"), val = tensor("EXACT")]; + tensor var_1114_cast = gelu(mode = var_1114_mode_0, x = var_1112_cast_1)[name = tensor("op_1114_cast")]; + tensor input_31_cast = mul(x = var_1112_cast_0, y = var_1114_cast)[name = tensor("input_31_cast")]; + tensor var_1118 = const()[name = tensor("op_1118"), val = tensor([1, 1])]; + tensor var_1120 = const()[name = tensor("op_1120"), val = tensor([1, 1])]; + tensor var_1122_pad_type_0 = const()[name = tensor("op_1122_pad_type_0"), val = tensor("custom")]; + tensor var_1122_pad_0 = const()[name = tensor("op_1122_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_0_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4904512))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5211776))), name = tensor("down_blocks_0_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16_palettized"), shape = tensor([320, 1280, 1, 1])]; + tensor down_blocks_0_attentions_0_transformer_blocks_0_ff_net_2_bias_to_fp16 = const()[name = tensor("down_blocks_0_attentions_0_transformer_blocks_0_ff_net_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5211968)))]; + tensor var_1122_cast = conv(bias = down_blocks_0_attentions_0_transformer_blocks_0_ff_net_2_bias_to_fp16, dilations = var_1120, groups = var_120, pad = var_1122_pad_0, pad_type = var_1122_pad_type_0, strides = var_1118, weight = down_blocks_0_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16_palettized, x = input_31_cast)[name = tensor("op_1122_cast")]; + tensor hidden_states_15_cast = add(x = var_1122_cast, y = inputs_5_cast)[name = tensor("hidden_states_15_cast")]; + tensor var_1124 = const()[name = tensor("op_1124"), val = tensor([2, 320, 64, 64])]; + tensor input_33_cast = reshape(shape = var_1124, x = hidden_states_15_cast)[name = tensor("input_33_cast")]; + tensor var_1128 = const()[name = tensor("op_1128"), val = tensor([1, 1])]; + tensor var_1130 = const()[name = tensor("op_1130"), val = tensor([1, 1])]; + tensor hidden_states_17_pad_type_0 = const()[name = tensor("hidden_states_17_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_17_pad_0 = const()[name = tensor("hidden_states_17_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_0_attentions_0_proj_out_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5212672))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5289536))), name = tensor("down_blocks_0_attentions_0_proj_out_weight_to_fp16_palettized"), shape = tensor([320, 320, 1, 1])]; + tensor down_blocks_0_attentions_0_proj_out_bias_to_fp16 = const()[name = tensor("down_blocks_0_attentions_0_proj_out_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5289728)))]; + tensor hidden_states_17_cast = conv(bias = down_blocks_0_attentions_0_proj_out_bias_to_fp16, dilations = var_1130, groups = var_120, pad = hidden_states_17_pad_0, pad_type = hidden_states_17_pad_type_0, strides = var_1128, weight = down_blocks_0_attentions_0_proj_out_weight_to_fp16_palettized, x = input_33_cast)[name = tensor("hidden_states_17_cast")]; + tensor input_35_cast = add(x = hidden_states_17_cast, y = hidden_states_5_cast)[name = tensor("input_35_cast")]; + tensor reshape_12_shape_0 = const()[name = tensor("reshape_12_shape_0"), val = tensor([2, 32, 10, 64, 64])]; + tensor reshape_12_cast = reshape(shape = reshape_12_shape_0, x = input_35_cast)[name = tensor("reshape_12_cast")]; + tensor reduce_mean_9_axes_0 = const()[name = tensor("reduce_mean_9_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_9_keep_dims_0 = const()[name = tensor("reduce_mean_9_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_9_cast = reduce_mean(axes = reduce_mean_9_axes_0, keep_dims = reduce_mean_9_keep_dims_0, x = reshape_12_cast)[name = tensor("reduce_mean_9_cast")]; + tensor sub_6_cast = sub(x = reshape_12_cast, y = reduce_mean_9_cast)[name = tensor("sub_6_cast")]; + tensor square_3_cast = square(x = sub_6_cast)[name = tensor("square_3_cast")]; + tensor reduce_mean_11_axes_0 = const()[name = tensor("reduce_mean_11_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_11_keep_dims_0 = const()[name = tensor("reduce_mean_11_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_11_cast = reduce_mean(axes = reduce_mean_11_axes_0, keep_dims = reduce_mean_11_keep_dims_0, x = square_3_cast)[name = tensor("reduce_mean_11_cast")]; + tensor add_6_y_0_to_fp16 = const()[name = tensor("add_6_y_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_6_cast = add(x = reduce_mean_11_cast, y = add_6_y_0_to_fp16)[name = tensor("add_6_cast")]; + tensor sqrt_3_cast = sqrt(x = add_6_cast)[name = tensor("sqrt_3_cast")]; + tensor real_div_3_cast = real_div(x = sub_6_cast, y = sqrt_3_cast)[name = tensor("real_div_3_cast")]; + tensor reshape_13_shape_0 = const()[name = tensor("reshape_13_shape_0"), val = tensor([2, 320, 64, 64])]; + tensor reshape_13_cast = reshape(shape = reshape_13_shape_0, x = real_div_3_cast)[name = tensor("reshape_13_cast")]; + tensor add_7_gamma_0_to_fp16 = const()[name = tensor("add_7_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5290432)))]; + tensor add_7_beta_0_to_fp16 = const()[name = tensor("add_7_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5291136)))]; + tensor add_7_epsilon_0_to_fp16 = const()[name = tensor("add_7_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_7_cast = batch_norm(beta = add_7_beta_0_to_fp16, epsilon = add_7_epsilon_0_to_fp16, gamma = add_7_gamma_0_to_fp16, mean = add_1_mean_0_to_fp16, variance = add_1_variance_0_to_fp16, x = reshape_13_cast)[name = tensor("add_7_cast")]; + tensor input_39_cast = silu(x = add_7_cast)[name = tensor("input_39_cast")]; + tensor var_1145 = const()[name = tensor("op_1145"), val = tensor([1, 1])]; + tensor var_1147 = const()[name = tensor("op_1147"), val = tensor([1, 1])]; + tensor hidden_states_19_pad_type_0 = const()[name = tensor("hidden_states_19_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_19_pad_0 = const()[name = tensor("hidden_states_19_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor down_blocks_0_resnets_1_conv1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5291840))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5983104))), name = tensor("down_blocks_0_resnets_1_conv1_weight_to_fp16_palettized"), shape = tensor([320, 320, 3, 3])]; + tensor down_blocks_0_resnets_1_conv1_bias_to_fp16 = const()[name = tensor("down_blocks_0_resnets_1_conv1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5983296)))]; + tensor hidden_states_19_cast = conv(bias = down_blocks_0_resnets_1_conv1_bias_to_fp16, dilations = var_1147, groups = var_120, pad = hidden_states_19_pad_0, pad_type = hidden_states_19_pad_type_0, strides = var_1145, weight = down_blocks_0_resnets_1_conv1_weight_to_fp16_palettized, x = input_39_cast)[name = tensor("hidden_states_19_cast")]; + tensor var_1153 = const()[name = tensor("op_1153"), val = tensor([1, 1])]; + tensor var_1155 = const()[name = tensor("op_1155"), val = tensor([1, 1])]; + tensor temb_3_pad_type_0 = const()[name = tensor("temb_3_pad_type_0"), val = tensor("custom")]; + tensor temb_3_pad_0 = const()[name = tensor("temb_3_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_0_resnets_1_time_emb_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5984000))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(6291264))), name = tensor("down_blocks_0_resnets_1_time_emb_proj_weight_to_fp16_palettized"), shape = tensor([320, 1280, 1, 1])]; + tensor down_blocks_0_resnets_1_time_emb_proj_bias_to_fp16 = const()[name = tensor("down_blocks_0_resnets_1_time_emb_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(6291456)))]; + tensor temb_3_cast = conv(bias = down_blocks_0_resnets_1_time_emb_proj_bias_to_fp16, dilations = var_1155, groups = var_120, pad = temb_3_pad_0, pad_type = temb_3_pad_type_0, strides = var_1153, weight = down_blocks_0_resnets_1_time_emb_proj_weight_to_fp16_palettized, x = input_15_cast)[name = tensor("temb_3_cast")]; + tensor input_43_cast = add(x = hidden_states_19_cast, y = temb_3_cast)[name = tensor("input_43_cast")]; + tensor reshape_16_shape_0 = const()[name = tensor("reshape_16_shape_0"), val = tensor([2, 32, 10, 64, 64])]; + tensor reshape_16_cast = reshape(shape = reshape_16_shape_0, x = input_43_cast)[name = tensor("reshape_16_cast")]; + tensor reduce_mean_12_axes_0 = const()[name = tensor("reduce_mean_12_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_12_keep_dims_0 = const()[name = tensor("reduce_mean_12_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_12_cast = reduce_mean(axes = reduce_mean_12_axes_0, keep_dims = reduce_mean_12_keep_dims_0, x = reshape_16_cast)[name = tensor("reduce_mean_12_cast")]; + tensor sub_8_cast = sub(x = reshape_16_cast, y = reduce_mean_12_cast)[name = tensor("sub_8_cast")]; + tensor square_4_cast = square(x = sub_8_cast)[name = tensor("square_4_cast")]; + tensor reduce_mean_14_axes_0 = const()[name = tensor("reduce_mean_14_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_14_keep_dims_0 = const()[name = tensor("reduce_mean_14_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_14_cast = reduce_mean(axes = reduce_mean_14_axes_0, keep_dims = reduce_mean_14_keep_dims_0, x = square_4_cast)[name = tensor("reduce_mean_14_cast")]; + tensor add_8_y_0_to_fp16 = const()[name = tensor("add_8_y_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_8_cast = add(x = reduce_mean_14_cast, y = add_8_y_0_to_fp16)[name = tensor("add_8_cast")]; + tensor sqrt_4_cast = sqrt(x = add_8_cast)[name = tensor("sqrt_4_cast")]; + tensor real_div_4_cast = real_div(x = sub_8_cast, y = sqrt_4_cast)[name = tensor("real_div_4_cast")]; + tensor reshape_17_shape_0 = const()[name = tensor("reshape_17_shape_0"), val = tensor([2, 320, 64, 64])]; + tensor reshape_17_cast = reshape(shape = reshape_17_shape_0, x = real_div_4_cast)[name = tensor("reshape_17_cast")]; + tensor add_9_gamma_0_to_fp16 = const()[name = tensor("add_9_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(6292160)))]; + tensor add_9_beta_0_to_fp16 = const()[name = tensor("add_9_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(6292864)))]; + tensor add_9_epsilon_0_to_fp16 = const()[name = tensor("add_9_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_9_cast = batch_norm(beta = add_9_beta_0_to_fp16, epsilon = add_9_epsilon_0_to_fp16, gamma = add_9_gamma_0_to_fp16, mean = add_1_mean_0_to_fp16, variance = add_1_variance_0_to_fp16, x = reshape_17_cast)[name = tensor("add_9_cast")]; + tensor input_47_cast = silu(x = add_9_cast)[name = tensor("input_47_cast")]; + tensor var_1165 = const()[name = tensor("op_1165"), val = tensor([1, 1])]; + tensor var_1167 = const()[name = tensor("op_1167"), val = tensor([1, 1])]; + tensor hidden_states_21_pad_type_0 = const()[name = tensor("hidden_states_21_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_21_pad_0 = const()[name = tensor("hidden_states_21_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor down_blocks_0_resnets_1_conv2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(6293568))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(6984832))), name = tensor("down_blocks_0_resnets_1_conv2_weight_to_fp16_palettized"), shape = tensor([320, 320, 3, 3])]; + tensor down_blocks_0_resnets_1_conv2_bias_to_fp16 = const()[name = tensor("down_blocks_0_resnets_1_conv2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(6985024)))]; + tensor hidden_states_21_cast = conv(bias = down_blocks_0_resnets_1_conv2_bias_to_fp16, dilations = var_1167, groups = var_120, pad = hidden_states_21_pad_0, pad_type = hidden_states_21_pad_type_0, strides = var_1165, weight = down_blocks_0_resnets_1_conv2_weight_to_fp16_palettized, x = input_47_cast)[name = tensor("hidden_states_21_cast")]; + tensor hidden_states_23_cast = add(x = input_35_cast, y = hidden_states_21_cast)[name = tensor("hidden_states_23_cast")]; + tensor reshape_20_shape_0 = const()[name = tensor("reshape_20_shape_0"), val = tensor([2, 32, 10, 64, 64])]; + tensor reshape_20_cast = reshape(shape = reshape_20_shape_0, x = hidden_states_23_cast)[name = tensor("reshape_20_cast")]; + tensor reduce_mean_15_axes_0 = const()[name = tensor("reduce_mean_15_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_15_keep_dims_0 = const()[name = tensor("reduce_mean_15_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_15_cast = reduce_mean(axes = reduce_mean_15_axes_0, keep_dims = reduce_mean_15_keep_dims_0, x = reshape_20_cast)[name = tensor("reduce_mean_15_cast")]; + tensor sub_10_cast = sub(x = reshape_20_cast, y = reduce_mean_15_cast)[name = tensor("sub_10_cast")]; + tensor square_5_cast = square(x = sub_10_cast)[name = tensor("square_5_cast")]; + tensor reduce_mean_17_axes_0 = const()[name = tensor("reduce_mean_17_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_17_keep_dims_0 = const()[name = tensor("reduce_mean_17_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_17_cast = reduce_mean(axes = reduce_mean_17_axes_0, keep_dims = reduce_mean_17_keep_dims_0, x = square_5_cast)[name = tensor("reduce_mean_17_cast")]; + tensor add_10_y_0_to_fp16 = const()[name = tensor("add_10_y_0_to_fp16"), val = tensor(0x1.1p-20)]; + tensor add_10_cast = add(x = reduce_mean_17_cast, y = add_10_y_0_to_fp16)[name = tensor("add_10_cast")]; + tensor sqrt_5_cast = sqrt(x = add_10_cast)[name = tensor("sqrt_5_cast")]; + tensor real_div_5_cast = real_div(x = sub_10_cast, y = sqrt_5_cast)[name = tensor("real_div_5_cast")]; + tensor reshape_21_shape_0 = const()[name = tensor("reshape_21_shape_0"), val = tensor([2, 320, 64, 64])]; + tensor reshape_21_cast = reshape(shape = reshape_21_shape_0, x = real_div_5_cast)[name = tensor("reshape_21_cast")]; + tensor add_11_gamma_0_to_fp16 = const()[name = tensor("add_11_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(6985728)))]; + tensor add_11_beta_0_to_fp16 = const()[name = tensor("add_11_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(6986432)))]; + tensor add_11_epsilon_0_to_fp16 = const()[name = tensor("add_11_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_11_cast = batch_norm(beta = add_11_beta_0_to_fp16, epsilon = add_11_epsilon_0_to_fp16, gamma = add_11_gamma_0_to_fp16, mean = add_1_mean_0_to_fp16, variance = add_1_variance_0_to_fp16, x = reshape_21_cast)[name = tensor("add_11_cast")]; + tensor var_1187 = const()[name = tensor("op_1187"), val = tensor([1, 1])]; + tensor var_1189 = const()[name = tensor("op_1189"), val = tensor([1, 1])]; + tensor hidden_states_25_pad_type_0 = const()[name = tensor("hidden_states_25_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_25_pad_0 = const()[name = tensor("hidden_states_25_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_0_attentions_1_proj_in_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(6987136))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(7064000))), name = tensor("down_blocks_0_attentions_1_proj_in_weight_to_fp16_palettized"), shape = tensor([320, 320, 1, 1])]; + tensor down_blocks_0_attentions_1_proj_in_bias_to_fp16 = const()[name = tensor("down_blocks_0_attentions_1_proj_in_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(7064192)))]; + tensor hidden_states_25_cast = conv(bias = down_blocks_0_attentions_1_proj_in_bias_to_fp16, dilations = var_1189, groups = var_120, pad = hidden_states_25_pad_0, pad_type = hidden_states_25_pad_type_0, strides = var_1187, weight = down_blocks_0_attentions_1_proj_in_weight_to_fp16_palettized, x = add_11_cast)[name = tensor("hidden_states_25_cast")]; + tensor var_1194 = const()[name = tensor("op_1194"), val = tensor([2, 320, 1, 4096])]; + tensor inputs_7_cast = reshape(shape = var_1194, x = hidden_states_25_cast)[name = tensor("inputs_7_cast")]; + tensor var_1204 = const()[name = tensor("op_1204"), val = tensor([1])]; + tensor channels_mean_7_cast = reduce_mean(axes = var_1204, keep_dims = var_115, x = inputs_7_cast)[name = tensor("channels_mean_7_cast")]; + tensor zero_mean_7_cast = sub(x = inputs_7_cast, y = channels_mean_7_cast)[name = tensor("zero_mean_7_cast")]; + tensor zero_mean_sq_7_cast = mul(x = zero_mean_7_cast, y = zero_mean_7_cast)[name = tensor("zero_mean_sq_7_cast")]; + tensor var_1208 = const()[name = tensor("op_1208"), val = tensor([1])]; + tensor var_1209_cast = reduce_mean(axes = var_1208, keep_dims = var_115, x = zero_mean_sq_7_cast)[name = tensor("op_1209_cast")]; + tensor var_1210_to_fp16 = const()[name = tensor("op_1210_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_1211_cast = add(x = var_1209_cast, y = var_1210_to_fp16)[name = tensor("op_1211_cast")]; + tensor denom_7_epsilon_0_to_fp16 = const()[name = tensor("denom_7_epsilon_0_to_fp16"), val = tensor(0x1p-24)]; + tensor denom_7_cast = rsqrt(epsilon = denom_7_epsilon_0_to_fp16, x = var_1211_cast)[name = tensor("denom_7_cast")]; + tensor out_7_cast = mul(x = zero_mean_7_cast, y = denom_7_cast)[name = tensor("out_7_cast")]; + tensor var_1215_to_fp16 = const()[name = tensor("op_1215_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(7064896)))]; + tensor var_1216_cast = add(x = out_7_cast, y = var_1215_to_fp16)[name = tensor("op_1216_cast")]; + tensor var_1218_to_fp16 = const()[name = tensor("op_1218_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(7065600)))]; + tensor hidden_states_27_cast = mul(x = var_1216_cast, y = var_1218_to_fp16)[name = tensor("hidden_states_27_cast")]; + tensor var_1225 = const()[name = tensor("op_1225"), val = tensor([1, 1])]; + tensor var_1227 = const()[name = tensor("op_1227"), val = tensor([1, 1])]; + tensor q_5_pad_type_0 = const()[name = tensor("q_5_pad_type_0"), val = tensor("custom")]; + tensor q_5_pad_0 = const()[name = tensor("q_5_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_0_attentions_1_transformer_blocks_0_attn1_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(7066304))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(7143168))), name = tensor("down_blocks_0_attentions_1_transformer_blocks_0_attn1_to_q_weight_to_fp16_palettized"), shape = tensor([320, 320, 1, 1])]; + tensor q_5_cast = conv(dilations = var_1227, groups = var_120, pad = q_5_pad_0, pad_type = q_5_pad_type_0, strides = var_1225, weight = down_blocks_0_attentions_1_transformer_blocks_0_attn1_to_q_weight_to_fp16_palettized, x = hidden_states_27_cast)[name = tensor("q_5_cast")]; + tensor var_1231 = const()[name = tensor("op_1231"), val = tensor([1, 1])]; + tensor var_1233 = const()[name = tensor("op_1233"), val = tensor([1, 1])]; + tensor k_9_pad_type_0 = const()[name = tensor("k_9_pad_type_0"), val = tensor("custom")]; + tensor k_9_pad_0 = const()[name = tensor("k_9_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_0_attentions_1_transformer_blocks_0_attn1_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(7143360))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(7220224))), name = tensor("down_blocks_0_attentions_1_transformer_blocks_0_attn1_to_k_weight_to_fp16_palettized"), shape = tensor([320, 320, 1, 1])]; + tensor k_9_cast = conv(dilations = var_1233, groups = var_120, pad = k_9_pad_0, pad_type = k_9_pad_type_0, strides = var_1231, weight = down_blocks_0_attentions_1_transformer_blocks_0_attn1_to_k_weight_to_fp16_palettized, x = hidden_states_27_cast)[name = tensor("k_9_cast")]; + tensor var_1237 = const()[name = tensor("op_1237"), val = tensor([1, 1])]; + tensor var_1239 = const()[name = tensor("op_1239"), val = tensor([1, 1])]; + tensor v_5_pad_type_0 = const()[name = tensor("v_5_pad_type_0"), val = tensor("custom")]; + tensor v_5_pad_0 = const()[name = tensor("v_5_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_0_attentions_1_transformer_blocks_0_attn1_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(7220416))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(7297280))), name = tensor("down_blocks_0_attentions_1_transformer_blocks_0_attn1_to_v_weight_to_fp16_palettized"), shape = tensor([320, 320, 1, 1])]; + tensor v_5_cast = conv(dilations = var_1239, groups = var_120, pad = v_5_pad_0, pad_type = v_5_pad_type_0, strides = var_1237, weight = down_blocks_0_attentions_1_transformer_blocks_0_attn1_to_v_weight_to_fp16_palettized, x = hidden_states_27_cast)[name = tensor("v_5_cast")]; + tensor var_1243_begin_0 = const()[name = tensor("op_1243_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1243_end_0 = const()[name = tensor("op_1243_end_0"), val = tensor([2, 64, 1, 4096])]; + tensor var_1243_end_mask_0 = const()[name = tensor("op_1243_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1243_cast = slice_by_index(begin = var_1243_begin_0, end = var_1243_end_0, end_mask = var_1243_end_mask_0, x = q_5_cast)[name = tensor("op_1243_cast")]; + tensor var_1247_begin_0 = const()[name = tensor("op_1247_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_1247_end_0 = const()[name = tensor("op_1247_end_0"), val = tensor([2, 128, 1, 4096])]; + tensor var_1247_end_mask_0 = const()[name = tensor("op_1247_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1247_cast = slice_by_index(begin = var_1247_begin_0, end = var_1247_end_0, end_mask = var_1247_end_mask_0, x = q_5_cast)[name = tensor("op_1247_cast")]; + tensor var_1251_begin_0 = const()[name = tensor("op_1251_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_1251_end_0 = const()[name = tensor("op_1251_end_0"), val = tensor([2, 192, 1, 4096])]; + tensor var_1251_end_mask_0 = const()[name = tensor("op_1251_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1251_cast = slice_by_index(begin = var_1251_begin_0, end = var_1251_end_0, end_mask = var_1251_end_mask_0, x = q_5_cast)[name = tensor("op_1251_cast")]; + tensor var_1255_begin_0 = const()[name = tensor("op_1255_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_1255_end_0 = const()[name = tensor("op_1255_end_0"), val = tensor([2, 256, 1, 4096])]; + tensor var_1255_end_mask_0 = const()[name = tensor("op_1255_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1255_cast = slice_by_index(begin = var_1255_begin_0, end = var_1255_end_0, end_mask = var_1255_end_mask_0, x = q_5_cast)[name = tensor("op_1255_cast")]; + tensor var_1259_begin_0 = const()[name = tensor("op_1259_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_1259_end_0 = const()[name = tensor("op_1259_end_0"), val = tensor([2, 320, 1, 4096])]; + tensor var_1259_end_mask_0 = const()[name = tensor("op_1259_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1259_cast = slice_by_index(begin = var_1259_begin_0, end = var_1259_end_0, end_mask = var_1259_end_mask_0, x = q_5_cast)[name = tensor("op_1259_cast")]; + tensor var_1262_begin_0 = const()[name = tensor("op_1262_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1262_end_0 = const()[name = tensor("op_1262_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_1262_end_mask_0 = const()[name = tensor("op_1262_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1262_cast = slice_by_index(begin = var_1262_begin_0, end = var_1262_end_0, end_mask = var_1262_end_mask_0, x = var_1243_cast)[name = tensor("op_1262_cast")]; + tensor var_1263_begin_0 = const()[name = tensor("op_1263_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_1263_end_0 = const()[name = tensor("op_1263_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_1263_end_mask_0 = const()[name = tensor("op_1263_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1263_cast = slice_by_index(begin = var_1263_begin_0, end = var_1263_end_0, end_mask = var_1263_end_mask_0, x = var_1243_cast)[name = tensor("op_1263_cast")]; + tensor var_1264_begin_0 = const()[name = tensor("op_1264_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_1264_end_0 = const()[name = tensor("op_1264_end_0"), val = tensor([2, 64, 1, 1536])]; + tensor var_1264_end_mask_0 = const()[name = tensor("op_1264_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1264_cast = slice_by_index(begin = var_1264_begin_0, end = var_1264_end_0, end_mask = var_1264_end_mask_0, x = var_1243_cast)[name = tensor("op_1264_cast")]; + tensor var_1265_begin_0 = const()[name = tensor("op_1265_begin_0"), val = tensor([0, 0, 0, 1536])]; + tensor var_1265_end_0 = const()[name = tensor("op_1265_end_0"), val = tensor([2, 64, 1, 2048])]; + tensor var_1265_end_mask_0 = const()[name = tensor("op_1265_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1265_cast = slice_by_index(begin = var_1265_begin_0, end = var_1265_end_0, end_mask = var_1265_end_mask_0, x = var_1243_cast)[name = tensor("op_1265_cast")]; + tensor var_1266_begin_0 = const()[name = tensor("op_1266_begin_0"), val = tensor([0, 0, 0, 2048])]; + tensor var_1266_end_0 = const()[name = tensor("op_1266_end_0"), val = tensor([2, 64, 1, 2560])]; + tensor var_1266_end_mask_0 = const()[name = tensor("op_1266_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1266_cast = slice_by_index(begin = var_1266_begin_0, end = var_1266_end_0, end_mask = var_1266_end_mask_0, x = var_1243_cast)[name = tensor("op_1266_cast")]; + tensor var_1267_begin_0 = const()[name = tensor("op_1267_begin_0"), val = tensor([0, 0, 0, 2560])]; + tensor var_1267_end_0 = const()[name = tensor("op_1267_end_0"), val = tensor([2, 64, 1, 3072])]; + tensor var_1267_end_mask_0 = const()[name = tensor("op_1267_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1267_cast = slice_by_index(begin = var_1267_begin_0, end = var_1267_end_0, end_mask = var_1267_end_mask_0, x = var_1243_cast)[name = tensor("op_1267_cast")]; + tensor var_1268_begin_0 = const()[name = tensor("op_1268_begin_0"), val = tensor([0, 0, 0, 3072])]; + tensor var_1268_end_0 = const()[name = tensor("op_1268_end_0"), val = tensor([2, 64, 1, 3584])]; + tensor var_1268_end_mask_0 = const()[name = tensor("op_1268_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1268_cast = slice_by_index(begin = var_1268_begin_0, end = var_1268_end_0, end_mask = var_1268_end_mask_0, x = var_1243_cast)[name = tensor("op_1268_cast")]; + tensor var_1269_begin_0 = const()[name = tensor("op_1269_begin_0"), val = tensor([0, 0, 0, 3584])]; + tensor var_1269_end_0 = const()[name = tensor("op_1269_end_0"), val = tensor([2, 64, 1, 4096])]; + tensor var_1269_end_mask_0 = const()[name = tensor("op_1269_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1269_cast = slice_by_index(begin = var_1269_begin_0, end = var_1269_end_0, end_mask = var_1269_end_mask_0, x = var_1243_cast)[name = tensor("op_1269_cast")]; + tensor var_1270_begin_0 = const()[name = tensor("op_1270_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1270_end_0 = const()[name = tensor("op_1270_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_1270_end_mask_0 = const()[name = tensor("op_1270_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1270_cast = slice_by_index(begin = var_1270_begin_0, end = var_1270_end_0, end_mask = var_1270_end_mask_0, x = var_1247_cast)[name = tensor("op_1270_cast")]; + tensor var_1271_begin_0 = const()[name = tensor("op_1271_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_1271_end_0 = const()[name = tensor("op_1271_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_1271_end_mask_0 = const()[name = tensor("op_1271_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1271_cast = slice_by_index(begin = var_1271_begin_0, end = var_1271_end_0, end_mask = var_1271_end_mask_0, x = var_1247_cast)[name = tensor("op_1271_cast")]; + tensor var_1272_begin_0 = const()[name = tensor("op_1272_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_1272_end_0 = const()[name = tensor("op_1272_end_0"), val = tensor([2, 64, 1, 1536])]; + tensor var_1272_end_mask_0 = const()[name = tensor("op_1272_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1272_cast = slice_by_index(begin = var_1272_begin_0, end = var_1272_end_0, end_mask = var_1272_end_mask_0, x = var_1247_cast)[name = tensor("op_1272_cast")]; + tensor var_1273_begin_0 = const()[name = tensor("op_1273_begin_0"), val = tensor([0, 0, 0, 1536])]; + tensor var_1273_end_0 = const()[name = tensor("op_1273_end_0"), val = tensor([2, 64, 1, 2048])]; + tensor var_1273_end_mask_0 = const()[name = tensor("op_1273_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1273_cast = slice_by_index(begin = var_1273_begin_0, end = var_1273_end_0, end_mask = var_1273_end_mask_0, x = var_1247_cast)[name = tensor("op_1273_cast")]; + tensor var_1274_begin_0 = const()[name = tensor("op_1274_begin_0"), val = tensor([0, 0, 0, 2048])]; + tensor var_1274_end_0 = const()[name = tensor("op_1274_end_0"), val = tensor([2, 64, 1, 2560])]; + tensor var_1274_end_mask_0 = const()[name = tensor("op_1274_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1274_cast = slice_by_index(begin = var_1274_begin_0, end = var_1274_end_0, end_mask = var_1274_end_mask_0, x = var_1247_cast)[name = tensor("op_1274_cast")]; + tensor var_1275_begin_0 = const()[name = tensor("op_1275_begin_0"), val = tensor([0, 0, 0, 2560])]; + tensor var_1275_end_0 = const()[name = tensor("op_1275_end_0"), val = tensor([2, 64, 1, 3072])]; + tensor var_1275_end_mask_0 = const()[name = tensor("op_1275_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1275_cast = slice_by_index(begin = var_1275_begin_0, end = var_1275_end_0, end_mask = var_1275_end_mask_0, x = var_1247_cast)[name = tensor("op_1275_cast")]; + tensor var_1276_begin_0 = const()[name = tensor("op_1276_begin_0"), val = tensor([0, 0, 0, 3072])]; + tensor var_1276_end_0 = const()[name = tensor("op_1276_end_0"), val = tensor([2, 64, 1, 3584])]; + tensor var_1276_end_mask_0 = const()[name = tensor("op_1276_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1276_cast = slice_by_index(begin = var_1276_begin_0, end = var_1276_end_0, end_mask = var_1276_end_mask_0, x = var_1247_cast)[name = tensor("op_1276_cast")]; + tensor var_1277_begin_0 = const()[name = tensor("op_1277_begin_0"), val = tensor([0, 0, 0, 3584])]; + tensor var_1277_end_0 = const()[name = tensor("op_1277_end_0"), val = tensor([2, 64, 1, 4096])]; + tensor var_1277_end_mask_0 = const()[name = tensor("op_1277_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1277_cast = slice_by_index(begin = var_1277_begin_0, end = var_1277_end_0, end_mask = var_1277_end_mask_0, x = var_1247_cast)[name = tensor("op_1277_cast")]; + tensor var_1278_begin_0 = const()[name = tensor("op_1278_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1278_end_0 = const()[name = tensor("op_1278_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_1278_end_mask_0 = const()[name = tensor("op_1278_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1278_cast = slice_by_index(begin = var_1278_begin_0, end = var_1278_end_0, end_mask = var_1278_end_mask_0, x = var_1251_cast)[name = tensor("op_1278_cast")]; + tensor var_1279_begin_0 = const()[name = tensor("op_1279_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_1279_end_0 = const()[name = tensor("op_1279_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_1279_end_mask_0 = const()[name = tensor("op_1279_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1279_cast = slice_by_index(begin = var_1279_begin_0, end = var_1279_end_0, end_mask = var_1279_end_mask_0, x = var_1251_cast)[name = tensor("op_1279_cast")]; + tensor var_1280_begin_0 = const()[name = tensor("op_1280_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_1280_end_0 = const()[name = tensor("op_1280_end_0"), val = tensor([2, 64, 1, 1536])]; + tensor var_1280_end_mask_0 = const()[name = tensor("op_1280_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1280_cast = slice_by_index(begin = var_1280_begin_0, end = var_1280_end_0, end_mask = var_1280_end_mask_0, x = var_1251_cast)[name = tensor("op_1280_cast")]; + tensor var_1281_begin_0 = const()[name = tensor("op_1281_begin_0"), val = tensor([0, 0, 0, 1536])]; + tensor var_1281_end_0 = const()[name = tensor("op_1281_end_0"), val = tensor([2, 64, 1, 2048])]; + tensor var_1281_end_mask_0 = const()[name = tensor("op_1281_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1281_cast = slice_by_index(begin = var_1281_begin_0, end = var_1281_end_0, end_mask = var_1281_end_mask_0, x = var_1251_cast)[name = tensor("op_1281_cast")]; + tensor var_1282_begin_0 = const()[name = tensor("op_1282_begin_0"), val = tensor([0, 0, 0, 2048])]; + tensor var_1282_end_0 = const()[name = tensor("op_1282_end_0"), val = tensor([2, 64, 1, 2560])]; + tensor var_1282_end_mask_0 = const()[name = tensor("op_1282_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1282_cast = slice_by_index(begin = var_1282_begin_0, end = var_1282_end_0, end_mask = var_1282_end_mask_0, x = var_1251_cast)[name = tensor("op_1282_cast")]; + tensor var_1283_begin_0 = const()[name = tensor("op_1283_begin_0"), val = tensor([0, 0, 0, 2560])]; + tensor var_1283_end_0 = const()[name = tensor("op_1283_end_0"), val = tensor([2, 64, 1, 3072])]; + tensor var_1283_end_mask_0 = const()[name = tensor("op_1283_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1283_cast = slice_by_index(begin = var_1283_begin_0, end = var_1283_end_0, end_mask = var_1283_end_mask_0, x = var_1251_cast)[name = tensor("op_1283_cast")]; + tensor var_1284_begin_0 = const()[name = tensor("op_1284_begin_0"), val = tensor([0, 0, 0, 3072])]; + tensor var_1284_end_0 = const()[name = tensor("op_1284_end_0"), val = tensor([2, 64, 1, 3584])]; + tensor var_1284_end_mask_0 = const()[name = tensor("op_1284_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1284_cast = slice_by_index(begin = var_1284_begin_0, end = var_1284_end_0, end_mask = var_1284_end_mask_0, x = var_1251_cast)[name = tensor("op_1284_cast")]; + tensor var_1285_begin_0 = const()[name = tensor("op_1285_begin_0"), val = tensor([0, 0, 0, 3584])]; + tensor var_1285_end_0 = const()[name = tensor("op_1285_end_0"), val = tensor([2, 64, 1, 4096])]; + tensor var_1285_end_mask_0 = const()[name = tensor("op_1285_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1285_cast = slice_by_index(begin = var_1285_begin_0, end = var_1285_end_0, end_mask = var_1285_end_mask_0, x = var_1251_cast)[name = tensor("op_1285_cast")]; + tensor var_1286_begin_0 = const()[name = tensor("op_1286_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1286_end_0 = const()[name = tensor("op_1286_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_1286_end_mask_0 = const()[name = tensor("op_1286_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1286_cast = slice_by_index(begin = var_1286_begin_0, end = var_1286_end_0, end_mask = var_1286_end_mask_0, x = var_1255_cast)[name = tensor("op_1286_cast")]; + tensor var_1287_begin_0 = const()[name = tensor("op_1287_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_1287_end_0 = const()[name = tensor("op_1287_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_1287_end_mask_0 = const()[name = tensor("op_1287_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1287_cast = slice_by_index(begin = var_1287_begin_0, end = var_1287_end_0, end_mask = var_1287_end_mask_0, x = var_1255_cast)[name = tensor("op_1287_cast")]; + tensor var_1288_begin_0 = const()[name = tensor("op_1288_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_1288_end_0 = const()[name = tensor("op_1288_end_0"), val = tensor([2, 64, 1, 1536])]; + tensor var_1288_end_mask_0 = const()[name = tensor("op_1288_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1288_cast = slice_by_index(begin = var_1288_begin_0, end = var_1288_end_0, end_mask = var_1288_end_mask_0, x = var_1255_cast)[name = tensor("op_1288_cast")]; + tensor var_1289_begin_0 = const()[name = tensor("op_1289_begin_0"), val = tensor([0, 0, 0, 1536])]; + tensor var_1289_end_0 = const()[name = tensor("op_1289_end_0"), val = tensor([2, 64, 1, 2048])]; + tensor var_1289_end_mask_0 = const()[name = tensor("op_1289_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1289_cast = slice_by_index(begin = var_1289_begin_0, end = var_1289_end_0, end_mask = var_1289_end_mask_0, x = var_1255_cast)[name = tensor("op_1289_cast")]; + tensor var_1290_begin_0 = const()[name = tensor("op_1290_begin_0"), val = tensor([0, 0, 0, 2048])]; + tensor var_1290_end_0 = const()[name = tensor("op_1290_end_0"), val = tensor([2, 64, 1, 2560])]; + tensor var_1290_end_mask_0 = const()[name = tensor("op_1290_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1290_cast = slice_by_index(begin = var_1290_begin_0, end = var_1290_end_0, end_mask = var_1290_end_mask_0, x = var_1255_cast)[name = tensor("op_1290_cast")]; + tensor var_1291_begin_0 = const()[name = tensor("op_1291_begin_0"), val = tensor([0, 0, 0, 2560])]; + tensor var_1291_end_0 = const()[name = tensor("op_1291_end_0"), val = tensor([2, 64, 1, 3072])]; + tensor var_1291_end_mask_0 = const()[name = tensor("op_1291_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1291_cast = slice_by_index(begin = var_1291_begin_0, end = var_1291_end_0, end_mask = var_1291_end_mask_0, x = var_1255_cast)[name = tensor("op_1291_cast")]; + tensor var_1292_begin_0 = const()[name = tensor("op_1292_begin_0"), val = tensor([0, 0, 0, 3072])]; + tensor var_1292_end_0 = const()[name = tensor("op_1292_end_0"), val = tensor([2, 64, 1, 3584])]; + tensor var_1292_end_mask_0 = const()[name = tensor("op_1292_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1292_cast = slice_by_index(begin = var_1292_begin_0, end = var_1292_end_0, end_mask = var_1292_end_mask_0, x = var_1255_cast)[name = tensor("op_1292_cast")]; + tensor var_1293_begin_0 = const()[name = tensor("op_1293_begin_0"), val = tensor([0, 0, 0, 3584])]; + tensor var_1293_end_0 = const()[name = tensor("op_1293_end_0"), val = tensor([2, 64, 1, 4096])]; + tensor var_1293_end_mask_0 = const()[name = tensor("op_1293_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1293_cast = slice_by_index(begin = var_1293_begin_0, end = var_1293_end_0, end_mask = var_1293_end_mask_0, x = var_1255_cast)[name = tensor("op_1293_cast")]; + tensor var_1294_begin_0 = const()[name = tensor("op_1294_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1294_end_0 = const()[name = tensor("op_1294_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_1294_end_mask_0 = const()[name = tensor("op_1294_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1294_cast = slice_by_index(begin = var_1294_begin_0, end = var_1294_end_0, end_mask = var_1294_end_mask_0, x = var_1259_cast)[name = tensor("op_1294_cast")]; + tensor var_1295_begin_0 = const()[name = tensor("op_1295_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_1295_end_0 = const()[name = tensor("op_1295_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_1295_end_mask_0 = const()[name = tensor("op_1295_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1295_cast = slice_by_index(begin = var_1295_begin_0, end = var_1295_end_0, end_mask = var_1295_end_mask_0, x = var_1259_cast)[name = tensor("op_1295_cast")]; + tensor var_1296_begin_0 = const()[name = tensor("op_1296_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_1296_end_0 = const()[name = tensor("op_1296_end_0"), val = tensor([2, 64, 1, 1536])]; + tensor var_1296_end_mask_0 = const()[name = tensor("op_1296_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1296_cast = slice_by_index(begin = var_1296_begin_0, end = var_1296_end_0, end_mask = var_1296_end_mask_0, x = var_1259_cast)[name = tensor("op_1296_cast")]; + tensor var_1297_begin_0 = const()[name = tensor("op_1297_begin_0"), val = tensor([0, 0, 0, 1536])]; + tensor var_1297_end_0 = const()[name = tensor("op_1297_end_0"), val = tensor([2, 64, 1, 2048])]; + tensor var_1297_end_mask_0 = const()[name = tensor("op_1297_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1297_cast = slice_by_index(begin = var_1297_begin_0, end = var_1297_end_0, end_mask = var_1297_end_mask_0, x = var_1259_cast)[name = tensor("op_1297_cast")]; + tensor var_1298_begin_0 = const()[name = tensor("op_1298_begin_0"), val = tensor([0, 0, 0, 2048])]; + tensor var_1298_end_0 = const()[name = tensor("op_1298_end_0"), val = tensor([2, 64, 1, 2560])]; + tensor var_1298_end_mask_0 = const()[name = tensor("op_1298_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1298_cast = slice_by_index(begin = var_1298_begin_0, end = var_1298_end_0, end_mask = var_1298_end_mask_0, x = var_1259_cast)[name = tensor("op_1298_cast")]; + tensor var_1299_begin_0 = const()[name = tensor("op_1299_begin_0"), val = tensor([0, 0, 0, 2560])]; + tensor var_1299_end_0 = const()[name = tensor("op_1299_end_0"), val = tensor([2, 64, 1, 3072])]; + tensor var_1299_end_mask_0 = const()[name = tensor("op_1299_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1299_cast = slice_by_index(begin = var_1299_begin_0, end = var_1299_end_0, end_mask = var_1299_end_mask_0, x = var_1259_cast)[name = tensor("op_1299_cast")]; + tensor var_1300_begin_0 = const()[name = tensor("op_1300_begin_0"), val = tensor([0, 0, 0, 3072])]; + tensor var_1300_end_0 = const()[name = tensor("op_1300_end_0"), val = tensor([2, 64, 1, 3584])]; + tensor var_1300_end_mask_0 = const()[name = tensor("op_1300_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1300_cast = slice_by_index(begin = var_1300_begin_0, end = var_1300_end_0, end_mask = var_1300_end_mask_0, x = var_1259_cast)[name = tensor("op_1300_cast")]; + tensor var_1301_begin_0 = const()[name = tensor("op_1301_begin_0"), val = tensor([0, 0, 0, 3584])]; + tensor var_1301_end_0 = const()[name = tensor("op_1301_end_0"), val = tensor([2, 64, 1, 4096])]; + tensor var_1301_end_mask_0 = const()[name = tensor("op_1301_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1301_cast = slice_by_index(begin = var_1301_begin_0, end = var_1301_end_0, end_mask = var_1301_end_mask_0, x = var_1259_cast)[name = tensor("op_1301_cast")]; + tensor k_11_perm_0 = const()[name = tensor("k_11_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_1306_begin_0 = const()[name = tensor("op_1306_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1306_end_0 = const()[name = tensor("op_1306_end_0"), val = tensor([2, 4096, 1, 64])]; + tensor var_1306_end_mask_0 = const()[name = tensor("op_1306_end_mask_0"), val = tensor([true, true, true, false])]; + tensor transpose_29 = transpose(perm = k_11_perm_0, x = k_9_cast)[name = tensor("transpose_29")]; + tensor var_1306_cast = slice_by_index(begin = var_1306_begin_0, end = var_1306_end_0, end_mask = var_1306_end_mask_0, x = transpose_29)[name = tensor("op_1306_cast")]; + tensor var_1310_begin_0 = const()[name = tensor("op_1310_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_1310_end_0 = const()[name = tensor("op_1310_end_0"), val = tensor([2, 4096, 1, 128])]; + tensor var_1310_end_mask_0 = const()[name = tensor("op_1310_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1310_cast = slice_by_index(begin = var_1310_begin_0, end = var_1310_end_0, end_mask = var_1310_end_mask_0, x = transpose_29)[name = tensor("op_1310_cast")]; + tensor var_1314_begin_0 = const()[name = tensor("op_1314_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_1314_end_0 = const()[name = tensor("op_1314_end_0"), val = tensor([2, 4096, 1, 192])]; + tensor var_1314_end_mask_0 = const()[name = tensor("op_1314_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1314_cast = slice_by_index(begin = var_1314_begin_0, end = var_1314_end_0, end_mask = var_1314_end_mask_0, x = transpose_29)[name = tensor("op_1314_cast")]; + tensor var_1318_begin_0 = const()[name = tensor("op_1318_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_1318_end_0 = const()[name = tensor("op_1318_end_0"), val = tensor([2, 4096, 1, 256])]; + tensor var_1318_end_mask_0 = const()[name = tensor("op_1318_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1318_cast = slice_by_index(begin = var_1318_begin_0, end = var_1318_end_0, end_mask = var_1318_end_mask_0, x = transpose_29)[name = tensor("op_1318_cast")]; + tensor var_1322_begin_0 = const()[name = tensor("op_1322_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_1322_end_0 = const()[name = tensor("op_1322_end_0"), val = tensor([2, 4096, 1, 320])]; + tensor var_1322_end_mask_0 = const()[name = tensor("op_1322_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1322_cast = slice_by_index(begin = var_1322_begin_0, end = var_1322_end_0, end_mask = var_1322_end_mask_0, x = transpose_29)[name = tensor("op_1322_cast")]; + tensor var_1324_begin_0 = const()[name = tensor("op_1324_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1324_end_0 = const()[name = tensor("op_1324_end_0"), val = tensor([2, 64, 1, 4096])]; + tensor var_1324_end_mask_0 = const()[name = tensor("op_1324_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1324_cast = slice_by_index(begin = var_1324_begin_0, end = var_1324_end_0, end_mask = var_1324_end_mask_0, x = v_5_cast)[name = tensor("op_1324_cast")]; + tensor var_1328_begin_0 = const()[name = tensor("op_1328_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_1328_end_0 = const()[name = tensor("op_1328_end_0"), val = tensor([2, 128, 1, 4096])]; + tensor var_1328_end_mask_0 = const()[name = tensor("op_1328_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1328_cast = slice_by_index(begin = var_1328_begin_0, end = var_1328_end_0, end_mask = var_1328_end_mask_0, x = v_5_cast)[name = tensor("op_1328_cast")]; + tensor var_1332_begin_0 = const()[name = tensor("op_1332_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_1332_end_0 = const()[name = tensor("op_1332_end_0"), val = tensor([2, 192, 1, 4096])]; + tensor var_1332_end_mask_0 = const()[name = tensor("op_1332_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1332_cast = slice_by_index(begin = var_1332_begin_0, end = var_1332_end_0, end_mask = var_1332_end_mask_0, x = v_5_cast)[name = tensor("op_1332_cast")]; + tensor var_1336_begin_0 = const()[name = tensor("op_1336_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_1336_end_0 = const()[name = tensor("op_1336_end_0"), val = tensor([2, 256, 1, 4096])]; + tensor var_1336_end_mask_0 = const()[name = tensor("op_1336_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1336_cast = slice_by_index(begin = var_1336_begin_0, end = var_1336_end_0, end_mask = var_1336_end_mask_0, x = v_5_cast)[name = tensor("op_1336_cast")]; + tensor var_1340_begin_0 = const()[name = tensor("op_1340_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_1340_end_0 = const()[name = tensor("op_1340_end_0"), val = tensor([2, 320, 1, 4096])]; + tensor var_1340_end_mask_0 = const()[name = tensor("op_1340_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1340_cast = slice_by_index(begin = var_1340_begin_0, end = var_1340_end_0, end_mask = var_1340_end_mask_0, x = v_5_cast)[name = tensor("op_1340_cast")]; + tensor var_1344_equation_0 = const()[name = tensor("op_1344_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1344_cast = einsum(equation = var_1344_equation_0, values = (var_1306_cast, var_1262_cast))[name = tensor("op_1344_cast")]; + tensor var_1345_to_fp16 = const()[name = tensor("op_1345_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_161_cast = mul(x = var_1344_cast, y = var_1345_to_fp16)[name = tensor("aw_chunk_161_cast")]; + tensor var_1348_equation_0 = const()[name = tensor("op_1348_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1348_cast = einsum(equation = var_1348_equation_0, values = (var_1306_cast, var_1263_cast))[name = tensor("op_1348_cast")]; + tensor var_1349_to_fp16 = const()[name = tensor("op_1349_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_163_cast = mul(x = var_1348_cast, y = var_1349_to_fp16)[name = tensor("aw_chunk_163_cast")]; + tensor var_1352_equation_0 = const()[name = tensor("op_1352_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1352_cast = einsum(equation = var_1352_equation_0, values = (var_1306_cast, var_1264_cast))[name = tensor("op_1352_cast")]; + tensor var_1353_to_fp16 = const()[name = tensor("op_1353_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_165_cast = mul(x = var_1352_cast, y = var_1353_to_fp16)[name = tensor("aw_chunk_165_cast")]; + tensor var_1356_equation_0 = const()[name = tensor("op_1356_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1356_cast = einsum(equation = var_1356_equation_0, values = (var_1306_cast, var_1265_cast))[name = tensor("op_1356_cast")]; + tensor var_1357_to_fp16 = const()[name = tensor("op_1357_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_167_cast = mul(x = var_1356_cast, y = var_1357_to_fp16)[name = tensor("aw_chunk_167_cast")]; + tensor var_1360_equation_0 = const()[name = tensor("op_1360_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1360_cast = einsum(equation = var_1360_equation_0, values = (var_1306_cast, var_1266_cast))[name = tensor("op_1360_cast")]; + tensor var_1361_to_fp16 = const()[name = tensor("op_1361_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_169_cast = mul(x = var_1360_cast, y = var_1361_to_fp16)[name = tensor("aw_chunk_169_cast")]; + tensor var_1364_equation_0 = const()[name = tensor("op_1364_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1364_cast = einsum(equation = var_1364_equation_0, values = (var_1306_cast, var_1267_cast))[name = tensor("op_1364_cast")]; + tensor var_1365_to_fp16 = const()[name = tensor("op_1365_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_171_cast = mul(x = var_1364_cast, y = var_1365_to_fp16)[name = tensor("aw_chunk_171_cast")]; + tensor var_1368_equation_0 = const()[name = tensor("op_1368_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1368_cast = einsum(equation = var_1368_equation_0, values = (var_1306_cast, var_1268_cast))[name = tensor("op_1368_cast")]; + tensor var_1369_to_fp16 = const()[name = tensor("op_1369_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_173_cast = mul(x = var_1368_cast, y = var_1369_to_fp16)[name = tensor("aw_chunk_173_cast")]; + tensor var_1372_equation_0 = const()[name = tensor("op_1372_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1372_cast = einsum(equation = var_1372_equation_0, values = (var_1306_cast, var_1269_cast))[name = tensor("op_1372_cast")]; + tensor var_1373_to_fp16 = const()[name = tensor("op_1373_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_175_cast = mul(x = var_1372_cast, y = var_1373_to_fp16)[name = tensor("aw_chunk_175_cast")]; + tensor var_1376_equation_0 = const()[name = tensor("op_1376_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1376_cast = einsum(equation = var_1376_equation_0, values = (var_1310_cast, var_1270_cast))[name = tensor("op_1376_cast")]; + tensor var_1377_to_fp16 = const()[name = tensor("op_1377_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_177_cast = mul(x = var_1376_cast, y = var_1377_to_fp16)[name = tensor("aw_chunk_177_cast")]; + tensor var_1380_equation_0 = const()[name = tensor("op_1380_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1380_cast = einsum(equation = var_1380_equation_0, values = (var_1310_cast, var_1271_cast))[name = tensor("op_1380_cast")]; + tensor var_1381_to_fp16 = const()[name = tensor("op_1381_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_179_cast = mul(x = var_1380_cast, y = var_1381_to_fp16)[name = tensor("aw_chunk_179_cast")]; + tensor var_1384_equation_0 = const()[name = tensor("op_1384_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1384_cast = einsum(equation = var_1384_equation_0, values = (var_1310_cast, var_1272_cast))[name = tensor("op_1384_cast")]; + tensor var_1385_to_fp16 = const()[name = tensor("op_1385_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_181_cast = mul(x = var_1384_cast, y = var_1385_to_fp16)[name = tensor("aw_chunk_181_cast")]; + tensor var_1388_equation_0 = const()[name = tensor("op_1388_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1388_cast = einsum(equation = var_1388_equation_0, values = (var_1310_cast, var_1273_cast))[name = tensor("op_1388_cast")]; + tensor var_1389_to_fp16 = const()[name = tensor("op_1389_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_183_cast = mul(x = var_1388_cast, y = var_1389_to_fp16)[name = tensor("aw_chunk_183_cast")]; + tensor var_1392_equation_0 = const()[name = tensor("op_1392_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1392_cast = einsum(equation = var_1392_equation_0, values = (var_1310_cast, var_1274_cast))[name = tensor("op_1392_cast")]; + tensor var_1393_to_fp16 = const()[name = tensor("op_1393_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_185_cast = mul(x = var_1392_cast, y = var_1393_to_fp16)[name = tensor("aw_chunk_185_cast")]; + tensor var_1396_equation_0 = const()[name = tensor("op_1396_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1396_cast = einsum(equation = var_1396_equation_0, values = (var_1310_cast, var_1275_cast))[name = tensor("op_1396_cast")]; + tensor var_1397_to_fp16 = const()[name = tensor("op_1397_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_187_cast = mul(x = var_1396_cast, y = var_1397_to_fp16)[name = tensor("aw_chunk_187_cast")]; + tensor var_1400_equation_0 = const()[name = tensor("op_1400_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1400_cast = einsum(equation = var_1400_equation_0, values = (var_1310_cast, var_1276_cast))[name = tensor("op_1400_cast")]; + tensor var_1401_to_fp16 = const()[name = tensor("op_1401_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_189_cast = mul(x = var_1400_cast, y = var_1401_to_fp16)[name = tensor("aw_chunk_189_cast")]; + tensor var_1404_equation_0 = const()[name = tensor("op_1404_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1404_cast = einsum(equation = var_1404_equation_0, values = (var_1310_cast, var_1277_cast))[name = tensor("op_1404_cast")]; + tensor var_1405_to_fp16 = const()[name = tensor("op_1405_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_191_cast = mul(x = var_1404_cast, y = var_1405_to_fp16)[name = tensor("aw_chunk_191_cast")]; + tensor var_1408_equation_0 = const()[name = tensor("op_1408_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1408_cast = einsum(equation = var_1408_equation_0, values = (var_1314_cast, var_1278_cast))[name = tensor("op_1408_cast")]; + tensor var_1409_to_fp16 = const()[name = tensor("op_1409_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_193_cast = mul(x = var_1408_cast, y = var_1409_to_fp16)[name = tensor("aw_chunk_193_cast")]; + tensor var_1412_equation_0 = const()[name = tensor("op_1412_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1412_cast = einsum(equation = var_1412_equation_0, values = (var_1314_cast, var_1279_cast))[name = tensor("op_1412_cast")]; + tensor var_1413_to_fp16 = const()[name = tensor("op_1413_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_195_cast = mul(x = var_1412_cast, y = var_1413_to_fp16)[name = tensor("aw_chunk_195_cast")]; + tensor var_1416_equation_0 = const()[name = tensor("op_1416_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1416_cast = einsum(equation = var_1416_equation_0, values = (var_1314_cast, var_1280_cast))[name = tensor("op_1416_cast")]; + tensor var_1417_to_fp16 = const()[name = tensor("op_1417_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_197_cast = mul(x = var_1416_cast, y = var_1417_to_fp16)[name = tensor("aw_chunk_197_cast")]; + tensor var_1420_equation_0 = const()[name = tensor("op_1420_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1420_cast = einsum(equation = var_1420_equation_0, values = (var_1314_cast, var_1281_cast))[name = tensor("op_1420_cast")]; + tensor var_1421_to_fp16 = const()[name = tensor("op_1421_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_199_cast = mul(x = var_1420_cast, y = var_1421_to_fp16)[name = tensor("aw_chunk_199_cast")]; + tensor var_1424_equation_0 = const()[name = tensor("op_1424_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1424_cast = einsum(equation = var_1424_equation_0, values = (var_1314_cast, var_1282_cast))[name = tensor("op_1424_cast")]; + tensor var_1425_to_fp16 = const()[name = tensor("op_1425_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_201_cast = mul(x = var_1424_cast, y = var_1425_to_fp16)[name = tensor("aw_chunk_201_cast")]; + tensor var_1428_equation_0 = const()[name = tensor("op_1428_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1428_cast = einsum(equation = var_1428_equation_0, values = (var_1314_cast, var_1283_cast))[name = tensor("op_1428_cast")]; + tensor var_1429_to_fp16 = const()[name = tensor("op_1429_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_203_cast = mul(x = var_1428_cast, y = var_1429_to_fp16)[name = tensor("aw_chunk_203_cast")]; + tensor var_1432_equation_0 = const()[name = tensor("op_1432_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1432_cast = einsum(equation = var_1432_equation_0, values = (var_1314_cast, var_1284_cast))[name = tensor("op_1432_cast")]; + tensor var_1433_to_fp16 = const()[name = tensor("op_1433_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_205_cast = mul(x = var_1432_cast, y = var_1433_to_fp16)[name = tensor("aw_chunk_205_cast")]; + tensor var_1436_equation_0 = const()[name = tensor("op_1436_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1436_cast = einsum(equation = var_1436_equation_0, values = (var_1314_cast, var_1285_cast))[name = tensor("op_1436_cast")]; + tensor var_1437_to_fp16 = const()[name = tensor("op_1437_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_207_cast = mul(x = var_1436_cast, y = var_1437_to_fp16)[name = tensor("aw_chunk_207_cast")]; + tensor var_1440_equation_0 = const()[name = tensor("op_1440_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1440_cast = einsum(equation = var_1440_equation_0, values = (var_1318_cast, var_1286_cast))[name = tensor("op_1440_cast")]; + tensor var_1441_to_fp16 = const()[name = tensor("op_1441_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_209_cast = mul(x = var_1440_cast, y = var_1441_to_fp16)[name = tensor("aw_chunk_209_cast")]; + tensor var_1444_equation_0 = const()[name = tensor("op_1444_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1444_cast = einsum(equation = var_1444_equation_0, values = (var_1318_cast, var_1287_cast))[name = tensor("op_1444_cast")]; + tensor var_1445_to_fp16 = const()[name = tensor("op_1445_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_211_cast = mul(x = var_1444_cast, y = var_1445_to_fp16)[name = tensor("aw_chunk_211_cast")]; + tensor var_1448_equation_0 = const()[name = tensor("op_1448_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1448_cast = einsum(equation = var_1448_equation_0, values = (var_1318_cast, var_1288_cast))[name = tensor("op_1448_cast")]; + tensor var_1449_to_fp16 = const()[name = tensor("op_1449_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_213_cast = mul(x = var_1448_cast, y = var_1449_to_fp16)[name = tensor("aw_chunk_213_cast")]; + tensor var_1452_equation_0 = const()[name = tensor("op_1452_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1452_cast = einsum(equation = var_1452_equation_0, values = (var_1318_cast, var_1289_cast))[name = tensor("op_1452_cast")]; + tensor var_1453_to_fp16 = const()[name = tensor("op_1453_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_215_cast = mul(x = var_1452_cast, y = var_1453_to_fp16)[name = tensor("aw_chunk_215_cast")]; + tensor var_1456_equation_0 = const()[name = tensor("op_1456_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1456_cast = einsum(equation = var_1456_equation_0, values = (var_1318_cast, var_1290_cast))[name = tensor("op_1456_cast")]; + tensor var_1457_to_fp16 = const()[name = tensor("op_1457_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_217_cast = mul(x = var_1456_cast, y = var_1457_to_fp16)[name = tensor("aw_chunk_217_cast")]; + tensor var_1460_equation_0 = const()[name = tensor("op_1460_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1460_cast = einsum(equation = var_1460_equation_0, values = (var_1318_cast, var_1291_cast))[name = tensor("op_1460_cast")]; + tensor var_1461_to_fp16 = const()[name = tensor("op_1461_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_219_cast = mul(x = var_1460_cast, y = var_1461_to_fp16)[name = tensor("aw_chunk_219_cast")]; + tensor var_1464_equation_0 = const()[name = tensor("op_1464_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1464_cast = einsum(equation = var_1464_equation_0, values = (var_1318_cast, var_1292_cast))[name = tensor("op_1464_cast")]; + tensor var_1465_to_fp16 = const()[name = tensor("op_1465_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_221_cast = mul(x = var_1464_cast, y = var_1465_to_fp16)[name = tensor("aw_chunk_221_cast")]; + tensor var_1468_equation_0 = const()[name = tensor("op_1468_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1468_cast = einsum(equation = var_1468_equation_0, values = (var_1318_cast, var_1293_cast))[name = tensor("op_1468_cast")]; + tensor var_1469_to_fp16 = const()[name = tensor("op_1469_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_223_cast = mul(x = var_1468_cast, y = var_1469_to_fp16)[name = tensor("aw_chunk_223_cast")]; + tensor var_1472_equation_0 = const()[name = tensor("op_1472_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1472_cast = einsum(equation = var_1472_equation_0, values = (var_1322_cast, var_1294_cast))[name = tensor("op_1472_cast")]; + tensor var_1473_to_fp16 = const()[name = tensor("op_1473_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_225_cast = mul(x = var_1472_cast, y = var_1473_to_fp16)[name = tensor("aw_chunk_225_cast")]; + tensor var_1476_equation_0 = const()[name = tensor("op_1476_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1476_cast = einsum(equation = var_1476_equation_0, values = (var_1322_cast, var_1295_cast))[name = tensor("op_1476_cast")]; + tensor var_1477_to_fp16 = const()[name = tensor("op_1477_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_227_cast = mul(x = var_1476_cast, y = var_1477_to_fp16)[name = tensor("aw_chunk_227_cast")]; + tensor var_1480_equation_0 = const()[name = tensor("op_1480_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1480_cast = einsum(equation = var_1480_equation_0, values = (var_1322_cast, var_1296_cast))[name = tensor("op_1480_cast")]; + tensor var_1481_to_fp16 = const()[name = tensor("op_1481_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_229_cast = mul(x = var_1480_cast, y = var_1481_to_fp16)[name = tensor("aw_chunk_229_cast")]; + tensor var_1484_equation_0 = const()[name = tensor("op_1484_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1484_cast = einsum(equation = var_1484_equation_0, values = (var_1322_cast, var_1297_cast))[name = tensor("op_1484_cast")]; + tensor var_1485_to_fp16 = const()[name = tensor("op_1485_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_231_cast = mul(x = var_1484_cast, y = var_1485_to_fp16)[name = tensor("aw_chunk_231_cast")]; + tensor var_1488_equation_0 = const()[name = tensor("op_1488_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1488_cast = einsum(equation = var_1488_equation_0, values = (var_1322_cast, var_1298_cast))[name = tensor("op_1488_cast")]; + tensor var_1489_to_fp16 = const()[name = tensor("op_1489_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_233_cast = mul(x = var_1488_cast, y = var_1489_to_fp16)[name = tensor("aw_chunk_233_cast")]; + tensor var_1492_equation_0 = const()[name = tensor("op_1492_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1492_cast = einsum(equation = var_1492_equation_0, values = (var_1322_cast, var_1299_cast))[name = tensor("op_1492_cast")]; + tensor var_1493_to_fp16 = const()[name = tensor("op_1493_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_235_cast = mul(x = var_1492_cast, y = var_1493_to_fp16)[name = tensor("aw_chunk_235_cast")]; + tensor var_1496_equation_0 = const()[name = tensor("op_1496_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1496_cast = einsum(equation = var_1496_equation_0, values = (var_1322_cast, var_1300_cast))[name = tensor("op_1496_cast")]; + tensor var_1497_to_fp16 = const()[name = tensor("op_1497_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_237_cast = mul(x = var_1496_cast, y = var_1497_to_fp16)[name = tensor("aw_chunk_237_cast")]; + tensor var_1500_equation_0 = const()[name = tensor("op_1500_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1500_cast = einsum(equation = var_1500_equation_0, values = (var_1322_cast, var_1301_cast))[name = tensor("op_1500_cast")]; + tensor var_1501_to_fp16 = const()[name = tensor("op_1501_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_239_cast = mul(x = var_1500_cast, y = var_1501_to_fp16)[name = tensor("aw_chunk_239_cast")]; + tensor var_1503_cast = softmax(axis = var_120, x = aw_chunk_161_cast)[name = tensor("op_1503_cast")]; + tensor var_1504_cast = softmax(axis = var_120, x = aw_chunk_163_cast)[name = tensor("op_1504_cast")]; + tensor var_1505_cast = softmax(axis = var_120, x = aw_chunk_165_cast)[name = tensor("op_1505_cast")]; + tensor var_1506_cast = softmax(axis = var_120, x = aw_chunk_167_cast)[name = tensor("op_1506_cast")]; + tensor var_1507_cast = softmax(axis = var_120, x = aw_chunk_169_cast)[name = tensor("op_1507_cast")]; + tensor var_1508_cast = softmax(axis = var_120, x = aw_chunk_171_cast)[name = tensor("op_1508_cast")]; + tensor var_1509_cast = softmax(axis = var_120, x = aw_chunk_173_cast)[name = tensor("op_1509_cast")]; + tensor var_1510_cast = softmax(axis = var_120, x = aw_chunk_175_cast)[name = tensor("op_1510_cast")]; + tensor var_1511_cast = softmax(axis = var_120, x = aw_chunk_177_cast)[name = tensor("op_1511_cast")]; + tensor var_1512_cast = softmax(axis = var_120, x = aw_chunk_179_cast)[name = tensor("op_1512_cast")]; + tensor var_1513_cast = softmax(axis = var_120, x = aw_chunk_181_cast)[name = tensor("op_1513_cast")]; + tensor var_1514_cast = softmax(axis = var_120, x = aw_chunk_183_cast)[name = tensor("op_1514_cast")]; + tensor var_1515_cast = softmax(axis = var_120, x = aw_chunk_185_cast)[name = tensor("op_1515_cast")]; + tensor var_1516_cast = softmax(axis = var_120, x = aw_chunk_187_cast)[name = tensor("op_1516_cast")]; + tensor var_1517_cast = softmax(axis = var_120, x = aw_chunk_189_cast)[name = tensor("op_1517_cast")]; + tensor var_1518_cast = softmax(axis = var_120, x = aw_chunk_191_cast)[name = tensor("op_1518_cast")]; + tensor var_1519_cast = softmax(axis = var_120, x = aw_chunk_193_cast)[name = tensor("op_1519_cast")]; + tensor var_1520_cast = softmax(axis = var_120, x = aw_chunk_195_cast)[name = tensor("op_1520_cast")]; + tensor var_1521_cast = softmax(axis = var_120, x = aw_chunk_197_cast)[name = tensor("op_1521_cast")]; + tensor var_1522_cast = softmax(axis = var_120, x = aw_chunk_199_cast)[name = tensor("op_1522_cast")]; + tensor var_1523_cast = softmax(axis = var_120, x = aw_chunk_201_cast)[name = tensor("op_1523_cast")]; + tensor var_1524_cast = softmax(axis = var_120, x = aw_chunk_203_cast)[name = tensor("op_1524_cast")]; + tensor var_1525_cast = softmax(axis = var_120, x = aw_chunk_205_cast)[name = tensor("op_1525_cast")]; + tensor var_1526_cast = softmax(axis = var_120, x = aw_chunk_207_cast)[name = tensor("op_1526_cast")]; + tensor var_1527_cast = softmax(axis = var_120, x = aw_chunk_209_cast)[name = tensor("op_1527_cast")]; + tensor var_1528_cast = softmax(axis = var_120, x = aw_chunk_211_cast)[name = tensor("op_1528_cast")]; + tensor var_1529_cast = softmax(axis = var_120, x = aw_chunk_213_cast)[name = tensor("op_1529_cast")]; + tensor var_1530_cast = softmax(axis = var_120, x = aw_chunk_215_cast)[name = tensor("op_1530_cast")]; + tensor var_1531_cast = softmax(axis = var_120, x = aw_chunk_217_cast)[name = tensor("op_1531_cast")]; + tensor var_1532_cast = softmax(axis = var_120, x = aw_chunk_219_cast)[name = tensor("op_1532_cast")]; + tensor var_1533_cast = softmax(axis = var_120, x = aw_chunk_221_cast)[name = tensor("op_1533_cast")]; + tensor var_1534_cast = softmax(axis = var_120, x = aw_chunk_223_cast)[name = tensor("op_1534_cast")]; + tensor var_1535_cast = softmax(axis = var_120, x = aw_chunk_225_cast)[name = tensor("op_1535_cast")]; + tensor var_1536_cast = softmax(axis = var_120, x = aw_chunk_227_cast)[name = tensor("op_1536_cast")]; + tensor var_1537_cast = softmax(axis = var_120, x = aw_chunk_229_cast)[name = tensor("op_1537_cast")]; + tensor var_1538_cast = softmax(axis = var_120, x = aw_chunk_231_cast)[name = tensor("op_1538_cast")]; + tensor var_1539_cast = softmax(axis = var_120, x = aw_chunk_233_cast)[name = tensor("op_1539_cast")]; + tensor var_1540_cast = softmax(axis = var_120, x = aw_chunk_235_cast)[name = tensor("op_1540_cast")]; + tensor var_1541_cast = softmax(axis = var_120, x = aw_chunk_237_cast)[name = tensor("op_1541_cast")]; + tensor var_1542_cast = softmax(axis = var_120, x = aw_chunk_239_cast)[name = tensor("op_1542_cast")]; + tensor var_1544_equation_0 = const()[name = tensor("op_1544_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1544_cast = einsum(equation = var_1544_equation_0, values = (var_1324_cast, var_1503_cast))[name = tensor("op_1544_cast")]; + tensor var_1546_equation_0 = const()[name = tensor("op_1546_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1546_cast = einsum(equation = var_1546_equation_0, values = (var_1324_cast, var_1504_cast))[name = tensor("op_1546_cast")]; + tensor var_1548_equation_0 = const()[name = tensor("op_1548_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1548_cast = einsum(equation = var_1548_equation_0, values = (var_1324_cast, var_1505_cast))[name = tensor("op_1548_cast")]; + tensor var_1550_equation_0 = const()[name = tensor("op_1550_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1550_cast = einsum(equation = var_1550_equation_0, values = (var_1324_cast, var_1506_cast))[name = tensor("op_1550_cast")]; + tensor var_1552_equation_0 = const()[name = tensor("op_1552_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1552_cast = einsum(equation = var_1552_equation_0, values = (var_1324_cast, var_1507_cast))[name = tensor("op_1552_cast")]; + tensor var_1554_equation_0 = const()[name = tensor("op_1554_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1554_cast = einsum(equation = var_1554_equation_0, values = (var_1324_cast, var_1508_cast))[name = tensor("op_1554_cast")]; + tensor var_1556_equation_0 = const()[name = tensor("op_1556_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1556_cast = einsum(equation = var_1556_equation_0, values = (var_1324_cast, var_1509_cast))[name = tensor("op_1556_cast")]; + tensor var_1558_equation_0 = const()[name = tensor("op_1558_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1558_cast = einsum(equation = var_1558_equation_0, values = (var_1324_cast, var_1510_cast))[name = tensor("op_1558_cast")]; + tensor var_1560_equation_0 = const()[name = tensor("op_1560_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1560_cast = einsum(equation = var_1560_equation_0, values = (var_1328_cast, var_1511_cast))[name = tensor("op_1560_cast")]; + tensor var_1562_equation_0 = const()[name = tensor("op_1562_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1562_cast = einsum(equation = var_1562_equation_0, values = (var_1328_cast, var_1512_cast))[name = tensor("op_1562_cast")]; + tensor var_1564_equation_0 = const()[name = tensor("op_1564_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1564_cast = einsum(equation = var_1564_equation_0, values = (var_1328_cast, var_1513_cast))[name = tensor("op_1564_cast")]; + tensor var_1566_equation_0 = const()[name = tensor("op_1566_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1566_cast = einsum(equation = var_1566_equation_0, values = (var_1328_cast, var_1514_cast))[name = tensor("op_1566_cast")]; + tensor var_1568_equation_0 = const()[name = tensor("op_1568_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1568_cast = einsum(equation = var_1568_equation_0, values = (var_1328_cast, var_1515_cast))[name = tensor("op_1568_cast")]; + tensor var_1570_equation_0 = const()[name = tensor("op_1570_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1570_cast = einsum(equation = var_1570_equation_0, values = (var_1328_cast, var_1516_cast))[name = tensor("op_1570_cast")]; + tensor var_1572_equation_0 = const()[name = tensor("op_1572_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1572_cast = einsum(equation = var_1572_equation_0, values = (var_1328_cast, var_1517_cast))[name = tensor("op_1572_cast")]; + tensor var_1574_equation_0 = const()[name = tensor("op_1574_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1574_cast = einsum(equation = var_1574_equation_0, values = (var_1328_cast, var_1518_cast))[name = tensor("op_1574_cast")]; + tensor var_1576_equation_0 = const()[name = tensor("op_1576_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1576_cast = einsum(equation = var_1576_equation_0, values = (var_1332_cast, var_1519_cast))[name = tensor("op_1576_cast")]; + tensor var_1578_equation_0 = const()[name = tensor("op_1578_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1578_cast = einsum(equation = var_1578_equation_0, values = (var_1332_cast, var_1520_cast))[name = tensor("op_1578_cast")]; + tensor var_1580_equation_0 = const()[name = tensor("op_1580_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1580_cast = einsum(equation = var_1580_equation_0, values = (var_1332_cast, var_1521_cast))[name = tensor("op_1580_cast")]; + tensor var_1582_equation_0 = const()[name = tensor("op_1582_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1582_cast = einsum(equation = var_1582_equation_0, values = (var_1332_cast, var_1522_cast))[name = tensor("op_1582_cast")]; + tensor var_1584_equation_0 = const()[name = tensor("op_1584_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1584_cast = einsum(equation = var_1584_equation_0, values = (var_1332_cast, var_1523_cast))[name = tensor("op_1584_cast")]; + tensor var_1586_equation_0 = const()[name = tensor("op_1586_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1586_cast = einsum(equation = var_1586_equation_0, values = (var_1332_cast, var_1524_cast))[name = tensor("op_1586_cast")]; + tensor var_1588_equation_0 = const()[name = tensor("op_1588_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1588_cast = einsum(equation = var_1588_equation_0, values = (var_1332_cast, var_1525_cast))[name = tensor("op_1588_cast")]; + tensor var_1590_equation_0 = const()[name = tensor("op_1590_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1590_cast = einsum(equation = var_1590_equation_0, values = (var_1332_cast, var_1526_cast))[name = tensor("op_1590_cast")]; + tensor var_1592_equation_0 = const()[name = tensor("op_1592_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1592_cast = einsum(equation = var_1592_equation_0, values = (var_1336_cast, var_1527_cast))[name = tensor("op_1592_cast")]; + tensor var_1594_equation_0 = const()[name = tensor("op_1594_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1594_cast = einsum(equation = var_1594_equation_0, values = (var_1336_cast, var_1528_cast))[name = tensor("op_1594_cast")]; + tensor var_1596_equation_0 = const()[name = tensor("op_1596_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1596_cast = einsum(equation = var_1596_equation_0, values = (var_1336_cast, var_1529_cast))[name = tensor("op_1596_cast")]; + tensor var_1598_equation_0 = const()[name = tensor("op_1598_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1598_cast = einsum(equation = var_1598_equation_0, values = (var_1336_cast, var_1530_cast))[name = tensor("op_1598_cast")]; + tensor var_1600_equation_0 = const()[name = tensor("op_1600_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1600_cast = einsum(equation = var_1600_equation_0, values = (var_1336_cast, var_1531_cast))[name = tensor("op_1600_cast")]; + tensor var_1602_equation_0 = const()[name = tensor("op_1602_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1602_cast = einsum(equation = var_1602_equation_0, values = (var_1336_cast, var_1532_cast))[name = tensor("op_1602_cast")]; + tensor var_1604_equation_0 = const()[name = tensor("op_1604_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1604_cast = einsum(equation = var_1604_equation_0, values = (var_1336_cast, var_1533_cast))[name = tensor("op_1604_cast")]; + tensor var_1606_equation_0 = const()[name = tensor("op_1606_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1606_cast = einsum(equation = var_1606_equation_0, values = (var_1336_cast, var_1534_cast))[name = tensor("op_1606_cast")]; + tensor var_1608_equation_0 = const()[name = tensor("op_1608_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1608_cast = einsum(equation = var_1608_equation_0, values = (var_1340_cast, var_1535_cast))[name = tensor("op_1608_cast")]; + tensor var_1610_equation_0 = const()[name = tensor("op_1610_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1610_cast = einsum(equation = var_1610_equation_0, values = (var_1340_cast, var_1536_cast))[name = tensor("op_1610_cast")]; + tensor var_1612_equation_0 = const()[name = tensor("op_1612_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1612_cast = einsum(equation = var_1612_equation_0, values = (var_1340_cast, var_1537_cast))[name = tensor("op_1612_cast")]; + tensor var_1614_equation_0 = const()[name = tensor("op_1614_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1614_cast = einsum(equation = var_1614_equation_0, values = (var_1340_cast, var_1538_cast))[name = tensor("op_1614_cast")]; + tensor var_1616_equation_0 = const()[name = tensor("op_1616_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1616_cast = einsum(equation = var_1616_equation_0, values = (var_1340_cast, var_1539_cast))[name = tensor("op_1616_cast")]; + tensor var_1618_equation_0 = const()[name = tensor("op_1618_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1618_cast = einsum(equation = var_1618_equation_0, values = (var_1340_cast, var_1540_cast))[name = tensor("op_1618_cast")]; + tensor var_1620_equation_0 = const()[name = tensor("op_1620_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1620_cast = einsum(equation = var_1620_equation_0, values = (var_1340_cast, var_1541_cast))[name = tensor("op_1620_cast")]; + tensor var_1622_equation_0 = const()[name = tensor("op_1622_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1622_cast = einsum(equation = var_1622_equation_0, values = (var_1340_cast, var_1542_cast))[name = tensor("op_1622_cast")]; + tensor var_1624_interleave_0 = const()[name = tensor("op_1624_interleave_0"), val = tensor(false)]; + tensor var_1624_cast = concat(axis = var_95, interleave = var_1624_interleave_0, values = (var_1544_cast, var_1546_cast, var_1548_cast, var_1550_cast, var_1552_cast, var_1554_cast, var_1556_cast, var_1558_cast))[name = tensor("op_1624_cast")]; + tensor var_1626_interleave_0 = const()[name = tensor("op_1626_interleave_0"), val = tensor(false)]; + tensor var_1626_cast = concat(axis = var_95, interleave = var_1626_interleave_0, values = (var_1560_cast, var_1562_cast, var_1564_cast, var_1566_cast, var_1568_cast, var_1570_cast, var_1572_cast, var_1574_cast))[name = tensor("op_1626_cast")]; + tensor var_1628_interleave_0 = const()[name = tensor("op_1628_interleave_0"), val = tensor(false)]; + tensor var_1628_cast = concat(axis = var_95, interleave = var_1628_interleave_0, values = (var_1576_cast, var_1578_cast, var_1580_cast, var_1582_cast, var_1584_cast, var_1586_cast, var_1588_cast, var_1590_cast))[name = tensor("op_1628_cast")]; + tensor var_1630_interleave_0 = const()[name = tensor("op_1630_interleave_0"), val = tensor(false)]; + tensor var_1630_cast = concat(axis = var_95, interleave = var_1630_interleave_0, values = (var_1592_cast, var_1594_cast, var_1596_cast, var_1598_cast, var_1600_cast, var_1602_cast, var_1604_cast, var_1606_cast))[name = tensor("op_1630_cast")]; + tensor var_1632_interleave_0 = const()[name = tensor("op_1632_interleave_0"), val = tensor(false)]; + tensor var_1632_cast = concat(axis = var_95, interleave = var_1632_interleave_0, values = (var_1608_cast, var_1610_cast, var_1612_cast, var_1614_cast, var_1616_cast, var_1618_cast, var_1620_cast, var_1622_cast))[name = tensor("op_1632_cast")]; + tensor input_51_interleave_0 = const()[name = tensor("input_51_interleave_0"), val = tensor(false)]; + tensor input_51_cast = concat(axis = var_120, interleave = input_51_interleave_0, values = (var_1624_cast, var_1626_cast, var_1628_cast, var_1630_cast, var_1632_cast))[name = tensor("input_51_cast")]; + tensor var_1638 = const()[name = tensor("op_1638"), val = tensor([1, 1])]; + tensor var_1640 = const()[name = tensor("op_1640"), val = tensor([1, 1])]; + tensor var_1642_pad_type_0 = const()[name = tensor("op_1642_pad_type_0"), val = tensor("custom")]; + tensor var_1642_pad_0 = const()[name = tensor("op_1642_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_0_attentions_1_transformer_blocks_0_attn1_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(7297472))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(7374336))), name = tensor("down_blocks_0_attentions_1_transformer_blocks_0_attn1_to_out_0_weight_to_fp16_palettized"), shape = tensor([320, 320, 1, 1])]; + tensor down_blocks_0_attentions_1_transformer_blocks_0_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("down_blocks_0_attentions_1_transformer_blocks_0_attn1_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(7374528)))]; + tensor var_1642_cast = conv(bias = down_blocks_0_attentions_1_transformer_blocks_0_attn1_to_out_0_bias_to_fp16, dilations = var_1640, groups = var_120, pad = var_1642_pad_0, pad_type = var_1642_pad_type_0, strides = var_1638, weight = down_blocks_0_attentions_1_transformer_blocks_0_attn1_to_out_0_weight_to_fp16_palettized, x = input_51_cast)[name = tensor("op_1642_cast")]; + tensor inputs_9_cast = add(x = var_1642_cast, y = inputs_7_cast)[name = tensor("inputs_9_cast")]; + tensor var_1646 = const()[name = tensor("op_1646"), val = tensor([1])]; + tensor channels_mean_9_cast = reduce_mean(axes = var_1646, keep_dims = var_115, x = inputs_9_cast)[name = tensor("channels_mean_9_cast")]; + tensor zero_mean_9_cast = sub(x = inputs_9_cast, y = channels_mean_9_cast)[name = tensor("zero_mean_9_cast")]; + tensor zero_mean_sq_9_cast = mul(x = zero_mean_9_cast, y = zero_mean_9_cast)[name = tensor("zero_mean_sq_9_cast")]; + tensor var_1650 = const()[name = tensor("op_1650"), val = tensor([1])]; + tensor var_1651_cast = reduce_mean(axes = var_1650, keep_dims = var_115, x = zero_mean_sq_9_cast)[name = tensor("op_1651_cast")]; + tensor var_1652_to_fp16 = const()[name = tensor("op_1652_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_1653_cast = add(x = var_1651_cast, y = var_1652_to_fp16)[name = tensor("op_1653_cast")]; + tensor denom_9_epsilon_0_to_fp16 = const()[name = tensor("denom_9_epsilon_0_to_fp16"), val = tensor(0x1p-24)]; + tensor denom_9_cast = rsqrt(epsilon = denom_9_epsilon_0_to_fp16, x = var_1653_cast)[name = tensor("denom_9_cast")]; + tensor out_9_cast = mul(x = zero_mean_9_cast, y = denom_9_cast)[name = tensor("out_9_cast")]; + tensor var_1657_to_fp16 = const()[name = tensor("op_1657_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(7375232)))]; + tensor var_1658_cast = add(x = out_9_cast, y = var_1657_to_fp16)[name = tensor("op_1658_cast")]; + tensor var_1660_to_fp16 = const()[name = tensor("op_1660_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(7375936)))]; + tensor hidden_states_29_cast = mul(x = var_1658_cast, y = var_1660_to_fp16)[name = tensor("hidden_states_29_cast")]; + tensor var_1667 = const()[name = tensor("op_1667"), val = tensor([1, 1])]; + tensor var_1669 = const()[name = tensor("op_1669"), val = tensor([1, 1])]; + tensor q_7_pad_type_0 = const()[name = tensor("q_7_pad_type_0"), val = tensor("custom")]; + tensor q_7_pad_0 = const()[name = tensor("q_7_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_0_attentions_1_transformer_blocks_0_attn2_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(7376640))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(7453504))), name = tensor("down_blocks_0_attentions_1_transformer_blocks_0_attn2_to_q_weight_to_fp16_palettized"), shape = tensor([320, 320, 1, 1])]; + tensor q_7_cast = conv(dilations = var_1669, groups = var_120, pad = q_7_pad_0, pad_type = q_7_pad_type_0, strides = var_1667, weight = down_blocks_0_attentions_1_transformer_blocks_0_attn2_to_q_weight_to_fp16_palettized, x = hidden_states_29_cast)[name = tensor("q_7_cast")]; + tensor var_1673 = const()[name = tensor("op_1673"), val = tensor([1, 1])]; + tensor var_1675 = const()[name = tensor("op_1675"), val = tensor([1, 1])]; + tensor k_13_pad_type_0 = const()[name = tensor("k_13_pad_type_0"), val = tensor("custom")]; + tensor k_13_pad_0 = const()[name = tensor("k_13_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_0_attentions_1_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(7453696))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(7699520))), name = tensor("down_blocks_0_attentions_1_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized"), shape = tensor([320, 1024, 1, 1])]; + tensor k_13_cast = conv(dilations = var_1675, groups = var_120, pad = k_13_pad_0, pad_type = k_13_pad_type_0, strides = var_1673, weight = down_blocks_0_attentions_1_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor("k_13_cast")]; + tensor var_1679 = const()[name = tensor("op_1679"), val = tensor([1, 1])]; + tensor var_1681 = const()[name = tensor("op_1681"), val = tensor([1, 1])]; + tensor v_7_pad_type_0 = const()[name = tensor("v_7_pad_type_0"), val = tensor("custom")]; + tensor v_7_pad_0 = const()[name = tensor("v_7_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_0_attentions_1_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(7699712))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(7945536))), name = tensor("down_blocks_0_attentions_1_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized"), shape = tensor([320, 1024, 1, 1])]; + tensor v_7_cast = conv(dilations = var_1681, groups = var_120, pad = v_7_pad_0, pad_type = v_7_pad_type_0, strides = var_1679, weight = down_blocks_0_attentions_1_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor("v_7_cast")]; + tensor var_1685_begin_0 = const()[name = tensor("op_1685_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1685_end_0 = const()[name = tensor("op_1685_end_0"), val = tensor([2, 64, 1, 4096])]; + tensor var_1685_end_mask_0 = const()[name = tensor("op_1685_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1685_cast = slice_by_index(begin = var_1685_begin_0, end = var_1685_end_0, end_mask = var_1685_end_mask_0, x = q_7_cast)[name = tensor("op_1685_cast")]; + tensor var_1689_begin_0 = const()[name = tensor("op_1689_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_1689_end_0 = const()[name = tensor("op_1689_end_0"), val = tensor([2, 128, 1, 4096])]; + tensor var_1689_end_mask_0 = const()[name = tensor("op_1689_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1689_cast = slice_by_index(begin = var_1689_begin_0, end = var_1689_end_0, end_mask = var_1689_end_mask_0, x = q_7_cast)[name = tensor("op_1689_cast")]; + tensor var_1693_begin_0 = const()[name = tensor("op_1693_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_1693_end_0 = const()[name = tensor("op_1693_end_0"), val = tensor([2, 192, 1, 4096])]; + tensor var_1693_end_mask_0 = const()[name = tensor("op_1693_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1693_cast = slice_by_index(begin = var_1693_begin_0, end = var_1693_end_0, end_mask = var_1693_end_mask_0, x = q_7_cast)[name = tensor("op_1693_cast")]; + tensor var_1697_begin_0 = const()[name = tensor("op_1697_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_1697_end_0 = const()[name = tensor("op_1697_end_0"), val = tensor([2, 256, 1, 4096])]; + tensor var_1697_end_mask_0 = const()[name = tensor("op_1697_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1697_cast = slice_by_index(begin = var_1697_begin_0, end = var_1697_end_0, end_mask = var_1697_end_mask_0, x = q_7_cast)[name = tensor("op_1697_cast")]; + tensor var_1701_begin_0 = const()[name = tensor("op_1701_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_1701_end_0 = const()[name = tensor("op_1701_end_0"), val = tensor([2, 320, 1, 4096])]; + tensor var_1701_end_mask_0 = const()[name = tensor("op_1701_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1701_cast = slice_by_index(begin = var_1701_begin_0, end = var_1701_end_0, end_mask = var_1701_end_mask_0, x = q_7_cast)[name = tensor("op_1701_cast")]; + tensor var_1704_begin_0 = const()[name = tensor("op_1704_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1704_end_0 = const()[name = tensor("op_1704_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_1704_end_mask_0 = const()[name = tensor("op_1704_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1704_cast = slice_by_index(begin = var_1704_begin_0, end = var_1704_end_0, end_mask = var_1704_end_mask_0, x = var_1685_cast)[name = tensor("op_1704_cast")]; + tensor var_1705_begin_0 = const()[name = tensor("op_1705_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_1705_end_0 = const()[name = tensor("op_1705_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_1705_end_mask_0 = const()[name = tensor("op_1705_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1705_cast = slice_by_index(begin = var_1705_begin_0, end = var_1705_end_0, end_mask = var_1705_end_mask_0, x = var_1685_cast)[name = tensor("op_1705_cast")]; + tensor var_1706_begin_0 = const()[name = tensor("op_1706_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_1706_end_0 = const()[name = tensor("op_1706_end_0"), val = tensor([2, 64, 1, 1536])]; + tensor var_1706_end_mask_0 = const()[name = tensor("op_1706_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1706_cast = slice_by_index(begin = var_1706_begin_0, end = var_1706_end_0, end_mask = var_1706_end_mask_0, x = var_1685_cast)[name = tensor("op_1706_cast")]; + tensor var_1707_begin_0 = const()[name = tensor("op_1707_begin_0"), val = tensor([0, 0, 0, 1536])]; + tensor var_1707_end_0 = const()[name = tensor("op_1707_end_0"), val = tensor([2, 64, 1, 2048])]; + tensor var_1707_end_mask_0 = const()[name = tensor("op_1707_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1707_cast = slice_by_index(begin = var_1707_begin_0, end = var_1707_end_0, end_mask = var_1707_end_mask_0, x = var_1685_cast)[name = tensor("op_1707_cast")]; + tensor var_1708_begin_0 = const()[name = tensor("op_1708_begin_0"), val = tensor([0, 0, 0, 2048])]; + tensor var_1708_end_0 = const()[name = tensor("op_1708_end_0"), val = tensor([2, 64, 1, 2560])]; + tensor var_1708_end_mask_0 = const()[name = tensor("op_1708_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1708_cast = slice_by_index(begin = var_1708_begin_0, end = var_1708_end_0, end_mask = var_1708_end_mask_0, x = var_1685_cast)[name = tensor("op_1708_cast")]; + tensor var_1709_begin_0 = const()[name = tensor("op_1709_begin_0"), val = tensor([0, 0, 0, 2560])]; + tensor var_1709_end_0 = const()[name = tensor("op_1709_end_0"), val = tensor([2, 64, 1, 3072])]; + tensor var_1709_end_mask_0 = const()[name = tensor("op_1709_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1709_cast = slice_by_index(begin = var_1709_begin_0, end = var_1709_end_0, end_mask = var_1709_end_mask_0, x = var_1685_cast)[name = tensor("op_1709_cast")]; + tensor var_1710_begin_0 = const()[name = tensor("op_1710_begin_0"), val = tensor([0, 0, 0, 3072])]; + tensor var_1710_end_0 = const()[name = tensor("op_1710_end_0"), val = tensor([2, 64, 1, 3584])]; + tensor var_1710_end_mask_0 = const()[name = tensor("op_1710_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1710_cast = slice_by_index(begin = var_1710_begin_0, end = var_1710_end_0, end_mask = var_1710_end_mask_0, x = var_1685_cast)[name = tensor("op_1710_cast")]; + tensor var_1711_begin_0 = const()[name = tensor("op_1711_begin_0"), val = tensor([0, 0, 0, 3584])]; + tensor var_1711_end_0 = const()[name = tensor("op_1711_end_0"), val = tensor([2, 64, 1, 4096])]; + tensor var_1711_end_mask_0 = const()[name = tensor("op_1711_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1711_cast = slice_by_index(begin = var_1711_begin_0, end = var_1711_end_0, end_mask = var_1711_end_mask_0, x = var_1685_cast)[name = tensor("op_1711_cast")]; + tensor var_1712_begin_0 = const()[name = tensor("op_1712_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1712_end_0 = const()[name = tensor("op_1712_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_1712_end_mask_0 = const()[name = tensor("op_1712_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1712_cast = slice_by_index(begin = var_1712_begin_0, end = var_1712_end_0, end_mask = var_1712_end_mask_0, x = var_1689_cast)[name = tensor("op_1712_cast")]; + tensor var_1713_begin_0 = const()[name = tensor("op_1713_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_1713_end_0 = const()[name = tensor("op_1713_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_1713_end_mask_0 = const()[name = tensor("op_1713_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1713_cast = slice_by_index(begin = var_1713_begin_0, end = var_1713_end_0, end_mask = var_1713_end_mask_0, x = var_1689_cast)[name = tensor("op_1713_cast")]; + tensor var_1714_begin_0 = const()[name = tensor("op_1714_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_1714_end_0 = const()[name = tensor("op_1714_end_0"), val = tensor([2, 64, 1, 1536])]; + tensor var_1714_end_mask_0 = const()[name = tensor("op_1714_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1714_cast = slice_by_index(begin = var_1714_begin_0, end = var_1714_end_0, end_mask = var_1714_end_mask_0, x = var_1689_cast)[name = tensor("op_1714_cast")]; + tensor var_1715_begin_0 = const()[name = tensor("op_1715_begin_0"), val = tensor([0, 0, 0, 1536])]; + tensor var_1715_end_0 = const()[name = tensor("op_1715_end_0"), val = tensor([2, 64, 1, 2048])]; + tensor var_1715_end_mask_0 = const()[name = tensor("op_1715_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1715_cast = slice_by_index(begin = var_1715_begin_0, end = var_1715_end_0, end_mask = var_1715_end_mask_0, x = var_1689_cast)[name = tensor("op_1715_cast")]; + tensor var_1716_begin_0 = const()[name = tensor("op_1716_begin_0"), val = tensor([0, 0, 0, 2048])]; + tensor var_1716_end_0 = const()[name = tensor("op_1716_end_0"), val = tensor([2, 64, 1, 2560])]; + tensor var_1716_end_mask_0 = const()[name = tensor("op_1716_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1716_cast = slice_by_index(begin = var_1716_begin_0, end = var_1716_end_0, end_mask = var_1716_end_mask_0, x = var_1689_cast)[name = tensor("op_1716_cast")]; + tensor var_1717_begin_0 = const()[name = tensor("op_1717_begin_0"), val = tensor([0, 0, 0, 2560])]; + tensor var_1717_end_0 = const()[name = tensor("op_1717_end_0"), val = tensor([2, 64, 1, 3072])]; + tensor var_1717_end_mask_0 = const()[name = tensor("op_1717_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1717_cast = slice_by_index(begin = var_1717_begin_0, end = var_1717_end_0, end_mask = var_1717_end_mask_0, x = var_1689_cast)[name = tensor("op_1717_cast")]; + tensor var_1718_begin_0 = const()[name = tensor("op_1718_begin_0"), val = tensor([0, 0, 0, 3072])]; + tensor var_1718_end_0 = const()[name = tensor("op_1718_end_0"), val = tensor([2, 64, 1, 3584])]; + tensor var_1718_end_mask_0 = const()[name = tensor("op_1718_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1718_cast = slice_by_index(begin = var_1718_begin_0, end = var_1718_end_0, end_mask = var_1718_end_mask_0, x = var_1689_cast)[name = tensor("op_1718_cast")]; + tensor var_1719_begin_0 = const()[name = tensor("op_1719_begin_0"), val = tensor([0, 0, 0, 3584])]; + tensor var_1719_end_0 = const()[name = tensor("op_1719_end_0"), val = tensor([2, 64, 1, 4096])]; + tensor var_1719_end_mask_0 = const()[name = tensor("op_1719_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1719_cast = slice_by_index(begin = var_1719_begin_0, end = var_1719_end_0, end_mask = var_1719_end_mask_0, x = var_1689_cast)[name = tensor("op_1719_cast")]; + tensor var_1720_begin_0 = const()[name = tensor("op_1720_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1720_end_0 = const()[name = tensor("op_1720_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_1720_end_mask_0 = const()[name = tensor("op_1720_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1720_cast = slice_by_index(begin = var_1720_begin_0, end = var_1720_end_0, end_mask = var_1720_end_mask_0, x = var_1693_cast)[name = tensor("op_1720_cast")]; + tensor var_1721_begin_0 = const()[name = tensor("op_1721_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_1721_end_0 = const()[name = tensor("op_1721_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_1721_end_mask_0 = const()[name = tensor("op_1721_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1721_cast = slice_by_index(begin = var_1721_begin_0, end = var_1721_end_0, end_mask = var_1721_end_mask_0, x = var_1693_cast)[name = tensor("op_1721_cast")]; + tensor var_1722_begin_0 = const()[name = tensor("op_1722_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_1722_end_0 = const()[name = tensor("op_1722_end_0"), val = tensor([2, 64, 1, 1536])]; + tensor var_1722_end_mask_0 = const()[name = tensor("op_1722_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1722_cast = slice_by_index(begin = var_1722_begin_0, end = var_1722_end_0, end_mask = var_1722_end_mask_0, x = var_1693_cast)[name = tensor("op_1722_cast")]; + tensor var_1723_begin_0 = const()[name = tensor("op_1723_begin_0"), val = tensor([0, 0, 0, 1536])]; + tensor var_1723_end_0 = const()[name = tensor("op_1723_end_0"), val = tensor([2, 64, 1, 2048])]; + tensor var_1723_end_mask_0 = const()[name = tensor("op_1723_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1723_cast = slice_by_index(begin = var_1723_begin_0, end = var_1723_end_0, end_mask = var_1723_end_mask_0, x = var_1693_cast)[name = tensor("op_1723_cast")]; + tensor var_1724_begin_0 = const()[name = tensor("op_1724_begin_0"), val = tensor([0, 0, 0, 2048])]; + tensor var_1724_end_0 = const()[name = tensor("op_1724_end_0"), val = tensor([2, 64, 1, 2560])]; + tensor var_1724_end_mask_0 = const()[name = tensor("op_1724_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1724_cast = slice_by_index(begin = var_1724_begin_0, end = var_1724_end_0, end_mask = var_1724_end_mask_0, x = var_1693_cast)[name = tensor("op_1724_cast")]; + tensor var_1725_begin_0 = const()[name = tensor("op_1725_begin_0"), val = tensor([0, 0, 0, 2560])]; + tensor var_1725_end_0 = const()[name = tensor("op_1725_end_0"), val = tensor([2, 64, 1, 3072])]; + tensor var_1725_end_mask_0 = const()[name = tensor("op_1725_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1725_cast = slice_by_index(begin = var_1725_begin_0, end = var_1725_end_0, end_mask = var_1725_end_mask_0, x = var_1693_cast)[name = tensor("op_1725_cast")]; + tensor var_1726_begin_0 = const()[name = tensor("op_1726_begin_0"), val = tensor([0, 0, 0, 3072])]; + tensor var_1726_end_0 = const()[name = tensor("op_1726_end_0"), val = tensor([2, 64, 1, 3584])]; + tensor var_1726_end_mask_0 = const()[name = tensor("op_1726_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1726_cast = slice_by_index(begin = var_1726_begin_0, end = var_1726_end_0, end_mask = var_1726_end_mask_0, x = var_1693_cast)[name = tensor("op_1726_cast")]; + tensor var_1727_begin_0 = const()[name = tensor("op_1727_begin_0"), val = tensor([0, 0, 0, 3584])]; + tensor var_1727_end_0 = const()[name = tensor("op_1727_end_0"), val = tensor([2, 64, 1, 4096])]; + tensor var_1727_end_mask_0 = const()[name = tensor("op_1727_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1727_cast = slice_by_index(begin = var_1727_begin_0, end = var_1727_end_0, end_mask = var_1727_end_mask_0, x = var_1693_cast)[name = tensor("op_1727_cast")]; + tensor var_1728_begin_0 = const()[name = tensor("op_1728_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1728_end_0 = const()[name = tensor("op_1728_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_1728_end_mask_0 = const()[name = tensor("op_1728_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1728_cast = slice_by_index(begin = var_1728_begin_0, end = var_1728_end_0, end_mask = var_1728_end_mask_0, x = var_1697_cast)[name = tensor("op_1728_cast")]; + tensor var_1729_begin_0 = const()[name = tensor("op_1729_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_1729_end_0 = const()[name = tensor("op_1729_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_1729_end_mask_0 = const()[name = tensor("op_1729_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1729_cast = slice_by_index(begin = var_1729_begin_0, end = var_1729_end_0, end_mask = var_1729_end_mask_0, x = var_1697_cast)[name = tensor("op_1729_cast")]; + tensor var_1730_begin_0 = const()[name = tensor("op_1730_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_1730_end_0 = const()[name = tensor("op_1730_end_0"), val = tensor([2, 64, 1, 1536])]; + tensor var_1730_end_mask_0 = const()[name = tensor("op_1730_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1730_cast = slice_by_index(begin = var_1730_begin_0, end = var_1730_end_0, end_mask = var_1730_end_mask_0, x = var_1697_cast)[name = tensor("op_1730_cast")]; + tensor var_1731_begin_0 = const()[name = tensor("op_1731_begin_0"), val = tensor([0, 0, 0, 1536])]; + tensor var_1731_end_0 = const()[name = tensor("op_1731_end_0"), val = tensor([2, 64, 1, 2048])]; + tensor var_1731_end_mask_0 = const()[name = tensor("op_1731_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1731_cast = slice_by_index(begin = var_1731_begin_0, end = var_1731_end_0, end_mask = var_1731_end_mask_0, x = var_1697_cast)[name = tensor("op_1731_cast")]; + tensor var_1732_begin_0 = const()[name = tensor("op_1732_begin_0"), val = tensor([0, 0, 0, 2048])]; + tensor var_1732_end_0 = const()[name = tensor("op_1732_end_0"), val = tensor([2, 64, 1, 2560])]; + tensor var_1732_end_mask_0 = const()[name = tensor("op_1732_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1732_cast = slice_by_index(begin = var_1732_begin_0, end = var_1732_end_0, end_mask = var_1732_end_mask_0, x = var_1697_cast)[name = tensor("op_1732_cast")]; + tensor var_1733_begin_0 = const()[name = tensor("op_1733_begin_0"), val = tensor([0, 0, 0, 2560])]; + tensor var_1733_end_0 = const()[name = tensor("op_1733_end_0"), val = tensor([2, 64, 1, 3072])]; + tensor var_1733_end_mask_0 = const()[name = tensor("op_1733_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1733_cast = slice_by_index(begin = var_1733_begin_0, end = var_1733_end_0, end_mask = var_1733_end_mask_0, x = var_1697_cast)[name = tensor("op_1733_cast")]; + tensor var_1734_begin_0 = const()[name = tensor("op_1734_begin_0"), val = tensor([0, 0, 0, 3072])]; + tensor var_1734_end_0 = const()[name = tensor("op_1734_end_0"), val = tensor([2, 64, 1, 3584])]; + tensor var_1734_end_mask_0 = const()[name = tensor("op_1734_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1734_cast = slice_by_index(begin = var_1734_begin_0, end = var_1734_end_0, end_mask = var_1734_end_mask_0, x = var_1697_cast)[name = tensor("op_1734_cast")]; + tensor var_1735_begin_0 = const()[name = tensor("op_1735_begin_0"), val = tensor([0, 0, 0, 3584])]; + tensor var_1735_end_0 = const()[name = tensor("op_1735_end_0"), val = tensor([2, 64, 1, 4096])]; + tensor var_1735_end_mask_0 = const()[name = tensor("op_1735_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1735_cast = slice_by_index(begin = var_1735_begin_0, end = var_1735_end_0, end_mask = var_1735_end_mask_0, x = var_1697_cast)[name = tensor("op_1735_cast")]; + tensor var_1736_begin_0 = const()[name = tensor("op_1736_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1736_end_0 = const()[name = tensor("op_1736_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_1736_end_mask_0 = const()[name = tensor("op_1736_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1736_cast = slice_by_index(begin = var_1736_begin_0, end = var_1736_end_0, end_mask = var_1736_end_mask_0, x = var_1701_cast)[name = tensor("op_1736_cast")]; + tensor var_1737_begin_0 = const()[name = tensor("op_1737_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_1737_end_0 = const()[name = tensor("op_1737_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_1737_end_mask_0 = const()[name = tensor("op_1737_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1737_cast = slice_by_index(begin = var_1737_begin_0, end = var_1737_end_0, end_mask = var_1737_end_mask_0, x = var_1701_cast)[name = tensor("op_1737_cast")]; + tensor var_1738_begin_0 = const()[name = tensor("op_1738_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_1738_end_0 = const()[name = tensor("op_1738_end_0"), val = tensor([2, 64, 1, 1536])]; + tensor var_1738_end_mask_0 = const()[name = tensor("op_1738_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1738_cast = slice_by_index(begin = var_1738_begin_0, end = var_1738_end_0, end_mask = var_1738_end_mask_0, x = var_1701_cast)[name = tensor("op_1738_cast")]; + tensor var_1739_begin_0 = const()[name = tensor("op_1739_begin_0"), val = tensor([0, 0, 0, 1536])]; + tensor var_1739_end_0 = const()[name = tensor("op_1739_end_0"), val = tensor([2, 64, 1, 2048])]; + tensor var_1739_end_mask_0 = const()[name = tensor("op_1739_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1739_cast = slice_by_index(begin = var_1739_begin_0, end = var_1739_end_0, end_mask = var_1739_end_mask_0, x = var_1701_cast)[name = tensor("op_1739_cast")]; + tensor var_1740_begin_0 = const()[name = tensor("op_1740_begin_0"), val = tensor([0, 0, 0, 2048])]; + tensor var_1740_end_0 = const()[name = tensor("op_1740_end_0"), val = tensor([2, 64, 1, 2560])]; + tensor var_1740_end_mask_0 = const()[name = tensor("op_1740_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1740_cast = slice_by_index(begin = var_1740_begin_0, end = var_1740_end_0, end_mask = var_1740_end_mask_0, x = var_1701_cast)[name = tensor("op_1740_cast")]; + tensor var_1741_begin_0 = const()[name = tensor("op_1741_begin_0"), val = tensor([0, 0, 0, 2560])]; + tensor var_1741_end_0 = const()[name = tensor("op_1741_end_0"), val = tensor([2, 64, 1, 3072])]; + tensor var_1741_end_mask_0 = const()[name = tensor("op_1741_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1741_cast = slice_by_index(begin = var_1741_begin_0, end = var_1741_end_0, end_mask = var_1741_end_mask_0, x = var_1701_cast)[name = tensor("op_1741_cast")]; + tensor var_1742_begin_0 = const()[name = tensor("op_1742_begin_0"), val = tensor([0, 0, 0, 3072])]; + tensor var_1742_end_0 = const()[name = tensor("op_1742_end_0"), val = tensor([2, 64, 1, 3584])]; + tensor var_1742_end_mask_0 = const()[name = tensor("op_1742_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1742_cast = slice_by_index(begin = var_1742_begin_0, end = var_1742_end_0, end_mask = var_1742_end_mask_0, x = var_1701_cast)[name = tensor("op_1742_cast")]; + tensor var_1743_begin_0 = const()[name = tensor("op_1743_begin_0"), val = tensor([0, 0, 0, 3584])]; + tensor var_1743_end_0 = const()[name = tensor("op_1743_end_0"), val = tensor([2, 64, 1, 4096])]; + tensor var_1743_end_mask_0 = const()[name = tensor("op_1743_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1743_cast = slice_by_index(begin = var_1743_begin_0, end = var_1743_end_0, end_mask = var_1743_end_mask_0, x = var_1701_cast)[name = tensor("op_1743_cast")]; + tensor k_15_perm_0 = const()[name = tensor("k_15_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_1748_begin_0 = const()[name = tensor("op_1748_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1748_end_0 = const()[name = tensor("op_1748_end_0"), val = tensor([2, 77, 1, 64])]; + tensor var_1748_end_mask_0 = const()[name = tensor("op_1748_end_mask_0"), val = tensor([true, true, true, false])]; + tensor transpose_28 = transpose(perm = k_15_perm_0, x = k_13_cast)[name = tensor("transpose_28")]; + tensor var_1748_cast = slice_by_index(begin = var_1748_begin_0, end = var_1748_end_0, end_mask = var_1748_end_mask_0, x = transpose_28)[name = tensor("op_1748_cast")]; + tensor var_1752_begin_0 = const()[name = tensor("op_1752_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_1752_end_0 = const()[name = tensor("op_1752_end_0"), val = tensor([2, 77, 1, 128])]; + tensor var_1752_end_mask_0 = const()[name = tensor("op_1752_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1752_cast = slice_by_index(begin = var_1752_begin_0, end = var_1752_end_0, end_mask = var_1752_end_mask_0, x = transpose_28)[name = tensor("op_1752_cast")]; + tensor var_1756_begin_0 = const()[name = tensor("op_1756_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_1756_end_0 = const()[name = tensor("op_1756_end_0"), val = tensor([2, 77, 1, 192])]; + tensor var_1756_end_mask_0 = const()[name = tensor("op_1756_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1756_cast = slice_by_index(begin = var_1756_begin_0, end = var_1756_end_0, end_mask = var_1756_end_mask_0, x = transpose_28)[name = tensor("op_1756_cast")]; + tensor var_1760_begin_0 = const()[name = tensor("op_1760_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_1760_end_0 = const()[name = tensor("op_1760_end_0"), val = tensor([2, 77, 1, 256])]; + tensor var_1760_end_mask_0 = const()[name = tensor("op_1760_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1760_cast = slice_by_index(begin = var_1760_begin_0, end = var_1760_end_0, end_mask = var_1760_end_mask_0, x = transpose_28)[name = tensor("op_1760_cast")]; + tensor var_1764_begin_0 = const()[name = tensor("op_1764_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_1764_end_0 = const()[name = tensor("op_1764_end_0"), val = tensor([2, 77, 1, 320])]; + tensor var_1764_end_mask_0 = const()[name = tensor("op_1764_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1764_cast = slice_by_index(begin = var_1764_begin_0, end = var_1764_end_0, end_mask = var_1764_end_mask_0, x = transpose_28)[name = tensor("op_1764_cast")]; + tensor var_1766_begin_0 = const()[name = tensor("op_1766_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1766_end_0 = const()[name = tensor("op_1766_end_0"), val = tensor([2, 64, 1, 77])]; + tensor var_1766_end_mask_0 = const()[name = tensor("op_1766_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1766_cast = slice_by_index(begin = var_1766_begin_0, end = var_1766_end_0, end_mask = var_1766_end_mask_0, x = v_7_cast)[name = tensor("op_1766_cast")]; + tensor var_1770_begin_0 = const()[name = tensor("op_1770_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_1770_end_0 = const()[name = tensor("op_1770_end_0"), val = tensor([2, 128, 1, 77])]; + tensor var_1770_end_mask_0 = const()[name = tensor("op_1770_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1770_cast = slice_by_index(begin = var_1770_begin_0, end = var_1770_end_0, end_mask = var_1770_end_mask_0, x = v_7_cast)[name = tensor("op_1770_cast")]; + tensor var_1774_begin_0 = const()[name = tensor("op_1774_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_1774_end_0 = const()[name = tensor("op_1774_end_0"), val = tensor([2, 192, 1, 77])]; + tensor var_1774_end_mask_0 = const()[name = tensor("op_1774_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1774_cast = slice_by_index(begin = var_1774_begin_0, end = var_1774_end_0, end_mask = var_1774_end_mask_0, x = v_7_cast)[name = tensor("op_1774_cast")]; + tensor var_1778_begin_0 = const()[name = tensor("op_1778_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_1778_end_0 = const()[name = tensor("op_1778_end_0"), val = tensor([2, 256, 1, 77])]; + tensor var_1778_end_mask_0 = const()[name = tensor("op_1778_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1778_cast = slice_by_index(begin = var_1778_begin_0, end = var_1778_end_0, end_mask = var_1778_end_mask_0, x = v_7_cast)[name = tensor("op_1778_cast")]; + tensor var_1782_begin_0 = const()[name = tensor("op_1782_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_1782_end_0 = const()[name = tensor("op_1782_end_0"), val = tensor([2, 320, 1, 77])]; + tensor var_1782_end_mask_0 = const()[name = tensor("op_1782_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1782_cast = slice_by_index(begin = var_1782_begin_0, end = var_1782_end_0, end_mask = var_1782_end_mask_0, x = v_7_cast)[name = tensor("op_1782_cast")]; + tensor var_1786_equation_0 = const()[name = tensor("op_1786_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1786_cast = einsum(equation = var_1786_equation_0, values = (var_1748_cast, var_1704_cast))[name = tensor("op_1786_cast")]; + tensor var_1787_to_fp16 = const()[name = tensor("op_1787_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_241_cast = mul(x = var_1786_cast, y = var_1787_to_fp16)[name = tensor("aw_chunk_241_cast")]; + tensor var_1790_equation_0 = const()[name = tensor("op_1790_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1790_cast = einsum(equation = var_1790_equation_0, values = (var_1748_cast, var_1705_cast))[name = tensor("op_1790_cast")]; + tensor var_1791_to_fp16 = const()[name = tensor("op_1791_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_243_cast = mul(x = var_1790_cast, y = var_1791_to_fp16)[name = tensor("aw_chunk_243_cast")]; + tensor var_1794_equation_0 = const()[name = tensor("op_1794_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1794_cast = einsum(equation = var_1794_equation_0, values = (var_1748_cast, var_1706_cast))[name = tensor("op_1794_cast")]; + tensor var_1795_to_fp16 = const()[name = tensor("op_1795_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_245_cast = mul(x = var_1794_cast, y = var_1795_to_fp16)[name = tensor("aw_chunk_245_cast")]; + tensor var_1798_equation_0 = const()[name = tensor("op_1798_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1798_cast = einsum(equation = var_1798_equation_0, values = (var_1748_cast, var_1707_cast))[name = tensor("op_1798_cast")]; + tensor var_1799_to_fp16 = const()[name = tensor("op_1799_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_247_cast = mul(x = var_1798_cast, y = var_1799_to_fp16)[name = tensor("aw_chunk_247_cast")]; + tensor var_1802_equation_0 = const()[name = tensor("op_1802_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1802_cast = einsum(equation = var_1802_equation_0, values = (var_1748_cast, var_1708_cast))[name = tensor("op_1802_cast")]; + tensor var_1803_to_fp16 = const()[name = tensor("op_1803_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_249_cast = mul(x = var_1802_cast, y = var_1803_to_fp16)[name = tensor("aw_chunk_249_cast")]; + tensor var_1806_equation_0 = const()[name = tensor("op_1806_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1806_cast = einsum(equation = var_1806_equation_0, values = (var_1748_cast, var_1709_cast))[name = tensor("op_1806_cast")]; + tensor var_1807_to_fp16 = const()[name = tensor("op_1807_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_251_cast = mul(x = var_1806_cast, y = var_1807_to_fp16)[name = tensor("aw_chunk_251_cast")]; + tensor var_1810_equation_0 = const()[name = tensor("op_1810_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1810_cast = einsum(equation = var_1810_equation_0, values = (var_1748_cast, var_1710_cast))[name = tensor("op_1810_cast")]; + tensor var_1811_to_fp16 = const()[name = tensor("op_1811_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_253_cast = mul(x = var_1810_cast, y = var_1811_to_fp16)[name = tensor("aw_chunk_253_cast")]; + tensor var_1814_equation_0 = const()[name = tensor("op_1814_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1814_cast = einsum(equation = var_1814_equation_0, values = (var_1748_cast, var_1711_cast))[name = tensor("op_1814_cast")]; + tensor var_1815_to_fp16 = const()[name = tensor("op_1815_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_255_cast = mul(x = var_1814_cast, y = var_1815_to_fp16)[name = tensor("aw_chunk_255_cast")]; + tensor var_1818_equation_0 = const()[name = tensor("op_1818_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1818_cast = einsum(equation = var_1818_equation_0, values = (var_1752_cast, var_1712_cast))[name = tensor("op_1818_cast")]; + tensor var_1819_to_fp16 = const()[name = tensor("op_1819_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_257_cast = mul(x = var_1818_cast, y = var_1819_to_fp16)[name = tensor("aw_chunk_257_cast")]; + tensor var_1822_equation_0 = const()[name = tensor("op_1822_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1822_cast = einsum(equation = var_1822_equation_0, values = (var_1752_cast, var_1713_cast))[name = tensor("op_1822_cast")]; + tensor var_1823_to_fp16 = const()[name = tensor("op_1823_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_259_cast = mul(x = var_1822_cast, y = var_1823_to_fp16)[name = tensor("aw_chunk_259_cast")]; + tensor var_1826_equation_0 = const()[name = tensor("op_1826_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1826_cast = einsum(equation = var_1826_equation_0, values = (var_1752_cast, var_1714_cast))[name = tensor("op_1826_cast")]; + tensor var_1827_to_fp16 = const()[name = tensor("op_1827_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_261_cast = mul(x = var_1826_cast, y = var_1827_to_fp16)[name = tensor("aw_chunk_261_cast")]; + tensor var_1830_equation_0 = const()[name = tensor("op_1830_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1830_cast = einsum(equation = var_1830_equation_0, values = (var_1752_cast, var_1715_cast))[name = tensor("op_1830_cast")]; + tensor var_1831_to_fp16 = const()[name = tensor("op_1831_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_263_cast = mul(x = var_1830_cast, y = var_1831_to_fp16)[name = tensor("aw_chunk_263_cast")]; + tensor var_1834_equation_0 = const()[name = tensor("op_1834_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1834_cast = einsum(equation = var_1834_equation_0, values = (var_1752_cast, var_1716_cast))[name = tensor("op_1834_cast")]; + tensor var_1835_to_fp16 = const()[name = tensor("op_1835_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_265_cast = mul(x = var_1834_cast, y = var_1835_to_fp16)[name = tensor("aw_chunk_265_cast")]; + tensor var_1838_equation_0 = const()[name = tensor("op_1838_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1838_cast = einsum(equation = var_1838_equation_0, values = (var_1752_cast, var_1717_cast))[name = tensor("op_1838_cast")]; + tensor var_1839_to_fp16 = const()[name = tensor("op_1839_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_267_cast = mul(x = var_1838_cast, y = var_1839_to_fp16)[name = tensor("aw_chunk_267_cast")]; + tensor var_1842_equation_0 = const()[name = tensor("op_1842_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1842_cast = einsum(equation = var_1842_equation_0, values = (var_1752_cast, var_1718_cast))[name = tensor("op_1842_cast")]; + tensor var_1843_to_fp16 = const()[name = tensor("op_1843_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_269_cast = mul(x = var_1842_cast, y = var_1843_to_fp16)[name = tensor("aw_chunk_269_cast")]; + tensor var_1846_equation_0 = const()[name = tensor("op_1846_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1846_cast = einsum(equation = var_1846_equation_0, values = (var_1752_cast, var_1719_cast))[name = tensor("op_1846_cast")]; + tensor var_1847_to_fp16 = const()[name = tensor("op_1847_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_271_cast = mul(x = var_1846_cast, y = var_1847_to_fp16)[name = tensor("aw_chunk_271_cast")]; + tensor var_1850_equation_0 = const()[name = tensor("op_1850_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1850_cast = einsum(equation = var_1850_equation_0, values = (var_1756_cast, var_1720_cast))[name = tensor("op_1850_cast")]; + tensor var_1851_to_fp16 = const()[name = tensor("op_1851_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_273_cast = mul(x = var_1850_cast, y = var_1851_to_fp16)[name = tensor("aw_chunk_273_cast")]; + tensor var_1854_equation_0 = const()[name = tensor("op_1854_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1854_cast = einsum(equation = var_1854_equation_0, values = (var_1756_cast, var_1721_cast))[name = tensor("op_1854_cast")]; + tensor var_1855_to_fp16 = const()[name = tensor("op_1855_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_275_cast = mul(x = var_1854_cast, y = var_1855_to_fp16)[name = tensor("aw_chunk_275_cast")]; + tensor var_1858_equation_0 = const()[name = tensor("op_1858_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1858_cast = einsum(equation = var_1858_equation_0, values = (var_1756_cast, var_1722_cast))[name = tensor("op_1858_cast")]; + tensor var_1859_to_fp16 = const()[name = tensor("op_1859_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_277_cast = mul(x = var_1858_cast, y = var_1859_to_fp16)[name = tensor("aw_chunk_277_cast")]; + tensor var_1862_equation_0 = const()[name = tensor("op_1862_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1862_cast = einsum(equation = var_1862_equation_0, values = (var_1756_cast, var_1723_cast))[name = tensor("op_1862_cast")]; + tensor var_1863_to_fp16 = const()[name = tensor("op_1863_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_279_cast = mul(x = var_1862_cast, y = var_1863_to_fp16)[name = tensor("aw_chunk_279_cast")]; + tensor var_1866_equation_0 = const()[name = tensor("op_1866_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1866_cast = einsum(equation = var_1866_equation_0, values = (var_1756_cast, var_1724_cast))[name = tensor("op_1866_cast")]; + tensor var_1867_to_fp16 = const()[name = tensor("op_1867_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_281_cast = mul(x = var_1866_cast, y = var_1867_to_fp16)[name = tensor("aw_chunk_281_cast")]; + tensor var_1870_equation_0 = const()[name = tensor("op_1870_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1870_cast = einsum(equation = var_1870_equation_0, values = (var_1756_cast, var_1725_cast))[name = tensor("op_1870_cast")]; + tensor var_1871_to_fp16 = const()[name = tensor("op_1871_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_283_cast = mul(x = var_1870_cast, y = var_1871_to_fp16)[name = tensor("aw_chunk_283_cast")]; + tensor var_1874_equation_0 = const()[name = tensor("op_1874_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1874_cast = einsum(equation = var_1874_equation_0, values = (var_1756_cast, var_1726_cast))[name = tensor("op_1874_cast")]; + tensor var_1875_to_fp16 = const()[name = tensor("op_1875_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_285_cast = mul(x = var_1874_cast, y = var_1875_to_fp16)[name = tensor("aw_chunk_285_cast")]; + tensor var_1878_equation_0 = const()[name = tensor("op_1878_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1878_cast = einsum(equation = var_1878_equation_0, values = (var_1756_cast, var_1727_cast))[name = tensor("op_1878_cast")]; + tensor var_1879_to_fp16 = const()[name = tensor("op_1879_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_287_cast = mul(x = var_1878_cast, y = var_1879_to_fp16)[name = tensor("aw_chunk_287_cast")]; + tensor var_1882_equation_0 = const()[name = tensor("op_1882_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1882_cast = einsum(equation = var_1882_equation_0, values = (var_1760_cast, var_1728_cast))[name = tensor("op_1882_cast")]; + tensor var_1883_to_fp16 = const()[name = tensor("op_1883_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_289_cast = mul(x = var_1882_cast, y = var_1883_to_fp16)[name = tensor("aw_chunk_289_cast")]; + tensor var_1886_equation_0 = const()[name = tensor("op_1886_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1886_cast = einsum(equation = var_1886_equation_0, values = (var_1760_cast, var_1729_cast))[name = tensor("op_1886_cast")]; + tensor var_1887_to_fp16 = const()[name = tensor("op_1887_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_291_cast = mul(x = var_1886_cast, y = var_1887_to_fp16)[name = tensor("aw_chunk_291_cast")]; + tensor var_1890_equation_0 = const()[name = tensor("op_1890_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1890_cast = einsum(equation = var_1890_equation_0, values = (var_1760_cast, var_1730_cast))[name = tensor("op_1890_cast")]; + tensor var_1891_to_fp16 = const()[name = tensor("op_1891_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_293_cast = mul(x = var_1890_cast, y = var_1891_to_fp16)[name = tensor("aw_chunk_293_cast")]; + tensor var_1894_equation_0 = const()[name = tensor("op_1894_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1894_cast = einsum(equation = var_1894_equation_0, values = (var_1760_cast, var_1731_cast))[name = tensor("op_1894_cast")]; + tensor var_1895_to_fp16 = const()[name = tensor("op_1895_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_295_cast = mul(x = var_1894_cast, y = var_1895_to_fp16)[name = tensor("aw_chunk_295_cast")]; + tensor var_1898_equation_0 = const()[name = tensor("op_1898_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1898_cast = einsum(equation = var_1898_equation_0, values = (var_1760_cast, var_1732_cast))[name = tensor("op_1898_cast")]; + tensor var_1899_to_fp16 = const()[name = tensor("op_1899_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_297_cast = mul(x = var_1898_cast, y = var_1899_to_fp16)[name = tensor("aw_chunk_297_cast")]; + tensor var_1902_equation_0 = const()[name = tensor("op_1902_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1902_cast = einsum(equation = var_1902_equation_0, values = (var_1760_cast, var_1733_cast))[name = tensor("op_1902_cast")]; + tensor var_1903_to_fp16 = const()[name = tensor("op_1903_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_299_cast = mul(x = var_1902_cast, y = var_1903_to_fp16)[name = tensor("aw_chunk_299_cast")]; + tensor var_1906_equation_0 = const()[name = tensor("op_1906_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1906_cast = einsum(equation = var_1906_equation_0, values = (var_1760_cast, var_1734_cast))[name = tensor("op_1906_cast")]; + tensor var_1907_to_fp16 = const()[name = tensor("op_1907_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_301_cast = mul(x = var_1906_cast, y = var_1907_to_fp16)[name = tensor("aw_chunk_301_cast")]; + tensor var_1910_equation_0 = const()[name = tensor("op_1910_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1910_cast = einsum(equation = var_1910_equation_0, values = (var_1760_cast, var_1735_cast))[name = tensor("op_1910_cast")]; + tensor var_1911_to_fp16 = const()[name = tensor("op_1911_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_303_cast = mul(x = var_1910_cast, y = var_1911_to_fp16)[name = tensor("aw_chunk_303_cast")]; + tensor var_1914_equation_0 = const()[name = tensor("op_1914_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1914_cast = einsum(equation = var_1914_equation_0, values = (var_1764_cast, var_1736_cast))[name = tensor("op_1914_cast")]; + tensor var_1915_to_fp16 = const()[name = tensor("op_1915_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_305_cast = mul(x = var_1914_cast, y = var_1915_to_fp16)[name = tensor("aw_chunk_305_cast")]; + tensor var_1918_equation_0 = const()[name = tensor("op_1918_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1918_cast = einsum(equation = var_1918_equation_0, values = (var_1764_cast, var_1737_cast))[name = tensor("op_1918_cast")]; + tensor var_1919_to_fp16 = const()[name = tensor("op_1919_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_307_cast = mul(x = var_1918_cast, y = var_1919_to_fp16)[name = tensor("aw_chunk_307_cast")]; + tensor var_1922_equation_0 = const()[name = tensor("op_1922_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1922_cast = einsum(equation = var_1922_equation_0, values = (var_1764_cast, var_1738_cast))[name = tensor("op_1922_cast")]; + tensor var_1923_to_fp16 = const()[name = tensor("op_1923_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_309_cast = mul(x = var_1922_cast, y = var_1923_to_fp16)[name = tensor("aw_chunk_309_cast")]; + tensor var_1926_equation_0 = const()[name = tensor("op_1926_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1926_cast = einsum(equation = var_1926_equation_0, values = (var_1764_cast, var_1739_cast))[name = tensor("op_1926_cast")]; + tensor var_1927_to_fp16 = const()[name = tensor("op_1927_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_311_cast = mul(x = var_1926_cast, y = var_1927_to_fp16)[name = tensor("aw_chunk_311_cast")]; + tensor var_1930_equation_0 = const()[name = tensor("op_1930_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1930_cast = einsum(equation = var_1930_equation_0, values = (var_1764_cast, var_1740_cast))[name = tensor("op_1930_cast")]; + tensor var_1931_to_fp16 = const()[name = tensor("op_1931_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_313_cast = mul(x = var_1930_cast, y = var_1931_to_fp16)[name = tensor("aw_chunk_313_cast")]; + tensor var_1934_equation_0 = const()[name = tensor("op_1934_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1934_cast = einsum(equation = var_1934_equation_0, values = (var_1764_cast, var_1741_cast))[name = tensor("op_1934_cast")]; + tensor var_1935_to_fp16 = const()[name = tensor("op_1935_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_315_cast = mul(x = var_1934_cast, y = var_1935_to_fp16)[name = tensor("aw_chunk_315_cast")]; + tensor var_1938_equation_0 = const()[name = tensor("op_1938_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1938_cast = einsum(equation = var_1938_equation_0, values = (var_1764_cast, var_1742_cast))[name = tensor("op_1938_cast")]; + tensor var_1939_to_fp16 = const()[name = tensor("op_1939_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_317_cast = mul(x = var_1938_cast, y = var_1939_to_fp16)[name = tensor("aw_chunk_317_cast")]; + tensor var_1942_equation_0 = const()[name = tensor("op_1942_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1942_cast = einsum(equation = var_1942_equation_0, values = (var_1764_cast, var_1743_cast))[name = tensor("op_1942_cast")]; + tensor var_1943_to_fp16 = const()[name = tensor("op_1943_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_319_cast = mul(x = var_1942_cast, y = var_1943_to_fp16)[name = tensor("aw_chunk_319_cast")]; + tensor var_1945_cast = softmax(axis = var_120, x = aw_chunk_241_cast)[name = tensor("op_1945_cast")]; + tensor var_1946_cast = softmax(axis = var_120, x = aw_chunk_243_cast)[name = tensor("op_1946_cast")]; + tensor var_1947_cast = softmax(axis = var_120, x = aw_chunk_245_cast)[name = tensor("op_1947_cast")]; + tensor var_1948_cast = softmax(axis = var_120, x = aw_chunk_247_cast)[name = tensor("op_1948_cast")]; + tensor var_1949_cast = softmax(axis = var_120, x = aw_chunk_249_cast)[name = tensor("op_1949_cast")]; + tensor var_1950_cast = softmax(axis = var_120, x = aw_chunk_251_cast)[name = tensor("op_1950_cast")]; + tensor var_1951_cast = softmax(axis = var_120, x = aw_chunk_253_cast)[name = tensor("op_1951_cast")]; + tensor var_1952_cast = softmax(axis = var_120, x = aw_chunk_255_cast)[name = tensor("op_1952_cast")]; + tensor var_1953_cast = softmax(axis = var_120, x = aw_chunk_257_cast)[name = tensor("op_1953_cast")]; + tensor var_1954_cast = softmax(axis = var_120, x = aw_chunk_259_cast)[name = tensor("op_1954_cast")]; + tensor var_1955_cast = softmax(axis = var_120, x = aw_chunk_261_cast)[name = tensor("op_1955_cast")]; + tensor var_1956_cast = softmax(axis = var_120, x = aw_chunk_263_cast)[name = tensor("op_1956_cast")]; + tensor var_1957_cast = softmax(axis = var_120, x = aw_chunk_265_cast)[name = tensor("op_1957_cast")]; + tensor var_1958_cast = softmax(axis = var_120, x = aw_chunk_267_cast)[name = tensor("op_1958_cast")]; + tensor var_1959_cast = softmax(axis = var_120, x = aw_chunk_269_cast)[name = tensor("op_1959_cast")]; + tensor var_1960_cast = softmax(axis = var_120, x = aw_chunk_271_cast)[name = tensor("op_1960_cast")]; + tensor var_1961_cast = softmax(axis = var_120, x = aw_chunk_273_cast)[name = tensor("op_1961_cast")]; + tensor var_1962_cast = softmax(axis = var_120, x = aw_chunk_275_cast)[name = tensor("op_1962_cast")]; + tensor var_1963_cast = softmax(axis = var_120, x = aw_chunk_277_cast)[name = tensor("op_1963_cast")]; + tensor var_1964_cast = softmax(axis = var_120, x = aw_chunk_279_cast)[name = tensor("op_1964_cast")]; + tensor var_1965_cast = softmax(axis = var_120, x = aw_chunk_281_cast)[name = tensor("op_1965_cast")]; + tensor var_1966_cast = softmax(axis = var_120, x = aw_chunk_283_cast)[name = tensor("op_1966_cast")]; + tensor var_1967_cast = softmax(axis = var_120, x = aw_chunk_285_cast)[name = tensor("op_1967_cast")]; + tensor var_1968_cast = softmax(axis = var_120, x = aw_chunk_287_cast)[name = tensor("op_1968_cast")]; + tensor var_1969_cast = softmax(axis = var_120, x = aw_chunk_289_cast)[name = tensor("op_1969_cast")]; + tensor var_1970_cast = softmax(axis = var_120, x = aw_chunk_291_cast)[name = tensor("op_1970_cast")]; + tensor var_1971_cast = softmax(axis = var_120, x = aw_chunk_293_cast)[name = tensor("op_1971_cast")]; + tensor var_1972_cast = softmax(axis = var_120, x = aw_chunk_295_cast)[name = tensor("op_1972_cast")]; + tensor var_1973_cast = softmax(axis = var_120, x = aw_chunk_297_cast)[name = tensor("op_1973_cast")]; + tensor var_1974_cast = softmax(axis = var_120, x = aw_chunk_299_cast)[name = tensor("op_1974_cast")]; + tensor var_1975_cast = softmax(axis = var_120, x = aw_chunk_301_cast)[name = tensor("op_1975_cast")]; + tensor var_1976_cast = softmax(axis = var_120, x = aw_chunk_303_cast)[name = tensor("op_1976_cast")]; + tensor var_1977_cast = softmax(axis = var_120, x = aw_chunk_305_cast)[name = tensor("op_1977_cast")]; + tensor var_1978_cast = softmax(axis = var_120, x = aw_chunk_307_cast)[name = tensor("op_1978_cast")]; + tensor var_1979_cast = softmax(axis = var_120, x = aw_chunk_309_cast)[name = tensor("op_1979_cast")]; + tensor var_1980_cast = softmax(axis = var_120, x = aw_chunk_311_cast)[name = tensor("op_1980_cast")]; + tensor var_1981_cast = softmax(axis = var_120, x = aw_chunk_313_cast)[name = tensor("op_1981_cast")]; + tensor var_1982_cast = softmax(axis = var_120, x = aw_chunk_315_cast)[name = tensor("op_1982_cast")]; + tensor var_1983_cast = softmax(axis = var_120, x = aw_chunk_317_cast)[name = tensor("op_1983_cast")]; + tensor var_1984_cast = softmax(axis = var_120, x = aw_chunk_319_cast)[name = tensor("op_1984_cast")]; + tensor var_1986_equation_0 = const()[name = tensor("op_1986_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1986_cast = einsum(equation = var_1986_equation_0, values = (var_1766_cast, var_1945_cast))[name = tensor("op_1986_cast")]; + tensor var_1988_equation_0 = const()[name = tensor("op_1988_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1988_cast = einsum(equation = var_1988_equation_0, values = (var_1766_cast, var_1946_cast))[name = tensor("op_1988_cast")]; + tensor var_1990_equation_0 = const()[name = tensor("op_1990_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1990_cast = einsum(equation = var_1990_equation_0, values = (var_1766_cast, var_1947_cast))[name = tensor("op_1990_cast")]; + tensor var_1992_equation_0 = const()[name = tensor("op_1992_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1992_cast = einsum(equation = var_1992_equation_0, values = (var_1766_cast, var_1948_cast))[name = tensor("op_1992_cast")]; + tensor var_1994_equation_0 = const()[name = tensor("op_1994_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1994_cast = einsum(equation = var_1994_equation_0, values = (var_1766_cast, var_1949_cast))[name = tensor("op_1994_cast")]; + tensor var_1996_equation_0 = const()[name = tensor("op_1996_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1996_cast = einsum(equation = var_1996_equation_0, values = (var_1766_cast, var_1950_cast))[name = tensor("op_1996_cast")]; + tensor var_1998_equation_0 = const()[name = tensor("op_1998_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1998_cast = einsum(equation = var_1998_equation_0, values = (var_1766_cast, var_1951_cast))[name = tensor("op_1998_cast")]; + tensor var_2000_equation_0 = const()[name = tensor("op_2000_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2000_cast = einsum(equation = var_2000_equation_0, values = (var_1766_cast, var_1952_cast))[name = tensor("op_2000_cast")]; + tensor var_2002_equation_0 = const()[name = tensor("op_2002_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2002_cast = einsum(equation = var_2002_equation_0, values = (var_1770_cast, var_1953_cast))[name = tensor("op_2002_cast")]; + tensor var_2004_equation_0 = const()[name = tensor("op_2004_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2004_cast = einsum(equation = var_2004_equation_0, values = (var_1770_cast, var_1954_cast))[name = tensor("op_2004_cast")]; + tensor var_2006_equation_0 = const()[name = tensor("op_2006_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2006_cast = einsum(equation = var_2006_equation_0, values = (var_1770_cast, var_1955_cast))[name = tensor("op_2006_cast")]; + tensor var_2008_equation_0 = const()[name = tensor("op_2008_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2008_cast = einsum(equation = var_2008_equation_0, values = (var_1770_cast, var_1956_cast))[name = tensor("op_2008_cast")]; + tensor var_2010_equation_0 = const()[name = tensor("op_2010_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2010_cast = einsum(equation = var_2010_equation_0, values = (var_1770_cast, var_1957_cast))[name = tensor("op_2010_cast")]; + tensor var_2012_equation_0 = const()[name = tensor("op_2012_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2012_cast = einsum(equation = var_2012_equation_0, values = (var_1770_cast, var_1958_cast))[name = tensor("op_2012_cast")]; + tensor var_2014_equation_0 = const()[name = tensor("op_2014_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2014_cast = einsum(equation = var_2014_equation_0, values = (var_1770_cast, var_1959_cast))[name = tensor("op_2014_cast")]; + tensor var_2016_equation_0 = const()[name = tensor("op_2016_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2016_cast = einsum(equation = var_2016_equation_0, values = (var_1770_cast, var_1960_cast))[name = tensor("op_2016_cast")]; + tensor var_2018_equation_0 = const()[name = tensor("op_2018_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2018_cast = einsum(equation = var_2018_equation_0, values = (var_1774_cast, var_1961_cast))[name = tensor("op_2018_cast")]; + tensor var_2020_equation_0 = const()[name = tensor("op_2020_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2020_cast = einsum(equation = var_2020_equation_0, values = (var_1774_cast, var_1962_cast))[name = tensor("op_2020_cast")]; + tensor var_2022_equation_0 = const()[name = tensor("op_2022_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2022_cast = einsum(equation = var_2022_equation_0, values = (var_1774_cast, var_1963_cast))[name = tensor("op_2022_cast")]; + tensor var_2024_equation_0 = const()[name = tensor("op_2024_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2024_cast = einsum(equation = var_2024_equation_0, values = (var_1774_cast, var_1964_cast))[name = tensor("op_2024_cast")]; + tensor var_2026_equation_0 = const()[name = tensor("op_2026_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2026_cast = einsum(equation = var_2026_equation_0, values = (var_1774_cast, var_1965_cast))[name = tensor("op_2026_cast")]; + tensor var_2028_equation_0 = const()[name = tensor("op_2028_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2028_cast = einsum(equation = var_2028_equation_0, values = (var_1774_cast, var_1966_cast))[name = tensor("op_2028_cast")]; + tensor var_2030_equation_0 = const()[name = tensor("op_2030_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2030_cast = einsum(equation = var_2030_equation_0, values = (var_1774_cast, var_1967_cast))[name = tensor("op_2030_cast")]; + tensor var_2032_equation_0 = const()[name = tensor("op_2032_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2032_cast = einsum(equation = var_2032_equation_0, values = (var_1774_cast, var_1968_cast))[name = tensor("op_2032_cast")]; + tensor var_2034_equation_0 = const()[name = tensor("op_2034_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2034_cast = einsum(equation = var_2034_equation_0, values = (var_1778_cast, var_1969_cast))[name = tensor("op_2034_cast")]; + tensor var_2036_equation_0 = const()[name = tensor("op_2036_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2036_cast = einsum(equation = var_2036_equation_0, values = (var_1778_cast, var_1970_cast))[name = tensor("op_2036_cast")]; + tensor var_2038_equation_0 = const()[name = tensor("op_2038_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2038_cast = einsum(equation = var_2038_equation_0, values = (var_1778_cast, var_1971_cast))[name = tensor("op_2038_cast")]; + tensor var_2040_equation_0 = const()[name = tensor("op_2040_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2040_cast = einsum(equation = var_2040_equation_0, values = (var_1778_cast, var_1972_cast))[name = tensor("op_2040_cast")]; + tensor var_2042_equation_0 = const()[name = tensor("op_2042_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2042_cast = einsum(equation = var_2042_equation_0, values = (var_1778_cast, var_1973_cast))[name = tensor("op_2042_cast")]; + tensor var_2044_equation_0 = const()[name = tensor("op_2044_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2044_cast = einsum(equation = var_2044_equation_0, values = (var_1778_cast, var_1974_cast))[name = tensor("op_2044_cast")]; + tensor var_2046_equation_0 = const()[name = tensor("op_2046_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2046_cast = einsum(equation = var_2046_equation_0, values = (var_1778_cast, var_1975_cast))[name = tensor("op_2046_cast")]; + tensor var_2048_equation_0 = const()[name = tensor("op_2048_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2048_cast = einsum(equation = var_2048_equation_0, values = (var_1778_cast, var_1976_cast))[name = tensor("op_2048_cast")]; + tensor var_2050_equation_0 = const()[name = tensor("op_2050_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2050_cast = einsum(equation = var_2050_equation_0, values = (var_1782_cast, var_1977_cast))[name = tensor("op_2050_cast")]; + tensor var_2052_equation_0 = const()[name = tensor("op_2052_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2052_cast = einsum(equation = var_2052_equation_0, values = (var_1782_cast, var_1978_cast))[name = tensor("op_2052_cast")]; + tensor var_2054_equation_0 = const()[name = tensor("op_2054_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2054_cast = einsum(equation = var_2054_equation_0, values = (var_1782_cast, var_1979_cast))[name = tensor("op_2054_cast")]; + tensor var_2056_equation_0 = const()[name = tensor("op_2056_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2056_cast = einsum(equation = var_2056_equation_0, values = (var_1782_cast, var_1980_cast))[name = tensor("op_2056_cast")]; + tensor var_2058_equation_0 = const()[name = tensor("op_2058_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2058_cast = einsum(equation = var_2058_equation_0, values = (var_1782_cast, var_1981_cast))[name = tensor("op_2058_cast")]; + tensor var_2060_equation_0 = const()[name = tensor("op_2060_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2060_cast = einsum(equation = var_2060_equation_0, values = (var_1782_cast, var_1982_cast))[name = tensor("op_2060_cast")]; + tensor var_2062_equation_0 = const()[name = tensor("op_2062_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2062_cast = einsum(equation = var_2062_equation_0, values = (var_1782_cast, var_1983_cast))[name = tensor("op_2062_cast")]; + tensor var_2064_equation_0 = const()[name = tensor("op_2064_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2064_cast = einsum(equation = var_2064_equation_0, values = (var_1782_cast, var_1984_cast))[name = tensor("op_2064_cast")]; + tensor var_2066_interleave_0 = const()[name = tensor("op_2066_interleave_0"), val = tensor(false)]; + tensor var_2066_cast = concat(axis = var_95, interleave = var_2066_interleave_0, values = (var_1986_cast, var_1988_cast, var_1990_cast, var_1992_cast, var_1994_cast, var_1996_cast, var_1998_cast, var_2000_cast))[name = tensor("op_2066_cast")]; + tensor var_2068_interleave_0 = const()[name = tensor("op_2068_interleave_0"), val = tensor(false)]; + tensor var_2068_cast = concat(axis = var_95, interleave = var_2068_interleave_0, values = (var_2002_cast, var_2004_cast, var_2006_cast, var_2008_cast, var_2010_cast, var_2012_cast, var_2014_cast, var_2016_cast))[name = tensor("op_2068_cast")]; + tensor var_2070_interleave_0 = const()[name = tensor("op_2070_interleave_0"), val = tensor(false)]; + tensor var_2070_cast = concat(axis = var_95, interleave = var_2070_interleave_0, values = (var_2018_cast, var_2020_cast, var_2022_cast, var_2024_cast, var_2026_cast, var_2028_cast, var_2030_cast, var_2032_cast))[name = tensor("op_2070_cast")]; + tensor var_2072_interleave_0 = const()[name = tensor("op_2072_interleave_0"), val = tensor(false)]; + tensor var_2072_cast = concat(axis = var_95, interleave = var_2072_interleave_0, values = (var_2034_cast, var_2036_cast, var_2038_cast, var_2040_cast, var_2042_cast, var_2044_cast, var_2046_cast, var_2048_cast))[name = tensor("op_2072_cast")]; + tensor var_2074_interleave_0 = const()[name = tensor("op_2074_interleave_0"), val = tensor(false)]; + tensor var_2074_cast = concat(axis = var_95, interleave = var_2074_interleave_0, values = (var_2050_cast, var_2052_cast, var_2054_cast, var_2056_cast, var_2058_cast, var_2060_cast, var_2062_cast, var_2064_cast))[name = tensor("op_2074_cast")]; + tensor input_53_interleave_0 = const()[name = tensor("input_53_interleave_0"), val = tensor(false)]; + tensor input_53_cast = concat(axis = var_120, interleave = input_53_interleave_0, values = (var_2066_cast, var_2068_cast, var_2070_cast, var_2072_cast, var_2074_cast))[name = tensor("input_53_cast")]; + tensor var_2080 = const()[name = tensor("op_2080"), val = tensor([1, 1])]; + tensor var_2082 = const()[name = tensor("op_2082"), val = tensor([1, 1])]; + tensor var_2084_pad_type_0 = const()[name = tensor("op_2084_pad_type_0"), val = tensor("custom")]; + tensor var_2084_pad_0 = const()[name = tensor("op_2084_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_0_attentions_1_transformer_blocks_0_attn2_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(7945728))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(8022592))), name = tensor("down_blocks_0_attentions_1_transformer_blocks_0_attn2_to_out_0_weight_to_fp16_palettized"), shape = tensor([320, 320, 1, 1])]; + tensor down_blocks_0_attentions_1_transformer_blocks_0_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("down_blocks_0_attentions_1_transformer_blocks_0_attn2_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(8022784)))]; + tensor var_2084_cast = conv(bias = down_blocks_0_attentions_1_transformer_blocks_0_attn2_to_out_0_bias_to_fp16, dilations = var_2082, groups = var_120, pad = var_2084_pad_0, pad_type = var_2084_pad_type_0, strides = var_2080, weight = down_blocks_0_attentions_1_transformer_blocks_0_attn2_to_out_0_weight_to_fp16_palettized, x = input_53_cast)[name = tensor("op_2084_cast")]; + tensor inputs_11_cast = add(x = var_2084_cast, y = inputs_9_cast)[name = tensor("inputs_11_cast")]; + tensor var_2088 = const()[name = tensor("op_2088"), val = tensor([1])]; + tensor channels_mean_11_cast = reduce_mean(axes = var_2088, keep_dims = var_115, x = inputs_11_cast)[name = tensor("channels_mean_11_cast")]; + tensor zero_mean_11_cast = sub(x = inputs_11_cast, y = channels_mean_11_cast)[name = tensor("zero_mean_11_cast")]; + tensor zero_mean_sq_11_cast = mul(x = zero_mean_11_cast, y = zero_mean_11_cast)[name = tensor("zero_mean_sq_11_cast")]; + tensor var_2092 = const()[name = tensor("op_2092"), val = tensor([1])]; + tensor var_2093_cast = reduce_mean(axes = var_2092, keep_dims = var_115, x = zero_mean_sq_11_cast)[name = tensor("op_2093_cast")]; + tensor var_2094_to_fp16 = const()[name = tensor("op_2094_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_2095_cast = add(x = var_2093_cast, y = var_2094_to_fp16)[name = tensor("op_2095_cast")]; + tensor denom_11_epsilon_0_to_fp16 = const()[name = tensor("denom_11_epsilon_0_to_fp16"), val = tensor(0x1p-24)]; + tensor denom_11_cast = rsqrt(epsilon = denom_11_epsilon_0_to_fp16, x = var_2095_cast)[name = tensor("denom_11_cast")]; + tensor out_11_cast = mul(x = zero_mean_11_cast, y = denom_11_cast)[name = tensor("out_11_cast")]; + tensor var_2099_to_fp16 = const()[name = tensor("op_2099_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(8023488)))]; + tensor var_2100_cast = add(x = out_11_cast, y = var_2099_to_fp16)[name = tensor("op_2100_cast")]; + tensor var_2102_to_fp16 = const()[name = tensor("op_2102_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(8024192)))]; + tensor input_55_cast = mul(x = var_2100_cast, y = var_2102_to_fp16)[name = tensor("input_55_cast")]; + tensor var_2110 = const()[name = tensor("op_2110"), val = tensor([1, 1])]; + tensor var_2112 = const()[name = tensor("op_2112"), val = tensor([1, 1])]; + tensor var_2114_pad_type_0 = const()[name = tensor("op_2114_pad_type_0"), val = tensor("custom")]; + tensor var_2114_pad_0 = const()[name = tensor("op_2114_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_0_attentions_1_transformer_blocks_0_ff_net_0_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(8024896))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(8639360))), name = tensor("down_blocks_0_attentions_1_transformer_blocks_0_ff_net_0_proj_weight_to_fp16_palettized"), shape = tensor([2560, 320, 1, 1])]; + tensor down_blocks_0_attentions_1_transformer_blocks_0_ff_net_0_proj_bias_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(8639552))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(8641536))), name = tensor("down_blocks_0_attentions_1_transformer_blocks_0_ff_net_0_proj_bias_to_fp16_palettized"), shape = tensor([2560])]; + tensor var_2114_cast = conv(bias = down_blocks_0_attentions_1_transformer_blocks_0_ff_net_0_proj_bias_to_fp16_palettized, dilations = var_2112, groups = var_120, pad = var_2114_pad_0, pad_type = var_2114_pad_type_0, strides = var_2110, weight = down_blocks_0_attentions_1_transformer_blocks_0_ff_net_0_proj_weight_to_fp16_palettized, x = input_55_cast)[name = tensor("op_2114_cast")]; + tensor var_2115_split_sizes_0 = const()[name = tensor("op_2115_split_sizes_0"), val = tensor([1280, 1280])]; + tensor var_2115_axis_0 = const()[name = tensor("op_2115_axis_0"), val = tensor(1)]; + tensor var_2115_cast_0, tensor var_2115_cast_1 = split(axis = var_2115_axis_0, split_sizes = var_2115_split_sizes_0, x = var_2114_cast)[name = tensor("op_2115_cast")]; + tensor var_2117_mode_0 = const()[name = tensor("op_2117_mode_0"), val = tensor("EXACT")]; + tensor var_2117_cast = gelu(mode = var_2117_mode_0, x = var_2115_cast_1)[name = tensor("op_2117_cast")]; + tensor input_57_cast = mul(x = var_2115_cast_0, y = var_2117_cast)[name = tensor("input_57_cast")]; + tensor var_2121 = const()[name = tensor("op_2121"), val = tensor([1, 1])]; + tensor var_2123 = const()[name = tensor("op_2123"), val = tensor([1, 1])]; + tensor var_2125_pad_type_0 = const()[name = tensor("op_2125_pad_type_0"), val = tensor("custom")]; + tensor var_2125_pad_0 = const()[name = tensor("op_2125_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_0_attentions_1_transformer_blocks_0_ff_net_2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(8641728))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(8948992))), name = tensor("down_blocks_0_attentions_1_transformer_blocks_0_ff_net_2_weight_to_fp16_palettized"), shape = tensor([320, 1280, 1, 1])]; + tensor down_blocks_0_attentions_1_transformer_blocks_0_ff_net_2_bias_to_fp16 = const()[name = tensor("down_blocks_0_attentions_1_transformer_blocks_0_ff_net_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(8949184)))]; + tensor var_2125_cast = conv(bias = down_blocks_0_attentions_1_transformer_blocks_0_ff_net_2_bias_to_fp16, dilations = var_2123, groups = var_120, pad = var_2125_pad_0, pad_type = var_2125_pad_type_0, strides = var_2121, weight = down_blocks_0_attentions_1_transformer_blocks_0_ff_net_2_weight_to_fp16_palettized, x = input_57_cast)[name = tensor("op_2125_cast")]; + tensor hidden_states_33_cast = add(x = var_2125_cast, y = inputs_11_cast)[name = tensor("hidden_states_33_cast")]; + tensor var_2127 = const()[name = tensor("op_2127"), val = tensor([2, 320, 64, 64])]; + tensor input_59_cast = reshape(shape = var_2127, x = hidden_states_33_cast)[name = tensor("input_59_cast")]; + tensor var_2131 = const()[name = tensor("op_2131"), val = tensor([1, 1])]; + tensor var_2133 = const()[name = tensor("op_2133"), val = tensor([1, 1])]; + tensor hidden_states_35_pad_type_0 = const()[name = tensor("hidden_states_35_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_35_pad_0 = const()[name = tensor("hidden_states_35_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_0_attentions_1_proj_out_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(8949888))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(9026752))), name = tensor("down_blocks_0_attentions_1_proj_out_weight_to_fp16_palettized"), shape = tensor([320, 320, 1, 1])]; + tensor down_blocks_0_attentions_1_proj_out_bias_to_fp16 = const()[name = tensor("down_blocks_0_attentions_1_proj_out_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(9026944)))]; + tensor hidden_states_35_cast = conv(bias = down_blocks_0_attentions_1_proj_out_bias_to_fp16, dilations = var_2133, groups = var_120, pad = hidden_states_35_pad_0, pad_type = hidden_states_35_pad_type_0, strides = var_2131, weight = down_blocks_0_attentions_1_proj_out_weight_to_fp16_palettized, x = input_59_cast)[name = tensor("hidden_states_35_cast")]; + tensor input_61_cast = add(x = hidden_states_35_cast, y = hidden_states_23_cast)[name = tensor("input_61_cast")]; + tensor var_2140 = const()[name = tensor("op_2140"), val = tensor([2, 2])]; + tensor var_2142 = const()[name = tensor("op_2142"), val = tensor([1, 1])]; + tensor input_63_pad_type_0 = const()[name = tensor("input_63_pad_type_0"), val = tensor("custom")]; + tensor input_63_pad_0 = const()[name = tensor("input_63_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor down_blocks_0_downsamplers_0_conv_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(9027648))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(9718912))), name = tensor("down_blocks_0_downsamplers_0_conv_weight_to_fp16_palettized"), shape = tensor([320, 320, 3, 3])]; + tensor down_blocks_0_downsamplers_0_conv_bias_to_fp16 = const()[name = tensor("down_blocks_0_downsamplers_0_conv_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(9719104)))]; + tensor input_63_cast = conv(bias = down_blocks_0_downsamplers_0_conv_bias_to_fp16, dilations = var_2142, groups = var_120, pad = input_63_pad_0, pad_type = input_63_pad_type_0, strides = var_2140, weight = down_blocks_0_downsamplers_0_conv_weight_to_fp16_palettized, x = input_61_cast)[name = tensor("input_63_cast")]; + tensor var_2150 = const()[name = tensor("op_2150"), val = tensor(3)]; + tensor var_2168 = const()[name = tensor("op_2168"), val = tensor(true)]; + tensor var_2173 = const()[name = tensor("op_2173"), val = tensor(1)]; + tensor reshape_24_shape_0 = const()[name = tensor("reshape_24_shape_0"), val = tensor([2, 32, 10, 32, 32])]; + tensor reshape_24_cast = reshape(shape = reshape_24_shape_0, x = input_63_cast)[name = tensor("reshape_24_cast")]; + tensor reduce_mean_18_axes_0 = const()[name = tensor("reduce_mean_18_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_18_keep_dims_0 = const()[name = tensor("reduce_mean_18_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_18_cast = reduce_mean(axes = reduce_mean_18_axes_0, keep_dims = reduce_mean_18_keep_dims_0, x = reshape_24_cast)[name = tensor("reduce_mean_18_cast")]; + tensor sub_12_cast = sub(x = reshape_24_cast, y = reduce_mean_18_cast)[name = tensor("sub_12_cast")]; + tensor square_6_cast = square(x = sub_12_cast)[name = tensor("square_6_cast")]; + tensor reduce_mean_20_axes_0 = const()[name = tensor("reduce_mean_20_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_20_keep_dims_0 = const()[name = tensor("reduce_mean_20_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_20_cast = reduce_mean(axes = reduce_mean_20_axes_0, keep_dims = reduce_mean_20_keep_dims_0, x = square_6_cast)[name = tensor("reduce_mean_20_cast")]; + tensor add_12_y_0_to_fp16 = const()[name = tensor("add_12_y_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_12_cast = add(x = reduce_mean_20_cast, y = add_12_y_0_to_fp16)[name = tensor("add_12_cast")]; + tensor sqrt_6_cast = sqrt(x = add_12_cast)[name = tensor("sqrt_6_cast")]; + tensor real_div_6_cast = real_div(x = sub_12_cast, y = sqrt_6_cast)[name = tensor("real_div_6_cast")]; + tensor reshape_25_shape_0 = const()[name = tensor("reshape_25_shape_0"), val = tensor([2, 320, 32, 32])]; + tensor reshape_25_cast = reshape(shape = reshape_25_shape_0, x = real_div_6_cast)[name = tensor("reshape_25_cast")]; + tensor add_13_gamma_0_to_fp16 = const()[name = tensor("add_13_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(9719808)))]; + tensor add_13_beta_0_to_fp16 = const()[name = tensor("add_13_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(9720512)))]; + tensor add_13_epsilon_0_to_fp16 = const()[name = tensor("add_13_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_13_cast = batch_norm(beta = add_13_beta_0_to_fp16, epsilon = add_13_epsilon_0_to_fp16, gamma = add_13_gamma_0_to_fp16, mean = add_1_mean_0_to_fp16, variance = add_1_variance_0_to_fp16, x = reshape_25_cast)[name = tensor("add_13_cast")]; + tensor input_67_cast = silu(x = add_13_cast)[name = tensor("input_67_cast")]; + tensor var_2196 = const()[name = tensor("op_2196"), val = tensor([1, 1])]; + tensor var_2198 = const()[name = tensor("op_2198"), val = tensor([1, 1])]; + tensor hidden_states_37_pad_type_0 = const()[name = tensor("hidden_states_37_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_37_pad_0 = const()[name = tensor("hidden_states_37_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor down_blocks_1_resnets_0_conv1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(9721216))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(11103680))), name = tensor("down_blocks_1_resnets_0_conv1_weight_to_fp16_palettized"), shape = tensor([640, 320, 3, 3])]; + tensor down_blocks_1_resnets_0_conv1_bias_to_fp16 = const()[name = tensor("down_blocks_1_resnets_0_conv1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(11103872)))]; + tensor hidden_states_37_cast = conv(bias = down_blocks_1_resnets_0_conv1_bias_to_fp16, dilations = var_2198, groups = var_2173, pad = hidden_states_37_pad_0, pad_type = hidden_states_37_pad_type_0, strides = var_2196, weight = down_blocks_1_resnets_0_conv1_weight_to_fp16_palettized, x = input_67_cast)[name = tensor("hidden_states_37_cast")]; + tensor var_2204 = const()[name = tensor("op_2204"), val = tensor([1, 1])]; + tensor var_2206 = const()[name = tensor("op_2206"), val = tensor([1, 1])]; + tensor temb_5_pad_type_0 = const()[name = tensor("temb_5_pad_type_0"), val = tensor("custom")]; + tensor temb_5_pad_0 = const()[name = tensor("temb_5_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_1_resnets_0_time_emb_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(11105216))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(11719680))), name = tensor("down_blocks_1_resnets_0_time_emb_proj_weight_to_fp16_palettized"), shape = tensor([640, 1280, 1, 1])]; + tensor down_blocks_1_resnets_0_time_emb_proj_bias_to_fp16 = const()[name = tensor("down_blocks_1_resnets_0_time_emb_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(11719872)))]; + tensor temb_5_cast = conv(bias = down_blocks_1_resnets_0_time_emb_proj_bias_to_fp16, dilations = var_2206, groups = var_2173, pad = temb_5_pad_0, pad_type = temb_5_pad_type_0, strides = var_2204, weight = down_blocks_1_resnets_0_time_emb_proj_weight_to_fp16_palettized, x = input_15_cast)[name = tensor("temb_5_cast")]; + tensor input_71_cast = add(x = hidden_states_37_cast, y = temb_5_cast)[name = tensor("input_71_cast")]; + tensor reshape_28_shape_0 = const()[name = tensor("reshape_28_shape_0"), val = tensor([2, 32, 20, 32, 32])]; + tensor reshape_28_cast = reshape(shape = reshape_28_shape_0, x = input_71_cast)[name = tensor("reshape_28_cast")]; + tensor reduce_mean_21_axes_0 = const()[name = tensor("reduce_mean_21_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_21_keep_dims_0 = const()[name = tensor("reduce_mean_21_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_21_cast = reduce_mean(axes = reduce_mean_21_axes_0, keep_dims = reduce_mean_21_keep_dims_0, x = reshape_28_cast)[name = tensor("reduce_mean_21_cast")]; + tensor sub_14_cast = sub(x = reshape_28_cast, y = reduce_mean_21_cast)[name = tensor("sub_14_cast")]; + tensor square_7_cast = square(x = sub_14_cast)[name = tensor("square_7_cast")]; + tensor reduce_mean_23_axes_0 = const()[name = tensor("reduce_mean_23_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_23_keep_dims_0 = const()[name = tensor("reduce_mean_23_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_23_cast = reduce_mean(axes = reduce_mean_23_axes_0, keep_dims = reduce_mean_23_keep_dims_0, x = square_7_cast)[name = tensor("reduce_mean_23_cast")]; + tensor add_14_y_0_to_fp16 = const()[name = tensor("add_14_y_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_14_cast = add(x = reduce_mean_23_cast, y = add_14_y_0_to_fp16)[name = tensor("add_14_cast")]; + tensor sqrt_7_cast = sqrt(x = add_14_cast)[name = tensor("sqrt_7_cast")]; + tensor real_div_7_cast = real_div(x = sub_14_cast, y = sqrt_7_cast)[name = tensor("real_div_7_cast")]; + tensor reshape_29_shape_0 = const()[name = tensor("reshape_29_shape_0"), val = tensor([2, 640, 32, 32])]; + tensor reshape_29_cast = reshape(shape = reshape_29_shape_0, x = real_div_7_cast)[name = tensor("reshape_29_cast")]; + tensor add_15_mean_0_to_fp16 = const()[name = tensor("add_15_mean_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(11721216)))]; + tensor add_15_variance_0_to_fp16 = const()[name = tensor("add_15_variance_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(11722560)))]; + tensor add_15_gamma_0_to_fp16 = const()[name = tensor("add_15_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(11723904)))]; + tensor add_15_beta_0_to_fp16 = const()[name = tensor("add_15_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(11725248)))]; + tensor add_15_epsilon_0_to_fp16 = const()[name = tensor("add_15_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_15_cast = batch_norm(beta = add_15_beta_0_to_fp16, epsilon = add_15_epsilon_0_to_fp16, gamma = add_15_gamma_0_to_fp16, mean = add_15_mean_0_to_fp16, variance = add_15_variance_0_to_fp16, x = reshape_29_cast)[name = tensor("add_15_cast")]; + tensor input_75_cast = silu(x = add_15_cast)[name = tensor("input_75_cast")]; + tensor var_2216 = const()[name = tensor("op_2216"), val = tensor([1, 1])]; + tensor var_2218 = const()[name = tensor("op_2218"), val = tensor([1, 1])]; + tensor hidden_states_39_pad_type_0 = const()[name = tensor("hidden_states_39_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_39_pad_0 = const()[name = tensor("hidden_states_39_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor down_blocks_1_resnets_0_conv2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(11726592))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(14491456))), name = tensor("down_blocks_1_resnets_0_conv2_weight_to_fp16_palettized"), shape = tensor([640, 640, 3, 3])]; + tensor down_blocks_1_resnets_0_conv2_bias_to_fp16 = const()[name = tensor("down_blocks_1_resnets_0_conv2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(14491648)))]; + tensor hidden_states_39_cast = conv(bias = down_blocks_1_resnets_0_conv2_bias_to_fp16, dilations = var_2218, groups = var_2173, pad = hidden_states_39_pad_0, pad_type = hidden_states_39_pad_type_0, strides = var_2216, weight = down_blocks_1_resnets_0_conv2_weight_to_fp16_palettized, x = input_75_cast)[name = tensor("hidden_states_39_cast")]; + tensor var_2223 = const()[name = tensor("op_2223"), val = tensor([1, 1])]; + tensor var_2225 = const()[name = tensor("op_2225"), val = tensor([1, 1])]; + tensor x_1_pad_type_0 = const()[name = tensor("x_1_pad_type_0"), val = tensor("custom")]; + tensor x_1_pad_0 = const()[name = tensor("x_1_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_1_resnets_0_conv_shortcut_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(14492992))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(14646656))), name = tensor("down_blocks_1_resnets_0_conv_shortcut_weight_to_fp16_palettized"), shape = tensor([640, 320, 1, 1])]; + tensor down_blocks_1_resnets_0_conv_shortcut_bias_to_fp16 = const()[name = tensor("down_blocks_1_resnets_0_conv_shortcut_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(14646848)))]; + tensor x_1_cast = conv(bias = down_blocks_1_resnets_0_conv_shortcut_bias_to_fp16, dilations = var_2225, groups = var_2173, pad = x_1_pad_0, pad_type = x_1_pad_type_0, strides = var_2223, weight = down_blocks_1_resnets_0_conv_shortcut_weight_to_fp16_palettized, x = input_63_cast)[name = tensor("x_1_cast")]; + tensor hidden_states_41_cast = add(x = x_1_cast, y = hidden_states_39_cast)[name = tensor("hidden_states_41_cast")]; + tensor reshape_32_shape_0 = const()[name = tensor("reshape_32_shape_0"), val = tensor([2, 32, 20, 32, 32])]; + tensor reshape_32_cast = reshape(shape = reshape_32_shape_0, x = hidden_states_41_cast)[name = tensor("reshape_32_cast")]; + tensor reduce_mean_24_axes_0 = const()[name = tensor("reduce_mean_24_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_24_keep_dims_0 = const()[name = tensor("reduce_mean_24_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_24_cast = reduce_mean(axes = reduce_mean_24_axes_0, keep_dims = reduce_mean_24_keep_dims_0, x = reshape_32_cast)[name = tensor("reduce_mean_24_cast")]; + tensor sub_16_cast = sub(x = reshape_32_cast, y = reduce_mean_24_cast)[name = tensor("sub_16_cast")]; + tensor square_8_cast = square(x = sub_16_cast)[name = tensor("square_8_cast")]; + tensor reduce_mean_26_axes_0 = const()[name = tensor("reduce_mean_26_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_26_keep_dims_0 = const()[name = tensor("reduce_mean_26_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_26_cast = reduce_mean(axes = reduce_mean_26_axes_0, keep_dims = reduce_mean_26_keep_dims_0, x = square_8_cast)[name = tensor("reduce_mean_26_cast")]; + tensor add_16_y_0_to_fp16 = const()[name = tensor("add_16_y_0_to_fp16"), val = tensor(0x1.1p-20)]; + tensor add_16_cast = add(x = reduce_mean_26_cast, y = add_16_y_0_to_fp16)[name = tensor("add_16_cast")]; + tensor sqrt_8_cast = sqrt(x = add_16_cast)[name = tensor("sqrt_8_cast")]; + tensor real_div_8_cast = real_div(x = sub_16_cast, y = sqrt_8_cast)[name = tensor("real_div_8_cast")]; + tensor reshape_33_shape_0 = const()[name = tensor("reshape_33_shape_0"), val = tensor([2, 640, 32, 32])]; + tensor reshape_33_cast = reshape(shape = reshape_33_shape_0, x = real_div_8_cast)[name = tensor("reshape_33_cast")]; + tensor add_17_gamma_0_to_fp16 = const()[name = tensor("add_17_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(14648192)))]; + tensor add_17_beta_0_to_fp16 = const()[name = tensor("add_17_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(14649536)))]; + tensor add_17_epsilon_0_to_fp16 = const()[name = tensor("add_17_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_17_cast = batch_norm(beta = add_17_beta_0_to_fp16, epsilon = add_17_epsilon_0_to_fp16, gamma = add_17_gamma_0_to_fp16, mean = add_15_mean_0_to_fp16, variance = add_15_variance_0_to_fp16, x = reshape_33_cast)[name = tensor("add_17_cast")]; + tensor var_2245 = const()[name = tensor("op_2245"), val = tensor([1, 1])]; + tensor var_2247 = const()[name = tensor("op_2247"), val = tensor([1, 1])]; + tensor hidden_states_43_pad_type_0 = const()[name = tensor("hidden_states_43_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_43_pad_0 = const()[name = tensor("hidden_states_43_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_1_attentions_0_proj_in_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(14650880))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(14958144))), name = tensor("down_blocks_1_attentions_0_proj_in_weight_to_fp16_palettized"), shape = tensor([640, 640, 1, 1])]; + tensor down_blocks_1_attentions_0_proj_in_bias_to_fp16 = const()[name = tensor("down_blocks_1_attentions_0_proj_in_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(14958336)))]; + tensor hidden_states_43_cast = conv(bias = down_blocks_1_attentions_0_proj_in_bias_to_fp16, dilations = var_2247, groups = var_2173, pad = hidden_states_43_pad_0, pad_type = hidden_states_43_pad_type_0, strides = var_2245, weight = down_blocks_1_attentions_0_proj_in_weight_to_fp16_palettized, x = add_17_cast)[name = tensor("hidden_states_43_cast")]; + tensor var_2252 = const()[name = tensor("op_2252"), val = tensor([2, 640, 1, 1024])]; + tensor inputs_13_cast = reshape(shape = var_2252, x = hidden_states_43_cast)[name = tensor("inputs_13_cast")]; + tensor var_2262 = const()[name = tensor("op_2262"), val = tensor([1])]; + tensor channels_mean_13_cast = reduce_mean(axes = var_2262, keep_dims = var_2168, x = inputs_13_cast)[name = tensor("channels_mean_13_cast")]; + tensor zero_mean_13_cast = sub(x = inputs_13_cast, y = channels_mean_13_cast)[name = tensor("zero_mean_13_cast")]; + tensor zero_mean_sq_13_cast = mul(x = zero_mean_13_cast, y = zero_mean_13_cast)[name = tensor("zero_mean_sq_13_cast")]; + tensor var_2266 = const()[name = tensor("op_2266"), val = tensor([1])]; + tensor var_2267_cast = reduce_mean(axes = var_2266, keep_dims = var_2168, x = zero_mean_sq_13_cast)[name = tensor("op_2267_cast")]; + tensor var_2268_to_fp16 = const()[name = tensor("op_2268_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_2269_cast = add(x = var_2267_cast, y = var_2268_to_fp16)[name = tensor("op_2269_cast")]; + tensor denom_13_epsilon_0_to_fp16 = const()[name = tensor("denom_13_epsilon_0_to_fp16"), val = tensor(0x1p-24)]; + tensor denom_13_cast = rsqrt(epsilon = denom_13_epsilon_0_to_fp16, x = var_2269_cast)[name = tensor("denom_13_cast")]; + tensor out_13_cast = mul(x = zero_mean_13_cast, y = denom_13_cast)[name = tensor("out_13_cast")]; + tensor var_2273_to_fp16 = const()[name = tensor("op_2273_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(14959680)))]; + tensor var_2274_cast = add(x = out_13_cast, y = var_2273_to_fp16)[name = tensor("op_2274_cast")]; + tensor var_2276_to_fp16 = const()[name = tensor("op_2276_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(14961024)))]; + tensor hidden_states_45_cast = mul(x = var_2274_cast, y = var_2276_to_fp16)[name = tensor("hidden_states_45_cast")]; + tensor var_2283 = const()[name = tensor("op_2283"), val = tensor([1, 1])]; + tensor var_2285 = const()[name = tensor("op_2285"), val = tensor([1, 1])]; + tensor q_9_pad_type_0 = const()[name = tensor("q_9_pad_type_0"), val = tensor("custom")]; + tensor q_9_pad_0 = const()[name = tensor("q_9_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_1_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(14962368))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(15269632))), name = tensor("down_blocks_1_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16_palettized"), shape = tensor([640, 640, 1, 1])]; + tensor q_9_cast = conv(dilations = var_2285, groups = var_2173, pad = q_9_pad_0, pad_type = q_9_pad_type_0, strides = var_2283, weight = down_blocks_1_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16_palettized, x = hidden_states_45_cast)[name = tensor("q_9_cast")]; + tensor var_2289 = const()[name = tensor("op_2289"), val = tensor([1, 1])]; + tensor var_2291 = const()[name = tensor("op_2291"), val = tensor([1, 1])]; + tensor k_17_pad_type_0 = const()[name = tensor("k_17_pad_type_0"), val = tensor("custom")]; + tensor k_17_pad_0 = const()[name = tensor("k_17_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_1_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(15269824))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(15577088))), name = tensor("down_blocks_1_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16_palettized"), shape = tensor([640, 640, 1, 1])]; + tensor k_17_cast = conv(dilations = var_2291, groups = var_2173, pad = k_17_pad_0, pad_type = k_17_pad_type_0, strides = var_2289, weight = down_blocks_1_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16_palettized, x = hidden_states_45_cast)[name = tensor("k_17_cast")]; + tensor var_2295 = const()[name = tensor("op_2295"), val = tensor([1, 1])]; + tensor var_2297 = const()[name = tensor("op_2297"), val = tensor([1, 1])]; + tensor v_9_pad_type_0 = const()[name = tensor("v_9_pad_type_0"), val = tensor("custom")]; + tensor v_9_pad_0 = const()[name = tensor("v_9_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_1_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(15577280))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(15884544))), name = tensor("down_blocks_1_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16_palettized"), shape = tensor([640, 640, 1, 1])]; + tensor v_9_cast = conv(dilations = var_2297, groups = var_2173, pad = v_9_pad_0, pad_type = v_9_pad_type_0, strides = var_2295, weight = down_blocks_1_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16_palettized, x = hidden_states_45_cast)[name = tensor("v_9_cast")]; + tensor var_2301_begin_0 = const()[name = tensor("op_2301_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2301_end_0 = const()[name = tensor("op_2301_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_2301_end_mask_0 = const()[name = tensor("op_2301_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2301_cast = slice_by_index(begin = var_2301_begin_0, end = var_2301_end_0, end_mask = var_2301_end_mask_0, x = q_9_cast)[name = tensor("op_2301_cast")]; + tensor var_2305_begin_0 = const()[name = tensor("op_2305_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_2305_end_0 = const()[name = tensor("op_2305_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_2305_end_mask_0 = const()[name = tensor("op_2305_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2305_cast = slice_by_index(begin = var_2305_begin_0, end = var_2305_end_0, end_mask = var_2305_end_mask_0, x = q_9_cast)[name = tensor("op_2305_cast")]; + tensor var_2309_begin_0 = const()[name = tensor("op_2309_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_2309_end_0 = const()[name = tensor("op_2309_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_2309_end_mask_0 = const()[name = tensor("op_2309_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2309_cast = slice_by_index(begin = var_2309_begin_0, end = var_2309_end_0, end_mask = var_2309_end_mask_0, x = q_9_cast)[name = tensor("op_2309_cast")]; + tensor var_2313_begin_0 = const()[name = tensor("op_2313_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_2313_end_0 = const()[name = tensor("op_2313_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_2313_end_mask_0 = const()[name = tensor("op_2313_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2313_cast = slice_by_index(begin = var_2313_begin_0, end = var_2313_end_0, end_mask = var_2313_end_mask_0, x = q_9_cast)[name = tensor("op_2313_cast")]; + tensor var_2317_begin_0 = const()[name = tensor("op_2317_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_2317_end_0 = const()[name = tensor("op_2317_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_2317_end_mask_0 = const()[name = tensor("op_2317_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2317_cast = slice_by_index(begin = var_2317_begin_0, end = var_2317_end_0, end_mask = var_2317_end_mask_0, x = q_9_cast)[name = tensor("op_2317_cast")]; + tensor var_2321_begin_0 = const()[name = tensor("op_2321_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_2321_end_0 = const()[name = tensor("op_2321_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_2321_end_mask_0 = const()[name = tensor("op_2321_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2321_cast = slice_by_index(begin = var_2321_begin_0, end = var_2321_end_0, end_mask = var_2321_end_mask_0, x = q_9_cast)[name = tensor("op_2321_cast")]; + tensor var_2325_begin_0 = const()[name = tensor("op_2325_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_2325_end_0 = const()[name = tensor("op_2325_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_2325_end_mask_0 = const()[name = tensor("op_2325_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2325_cast = slice_by_index(begin = var_2325_begin_0, end = var_2325_end_0, end_mask = var_2325_end_mask_0, x = q_9_cast)[name = tensor("op_2325_cast")]; + tensor var_2329_begin_0 = const()[name = tensor("op_2329_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_2329_end_0 = const()[name = tensor("op_2329_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_2329_end_mask_0 = const()[name = tensor("op_2329_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2329_cast = slice_by_index(begin = var_2329_begin_0, end = var_2329_end_0, end_mask = var_2329_end_mask_0, x = q_9_cast)[name = tensor("op_2329_cast")]; + tensor var_2333_begin_0 = const()[name = tensor("op_2333_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_2333_end_0 = const()[name = tensor("op_2333_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_2333_end_mask_0 = const()[name = tensor("op_2333_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2333_cast = slice_by_index(begin = var_2333_begin_0, end = var_2333_end_0, end_mask = var_2333_end_mask_0, x = q_9_cast)[name = tensor("op_2333_cast")]; + tensor var_2337_begin_0 = const()[name = tensor("op_2337_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_2337_end_0 = const()[name = tensor("op_2337_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_2337_end_mask_0 = const()[name = tensor("op_2337_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2337_cast = slice_by_index(begin = var_2337_begin_0, end = var_2337_end_0, end_mask = var_2337_end_mask_0, x = q_9_cast)[name = tensor("op_2337_cast")]; + tensor var_2340_begin_0 = const()[name = tensor("op_2340_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2340_end_0 = const()[name = tensor("op_2340_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_2340_end_mask_0 = const()[name = tensor("op_2340_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2340_cast = slice_by_index(begin = var_2340_begin_0, end = var_2340_end_0, end_mask = var_2340_end_mask_0, x = var_2301_cast)[name = tensor("op_2340_cast")]; + tensor var_2341_begin_0 = const()[name = tensor("op_2341_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_2341_end_0 = const()[name = tensor("op_2341_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_2341_end_mask_0 = const()[name = tensor("op_2341_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2341_cast = slice_by_index(begin = var_2341_begin_0, end = var_2341_end_0, end_mask = var_2341_end_mask_0, x = var_2301_cast)[name = tensor("op_2341_cast")]; + tensor var_2342_begin_0 = const()[name = tensor("op_2342_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2342_end_0 = const()[name = tensor("op_2342_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_2342_end_mask_0 = const()[name = tensor("op_2342_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2342_cast = slice_by_index(begin = var_2342_begin_0, end = var_2342_end_0, end_mask = var_2342_end_mask_0, x = var_2305_cast)[name = tensor("op_2342_cast")]; + tensor var_2343_begin_0 = const()[name = tensor("op_2343_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_2343_end_0 = const()[name = tensor("op_2343_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_2343_end_mask_0 = const()[name = tensor("op_2343_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2343_cast = slice_by_index(begin = var_2343_begin_0, end = var_2343_end_0, end_mask = var_2343_end_mask_0, x = var_2305_cast)[name = tensor("op_2343_cast")]; + tensor var_2344_begin_0 = const()[name = tensor("op_2344_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2344_end_0 = const()[name = tensor("op_2344_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_2344_end_mask_0 = const()[name = tensor("op_2344_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2344_cast = slice_by_index(begin = var_2344_begin_0, end = var_2344_end_0, end_mask = var_2344_end_mask_0, x = var_2309_cast)[name = tensor("op_2344_cast")]; + tensor var_2345_begin_0 = const()[name = tensor("op_2345_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_2345_end_0 = const()[name = tensor("op_2345_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_2345_end_mask_0 = const()[name = tensor("op_2345_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2345_cast = slice_by_index(begin = var_2345_begin_0, end = var_2345_end_0, end_mask = var_2345_end_mask_0, x = var_2309_cast)[name = tensor("op_2345_cast")]; + tensor var_2346_begin_0 = const()[name = tensor("op_2346_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2346_end_0 = const()[name = tensor("op_2346_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_2346_end_mask_0 = const()[name = tensor("op_2346_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2346_cast = slice_by_index(begin = var_2346_begin_0, end = var_2346_end_0, end_mask = var_2346_end_mask_0, x = var_2313_cast)[name = tensor("op_2346_cast")]; + tensor var_2347_begin_0 = const()[name = tensor("op_2347_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_2347_end_0 = const()[name = tensor("op_2347_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_2347_end_mask_0 = const()[name = tensor("op_2347_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2347_cast = slice_by_index(begin = var_2347_begin_0, end = var_2347_end_0, end_mask = var_2347_end_mask_0, x = var_2313_cast)[name = tensor("op_2347_cast")]; + tensor var_2348_begin_0 = const()[name = tensor("op_2348_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2348_end_0 = const()[name = tensor("op_2348_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_2348_end_mask_0 = const()[name = tensor("op_2348_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2348_cast = slice_by_index(begin = var_2348_begin_0, end = var_2348_end_0, end_mask = var_2348_end_mask_0, x = var_2317_cast)[name = tensor("op_2348_cast")]; + tensor var_2349_begin_0 = const()[name = tensor("op_2349_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_2349_end_0 = const()[name = tensor("op_2349_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_2349_end_mask_0 = const()[name = tensor("op_2349_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2349_cast = slice_by_index(begin = var_2349_begin_0, end = var_2349_end_0, end_mask = var_2349_end_mask_0, x = var_2317_cast)[name = tensor("op_2349_cast")]; + tensor var_2350_begin_0 = const()[name = tensor("op_2350_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2350_end_0 = const()[name = tensor("op_2350_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_2350_end_mask_0 = const()[name = tensor("op_2350_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2350_cast = slice_by_index(begin = var_2350_begin_0, end = var_2350_end_0, end_mask = var_2350_end_mask_0, x = var_2321_cast)[name = tensor("op_2350_cast")]; + tensor var_2351_begin_0 = const()[name = tensor("op_2351_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_2351_end_0 = const()[name = tensor("op_2351_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_2351_end_mask_0 = const()[name = tensor("op_2351_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2351_cast = slice_by_index(begin = var_2351_begin_0, end = var_2351_end_0, end_mask = var_2351_end_mask_0, x = var_2321_cast)[name = tensor("op_2351_cast")]; + tensor var_2352_begin_0 = const()[name = tensor("op_2352_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2352_end_0 = const()[name = tensor("op_2352_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_2352_end_mask_0 = const()[name = tensor("op_2352_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2352_cast = slice_by_index(begin = var_2352_begin_0, end = var_2352_end_0, end_mask = var_2352_end_mask_0, x = var_2325_cast)[name = tensor("op_2352_cast")]; + tensor var_2353_begin_0 = const()[name = tensor("op_2353_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_2353_end_0 = const()[name = tensor("op_2353_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_2353_end_mask_0 = const()[name = tensor("op_2353_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2353_cast = slice_by_index(begin = var_2353_begin_0, end = var_2353_end_0, end_mask = var_2353_end_mask_0, x = var_2325_cast)[name = tensor("op_2353_cast")]; + tensor var_2354_begin_0 = const()[name = tensor("op_2354_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2354_end_0 = const()[name = tensor("op_2354_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_2354_end_mask_0 = const()[name = tensor("op_2354_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2354_cast = slice_by_index(begin = var_2354_begin_0, end = var_2354_end_0, end_mask = var_2354_end_mask_0, x = var_2329_cast)[name = tensor("op_2354_cast")]; + tensor var_2355_begin_0 = const()[name = tensor("op_2355_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_2355_end_0 = const()[name = tensor("op_2355_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_2355_end_mask_0 = const()[name = tensor("op_2355_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2355_cast = slice_by_index(begin = var_2355_begin_0, end = var_2355_end_0, end_mask = var_2355_end_mask_0, x = var_2329_cast)[name = tensor("op_2355_cast")]; + tensor var_2356_begin_0 = const()[name = tensor("op_2356_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2356_end_0 = const()[name = tensor("op_2356_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_2356_end_mask_0 = const()[name = tensor("op_2356_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2356_cast = slice_by_index(begin = var_2356_begin_0, end = var_2356_end_0, end_mask = var_2356_end_mask_0, x = var_2333_cast)[name = tensor("op_2356_cast")]; + tensor var_2357_begin_0 = const()[name = tensor("op_2357_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_2357_end_0 = const()[name = tensor("op_2357_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_2357_end_mask_0 = const()[name = tensor("op_2357_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2357_cast = slice_by_index(begin = var_2357_begin_0, end = var_2357_end_0, end_mask = var_2357_end_mask_0, x = var_2333_cast)[name = tensor("op_2357_cast")]; + tensor var_2358_begin_0 = const()[name = tensor("op_2358_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2358_end_0 = const()[name = tensor("op_2358_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_2358_end_mask_0 = const()[name = tensor("op_2358_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2358_cast = slice_by_index(begin = var_2358_begin_0, end = var_2358_end_0, end_mask = var_2358_end_mask_0, x = var_2337_cast)[name = tensor("op_2358_cast")]; + tensor var_2359_begin_0 = const()[name = tensor("op_2359_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_2359_end_0 = const()[name = tensor("op_2359_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_2359_end_mask_0 = const()[name = tensor("op_2359_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2359_cast = slice_by_index(begin = var_2359_begin_0, end = var_2359_end_0, end_mask = var_2359_end_mask_0, x = var_2337_cast)[name = tensor("op_2359_cast")]; + tensor k_19_perm_0 = const()[name = tensor("k_19_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_2364_begin_0 = const()[name = tensor("op_2364_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2364_end_0 = const()[name = tensor("op_2364_end_0"), val = tensor([2, 1024, 1, 64])]; + tensor var_2364_end_mask_0 = const()[name = tensor("op_2364_end_mask_0"), val = tensor([true, true, true, false])]; + tensor transpose_27 = transpose(perm = k_19_perm_0, x = k_17_cast)[name = tensor("transpose_27")]; + tensor var_2364_cast = slice_by_index(begin = var_2364_begin_0, end = var_2364_end_0, end_mask = var_2364_end_mask_0, x = transpose_27)[name = tensor("op_2364_cast")]; + tensor var_2368_begin_0 = const()[name = tensor("op_2368_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_2368_end_0 = const()[name = tensor("op_2368_end_0"), val = tensor([2, 1024, 1, 128])]; + tensor var_2368_end_mask_0 = const()[name = tensor("op_2368_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2368_cast = slice_by_index(begin = var_2368_begin_0, end = var_2368_end_0, end_mask = var_2368_end_mask_0, x = transpose_27)[name = tensor("op_2368_cast")]; + tensor var_2372_begin_0 = const()[name = tensor("op_2372_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_2372_end_0 = const()[name = tensor("op_2372_end_0"), val = tensor([2, 1024, 1, 192])]; + tensor var_2372_end_mask_0 = const()[name = tensor("op_2372_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2372_cast = slice_by_index(begin = var_2372_begin_0, end = var_2372_end_0, end_mask = var_2372_end_mask_0, x = transpose_27)[name = tensor("op_2372_cast")]; + tensor var_2376_begin_0 = const()[name = tensor("op_2376_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_2376_end_0 = const()[name = tensor("op_2376_end_0"), val = tensor([2, 1024, 1, 256])]; + tensor var_2376_end_mask_0 = const()[name = tensor("op_2376_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2376_cast = slice_by_index(begin = var_2376_begin_0, end = var_2376_end_0, end_mask = var_2376_end_mask_0, x = transpose_27)[name = tensor("op_2376_cast")]; + tensor var_2380_begin_0 = const()[name = tensor("op_2380_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_2380_end_0 = const()[name = tensor("op_2380_end_0"), val = tensor([2, 1024, 1, 320])]; + tensor var_2380_end_mask_0 = const()[name = tensor("op_2380_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2380_cast = slice_by_index(begin = var_2380_begin_0, end = var_2380_end_0, end_mask = var_2380_end_mask_0, x = transpose_27)[name = tensor("op_2380_cast")]; + tensor var_2384_begin_0 = const()[name = tensor("op_2384_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_2384_end_0 = const()[name = tensor("op_2384_end_0"), val = tensor([2, 1024, 1, 384])]; + tensor var_2384_end_mask_0 = const()[name = tensor("op_2384_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2384_cast = slice_by_index(begin = var_2384_begin_0, end = var_2384_end_0, end_mask = var_2384_end_mask_0, x = transpose_27)[name = tensor("op_2384_cast")]; + tensor var_2388_begin_0 = const()[name = tensor("op_2388_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_2388_end_0 = const()[name = tensor("op_2388_end_0"), val = tensor([2, 1024, 1, 448])]; + tensor var_2388_end_mask_0 = const()[name = tensor("op_2388_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2388_cast = slice_by_index(begin = var_2388_begin_0, end = var_2388_end_0, end_mask = var_2388_end_mask_0, x = transpose_27)[name = tensor("op_2388_cast")]; + tensor var_2392_begin_0 = const()[name = tensor("op_2392_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_2392_end_0 = const()[name = tensor("op_2392_end_0"), val = tensor([2, 1024, 1, 512])]; + tensor var_2392_end_mask_0 = const()[name = tensor("op_2392_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2392_cast = slice_by_index(begin = var_2392_begin_0, end = var_2392_end_0, end_mask = var_2392_end_mask_0, x = transpose_27)[name = tensor("op_2392_cast")]; + tensor var_2396_begin_0 = const()[name = tensor("op_2396_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_2396_end_0 = const()[name = tensor("op_2396_end_0"), val = tensor([2, 1024, 1, 576])]; + tensor var_2396_end_mask_0 = const()[name = tensor("op_2396_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2396_cast = slice_by_index(begin = var_2396_begin_0, end = var_2396_end_0, end_mask = var_2396_end_mask_0, x = transpose_27)[name = tensor("op_2396_cast")]; + tensor var_2400_begin_0 = const()[name = tensor("op_2400_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_2400_end_0 = const()[name = tensor("op_2400_end_0"), val = tensor([2, 1024, 1, 640])]; + tensor var_2400_end_mask_0 = const()[name = tensor("op_2400_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2400_cast = slice_by_index(begin = var_2400_begin_0, end = var_2400_end_0, end_mask = var_2400_end_mask_0, x = transpose_27)[name = tensor("op_2400_cast")]; + tensor var_2402_begin_0 = const()[name = tensor("op_2402_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2402_end_0 = const()[name = tensor("op_2402_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_2402_end_mask_0 = const()[name = tensor("op_2402_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2402_cast = slice_by_index(begin = var_2402_begin_0, end = var_2402_end_0, end_mask = var_2402_end_mask_0, x = v_9_cast)[name = tensor("op_2402_cast")]; + tensor var_2406_begin_0 = const()[name = tensor("op_2406_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_2406_end_0 = const()[name = tensor("op_2406_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_2406_end_mask_0 = const()[name = tensor("op_2406_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2406_cast = slice_by_index(begin = var_2406_begin_0, end = var_2406_end_0, end_mask = var_2406_end_mask_0, x = v_9_cast)[name = tensor("op_2406_cast")]; + tensor var_2410_begin_0 = const()[name = tensor("op_2410_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_2410_end_0 = const()[name = tensor("op_2410_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_2410_end_mask_0 = const()[name = tensor("op_2410_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2410_cast = slice_by_index(begin = var_2410_begin_0, end = var_2410_end_0, end_mask = var_2410_end_mask_0, x = v_9_cast)[name = tensor("op_2410_cast")]; + tensor var_2414_begin_0 = const()[name = tensor("op_2414_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_2414_end_0 = const()[name = tensor("op_2414_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_2414_end_mask_0 = const()[name = tensor("op_2414_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2414_cast = slice_by_index(begin = var_2414_begin_0, end = var_2414_end_0, end_mask = var_2414_end_mask_0, x = v_9_cast)[name = tensor("op_2414_cast")]; + tensor var_2418_begin_0 = const()[name = tensor("op_2418_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_2418_end_0 = const()[name = tensor("op_2418_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_2418_end_mask_0 = const()[name = tensor("op_2418_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2418_cast = slice_by_index(begin = var_2418_begin_0, end = var_2418_end_0, end_mask = var_2418_end_mask_0, x = v_9_cast)[name = tensor("op_2418_cast")]; + tensor var_2422_begin_0 = const()[name = tensor("op_2422_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_2422_end_0 = const()[name = tensor("op_2422_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_2422_end_mask_0 = const()[name = tensor("op_2422_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2422_cast = slice_by_index(begin = var_2422_begin_0, end = var_2422_end_0, end_mask = var_2422_end_mask_0, x = v_9_cast)[name = tensor("op_2422_cast")]; + tensor var_2426_begin_0 = const()[name = tensor("op_2426_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_2426_end_0 = const()[name = tensor("op_2426_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_2426_end_mask_0 = const()[name = tensor("op_2426_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2426_cast = slice_by_index(begin = var_2426_begin_0, end = var_2426_end_0, end_mask = var_2426_end_mask_0, x = v_9_cast)[name = tensor("op_2426_cast")]; + tensor var_2430_begin_0 = const()[name = tensor("op_2430_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_2430_end_0 = const()[name = tensor("op_2430_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_2430_end_mask_0 = const()[name = tensor("op_2430_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2430_cast = slice_by_index(begin = var_2430_begin_0, end = var_2430_end_0, end_mask = var_2430_end_mask_0, x = v_9_cast)[name = tensor("op_2430_cast")]; + tensor var_2434_begin_0 = const()[name = tensor("op_2434_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_2434_end_0 = const()[name = tensor("op_2434_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_2434_end_mask_0 = const()[name = tensor("op_2434_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2434_cast = slice_by_index(begin = var_2434_begin_0, end = var_2434_end_0, end_mask = var_2434_end_mask_0, x = v_9_cast)[name = tensor("op_2434_cast")]; + tensor var_2438_begin_0 = const()[name = tensor("op_2438_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_2438_end_0 = const()[name = tensor("op_2438_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_2438_end_mask_0 = const()[name = tensor("op_2438_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2438_cast = slice_by_index(begin = var_2438_begin_0, end = var_2438_end_0, end_mask = var_2438_end_mask_0, x = v_9_cast)[name = tensor("op_2438_cast")]; + tensor var_2442_equation_0 = const()[name = tensor("op_2442_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2442_cast = einsum(equation = var_2442_equation_0, values = (var_2364_cast, var_2340_cast))[name = tensor("op_2442_cast")]; + tensor var_2443_to_fp16 = const()[name = tensor("op_2443_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_321_cast = mul(x = var_2442_cast, y = var_2443_to_fp16)[name = tensor("aw_chunk_321_cast")]; + tensor var_2446_equation_0 = const()[name = tensor("op_2446_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2446_cast = einsum(equation = var_2446_equation_0, values = (var_2364_cast, var_2341_cast))[name = tensor("op_2446_cast")]; + tensor var_2447_to_fp16 = const()[name = tensor("op_2447_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_323_cast = mul(x = var_2446_cast, y = var_2447_to_fp16)[name = tensor("aw_chunk_323_cast")]; + tensor var_2450_equation_0 = const()[name = tensor("op_2450_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2450_cast = einsum(equation = var_2450_equation_0, values = (var_2368_cast, var_2342_cast))[name = tensor("op_2450_cast")]; + tensor var_2451_to_fp16 = const()[name = tensor("op_2451_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_325_cast = mul(x = var_2450_cast, y = var_2451_to_fp16)[name = tensor("aw_chunk_325_cast")]; + tensor var_2454_equation_0 = const()[name = tensor("op_2454_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2454_cast = einsum(equation = var_2454_equation_0, values = (var_2368_cast, var_2343_cast))[name = tensor("op_2454_cast")]; + tensor var_2455_to_fp16 = const()[name = tensor("op_2455_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_327_cast = mul(x = var_2454_cast, y = var_2455_to_fp16)[name = tensor("aw_chunk_327_cast")]; + tensor var_2458_equation_0 = const()[name = tensor("op_2458_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2458_cast = einsum(equation = var_2458_equation_0, values = (var_2372_cast, var_2344_cast))[name = tensor("op_2458_cast")]; + tensor var_2459_to_fp16 = const()[name = tensor("op_2459_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_329_cast = mul(x = var_2458_cast, y = var_2459_to_fp16)[name = tensor("aw_chunk_329_cast")]; + tensor var_2462_equation_0 = const()[name = tensor("op_2462_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2462_cast = einsum(equation = var_2462_equation_0, values = (var_2372_cast, var_2345_cast))[name = tensor("op_2462_cast")]; + tensor var_2463_to_fp16 = const()[name = tensor("op_2463_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_331_cast = mul(x = var_2462_cast, y = var_2463_to_fp16)[name = tensor("aw_chunk_331_cast")]; + tensor var_2466_equation_0 = const()[name = tensor("op_2466_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2466_cast = einsum(equation = var_2466_equation_0, values = (var_2376_cast, var_2346_cast))[name = tensor("op_2466_cast")]; + tensor var_2467_to_fp16 = const()[name = tensor("op_2467_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_333_cast = mul(x = var_2466_cast, y = var_2467_to_fp16)[name = tensor("aw_chunk_333_cast")]; + tensor var_2470_equation_0 = const()[name = tensor("op_2470_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2470_cast = einsum(equation = var_2470_equation_0, values = (var_2376_cast, var_2347_cast))[name = tensor("op_2470_cast")]; + tensor var_2471_to_fp16 = const()[name = tensor("op_2471_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_335_cast = mul(x = var_2470_cast, y = var_2471_to_fp16)[name = tensor("aw_chunk_335_cast")]; + tensor var_2474_equation_0 = const()[name = tensor("op_2474_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2474_cast = einsum(equation = var_2474_equation_0, values = (var_2380_cast, var_2348_cast))[name = tensor("op_2474_cast")]; + tensor var_2475_to_fp16 = const()[name = tensor("op_2475_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_337_cast = mul(x = var_2474_cast, y = var_2475_to_fp16)[name = tensor("aw_chunk_337_cast")]; + tensor var_2478_equation_0 = const()[name = tensor("op_2478_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2478_cast = einsum(equation = var_2478_equation_0, values = (var_2380_cast, var_2349_cast))[name = tensor("op_2478_cast")]; + tensor var_2479_to_fp16 = const()[name = tensor("op_2479_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_339_cast = mul(x = var_2478_cast, y = var_2479_to_fp16)[name = tensor("aw_chunk_339_cast")]; + tensor var_2482_equation_0 = const()[name = tensor("op_2482_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2482_cast = einsum(equation = var_2482_equation_0, values = (var_2384_cast, var_2350_cast))[name = tensor("op_2482_cast")]; + tensor var_2483_to_fp16 = const()[name = tensor("op_2483_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_341_cast = mul(x = var_2482_cast, y = var_2483_to_fp16)[name = tensor("aw_chunk_341_cast")]; + tensor var_2486_equation_0 = const()[name = tensor("op_2486_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2486_cast = einsum(equation = var_2486_equation_0, values = (var_2384_cast, var_2351_cast))[name = tensor("op_2486_cast")]; + tensor var_2487_to_fp16 = const()[name = tensor("op_2487_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_343_cast = mul(x = var_2486_cast, y = var_2487_to_fp16)[name = tensor("aw_chunk_343_cast")]; + tensor var_2490_equation_0 = const()[name = tensor("op_2490_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2490_cast = einsum(equation = var_2490_equation_0, values = (var_2388_cast, var_2352_cast))[name = tensor("op_2490_cast")]; + tensor var_2491_to_fp16 = const()[name = tensor("op_2491_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_345_cast = mul(x = var_2490_cast, y = var_2491_to_fp16)[name = tensor("aw_chunk_345_cast")]; + tensor var_2494_equation_0 = const()[name = tensor("op_2494_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2494_cast = einsum(equation = var_2494_equation_0, values = (var_2388_cast, var_2353_cast))[name = tensor("op_2494_cast")]; + tensor var_2495_to_fp16 = const()[name = tensor("op_2495_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_347_cast = mul(x = var_2494_cast, y = var_2495_to_fp16)[name = tensor("aw_chunk_347_cast")]; + tensor var_2498_equation_0 = const()[name = tensor("op_2498_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2498_cast = einsum(equation = var_2498_equation_0, values = (var_2392_cast, var_2354_cast))[name = tensor("op_2498_cast")]; + tensor var_2499_to_fp16 = const()[name = tensor("op_2499_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_349_cast = mul(x = var_2498_cast, y = var_2499_to_fp16)[name = tensor("aw_chunk_349_cast")]; + tensor var_2502_equation_0 = const()[name = tensor("op_2502_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2502_cast = einsum(equation = var_2502_equation_0, values = (var_2392_cast, var_2355_cast))[name = tensor("op_2502_cast")]; + tensor var_2503_to_fp16 = const()[name = tensor("op_2503_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_351_cast = mul(x = var_2502_cast, y = var_2503_to_fp16)[name = tensor("aw_chunk_351_cast")]; + tensor var_2506_equation_0 = const()[name = tensor("op_2506_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2506_cast = einsum(equation = var_2506_equation_0, values = (var_2396_cast, var_2356_cast))[name = tensor("op_2506_cast")]; + tensor var_2507_to_fp16 = const()[name = tensor("op_2507_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_353_cast = mul(x = var_2506_cast, y = var_2507_to_fp16)[name = tensor("aw_chunk_353_cast")]; + tensor var_2510_equation_0 = const()[name = tensor("op_2510_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2510_cast = einsum(equation = var_2510_equation_0, values = (var_2396_cast, var_2357_cast))[name = tensor("op_2510_cast")]; + tensor var_2511_to_fp16 = const()[name = tensor("op_2511_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_355_cast = mul(x = var_2510_cast, y = var_2511_to_fp16)[name = tensor("aw_chunk_355_cast")]; + tensor var_2514_equation_0 = const()[name = tensor("op_2514_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2514_cast = einsum(equation = var_2514_equation_0, values = (var_2400_cast, var_2358_cast))[name = tensor("op_2514_cast")]; + tensor var_2515_to_fp16 = const()[name = tensor("op_2515_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_357_cast = mul(x = var_2514_cast, y = var_2515_to_fp16)[name = tensor("aw_chunk_357_cast")]; + tensor var_2518_equation_0 = const()[name = tensor("op_2518_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2518_cast = einsum(equation = var_2518_equation_0, values = (var_2400_cast, var_2359_cast))[name = tensor("op_2518_cast")]; + tensor var_2519_to_fp16 = const()[name = tensor("op_2519_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_359_cast = mul(x = var_2518_cast, y = var_2519_to_fp16)[name = tensor("aw_chunk_359_cast")]; + tensor var_2521_cast = softmax(axis = var_2173, x = aw_chunk_321_cast)[name = tensor("op_2521_cast")]; + tensor var_2522_cast = softmax(axis = var_2173, x = aw_chunk_323_cast)[name = tensor("op_2522_cast")]; + tensor var_2523_cast = softmax(axis = var_2173, x = aw_chunk_325_cast)[name = tensor("op_2523_cast")]; + tensor var_2524_cast = softmax(axis = var_2173, x = aw_chunk_327_cast)[name = tensor("op_2524_cast")]; + tensor var_2525_cast = softmax(axis = var_2173, x = aw_chunk_329_cast)[name = tensor("op_2525_cast")]; + tensor var_2526_cast = softmax(axis = var_2173, x = aw_chunk_331_cast)[name = tensor("op_2526_cast")]; + tensor var_2527_cast = softmax(axis = var_2173, x = aw_chunk_333_cast)[name = tensor("op_2527_cast")]; + tensor var_2528_cast = softmax(axis = var_2173, x = aw_chunk_335_cast)[name = tensor("op_2528_cast")]; + tensor var_2529_cast = softmax(axis = var_2173, x = aw_chunk_337_cast)[name = tensor("op_2529_cast")]; + tensor var_2530_cast = softmax(axis = var_2173, x = aw_chunk_339_cast)[name = tensor("op_2530_cast")]; + tensor var_2531_cast = softmax(axis = var_2173, x = aw_chunk_341_cast)[name = tensor("op_2531_cast")]; + tensor var_2532_cast = softmax(axis = var_2173, x = aw_chunk_343_cast)[name = tensor("op_2532_cast")]; + tensor var_2533_cast = softmax(axis = var_2173, x = aw_chunk_345_cast)[name = tensor("op_2533_cast")]; + tensor var_2534_cast = softmax(axis = var_2173, x = aw_chunk_347_cast)[name = tensor("op_2534_cast")]; + tensor var_2535_cast = softmax(axis = var_2173, x = aw_chunk_349_cast)[name = tensor("op_2535_cast")]; + tensor var_2536_cast = softmax(axis = var_2173, x = aw_chunk_351_cast)[name = tensor("op_2536_cast")]; + tensor var_2537_cast = softmax(axis = var_2173, x = aw_chunk_353_cast)[name = tensor("op_2537_cast")]; + tensor var_2538_cast = softmax(axis = var_2173, x = aw_chunk_355_cast)[name = tensor("op_2538_cast")]; + tensor var_2539_cast = softmax(axis = var_2173, x = aw_chunk_357_cast)[name = tensor("op_2539_cast")]; + tensor var_2540_cast = softmax(axis = var_2173, x = aw_chunk_359_cast)[name = tensor("op_2540_cast")]; + tensor var_2542_equation_0 = const()[name = tensor("op_2542_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2542_cast = einsum(equation = var_2542_equation_0, values = (var_2402_cast, var_2521_cast))[name = tensor("op_2542_cast")]; + tensor var_2544_equation_0 = const()[name = tensor("op_2544_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2544_cast = einsum(equation = var_2544_equation_0, values = (var_2402_cast, var_2522_cast))[name = tensor("op_2544_cast")]; + tensor var_2546_equation_0 = const()[name = tensor("op_2546_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2546_cast = einsum(equation = var_2546_equation_0, values = (var_2406_cast, var_2523_cast))[name = tensor("op_2546_cast")]; + tensor var_2548_equation_0 = const()[name = tensor("op_2548_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2548_cast = einsum(equation = var_2548_equation_0, values = (var_2406_cast, var_2524_cast))[name = tensor("op_2548_cast")]; + tensor var_2550_equation_0 = const()[name = tensor("op_2550_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2550_cast = einsum(equation = var_2550_equation_0, values = (var_2410_cast, var_2525_cast))[name = tensor("op_2550_cast")]; + tensor var_2552_equation_0 = const()[name = tensor("op_2552_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2552_cast = einsum(equation = var_2552_equation_0, values = (var_2410_cast, var_2526_cast))[name = tensor("op_2552_cast")]; + tensor var_2554_equation_0 = const()[name = tensor("op_2554_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2554_cast = einsum(equation = var_2554_equation_0, values = (var_2414_cast, var_2527_cast))[name = tensor("op_2554_cast")]; + tensor var_2556_equation_0 = const()[name = tensor("op_2556_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2556_cast = einsum(equation = var_2556_equation_0, values = (var_2414_cast, var_2528_cast))[name = tensor("op_2556_cast")]; + tensor var_2558_equation_0 = const()[name = tensor("op_2558_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2558_cast = einsum(equation = var_2558_equation_0, values = (var_2418_cast, var_2529_cast))[name = tensor("op_2558_cast")]; + tensor var_2560_equation_0 = const()[name = tensor("op_2560_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2560_cast = einsum(equation = var_2560_equation_0, values = (var_2418_cast, var_2530_cast))[name = tensor("op_2560_cast")]; + tensor var_2562_equation_0 = const()[name = tensor("op_2562_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2562_cast = einsum(equation = var_2562_equation_0, values = (var_2422_cast, var_2531_cast))[name = tensor("op_2562_cast")]; + tensor var_2564_equation_0 = const()[name = tensor("op_2564_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2564_cast = einsum(equation = var_2564_equation_0, values = (var_2422_cast, var_2532_cast))[name = tensor("op_2564_cast")]; + tensor var_2566_equation_0 = const()[name = tensor("op_2566_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2566_cast = einsum(equation = var_2566_equation_0, values = (var_2426_cast, var_2533_cast))[name = tensor("op_2566_cast")]; + tensor var_2568_equation_0 = const()[name = tensor("op_2568_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2568_cast = einsum(equation = var_2568_equation_0, values = (var_2426_cast, var_2534_cast))[name = tensor("op_2568_cast")]; + tensor var_2570_equation_0 = const()[name = tensor("op_2570_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2570_cast = einsum(equation = var_2570_equation_0, values = (var_2430_cast, var_2535_cast))[name = tensor("op_2570_cast")]; + tensor var_2572_equation_0 = const()[name = tensor("op_2572_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2572_cast = einsum(equation = var_2572_equation_0, values = (var_2430_cast, var_2536_cast))[name = tensor("op_2572_cast")]; + tensor var_2574_equation_0 = const()[name = tensor("op_2574_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2574_cast = einsum(equation = var_2574_equation_0, values = (var_2434_cast, var_2537_cast))[name = tensor("op_2574_cast")]; + tensor var_2576_equation_0 = const()[name = tensor("op_2576_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2576_cast = einsum(equation = var_2576_equation_0, values = (var_2434_cast, var_2538_cast))[name = tensor("op_2576_cast")]; + tensor var_2578_equation_0 = const()[name = tensor("op_2578_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2578_cast = einsum(equation = var_2578_equation_0, values = (var_2438_cast, var_2539_cast))[name = tensor("op_2578_cast")]; + tensor var_2580_equation_0 = const()[name = tensor("op_2580_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2580_cast = einsum(equation = var_2580_equation_0, values = (var_2438_cast, var_2540_cast))[name = tensor("op_2580_cast")]; + tensor var_2582_interleave_0 = const()[name = tensor("op_2582_interleave_0"), val = tensor(false)]; + tensor var_2582_cast = concat(axis = var_2150, interleave = var_2582_interleave_0, values = (var_2542_cast, var_2544_cast))[name = tensor("op_2582_cast")]; + tensor var_2584_interleave_0 = const()[name = tensor("op_2584_interleave_0"), val = tensor(false)]; + tensor var_2584_cast = concat(axis = var_2150, interleave = var_2584_interleave_0, values = (var_2546_cast, var_2548_cast))[name = tensor("op_2584_cast")]; + tensor var_2586_interleave_0 = const()[name = tensor("op_2586_interleave_0"), val = tensor(false)]; + tensor var_2586_cast = concat(axis = var_2150, interleave = var_2586_interleave_0, values = (var_2550_cast, var_2552_cast))[name = tensor("op_2586_cast")]; + tensor var_2588_interleave_0 = const()[name = tensor("op_2588_interleave_0"), val = tensor(false)]; + tensor var_2588_cast = concat(axis = var_2150, interleave = var_2588_interleave_0, values = (var_2554_cast, var_2556_cast))[name = tensor("op_2588_cast")]; + tensor var_2590_interleave_0 = const()[name = tensor("op_2590_interleave_0"), val = tensor(false)]; + tensor var_2590_cast = concat(axis = var_2150, interleave = var_2590_interleave_0, values = (var_2558_cast, var_2560_cast))[name = tensor("op_2590_cast")]; + tensor var_2592_interleave_0 = const()[name = tensor("op_2592_interleave_0"), val = tensor(false)]; + tensor var_2592_cast = concat(axis = var_2150, interleave = var_2592_interleave_0, values = (var_2562_cast, var_2564_cast))[name = tensor("op_2592_cast")]; + tensor var_2594_interleave_0 = const()[name = tensor("op_2594_interleave_0"), val = tensor(false)]; + tensor var_2594_cast = concat(axis = var_2150, interleave = var_2594_interleave_0, values = (var_2566_cast, var_2568_cast))[name = tensor("op_2594_cast")]; + tensor var_2596_interleave_0 = const()[name = tensor("op_2596_interleave_0"), val = tensor(false)]; + tensor var_2596_cast = concat(axis = var_2150, interleave = var_2596_interleave_0, values = (var_2570_cast, var_2572_cast))[name = tensor("op_2596_cast")]; + tensor var_2598_interleave_0 = const()[name = tensor("op_2598_interleave_0"), val = tensor(false)]; + tensor var_2598_cast = concat(axis = var_2150, interleave = var_2598_interleave_0, values = (var_2574_cast, var_2576_cast))[name = tensor("op_2598_cast")]; + tensor var_2600_interleave_0 = const()[name = tensor("op_2600_interleave_0"), val = tensor(false)]; + tensor var_2600_cast = concat(axis = var_2150, interleave = var_2600_interleave_0, values = (var_2578_cast, var_2580_cast))[name = tensor("op_2600_cast")]; + tensor input_79_interleave_0 = const()[name = tensor("input_79_interleave_0"), val = tensor(false)]; + tensor input_79_cast = concat(axis = var_2173, interleave = input_79_interleave_0, values = (var_2582_cast, var_2584_cast, var_2586_cast, var_2588_cast, var_2590_cast, var_2592_cast, var_2594_cast, var_2596_cast, var_2598_cast, var_2600_cast))[name = tensor("input_79_cast")]; + tensor var_2606 = const()[name = tensor("op_2606"), val = tensor([1, 1])]; + tensor var_2608 = const()[name = tensor("op_2608"), val = tensor([1, 1])]; + tensor var_2610_pad_type_0 = const()[name = tensor("op_2610_pad_type_0"), val = tensor("custom")]; + tensor var_2610_pad_0 = const()[name = tensor("op_2610_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_1_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(15884736))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(16192000))), name = tensor("down_blocks_1_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16_palettized"), shape = tensor([640, 640, 1, 1])]; + tensor down_blocks_1_attentions_0_transformer_blocks_0_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("down_blocks_1_attentions_0_transformer_blocks_0_attn1_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(16192192)))]; + tensor var_2610_cast = conv(bias = down_blocks_1_attentions_0_transformer_blocks_0_attn1_to_out_0_bias_to_fp16, dilations = var_2608, groups = var_2173, pad = var_2610_pad_0, pad_type = var_2610_pad_type_0, strides = var_2606, weight = down_blocks_1_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16_palettized, x = input_79_cast)[name = tensor("op_2610_cast")]; + tensor inputs_15_cast = add(x = var_2610_cast, y = inputs_13_cast)[name = tensor("inputs_15_cast")]; + tensor var_2614 = const()[name = tensor("op_2614"), val = tensor([1])]; + tensor channels_mean_15_cast = reduce_mean(axes = var_2614, keep_dims = var_2168, x = inputs_15_cast)[name = tensor("channels_mean_15_cast")]; + tensor zero_mean_15_cast = sub(x = inputs_15_cast, y = channels_mean_15_cast)[name = tensor("zero_mean_15_cast")]; + tensor zero_mean_sq_15_cast = mul(x = zero_mean_15_cast, y = zero_mean_15_cast)[name = tensor("zero_mean_sq_15_cast")]; + tensor var_2618 = const()[name = tensor("op_2618"), val = tensor([1])]; + tensor var_2619_cast = reduce_mean(axes = var_2618, keep_dims = var_2168, x = zero_mean_sq_15_cast)[name = tensor("op_2619_cast")]; + tensor var_2620_to_fp16 = const()[name = tensor("op_2620_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_2621_cast = add(x = var_2619_cast, y = var_2620_to_fp16)[name = tensor("op_2621_cast")]; + tensor denom_15_epsilon_0_to_fp16 = const()[name = tensor("denom_15_epsilon_0_to_fp16"), val = tensor(0x1p-24)]; + tensor denom_15_cast = rsqrt(epsilon = denom_15_epsilon_0_to_fp16, x = var_2621_cast)[name = tensor("denom_15_cast")]; + tensor out_15_cast = mul(x = zero_mean_15_cast, y = denom_15_cast)[name = tensor("out_15_cast")]; + tensor var_2625_to_fp16 = const()[name = tensor("op_2625_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(16193536)))]; + tensor var_2626_cast = add(x = out_15_cast, y = var_2625_to_fp16)[name = tensor("op_2626_cast")]; + tensor var_2628_to_fp16 = const()[name = tensor("op_2628_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(16194880)))]; + tensor hidden_states_47_cast = mul(x = var_2626_cast, y = var_2628_to_fp16)[name = tensor("hidden_states_47_cast")]; + tensor var_2635 = const()[name = tensor("op_2635"), val = tensor([1, 1])]; + tensor var_2637 = const()[name = tensor("op_2637"), val = tensor([1, 1])]; + tensor q_11_pad_type_0 = const()[name = tensor("q_11_pad_type_0"), val = tensor("custom")]; + tensor q_11_pad_0 = const()[name = tensor("q_11_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_1_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(16196224))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(16503488))), name = tensor("down_blocks_1_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16_palettized"), shape = tensor([640, 640, 1, 1])]; + tensor q_11_cast = conv(dilations = var_2637, groups = var_2173, pad = q_11_pad_0, pad_type = q_11_pad_type_0, strides = var_2635, weight = down_blocks_1_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16_palettized, x = hidden_states_47_cast)[name = tensor("q_11_cast")]; + tensor var_2641 = const()[name = tensor("op_2641"), val = tensor([1, 1])]; + tensor var_2643 = const()[name = tensor("op_2643"), val = tensor([1, 1])]; + tensor k_21_pad_type_0 = const()[name = tensor("k_21_pad_type_0"), val = tensor("custom")]; + tensor k_21_pad_0 = const()[name = tensor("k_21_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_1_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(16503680))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(16995264))), name = tensor("down_blocks_1_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized"), shape = tensor([640, 1024, 1, 1])]; + tensor k_21_cast = conv(dilations = var_2643, groups = var_2173, pad = k_21_pad_0, pad_type = k_21_pad_type_0, strides = var_2641, weight = down_blocks_1_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor("k_21_cast")]; + tensor var_2647 = const()[name = tensor("op_2647"), val = tensor([1, 1])]; + tensor var_2649 = const()[name = tensor("op_2649"), val = tensor([1, 1])]; + tensor v_11_pad_type_0 = const()[name = tensor("v_11_pad_type_0"), val = tensor("custom")]; + tensor v_11_pad_0 = const()[name = tensor("v_11_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_1_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(16995456))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(17487040))), name = tensor("down_blocks_1_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized"), shape = tensor([640, 1024, 1, 1])]; + tensor v_11_cast = conv(dilations = var_2649, groups = var_2173, pad = v_11_pad_0, pad_type = v_11_pad_type_0, strides = var_2647, weight = down_blocks_1_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor("v_11_cast")]; + tensor var_2653_begin_0 = const()[name = tensor("op_2653_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2653_end_0 = const()[name = tensor("op_2653_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_2653_end_mask_0 = const()[name = tensor("op_2653_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2653_cast = slice_by_index(begin = var_2653_begin_0, end = var_2653_end_0, end_mask = var_2653_end_mask_0, x = q_11_cast)[name = tensor("op_2653_cast")]; + tensor var_2657_begin_0 = const()[name = tensor("op_2657_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_2657_end_0 = const()[name = tensor("op_2657_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_2657_end_mask_0 = const()[name = tensor("op_2657_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2657_cast = slice_by_index(begin = var_2657_begin_0, end = var_2657_end_0, end_mask = var_2657_end_mask_0, x = q_11_cast)[name = tensor("op_2657_cast")]; + tensor var_2661_begin_0 = const()[name = tensor("op_2661_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_2661_end_0 = const()[name = tensor("op_2661_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_2661_end_mask_0 = const()[name = tensor("op_2661_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2661_cast = slice_by_index(begin = var_2661_begin_0, end = var_2661_end_0, end_mask = var_2661_end_mask_0, x = q_11_cast)[name = tensor("op_2661_cast")]; + tensor var_2665_begin_0 = const()[name = tensor("op_2665_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_2665_end_0 = const()[name = tensor("op_2665_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_2665_end_mask_0 = const()[name = tensor("op_2665_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2665_cast = slice_by_index(begin = var_2665_begin_0, end = var_2665_end_0, end_mask = var_2665_end_mask_0, x = q_11_cast)[name = tensor("op_2665_cast")]; + tensor var_2669_begin_0 = const()[name = tensor("op_2669_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_2669_end_0 = const()[name = tensor("op_2669_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_2669_end_mask_0 = const()[name = tensor("op_2669_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2669_cast = slice_by_index(begin = var_2669_begin_0, end = var_2669_end_0, end_mask = var_2669_end_mask_0, x = q_11_cast)[name = tensor("op_2669_cast")]; + tensor var_2673_begin_0 = const()[name = tensor("op_2673_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_2673_end_0 = const()[name = tensor("op_2673_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_2673_end_mask_0 = const()[name = tensor("op_2673_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2673_cast = slice_by_index(begin = var_2673_begin_0, end = var_2673_end_0, end_mask = var_2673_end_mask_0, x = q_11_cast)[name = tensor("op_2673_cast")]; + tensor var_2677_begin_0 = const()[name = tensor("op_2677_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_2677_end_0 = const()[name = tensor("op_2677_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_2677_end_mask_0 = const()[name = tensor("op_2677_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2677_cast = slice_by_index(begin = var_2677_begin_0, end = var_2677_end_0, end_mask = var_2677_end_mask_0, x = q_11_cast)[name = tensor("op_2677_cast")]; + tensor var_2681_begin_0 = const()[name = tensor("op_2681_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_2681_end_0 = const()[name = tensor("op_2681_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_2681_end_mask_0 = const()[name = tensor("op_2681_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2681_cast = slice_by_index(begin = var_2681_begin_0, end = var_2681_end_0, end_mask = var_2681_end_mask_0, x = q_11_cast)[name = tensor("op_2681_cast")]; + tensor var_2685_begin_0 = const()[name = tensor("op_2685_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_2685_end_0 = const()[name = tensor("op_2685_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_2685_end_mask_0 = const()[name = tensor("op_2685_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2685_cast = slice_by_index(begin = var_2685_begin_0, end = var_2685_end_0, end_mask = var_2685_end_mask_0, x = q_11_cast)[name = tensor("op_2685_cast")]; + tensor var_2689_begin_0 = const()[name = tensor("op_2689_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_2689_end_0 = const()[name = tensor("op_2689_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_2689_end_mask_0 = const()[name = tensor("op_2689_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2689_cast = slice_by_index(begin = var_2689_begin_0, end = var_2689_end_0, end_mask = var_2689_end_mask_0, x = q_11_cast)[name = tensor("op_2689_cast")]; + tensor var_2692_begin_0 = const()[name = tensor("op_2692_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2692_end_0 = const()[name = tensor("op_2692_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_2692_end_mask_0 = const()[name = tensor("op_2692_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2692_cast = slice_by_index(begin = var_2692_begin_0, end = var_2692_end_0, end_mask = var_2692_end_mask_0, x = var_2653_cast)[name = tensor("op_2692_cast")]; + tensor var_2693_begin_0 = const()[name = tensor("op_2693_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_2693_end_0 = const()[name = tensor("op_2693_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_2693_end_mask_0 = const()[name = tensor("op_2693_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2693_cast = slice_by_index(begin = var_2693_begin_0, end = var_2693_end_0, end_mask = var_2693_end_mask_0, x = var_2653_cast)[name = tensor("op_2693_cast")]; + tensor var_2694_begin_0 = const()[name = tensor("op_2694_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2694_end_0 = const()[name = tensor("op_2694_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_2694_end_mask_0 = const()[name = tensor("op_2694_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2694_cast = slice_by_index(begin = var_2694_begin_0, end = var_2694_end_0, end_mask = var_2694_end_mask_0, x = var_2657_cast)[name = tensor("op_2694_cast")]; + tensor var_2695_begin_0 = const()[name = tensor("op_2695_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_2695_end_0 = const()[name = tensor("op_2695_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_2695_end_mask_0 = const()[name = tensor("op_2695_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2695_cast = slice_by_index(begin = var_2695_begin_0, end = var_2695_end_0, end_mask = var_2695_end_mask_0, x = var_2657_cast)[name = tensor("op_2695_cast")]; + tensor var_2696_begin_0 = const()[name = tensor("op_2696_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2696_end_0 = const()[name = tensor("op_2696_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_2696_end_mask_0 = const()[name = tensor("op_2696_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2696_cast = slice_by_index(begin = var_2696_begin_0, end = var_2696_end_0, end_mask = var_2696_end_mask_0, x = var_2661_cast)[name = tensor("op_2696_cast")]; + tensor var_2697_begin_0 = const()[name = tensor("op_2697_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_2697_end_0 = const()[name = tensor("op_2697_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_2697_end_mask_0 = const()[name = tensor("op_2697_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2697_cast = slice_by_index(begin = var_2697_begin_0, end = var_2697_end_0, end_mask = var_2697_end_mask_0, x = var_2661_cast)[name = tensor("op_2697_cast")]; + tensor var_2698_begin_0 = const()[name = tensor("op_2698_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2698_end_0 = const()[name = tensor("op_2698_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_2698_end_mask_0 = const()[name = tensor("op_2698_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2698_cast = slice_by_index(begin = var_2698_begin_0, end = var_2698_end_0, end_mask = var_2698_end_mask_0, x = var_2665_cast)[name = tensor("op_2698_cast")]; + tensor var_2699_begin_0 = const()[name = tensor("op_2699_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_2699_end_0 = const()[name = tensor("op_2699_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_2699_end_mask_0 = const()[name = tensor("op_2699_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2699_cast = slice_by_index(begin = var_2699_begin_0, end = var_2699_end_0, end_mask = var_2699_end_mask_0, x = var_2665_cast)[name = tensor("op_2699_cast")]; + tensor var_2700_begin_0 = const()[name = tensor("op_2700_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2700_end_0 = const()[name = tensor("op_2700_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_2700_end_mask_0 = const()[name = tensor("op_2700_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2700_cast = slice_by_index(begin = var_2700_begin_0, end = var_2700_end_0, end_mask = var_2700_end_mask_0, x = var_2669_cast)[name = tensor("op_2700_cast")]; + tensor var_2701_begin_0 = const()[name = tensor("op_2701_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_2701_end_0 = const()[name = tensor("op_2701_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_2701_end_mask_0 = const()[name = tensor("op_2701_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2701_cast = slice_by_index(begin = var_2701_begin_0, end = var_2701_end_0, end_mask = var_2701_end_mask_0, x = var_2669_cast)[name = tensor("op_2701_cast")]; + tensor var_2702_begin_0 = const()[name = tensor("op_2702_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2702_end_0 = const()[name = tensor("op_2702_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_2702_end_mask_0 = const()[name = tensor("op_2702_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2702_cast = slice_by_index(begin = var_2702_begin_0, end = var_2702_end_0, end_mask = var_2702_end_mask_0, x = var_2673_cast)[name = tensor("op_2702_cast")]; + tensor var_2703_begin_0 = const()[name = tensor("op_2703_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_2703_end_0 = const()[name = tensor("op_2703_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_2703_end_mask_0 = const()[name = tensor("op_2703_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2703_cast = slice_by_index(begin = var_2703_begin_0, end = var_2703_end_0, end_mask = var_2703_end_mask_0, x = var_2673_cast)[name = tensor("op_2703_cast")]; + tensor var_2704_begin_0 = const()[name = tensor("op_2704_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2704_end_0 = const()[name = tensor("op_2704_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_2704_end_mask_0 = const()[name = tensor("op_2704_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2704_cast = slice_by_index(begin = var_2704_begin_0, end = var_2704_end_0, end_mask = var_2704_end_mask_0, x = var_2677_cast)[name = tensor("op_2704_cast")]; + tensor var_2705_begin_0 = const()[name = tensor("op_2705_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_2705_end_0 = const()[name = tensor("op_2705_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_2705_end_mask_0 = const()[name = tensor("op_2705_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2705_cast = slice_by_index(begin = var_2705_begin_0, end = var_2705_end_0, end_mask = var_2705_end_mask_0, x = var_2677_cast)[name = tensor("op_2705_cast")]; + tensor var_2706_begin_0 = const()[name = tensor("op_2706_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2706_end_0 = const()[name = tensor("op_2706_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_2706_end_mask_0 = const()[name = tensor("op_2706_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2706_cast = slice_by_index(begin = var_2706_begin_0, end = var_2706_end_0, end_mask = var_2706_end_mask_0, x = var_2681_cast)[name = tensor("op_2706_cast")]; + tensor var_2707_begin_0 = const()[name = tensor("op_2707_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_2707_end_0 = const()[name = tensor("op_2707_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_2707_end_mask_0 = const()[name = tensor("op_2707_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2707_cast = slice_by_index(begin = var_2707_begin_0, end = var_2707_end_0, end_mask = var_2707_end_mask_0, x = var_2681_cast)[name = tensor("op_2707_cast")]; + tensor var_2708_begin_0 = const()[name = tensor("op_2708_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2708_end_0 = const()[name = tensor("op_2708_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_2708_end_mask_0 = const()[name = tensor("op_2708_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2708_cast = slice_by_index(begin = var_2708_begin_0, end = var_2708_end_0, end_mask = var_2708_end_mask_0, x = var_2685_cast)[name = tensor("op_2708_cast")]; + tensor var_2709_begin_0 = const()[name = tensor("op_2709_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_2709_end_0 = const()[name = tensor("op_2709_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_2709_end_mask_0 = const()[name = tensor("op_2709_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2709_cast = slice_by_index(begin = var_2709_begin_0, end = var_2709_end_0, end_mask = var_2709_end_mask_0, x = var_2685_cast)[name = tensor("op_2709_cast")]; + tensor var_2710_begin_0 = const()[name = tensor("op_2710_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2710_end_0 = const()[name = tensor("op_2710_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_2710_end_mask_0 = const()[name = tensor("op_2710_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2710_cast = slice_by_index(begin = var_2710_begin_0, end = var_2710_end_0, end_mask = var_2710_end_mask_0, x = var_2689_cast)[name = tensor("op_2710_cast")]; + tensor var_2711_begin_0 = const()[name = tensor("op_2711_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_2711_end_0 = const()[name = tensor("op_2711_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_2711_end_mask_0 = const()[name = tensor("op_2711_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2711_cast = slice_by_index(begin = var_2711_begin_0, end = var_2711_end_0, end_mask = var_2711_end_mask_0, x = var_2689_cast)[name = tensor("op_2711_cast")]; + tensor k_23_perm_0 = const()[name = tensor("k_23_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_2716_begin_0 = const()[name = tensor("op_2716_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2716_end_0 = const()[name = tensor("op_2716_end_0"), val = tensor([2, 77, 1, 64])]; + tensor var_2716_end_mask_0 = const()[name = tensor("op_2716_end_mask_0"), val = tensor([true, true, true, false])]; + tensor transpose_26 = transpose(perm = k_23_perm_0, x = k_21_cast)[name = tensor("transpose_26")]; + tensor var_2716_cast = slice_by_index(begin = var_2716_begin_0, end = var_2716_end_0, end_mask = var_2716_end_mask_0, x = transpose_26)[name = tensor("op_2716_cast")]; + tensor var_2720_begin_0 = const()[name = tensor("op_2720_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_2720_end_0 = const()[name = tensor("op_2720_end_0"), val = tensor([2, 77, 1, 128])]; + tensor var_2720_end_mask_0 = const()[name = tensor("op_2720_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2720_cast = slice_by_index(begin = var_2720_begin_0, end = var_2720_end_0, end_mask = var_2720_end_mask_0, x = transpose_26)[name = tensor("op_2720_cast")]; + tensor var_2724_begin_0 = const()[name = tensor("op_2724_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_2724_end_0 = const()[name = tensor("op_2724_end_0"), val = tensor([2, 77, 1, 192])]; + tensor var_2724_end_mask_0 = const()[name = tensor("op_2724_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2724_cast = slice_by_index(begin = var_2724_begin_0, end = var_2724_end_0, end_mask = var_2724_end_mask_0, x = transpose_26)[name = tensor("op_2724_cast")]; + tensor var_2728_begin_0 = const()[name = tensor("op_2728_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_2728_end_0 = const()[name = tensor("op_2728_end_0"), val = tensor([2, 77, 1, 256])]; + tensor var_2728_end_mask_0 = const()[name = tensor("op_2728_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2728_cast = slice_by_index(begin = var_2728_begin_0, end = var_2728_end_0, end_mask = var_2728_end_mask_0, x = transpose_26)[name = tensor("op_2728_cast")]; + tensor var_2732_begin_0 = const()[name = tensor("op_2732_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_2732_end_0 = const()[name = tensor("op_2732_end_0"), val = tensor([2, 77, 1, 320])]; + tensor var_2732_end_mask_0 = const()[name = tensor("op_2732_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2732_cast = slice_by_index(begin = var_2732_begin_0, end = var_2732_end_0, end_mask = var_2732_end_mask_0, x = transpose_26)[name = tensor("op_2732_cast")]; + tensor var_2736_begin_0 = const()[name = tensor("op_2736_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_2736_end_0 = const()[name = tensor("op_2736_end_0"), val = tensor([2, 77, 1, 384])]; + tensor var_2736_end_mask_0 = const()[name = tensor("op_2736_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2736_cast = slice_by_index(begin = var_2736_begin_0, end = var_2736_end_0, end_mask = var_2736_end_mask_0, x = transpose_26)[name = tensor("op_2736_cast")]; + tensor var_2740_begin_0 = const()[name = tensor("op_2740_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_2740_end_0 = const()[name = tensor("op_2740_end_0"), val = tensor([2, 77, 1, 448])]; + tensor var_2740_end_mask_0 = const()[name = tensor("op_2740_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2740_cast = slice_by_index(begin = var_2740_begin_0, end = var_2740_end_0, end_mask = var_2740_end_mask_0, x = transpose_26)[name = tensor("op_2740_cast")]; + tensor var_2744_begin_0 = const()[name = tensor("op_2744_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_2744_end_0 = const()[name = tensor("op_2744_end_0"), val = tensor([2, 77, 1, 512])]; + tensor var_2744_end_mask_0 = const()[name = tensor("op_2744_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2744_cast = slice_by_index(begin = var_2744_begin_0, end = var_2744_end_0, end_mask = var_2744_end_mask_0, x = transpose_26)[name = tensor("op_2744_cast")]; + tensor var_2748_begin_0 = const()[name = tensor("op_2748_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_2748_end_0 = const()[name = tensor("op_2748_end_0"), val = tensor([2, 77, 1, 576])]; + tensor var_2748_end_mask_0 = const()[name = tensor("op_2748_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2748_cast = slice_by_index(begin = var_2748_begin_0, end = var_2748_end_0, end_mask = var_2748_end_mask_0, x = transpose_26)[name = tensor("op_2748_cast")]; + tensor var_2752_begin_0 = const()[name = tensor("op_2752_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_2752_end_0 = const()[name = tensor("op_2752_end_0"), val = tensor([2, 77, 1, 640])]; + tensor var_2752_end_mask_0 = const()[name = tensor("op_2752_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2752_cast = slice_by_index(begin = var_2752_begin_0, end = var_2752_end_0, end_mask = var_2752_end_mask_0, x = transpose_26)[name = tensor("op_2752_cast")]; + tensor var_2754_begin_0 = const()[name = tensor("op_2754_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2754_end_0 = const()[name = tensor("op_2754_end_0"), val = tensor([2, 64, 1, 77])]; + tensor var_2754_end_mask_0 = const()[name = tensor("op_2754_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2754_cast = slice_by_index(begin = var_2754_begin_0, end = var_2754_end_0, end_mask = var_2754_end_mask_0, x = v_11_cast)[name = tensor("op_2754_cast")]; + tensor var_2758_begin_0 = const()[name = tensor("op_2758_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_2758_end_0 = const()[name = tensor("op_2758_end_0"), val = tensor([2, 128, 1, 77])]; + tensor var_2758_end_mask_0 = const()[name = tensor("op_2758_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2758_cast = slice_by_index(begin = var_2758_begin_0, end = var_2758_end_0, end_mask = var_2758_end_mask_0, x = v_11_cast)[name = tensor("op_2758_cast")]; + tensor var_2762_begin_0 = const()[name = tensor("op_2762_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_2762_end_0 = const()[name = tensor("op_2762_end_0"), val = tensor([2, 192, 1, 77])]; + tensor var_2762_end_mask_0 = const()[name = tensor("op_2762_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2762_cast = slice_by_index(begin = var_2762_begin_0, end = var_2762_end_0, end_mask = var_2762_end_mask_0, x = v_11_cast)[name = tensor("op_2762_cast")]; + tensor var_2766_begin_0 = const()[name = tensor("op_2766_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_2766_end_0 = const()[name = tensor("op_2766_end_0"), val = tensor([2, 256, 1, 77])]; + tensor var_2766_end_mask_0 = const()[name = tensor("op_2766_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2766_cast = slice_by_index(begin = var_2766_begin_0, end = var_2766_end_0, end_mask = var_2766_end_mask_0, x = v_11_cast)[name = tensor("op_2766_cast")]; + tensor var_2770_begin_0 = const()[name = tensor("op_2770_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_2770_end_0 = const()[name = tensor("op_2770_end_0"), val = tensor([2, 320, 1, 77])]; + tensor var_2770_end_mask_0 = const()[name = tensor("op_2770_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2770_cast = slice_by_index(begin = var_2770_begin_0, end = var_2770_end_0, end_mask = var_2770_end_mask_0, x = v_11_cast)[name = tensor("op_2770_cast")]; + tensor var_2774_begin_0 = const()[name = tensor("op_2774_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_2774_end_0 = const()[name = tensor("op_2774_end_0"), val = tensor([2, 384, 1, 77])]; + tensor var_2774_end_mask_0 = const()[name = tensor("op_2774_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2774_cast = slice_by_index(begin = var_2774_begin_0, end = var_2774_end_0, end_mask = var_2774_end_mask_0, x = v_11_cast)[name = tensor("op_2774_cast")]; + tensor var_2778_begin_0 = const()[name = tensor("op_2778_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_2778_end_0 = const()[name = tensor("op_2778_end_0"), val = tensor([2, 448, 1, 77])]; + tensor var_2778_end_mask_0 = const()[name = tensor("op_2778_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2778_cast = slice_by_index(begin = var_2778_begin_0, end = var_2778_end_0, end_mask = var_2778_end_mask_0, x = v_11_cast)[name = tensor("op_2778_cast")]; + tensor var_2782_begin_0 = const()[name = tensor("op_2782_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_2782_end_0 = const()[name = tensor("op_2782_end_0"), val = tensor([2, 512, 1, 77])]; + tensor var_2782_end_mask_0 = const()[name = tensor("op_2782_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2782_cast = slice_by_index(begin = var_2782_begin_0, end = var_2782_end_0, end_mask = var_2782_end_mask_0, x = v_11_cast)[name = tensor("op_2782_cast")]; + tensor var_2786_begin_0 = const()[name = tensor("op_2786_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_2786_end_0 = const()[name = tensor("op_2786_end_0"), val = tensor([2, 576, 1, 77])]; + tensor var_2786_end_mask_0 = const()[name = tensor("op_2786_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2786_cast = slice_by_index(begin = var_2786_begin_0, end = var_2786_end_0, end_mask = var_2786_end_mask_0, x = v_11_cast)[name = tensor("op_2786_cast")]; + tensor var_2790_begin_0 = const()[name = tensor("op_2790_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_2790_end_0 = const()[name = tensor("op_2790_end_0"), val = tensor([2, 640, 1, 77])]; + tensor var_2790_end_mask_0 = const()[name = tensor("op_2790_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2790_cast = slice_by_index(begin = var_2790_begin_0, end = var_2790_end_0, end_mask = var_2790_end_mask_0, x = v_11_cast)[name = tensor("op_2790_cast")]; + tensor var_2794_equation_0 = const()[name = tensor("op_2794_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2794_cast = einsum(equation = var_2794_equation_0, values = (var_2716_cast, var_2692_cast))[name = tensor("op_2794_cast")]; + tensor var_2795_to_fp16 = const()[name = tensor("op_2795_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_361_cast = mul(x = var_2794_cast, y = var_2795_to_fp16)[name = tensor("aw_chunk_361_cast")]; + tensor var_2798_equation_0 = const()[name = tensor("op_2798_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2798_cast = einsum(equation = var_2798_equation_0, values = (var_2716_cast, var_2693_cast))[name = tensor("op_2798_cast")]; + tensor var_2799_to_fp16 = const()[name = tensor("op_2799_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_363_cast = mul(x = var_2798_cast, y = var_2799_to_fp16)[name = tensor("aw_chunk_363_cast")]; + tensor var_2802_equation_0 = const()[name = tensor("op_2802_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2802_cast = einsum(equation = var_2802_equation_0, values = (var_2720_cast, var_2694_cast))[name = tensor("op_2802_cast")]; + tensor var_2803_to_fp16 = const()[name = tensor("op_2803_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_365_cast = mul(x = var_2802_cast, y = var_2803_to_fp16)[name = tensor("aw_chunk_365_cast")]; + tensor var_2806_equation_0 = const()[name = tensor("op_2806_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2806_cast = einsum(equation = var_2806_equation_0, values = (var_2720_cast, var_2695_cast))[name = tensor("op_2806_cast")]; + tensor var_2807_to_fp16 = const()[name = tensor("op_2807_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_367_cast = mul(x = var_2806_cast, y = var_2807_to_fp16)[name = tensor("aw_chunk_367_cast")]; + tensor var_2810_equation_0 = const()[name = tensor("op_2810_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2810_cast = einsum(equation = var_2810_equation_0, values = (var_2724_cast, var_2696_cast))[name = tensor("op_2810_cast")]; + tensor var_2811_to_fp16 = const()[name = tensor("op_2811_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_369_cast = mul(x = var_2810_cast, y = var_2811_to_fp16)[name = tensor("aw_chunk_369_cast")]; + tensor var_2814_equation_0 = const()[name = tensor("op_2814_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2814_cast = einsum(equation = var_2814_equation_0, values = (var_2724_cast, var_2697_cast))[name = tensor("op_2814_cast")]; + tensor var_2815_to_fp16 = const()[name = tensor("op_2815_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_371_cast = mul(x = var_2814_cast, y = var_2815_to_fp16)[name = tensor("aw_chunk_371_cast")]; + tensor var_2818_equation_0 = const()[name = tensor("op_2818_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2818_cast = einsum(equation = var_2818_equation_0, values = (var_2728_cast, var_2698_cast))[name = tensor("op_2818_cast")]; + tensor var_2819_to_fp16 = const()[name = tensor("op_2819_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_373_cast = mul(x = var_2818_cast, y = var_2819_to_fp16)[name = tensor("aw_chunk_373_cast")]; + tensor var_2822_equation_0 = const()[name = tensor("op_2822_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2822_cast = einsum(equation = var_2822_equation_0, values = (var_2728_cast, var_2699_cast))[name = tensor("op_2822_cast")]; + tensor var_2823_to_fp16 = const()[name = tensor("op_2823_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_375_cast = mul(x = var_2822_cast, y = var_2823_to_fp16)[name = tensor("aw_chunk_375_cast")]; + tensor var_2826_equation_0 = const()[name = tensor("op_2826_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2826_cast = einsum(equation = var_2826_equation_0, values = (var_2732_cast, var_2700_cast))[name = tensor("op_2826_cast")]; + tensor var_2827_to_fp16 = const()[name = tensor("op_2827_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_377_cast = mul(x = var_2826_cast, y = var_2827_to_fp16)[name = tensor("aw_chunk_377_cast")]; + tensor var_2830_equation_0 = const()[name = tensor("op_2830_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2830_cast = einsum(equation = var_2830_equation_0, values = (var_2732_cast, var_2701_cast))[name = tensor("op_2830_cast")]; + tensor var_2831_to_fp16 = const()[name = tensor("op_2831_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_379_cast = mul(x = var_2830_cast, y = var_2831_to_fp16)[name = tensor("aw_chunk_379_cast")]; + tensor var_2834_equation_0 = const()[name = tensor("op_2834_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2834_cast = einsum(equation = var_2834_equation_0, values = (var_2736_cast, var_2702_cast))[name = tensor("op_2834_cast")]; + tensor var_2835_to_fp16 = const()[name = tensor("op_2835_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_381_cast = mul(x = var_2834_cast, y = var_2835_to_fp16)[name = tensor("aw_chunk_381_cast")]; + tensor var_2838_equation_0 = const()[name = tensor("op_2838_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2838_cast = einsum(equation = var_2838_equation_0, values = (var_2736_cast, var_2703_cast))[name = tensor("op_2838_cast")]; + tensor var_2839_to_fp16 = const()[name = tensor("op_2839_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_383_cast = mul(x = var_2838_cast, y = var_2839_to_fp16)[name = tensor("aw_chunk_383_cast")]; + tensor var_2842_equation_0 = const()[name = tensor("op_2842_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2842_cast = einsum(equation = var_2842_equation_0, values = (var_2740_cast, var_2704_cast))[name = tensor("op_2842_cast")]; + tensor var_2843_to_fp16 = const()[name = tensor("op_2843_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_385_cast = mul(x = var_2842_cast, y = var_2843_to_fp16)[name = tensor("aw_chunk_385_cast")]; + tensor var_2846_equation_0 = const()[name = tensor("op_2846_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2846_cast = einsum(equation = var_2846_equation_0, values = (var_2740_cast, var_2705_cast))[name = tensor("op_2846_cast")]; + tensor var_2847_to_fp16 = const()[name = tensor("op_2847_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_387_cast = mul(x = var_2846_cast, y = var_2847_to_fp16)[name = tensor("aw_chunk_387_cast")]; + tensor var_2850_equation_0 = const()[name = tensor("op_2850_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2850_cast = einsum(equation = var_2850_equation_0, values = (var_2744_cast, var_2706_cast))[name = tensor("op_2850_cast")]; + tensor var_2851_to_fp16 = const()[name = tensor("op_2851_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_389_cast = mul(x = var_2850_cast, y = var_2851_to_fp16)[name = tensor("aw_chunk_389_cast")]; + tensor var_2854_equation_0 = const()[name = tensor("op_2854_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2854_cast = einsum(equation = var_2854_equation_0, values = (var_2744_cast, var_2707_cast))[name = tensor("op_2854_cast")]; + tensor var_2855_to_fp16 = const()[name = tensor("op_2855_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_391_cast = mul(x = var_2854_cast, y = var_2855_to_fp16)[name = tensor("aw_chunk_391_cast")]; + tensor var_2858_equation_0 = const()[name = tensor("op_2858_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2858_cast = einsum(equation = var_2858_equation_0, values = (var_2748_cast, var_2708_cast))[name = tensor("op_2858_cast")]; + tensor var_2859_to_fp16 = const()[name = tensor("op_2859_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_393_cast = mul(x = var_2858_cast, y = var_2859_to_fp16)[name = tensor("aw_chunk_393_cast")]; + tensor var_2862_equation_0 = const()[name = tensor("op_2862_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2862_cast = einsum(equation = var_2862_equation_0, values = (var_2748_cast, var_2709_cast))[name = tensor("op_2862_cast")]; + tensor var_2863_to_fp16 = const()[name = tensor("op_2863_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_395_cast = mul(x = var_2862_cast, y = var_2863_to_fp16)[name = tensor("aw_chunk_395_cast")]; + tensor var_2866_equation_0 = const()[name = tensor("op_2866_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2866_cast = einsum(equation = var_2866_equation_0, values = (var_2752_cast, var_2710_cast))[name = tensor("op_2866_cast")]; + tensor var_2867_to_fp16 = const()[name = tensor("op_2867_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_397_cast = mul(x = var_2866_cast, y = var_2867_to_fp16)[name = tensor("aw_chunk_397_cast")]; + tensor var_2870_equation_0 = const()[name = tensor("op_2870_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2870_cast = einsum(equation = var_2870_equation_0, values = (var_2752_cast, var_2711_cast))[name = tensor("op_2870_cast")]; + tensor var_2871_to_fp16 = const()[name = tensor("op_2871_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_399_cast = mul(x = var_2870_cast, y = var_2871_to_fp16)[name = tensor("aw_chunk_399_cast")]; + tensor var_2873_cast = softmax(axis = var_2173, x = aw_chunk_361_cast)[name = tensor("op_2873_cast")]; + tensor var_2874_cast = softmax(axis = var_2173, x = aw_chunk_363_cast)[name = tensor("op_2874_cast")]; + tensor var_2875_cast = softmax(axis = var_2173, x = aw_chunk_365_cast)[name = tensor("op_2875_cast")]; + tensor var_2876_cast = softmax(axis = var_2173, x = aw_chunk_367_cast)[name = tensor("op_2876_cast")]; + tensor var_2877_cast = softmax(axis = var_2173, x = aw_chunk_369_cast)[name = tensor("op_2877_cast")]; + tensor var_2878_cast = softmax(axis = var_2173, x = aw_chunk_371_cast)[name = tensor("op_2878_cast")]; + tensor var_2879_cast = softmax(axis = var_2173, x = aw_chunk_373_cast)[name = tensor("op_2879_cast")]; + tensor var_2880_cast = softmax(axis = var_2173, x = aw_chunk_375_cast)[name = tensor("op_2880_cast")]; + tensor var_2881_cast = softmax(axis = var_2173, x = aw_chunk_377_cast)[name = tensor("op_2881_cast")]; + tensor var_2882_cast = softmax(axis = var_2173, x = aw_chunk_379_cast)[name = tensor("op_2882_cast")]; + tensor var_2883_cast = softmax(axis = var_2173, x = aw_chunk_381_cast)[name = tensor("op_2883_cast")]; + tensor var_2884_cast = softmax(axis = var_2173, x = aw_chunk_383_cast)[name = tensor("op_2884_cast")]; + tensor var_2885_cast = softmax(axis = var_2173, x = aw_chunk_385_cast)[name = tensor("op_2885_cast")]; + tensor var_2886_cast = softmax(axis = var_2173, x = aw_chunk_387_cast)[name = tensor("op_2886_cast")]; + tensor var_2887_cast = softmax(axis = var_2173, x = aw_chunk_389_cast)[name = tensor("op_2887_cast")]; + tensor var_2888_cast = softmax(axis = var_2173, x = aw_chunk_391_cast)[name = tensor("op_2888_cast")]; + tensor var_2889_cast = softmax(axis = var_2173, x = aw_chunk_393_cast)[name = tensor("op_2889_cast")]; + tensor var_2890_cast = softmax(axis = var_2173, x = aw_chunk_395_cast)[name = tensor("op_2890_cast")]; + tensor var_2891_cast = softmax(axis = var_2173, x = aw_chunk_397_cast)[name = tensor("op_2891_cast")]; + tensor var_2892_cast = softmax(axis = var_2173, x = aw_chunk_399_cast)[name = tensor("op_2892_cast")]; + tensor var_2894_equation_0 = const()[name = tensor("op_2894_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2894_cast = einsum(equation = var_2894_equation_0, values = (var_2754_cast, var_2873_cast))[name = tensor("op_2894_cast")]; + tensor var_2896_equation_0 = const()[name = tensor("op_2896_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2896_cast = einsum(equation = var_2896_equation_0, values = (var_2754_cast, var_2874_cast))[name = tensor("op_2896_cast")]; + tensor var_2898_equation_0 = const()[name = tensor("op_2898_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2898_cast = einsum(equation = var_2898_equation_0, values = (var_2758_cast, var_2875_cast))[name = tensor("op_2898_cast")]; + tensor var_2900_equation_0 = const()[name = tensor("op_2900_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2900_cast = einsum(equation = var_2900_equation_0, values = (var_2758_cast, var_2876_cast))[name = tensor("op_2900_cast")]; + tensor var_2902_equation_0 = const()[name = tensor("op_2902_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2902_cast = einsum(equation = var_2902_equation_0, values = (var_2762_cast, var_2877_cast))[name = tensor("op_2902_cast")]; + tensor var_2904_equation_0 = const()[name = tensor("op_2904_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2904_cast = einsum(equation = var_2904_equation_0, values = (var_2762_cast, var_2878_cast))[name = tensor("op_2904_cast")]; + tensor var_2906_equation_0 = const()[name = tensor("op_2906_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2906_cast = einsum(equation = var_2906_equation_0, values = (var_2766_cast, var_2879_cast))[name = tensor("op_2906_cast")]; + tensor var_2908_equation_0 = const()[name = tensor("op_2908_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2908_cast = einsum(equation = var_2908_equation_0, values = (var_2766_cast, var_2880_cast))[name = tensor("op_2908_cast")]; + tensor var_2910_equation_0 = const()[name = tensor("op_2910_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2910_cast = einsum(equation = var_2910_equation_0, values = (var_2770_cast, var_2881_cast))[name = tensor("op_2910_cast")]; + tensor var_2912_equation_0 = const()[name = tensor("op_2912_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2912_cast = einsum(equation = var_2912_equation_0, values = (var_2770_cast, var_2882_cast))[name = tensor("op_2912_cast")]; + tensor var_2914_equation_0 = const()[name = tensor("op_2914_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2914_cast = einsum(equation = var_2914_equation_0, values = (var_2774_cast, var_2883_cast))[name = tensor("op_2914_cast")]; + tensor var_2916_equation_0 = const()[name = tensor("op_2916_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2916_cast = einsum(equation = var_2916_equation_0, values = (var_2774_cast, var_2884_cast))[name = tensor("op_2916_cast")]; + tensor var_2918_equation_0 = const()[name = tensor("op_2918_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2918_cast = einsum(equation = var_2918_equation_0, values = (var_2778_cast, var_2885_cast))[name = tensor("op_2918_cast")]; + tensor var_2920_equation_0 = const()[name = tensor("op_2920_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2920_cast = einsum(equation = var_2920_equation_0, values = (var_2778_cast, var_2886_cast))[name = tensor("op_2920_cast")]; + tensor var_2922_equation_0 = const()[name = tensor("op_2922_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2922_cast = einsum(equation = var_2922_equation_0, values = (var_2782_cast, var_2887_cast))[name = tensor("op_2922_cast")]; + tensor var_2924_equation_0 = const()[name = tensor("op_2924_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2924_cast = einsum(equation = var_2924_equation_0, values = (var_2782_cast, var_2888_cast))[name = tensor("op_2924_cast")]; + tensor var_2926_equation_0 = const()[name = tensor("op_2926_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2926_cast = einsum(equation = var_2926_equation_0, values = (var_2786_cast, var_2889_cast))[name = tensor("op_2926_cast")]; + tensor var_2928_equation_0 = const()[name = tensor("op_2928_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2928_cast = einsum(equation = var_2928_equation_0, values = (var_2786_cast, var_2890_cast))[name = tensor("op_2928_cast")]; + tensor var_2930_equation_0 = const()[name = tensor("op_2930_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2930_cast = einsum(equation = var_2930_equation_0, values = (var_2790_cast, var_2891_cast))[name = tensor("op_2930_cast")]; + tensor var_2932_equation_0 = const()[name = tensor("op_2932_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_2932_cast = einsum(equation = var_2932_equation_0, values = (var_2790_cast, var_2892_cast))[name = tensor("op_2932_cast")]; + tensor var_2934_interleave_0 = const()[name = tensor("op_2934_interleave_0"), val = tensor(false)]; + tensor var_2934_cast = concat(axis = var_2150, interleave = var_2934_interleave_0, values = (var_2894_cast, var_2896_cast))[name = tensor("op_2934_cast")]; + tensor var_2936_interleave_0 = const()[name = tensor("op_2936_interleave_0"), val = tensor(false)]; + tensor var_2936_cast = concat(axis = var_2150, interleave = var_2936_interleave_0, values = (var_2898_cast, var_2900_cast))[name = tensor("op_2936_cast")]; + tensor var_2938_interleave_0 = const()[name = tensor("op_2938_interleave_0"), val = tensor(false)]; + tensor var_2938_cast = concat(axis = var_2150, interleave = var_2938_interleave_0, values = (var_2902_cast, var_2904_cast))[name = tensor("op_2938_cast")]; + tensor var_2940_interleave_0 = const()[name = tensor("op_2940_interleave_0"), val = tensor(false)]; + tensor var_2940_cast = concat(axis = var_2150, interleave = var_2940_interleave_0, values = (var_2906_cast, var_2908_cast))[name = tensor("op_2940_cast")]; + tensor var_2942_interleave_0 = const()[name = tensor("op_2942_interleave_0"), val = tensor(false)]; + tensor var_2942_cast = concat(axis = var_2150, interleave = var_2942_interleave_0, values = (var_2910_cast, var_2912_cast))[name = tensor("op_2942_cast")]; + tensor var_2944_interleave_0 = const()[name = tensor("op_2944_interleave_0"), val = tensor(false)]; + tensor var_2944_cast = concat(axis = var_2150, interleave = var_2944_interleave_0, values = (var_2914_cast, var_2916_cast))[name = tensor("op_2944_cast")]; + tensor var_2946_interleave_0 = const()[name = tensor("op_2946_interleave_0"), val = tensor(false)]; + tensor var_2946_cast = concat(axis = var_2150, interleave = var_2946_interleave_0, values = (var_2918_cast, var_2920_cast))[name = tensor("op_2946_cast")]; + tensor var_2948_interleave_0 = const()[name = tensor("op_2948_interleave_0"), val = tensor(false)]; + tensor var_2948_cast = concat(axis = var_2150, interleave = var_2948_interleave_0, values = (var_2922_cast, var_2924_cast))[name = tensor("op_2948_cast")]; + tensor var_2950_interleave_0 = const()[name = tensor("op_2950_interleave_0"), val = tensor(false)]; + tensor var_2950_cast = concat(axis = var_2150, interleave = var_2950_interleave_0, values = (var_2926_cast, var_2928_cast))[name = tensor("op_2950_cast")]; + tensor var_2952_interleave_0 = const()[name = tensor("op_2952_interleave_0"), val = tensor(false)]; + tensor var_2952_cast = concat(axis = var_2150, interleave = var_2952_interleave_0, values = (var_2930_cast, var_2932_cast))[name = tensor("op_2952_cast")]; + tensor input_81_interleave_0 = const()[name = tensor("input_81_interleave_0"), val = tensor(false)]; + tensor input_81_cast = concat(axis = var_2173, interleave = input_81_interleave_0, values = (var_2934_cast, var_2936_cast, var_2938_cast, var_2940_cast, var_2942_cast, var_2944_cast, var_2946_cast, var_2948_cast, var_2950_cast, var_2952_cast))[name = tensor("input_81_cast")]; + tensor var_2958 = const()[name = tensor("op_2958"), val = tensor([1, 1])]; + tensor var_2960 = const()[name = tensor("op_2960"), val = tensor([1, 1])]; + tensor var_2962_pad_type_0 = const()[name = tensor("op_2962_pad_type_0"), val = tensor("custom")]; + tensor var_2962_pad_0 = const()[name = tensor("op_2962_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_1_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(17487232))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(17794496))), name = tensor("down_blocks_1_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16_palettized"), shape = tensor([640, 640, 1, 1])]; + tensor down_blocks_1_attentions_0_transformer_blocks_0_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("down_blocks_1_attentions_0_transformer_blocks_0_attn2_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(17794688)))]; + tensor var_2962_cast = conv(bias = down_blocks_1_attentions_0_transformer_blocks_0_attn2_to_out_0_bias_to_fp16, dilations = var_2960, groups = var_2173, pad = var_2962_pad_0, pad_type = var_2962_pad_type_0, strides = var_2958, weight = down_blocks_1_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16_palettized, x = input_81_cast)[name = tensor("op_2962_cast")]; + tensor inputs_17_cast = add(x = var_2962_cast, y = inputs_15_cast)[name = tensor("inputs_17_cast")]; + tensor var_2966 = const()[name = tensor("op_2966"), val = tensor([1])]; + tensor channels_mean_17_cast = reduce_mean(axes = var_2966, keep_dims = var_2168, x = inputs_17_cast)[name = tensor("channels_mean_17_cast")]; + tensor zero_mean_17_cast = sub(x = inputs_17_cast, y = channels_mean_17_cast)[name = tensor("zero_mean_17_cast")]; + tensor zero_mean_sq_17_cast = mul(x = zero_mean_17_cast, y = zero_mean_17_cast)[name = tensor("zero_mean_sq_17_cast")]; + tensor var_2970 = const()[name = tensor("op_2970"), val = tensor([1])]; + tensor var_2971_cast = reduce_mean(axes = var_2970, keep_dims = var_2168, x = zero_mean_sq_17_cast)[name = tensor("op_2971_cast")]; + tensor var_2972_to_fp16 = const()[name = tensor("op_2972_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_2973_cast = add(x = var_2971_cast, y = var_2972_to_fp16)[name = tensor("op_2973_cast")]; + tensor denom_17_epsilon_0_to_fp16 = const()[name = tensor("denom_17_epsilon_0_to_fp16"), val = tensor(0x1p-24)]; + tensor denom_17_cast = rsqrt(epsilon = denom_17_epsilon_0_to_fp16, x = var_2973_cast)[name = tensor("denom_17_cast")]; + tensor out_17_cast = mul(x = zero_mean_17_cast, y = denom_17_cast)[name = tensor("out_17_cast")]; + tensor var_2977_to_fp16 = const()[name = tensor("op_2977_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(17796032)))]; + tensor var_2978_cast = add(x = out_17_cast, y = var_2977_to_fp16)[name = tensor("op_2978_cast")]; + tensor var_2980_to_fp16 = const()[name = tensor("op_2980_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(17797376)))]; + tensor input_83_cast = mul(x = var_2978_cast, y = var_2980_to_fp16)[name = tensor("input_83_cast")]; + tensor var_2988 = const()[name = tensor("op_2988"), val = tensor([1, 1])]; + tensor var_2990 = const()[name = tensor("op_2990"), val = tensor([1, 1])]; + tensor var_2992_pad_type_0 = const()[name = tensor("op_2992_pad_type_0"), val = tensor("custom")]; + tensor var_2992_pad_0 = const()[name = tensor("op_2992_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_1_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(17798720))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(20256384))), name = tensor("down_blocks_1_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16_palettized"), shape = tensor([5120, 640, 1, 1])]; + tensor down_blocks_1_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(20256576))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(20260480))), name = tensor("down_blocks_1_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16_palettized"), shape = tensor([5120])]; + tensor var_2992_cast = conv(bias = down_blocks_1_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16_palettized, dilations = var_2990, groups = var_2173, pad = var_2992_pad_0, pad_type = var_2992_pad_type_0, strides = var_2988, weight = down_blocks_1_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16_palettized, x = input_83_cast)[name = tensor("op_2992_cast")]; + tensor var_2993_split_sizes_0 = const()[name = tensor("op_2993_split_sizes_0"), val = tensor([2560, 2560])]; + tensor var_2993_axis_0 = const()[name = tensor("op_2993_axis_0"), val = tensor(1)]; + tensor var_2993_cast_0, tensor var_2993_cast_1 = split(axis = var_2993_axis_0, split_sizes = var_2993_split_sizes_0, x = var_2992_cast)[name = tensor("op_2993_cast")]; + tensor var_2995_mode_0 = const()[name = tensor("op_2995_mode_0"), val = tensor("EXACT")]; + tensor var_2995_cast = gelu(mode = var_2995_mode_0, x = var_2993_cast_1)[name = tensor("op_2995_cast")]; + tensor input_85_cast = mul(x = var_2993_cast_0, y = var_2995_cast)[name = tensor("input_85_cast")]; + tensor var_2999 = const()[name = tensor("op_2999"), val = tensor([1, 1])]; + tensor var_3001 = const()[name = tensor("op_3001"), val = tensor([1, 1])]; + tensor var_3003_pad_type_0 = const()[name = tensor("op_3003_pad_type_0"), val = tensor("custom")]; + tensor var_3003_pad_0 = const()[name = tensor("op_3003_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_1_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(20260672))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(21489536))), name = tensor("down_blocks_1_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16_palettized"), shape = tensor([640, 2560, 1, 1])]; + tensor down_blocks_1_attentions_0_transformer_blocks_0_ff_net_2_bias_to_fp16 = const()[name = tensor("down_blocks_1_attentions_0_transformer_blocks_0_ff_net_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(21489728)))]; + tensor var_3003_cast = conv(bias = down_blocks_1_attentions_0_transformer_blocks_0_ff_net_2_bias_to_fp16, dilations = var_3001, groups = var_2173, pad = var_3003_pad_0, pad_type = var_3003_pad_type_0, strides = var_2999, weight = down_blocks_1_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16_palettized, x = input_85_cast)[name = tensor("op_3003_cast")]; + tensor hidden_states_51_cast = add(x = var_3003_cast, y = inputs_17_cast)[name = tensor("hidden_states_51_cast")]; + tensor var_3005 = const()[name = tensor("op_3005"), val = tensor([2, 640, 32, 32])]; + tensor input_87_cast = reshape(shape = var_3005, x = hidden_states_51_cast)[name = tensor("input_87_cast")]; + tensor var_3009 = const()[name = tensor("op_3009"), val = tensor([1, 1])]; + tensor var_3011 = const()[name = tensor("op_3011"), val = tensor([1, 1])]; + tensor hidden_states_53_pad_type_0 = const()[name = tensor("hidden_states_53_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_53_pad_0 = const()[name = tensor("hidden_states_53_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_1_attentions_0_proj_out_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(21491072))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(21798336))), name = tensor("down_blocks_1_attentions_0_proj_out_weight_to_fp16_palettized"), shape = tensor([640, 640, 1, 1])]; + tensor down_blocks_1_attentions_0_proj_out_bias_to_fp16 = const()[name = tensor("down_blocks_1_attentions_0_proj_out_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(21798528)))]; + tensor hidden_states_53_cast = conv(bias = down_blocks_1_attentions_0_proj_out_bias_to_fp16, dilations = var_3011, groups = var_2173, pad = hidden_states_53_pad_0, pad_type = hidden_states_53_pad_type_0, strides = var_3009, weight = down_blocks_1_attentions_0_proj_out_weight_to_fp16_palettized, x = input_87_cast)[name = tensor("hidden_states_53_cast")]; + tensor input_89_cast = add(x = hidden_states_53_cast, y = hidden_states_41_cast)[name = tensor("input_89_cast")]; + tensor reshape_36_shape_0 = const()[name = tensor("reshape_36_shape_0"), val = tensor([2, 32, 20, 32, 32])]; + tensor reshape_36_cast = reshape(shape = reshape_36_shape_0, x = input_89_cast)[name = tensor("reshape_36_cast")]; + tensor reduce_mean_27_axes_0 = const()[name = tensor("reduce_mean_27_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_27_keep_dims_0 = const()[name = tensor("reduce_mean_27_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_27_cast = reduce_mean(axes = reduce_mean_27_axes_0, keep_dims = reduce_mean_27_keep_dims_0, x = reshape_36_cast)[name = tensor("reduce_mean_27_cast")]; + tensor sub_18_cast = sub(x = reshape_36_cast, y = reduce_mean_27_cast)[name = tensor("sub_18_cast")]; + tensor square_9_cast = square(x = sub_18_cast)[name = tensor("square_9_cast")]; + tensor reduce_mean_29_axes_0 = const()[name = tensor("reduce_mean_29_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_29_keep_dims_0 = const()[name = tensor("reduce_mean_29_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_29_cast = reduce_mean(axes = reduce_mean_29_axes_0, keep_dims = reduce_mean_29_keep_dims_0, x = square_9_cast)[name = tensor("reduce_mean_29_cast")]; + tensor add_18_y_0_to_fp16 = const()[name = tensor("add_18_y_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_18_cast = add(x = reduce_mean_29_cast, y = add_18_y_0_to_fp16)[name = tensor("add_18_cast")]; + tensor sqrt_9_cast = sqrt(x = add_18_cast)[name = tensor("sqrt_9_cast")]; + tensor real_div_9_cast = real_div(x = sub_18_cast, y = sqrt_9_cast)[name = tensor("real_div_9_cast")]; + tensor reshape_37_shape_0 = const()[name = tensor("reshape_37_shape_0"), val = tensor([2, 640, 32, 32])]; + tensor reshape_37_cast = reshape(shape = reshape_37_shape_0, x = real_div_9_cast)[name = tensor("reshape_37_cast")]; + tensor add_19_gamma_0_to_fp16 = const()[name = tensor("add_19_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(21799872)))]; + tensor add_19_beta_0_to_fp16 = const()[name = tensor("add_19_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(21801216)))]; + tensor add_19_epsilon_0_to_fp16 = const()[name = tensor("add_19_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_19_cast = batch_norm(beta = add_19_beta_0_to_fp16, epsilon = add_19_epsilon_0_to_fp16, gamma = add_19_gamma_0_to_fp16, mean = add_15_mean_0_to_fp16, variance = add_15_variance_0_to_fp16, x = reshape_37_cast)[name = tensor("add_19_cast")]; + tensor input_93_cast = silu(x = add_19_cast)[name = tensor("input_93_cast")]; + tensor var_3026 = const()[name = tensor("op_3026"), val = tensor([1, 1])]; + tensor var_3028 = const()[name = tensor("op_3028"), val = tensor([1, 1])]; + tensor hidden_states_55_pad_type_0 = const()[name = tensor("hidden_states_55_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_55_pad_0 = const()[name = tensor("hidden_states_55_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor down_blocks_1_resnets_1_conv1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(21802560))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(24567424))), name = tensor("down_blocks_1_resnets_1_conv1_weight_to_fp16_palettized"), shape = tensor([640, 640, 3, 3])]; + tensor down_blocks_1_resnets_1_conv1_bias_to_fp16 = const()[name = tensor("down_blocks_1_resnets_1_conv1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(24567616)))]; + tensor hidden_states_55_cast = conv(bias = down_blocks_1_resnets_1_conv1_bias_to_fp16, dilations = var_3028, groups = var_2173, pad = hidden_states_55_pad_0, pad_type = hidden_states_55_pad_type_0, strides = var_3026, weight = down_blocks_1_resnets_1_conv1_weight_to_fp16_palettized, x = input_93_cast)[name = tensor("hidden_states_55_cast")]; + tensor var_3034 = const()[name = tensor("op_3034"), val = tensor([1, 1])]; + tensor var_3036 = const()[name = tensor("op_3036"), val = tensor([1, 1])]; + tensor temb_7_pad_type_0 = const()[name = tensor("temb_7_pad_type_0"), val = tensor("custom")]; + tensor temb_7_pad_0 = const()[name = tensor("temb_7_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_1_resnets_1_time_emb_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(24568960))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(25183424))), name = tensor("down_blocks_1_resnets_1_time_emb_proj_weight_to_fp16_palettized"), shape = tensor([640, 1280, 1, 1])]; + tensor down_blocks_1_resnets_1_time_emb_proj_bias_to_fp16 = const()[name = tensor("down_blocks_1_resnets_1_time_emb_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(25183616)))]; + tensor temb_7_cast = conv(bias = down_blocks_1_resnets_1_time_emb_proj_bias_to_fp16, dilations = var_3036, groups = var_2173, pad = temb_7_pad_0, pad_type = temb_7_pad_type_0, strides = var_3034, weight = down_blocks_1_resnets_1_time_emb_proj_weight_to_fp16_palettized, x = input_15_cast)[name = tensor("temb_7_cast")]; + tensor input_97_cast = add(x = hidden_states_55_cast, y = temb_7_cast)[name = tensor("input_97_cast")]; + tensor reshape_40_shape_0 = const()[name = tensor("reshape_40_shape_0"), val = tensor([2, 32, 20, 32, 32])]; + tensor reshape_40_cast = reshape(shape = reshape_40_shape_0, x = input_97_cast)[name = tensor("reshape_40_cast")]; + tensor reduce_mean_30_axes_0 = const()[name = tensor("reduce_mean_30_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_30_keep_dims_0 = const()[name = tensor("reduce_mean_30_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_30_cast = reduce_mean(axes = reduce_mean_30_axes_0, keep_dims = reduce_mean_30_keep_dims_0, x = reshape_40_cast)[name = tensor("reduce_mean_30_cast")]; + tensor sub_20_cast = sub(x = reshape_40_cast, y = reduce_mean_30_cast)[name = tensor("sub_20_cast")]; + tensor square_10_cast = square(x = sub_20_cast)[name = tensor("square_10_cast")]; + tensor reduce_mean_32_axes_0 = const()[name = tensor("reduce_mean_32_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_32_keep_dims_0 = const()[name = tensor("reduce_mean_32_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_32_cast = reduce_mean(axes = reduce_mean_32_axes_0, keep_dims = reduce_mean_32_keep_dims_0, x = square_10_cast)[name = tensor("reduce_mean_32_cast")]; + tensor add_20_y_0_to_fp16 = const()[name = tensor("add_20_y_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_20_cast = add(x = reduce_mean_32_cast, y = add_20_y_0_to_fp16)[name = tensor("add_20_cast")]; + tensor sqrt_10_cast = sqrt(x = add_20_cast)[name = tensor("sqrt_10_cast")]; + tensor real_div_10_cast = real_div(x = sub_20_cast, y = sqrt_10_cast)[name = tensor("real_div_10_cast")]; + tensor reshape_41_shape_0 = const()[name = tensor("reshape_41_shape_0"), val = tensor([2, 640, 32, 32])]; + tensor reshape_41_cast = reshape(shape = reshape_41_shape_0, x = real_div_10_cast)[name = tensor("reshape_41_cast")]; + tensor add_21_gamma_0_to_fp16 = const()[name = tensor("add_21_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(25184960)))]; + tensor add_21_beta_0_to_fp16 = const()[name = tensor("add_21_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(25186304)))]; + tensor add_21_epsilon_0_to_fp16 = const()[name = tensor("add_21_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_21_cast = batch_norm(beta = add_21_beta_0_to_fp16, epsilon = add_21_epsilon_0_to_fp16, gamma = add_21_gamma_0_to_fp16, mean = add_15_mean_0_to_fp16, variance = add_15_variance_0_to_fp16, x = reshape_41_cast)[name = tensor("add_21_cast")]; + tensor input_101_cast = silu(x = add_21_cast)[name = tensor("input_101_cast")]; + tensor var_3046 = const()[name = tensor("op_3046"), val = tensor([1, 1])]; + tensor var_3048 = const()[name = tensor("op_3048"), val = tensor([1, 1])]; + tensor hidden_states_57_pad_type_0 = const()[name = tensor("hidden_states_57_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_57_pad_0 = const()[name = tensor("hidden_states_57_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor down_blocks_1_resnets_1_conv2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(25187648))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(27952512))), name = tensor("down_blocks_1_resnets_1_conv2_weight_to_fp16_palettized"), shape = tensor([640, 640, 3, 3])]; + tensor down_blocks_1_resnets_1_conv2_bias_to_fp16 = const()[name = tensor("down_blocks_1_resnets_1_conv2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(27952704)))]; + tensor hidden_states_57_cast = conv(bias = down_blocks_1_resnets_1_conv2_bias_to_fp16, dilations = var_3048, groups = var_2173, pad = hidden_states_57_pad_0, pad_type = hidden_states_57_pad_type_0, strides = var_3046, weight = down_blocks_1_resnets_1_conv2_weight_to_fp16_palettized, x = input_101_cast)[name = tensor("hidden_states_57_cast")]; + tensor hidden_states_59_cast = add(x = input_89_cast, y = hidden_states_57_cast)[name = tensor("hidden_states_59_cast")]; + tensor reshape_44_shape_0 = const()[name = tensor("reshape_44_shape_0"), val = tensor([2, 32, 20, 32, 32])]; + tensor reshape_44_cast = reshape(shape = reshape_44_shape_0, x = hidden_states_59_cast)[name = tensor("reshape_44_cast")]; + tensor reduce_mean_33_axes_0 = const()[name = tensor("reduce_mean_33_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_33_keep_dims_0 = const()[name = tensor("reduce_mean_33_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_33_cast = reduce_mean(axes = reduce_mean_33_axes_0, keep_dims = reduce_mean_33_keep_dims_0, x = reshape_44_cast)[name = tensor("reduce_mean_33_cast")]; + tensor sub_22_cast = sub(x = reshape_44_cast, y = reduce_mean_33_cast)[name = tensor("sub_22_cast")]; + tensor square_11_cast = square(x = sub_22_cast)[name = tensor("square_11_cast")]; + tensor reduce_mean_35_axes_0 = const()[name = tensor("reduce_mean_35_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_35_keep_dims_0 = const()[name = tensor("reduce_mean_35_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_35_cast = reduce_mean(axes = reduce_mean_35_axes_0, keep_dims = reduce_mean_35_keep_dims_0, x = square_11_cast)[name = tensor("reduce_mean_35_cast")]; + tensor add_22_y_0_to_fp16 = const()[name = tensor("add_22_y_0_to_fp16"), val = tensor(0x1.1p-20)]; + tensor add_22_cast = add(x = reduce_mean_35_cast, y = add_22_y_0_to_fp16)[name = tensor("add_22_cast")]; + tensor sqrt_11_cast = sqrt(x = add_22_cast)[name = tensor("sqrt_11_cast")]; + tensor real_div_11_cast = real_div(x = sub_22_cast, y = sqrt_11_cast)[name = tensor("real_div_11_cast")]; + tensor reshape_45_shape_0 = const()[name = tensor("reshape_45_shape_0"), val = tensor([2, 640, 32, 32])]; + tensor reshape_45_cast = reshape(shape = reshape_45_shape_0, x = real_div_11_cast)[name = tensor("reshape_45_cast")]; + tensor add_23_gamma_0_to_fp16 = const()[name = tensor("add_23_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(27954048)))]; + tensor add_23_beta_0_to_fp16 = const()[name = tensor("add_23_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(27955392)))]; + tensor add_23_epsilon_0_to_fp16 = const()[name = tensor("add_23_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_23_cast = batch_norm(beta = add_23_beta_0_to_fp16, epsilon = add_23_epsilon_0_to_fp16, gamma = add_23_gamma_0_to_fp16, mean = add_15_mean_0_to_fp16, variance = add_15_variance_0_to_fp16, x = reshape_45_cast)[name = tensor("add_23_cast")]; + tensor var_3068 = const()[name = tensor("op_3068"), val = tensor([1, 1])]; + tensor var_3070 = const()[name = tensor("op_3070"), val = tensor([1, 1])]; + tensor hidden_states_61_pad_type_0 = const()[name = tensor("hidden_states_61_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_61_pad_0 = const()[name = tensor("hidden_states_61_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_1_attentions_1_proj_in_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(27956736))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(28264000))), name = tensor("down_blocks_1_attentions_1_proj_in_weight_to_fp16_palettized"), shape = tensor([640, 640, 1, 1])]; + tensor down_blocks_1_attentions_1_proj_in_bias_to_fp16 = const()[name = tensor("down_blocks_1_attentions_1_proj_in_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(28264192)))]; + tensor hidden_states_61_cast = conv(bias = down_blocks_1_attentions_1_proj_in_bias_to_fp16, dilations = var_3070, groups = var_2173, pad = hidden_states_61_pad_0, pad_type = hidden_states_61_pad_type_0, strides = var_3068, weight = down_blocks_1_attentions_1_proj_in_weight_to_fp16_palettized, x = add_23_cast)[name = tensor("hidden_states_61_cast")]; + tensor var_3075 = const()[name = tensor("op_3075"), val = tensor([2, 640, 1, 1024])]; + tensor inputs_19_cast = reshape(shape = var_3075, x = hidden_states_61_cast)[name = tensor("inputs_19_cast")]; + tensor var_3085 = const()[name = tensor("op_3085"), val = tensor([1])]; + tensor channels_mean_19_cast = reduce_mean(axes = var_3085, keep_dims = var_2168, x = inputs_19_cast)[name = tensor("channels_mean_19_cast")]; + tensor zero_mean_19_cast = sub(x = inputs_19_cast, y = channels_mean_19_cast)[name = tensor("zero_mean_19_cast")]; + tensor zero_mean_sq_19_cast = mul(x = zero_mean_19_cast, y = zero_mean_19_cast)[name = tensor("zero_mean_sq_19_cast")]; + tensor var_3089 = const()[name = tensor("op_3089"), val = tensor([1])]; + tensor var_3090_cast = reduce_mean(axes = var_3089, keep_dims = var_2168, x = zero_mean_sq_19_cast)[name = tensor("op_3090_cast")]; + tensor var_3091_to_fp16 = const()[name = tensor("op_3091_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_3092_cast = add(x = var_3090_cast, y = var_3091_to_fp16)[name = tensor("op_3092_cast")]; + tensor denom_19_epsilon_0_to_fp16 = const()[name = tensor("denom_19_epsilon_0_to_fp16"), val = tensor(0x1p-24)]; + tensor denom_19_cast = rsqrt(epsilon = denom_19_epsilon_0_to_fp16, x = var_3092_cast)[name = tensor("denom_19_cast")]; + tensor out_19_cast = mul(x = zero_mean_19_cast, y = denom_19_cast)[name = tensor("out_19_cast")]; + tensor var_3096_to_fp16 = const()[name = tensor("op_3096_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(28265536)))]; + tensor var_3097_cast = add(x = out_19_cast, y = var_3096_to_fp16)[name = tensor("op_3097_cast")]; + tensor var_3099_to_fp16 = const()[name = tensor("op_3099_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(28266880)))]; + tensor hidden_states_63_cast = mul(x = var_3097_cast, y = var_3099_to_fp16)[name = tensor("hidden_states_63_cast")]; + tensor var_3106 = const()[name = tensor("op_3106"), val = tensor([1, 1])]; + tensor var_3108 = const()[name = tensor("op_3108"), val = tensor([1, 1])]; + tensor q_13_pad_type_0 = const()[name = tensor("q_13_pad_type_0"), val = tensor("custom")]; + tensor q_13_pad_0 = const()[name = tensor("q_13_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_1_attentions_1_transformer_blocks_0_attn1_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(28268224))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(28575488))), name = tensor("down_blocks_1_attentions_1_transformer_blocks_0_attn1_to_q_weight_to_fp16_palettized"), shape = tensor([640, 640, 1, 1])]; + tensor q_13_cast = conv(dilations = var_3108, groups = var_2173, pad = q_13_pad_0, pad_type = q_13_pad_type_0, strides = var_3106, weight = down_blocks_1_attentions_1_transformer_blocks_0_attn1_to_q_weight_to_fp16_palettized, x = hidden_states_63_cast)[name = tensor("q_13_cast")]; + tensor var_3112 = const()[name = tensor("op_3112"), val = tensor([1, 1])]; + tensor var_3114 = const()[name = tensor("op_3114"), val = tensor([1, 1])]; + tensor k_25_pad_type_0 = const()[name = tensor("k_25_pad_type_0"), val = tensor("custom")]; + tensor k_25_pad_0 = const()[name = tensor("k_25_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_1_attentions_1_transformer_blocks_0_attn1_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(28575680))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(28882944))), name = tensor("down_blocks_1_attentions_1_transformer_blocks_0_attn1_to_k_weight_to_fp16_palettized"), shape = tensor([640, 640, 1, 1])]; + tensor k_25_cast = conv(dilations = var_3114, groups = var_2173, pad = k_25_pad_0, pad_type = k_25_pad_type_0, strides = var_3112, weight = down_blocks_1_attentions_1_transformer_blocks_0_attn1_to_k_weight_to_fp16_palettized, x = hidden_states_63_cast)[name = tensor("k_25_cast")]; + tensor var_3118 = const()[name = tensor("op_3118"), val = tensor([1, 1])]; + tensor var_3120 = const()[name = tensor("op_3120"), val = tensor([1, 1])]; + tensor v_13_pad_type_0 = const()[name = tensor("v_13_pad_type_0"), val = tensor("custom")]; + tensor v_13_pad_0 = const()[name = tensor("v_13_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_1_attentions_1_transformer_blocks_0_attn1_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(28883136))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(29190400))), name = tensor("down_blocks_1_attentions_1_transformer_blocks_0_attn1_to_v_weight_to_fp16_palettized"), shape = tensor([640, 640, 1, 1])]; + tensor v_13_cast = conv(dilations = var_3120, groups = var_2173, pad = v_13_pad_0, pad_type = v_13_pad_type_0, strides = var_3118, weight = down_blocks_1_attentions_1_transformer_blocks_0_attn1_to_v_weight_to_fp16_palettized, x = hidden_states_63_cast)[name = tensor("v_13_cast")]; + tensor var_3124_begin_0 = const()[name = tensor("op_3124_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3124_end_0 = const()[name = tensor("op_3124_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_3124_end_mask_0 = const()[name = tensor("op_3124_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3124_cast = slice_by_index(begin = var_3124_begin_0, end = var_3124_end_0, end_mask = var_3124_end_mask_0, x = q_13_cast)[name = tensor("op_3124_cast")]; + tensor var_3128_begin_0 = const()[name = tensor("op_3128_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_3128_end_0 = const()[name = tensor("op_3128_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_3128_end_mask_0 = const()[name = tensor("op_3128_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3128_cast = slice_by_index(begin = var_3128_begin_0, end = var_3128_end_0, end_mask = var_3128_end_mask_0, x = q_13_cast)[name = tensor("op_3128_cast")]; + tensor var_3132_begin_0 = const()[name = tensor("op_3132_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_3132_end_0 = const()[name = tensor("op_3132_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_3132_end_mask_0 = const()[name = tensor("op_3132_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3132_cast = slice_by_index(begin = var_3132_begin_0, end = var_3132_end_0, end_mask = var_3132_end_mask_0, x = q_13_cast)[name = tensor("op_3132_cast")]; + tensor var_3136_begin_0 = const()[name = tensor("op_3136_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_3136_end_0 = const()[name = tensor("op_3136_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_3136_end_mask_0 = const()[name = tensor("op_3136_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3136_cast = slice_by_index(begin = var_3136_begin_0, end = var_3136_end_0, end_mask = var_3136_end_mask_0, x = q_13_cast)[name = tensor("op_3136_cast")]; + tensor var_3140_begin_0 = const()[name = tensor("op_3140_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_3140_end_0 = const()[name = tensor("op_3140_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_3140_end_mask_0 = const()[name = tensor("op_3140_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3140_cast = slice_by_index(begin = var_3140_begin_0, end = var_3140_end_0, end_mask = var_3140_end_mask_0, x = q_13_cast)[name = tensor("op_3140_cast")]; + tensor var_3144_begin_0 = const()[name = tensor("op_3144_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_3144_end_0 = const()[name = tensor("op_3144_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_3144_end_mask_0 = const()[name = tensor("op_3144_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3144_cast = slice_by_index(begin = var_3144_begin_0, end = var_3144_end_0, end_mask = var_3144_end_mask_0, x = q_13_cast)[name = tensor("op_3144_cast")]; + tensor var_3148_begin_0 = const()[name = tensor("op_3148_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_3148_end_0 = const()[name = tensor("op_3148_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_3148_end_mask_0 = const()[name = tensor("op_3148_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3148_cast = slice_by_index(begin = var_3148_begin_0, end = var_3148_end_0, end_mask = var_3148_end_mask_0, x = q_13_cast)[name = tensor("op_3148_cast")]; + tensor var_3152_begin_0 = const()[name = tensor("op_3152_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_3152_end_0 = const()[name = tensor("op_3152_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_3152_end_mask_0 = const()[name = tensor("op_3152_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3152_cast = slice_by_index(begin = var_3152_begin_0, end = var_3152_end_0, end_mask = var_3152_end_mask_0, x = q_13_cast)[name = tensor("op_3152_cast")]; + tensor var_3156_begin_0 = const()[name = tensor("op_3156_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_3156_end_0 = const()[name = tensor("op_3156_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_3156_end_mask_0 = const()[name = tensor("op_3156_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3156_cast = slice_by_index(begin = var_3156_begin_0, end = var_3156_end_0, end_mask = var_3156_end_mask_0, x = q_13_cast)[name = tensor("op_3156_cast")]; + tensor var_3160_begin_0 = const()[name = tensor("op_3160_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_3160_end_0 = const()[name = tensor("op_3160_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_3160_end_mask_0 = const()[name = tensor("op_3160_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3160_cast = slice_by_index(begin = var_3160_begin_0, end = var_3160_end_0, end_mask = var_3160_end_mask_0, x = q_13_cast)[name = tensor("op_3160_cast")]; + tensor var_3163_begin_0 = const()[name = tensor("op_3163_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3163_end_0 = const()[name = tensor("op_3163_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_3163_end_mask_0 = const()[name = tensor("op_3163_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3163_cast = slice_by_index(begin = var_3163_begin_0, end = var_3163_end_0, end_mask = var_3163_end_mask_0, x = var_3124_cast)[name = tensor("op_3163_cast")]; + tensor var_3164_begin_0 = const()[name = tensor("op_3164_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_3164_end_0 = const()[name = tensor("op_3164_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_3164_end_mask_0 = const()[name = tensor("op_3164_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3164_cast = slice_by_index(begin = var_3164_begin_0, end = var_3164_end_0, end_mask = var_3164_end_mask_0, x = var_3124_cast)[name = tensor("op_3164_cast")]; + tensor var_3165_begin_0 = const()[name = tensor("op_3165_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3165_end_0 = const()[name = tensor("op_3165_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_3165_end_mask_0 = const()[name = tensor("op_3165_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3165_cast = slice_by_index(begin = var_3165_begin_0, end = var_3165_end_0, end_mask = var_3165_end_mask_0, x = var_3128_cast)[name = tensor("op_3165_cast")]; + tensor var_3166_begin_0 = const()[name = tensor("op_3166_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_3166_end_0 = const()[name = tensor("op_3166_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_3166_end_mask_0 = const()[name = tensor("op_3166_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3166_cast = slice_by_index(begin = var_3166_begin_0, end = var_3166_end_0, end_mask = var_3166_end_mask_0, x = var_3128_cast)[name = tensor("op_3166_cast")]; + tensor var_3167_begin_0 = const()[name = tensor("op_3167_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3167_end_0 = const()[name = tensor("op_3167_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_3167_end_mask_0 = const()[name = tensor("op_3167_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3167_cast = slice_by_index(begin = var_3167_begin_0, end = var_3167_end_0, end_mask = var_3167_end_mask_0, x = var_3132_cast)[name = tensor("op_3167_cast")]; + tensor var_3168_begin_0 = const()[name = tensor("op_3168_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_3168_end_0 = const()[name = tensor("op_3168_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_3168_end_mask_0 = const()[name = tensor("op_3168_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3168_cast = slice_by_index(begin = var_3168_begin_0, end = var_3168_end_0, end_mask = var_3168_end_mask_0, x = var_3132_cast)[name = tensor("op_3168_cast")]; + tensor var_3169_begin_0 = const()[name = tensor("op_3169_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3169_end_0 = const()[name = tensor("op_3169_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_3169_end_mask_0 = const()[name = tensor("op_3169_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3169_cast = slice_by_index(begin = var_3169_begin_0, end = var_3169_end_0, end_mask = var_3169_end_mask_0, x = var_3136_cast)[name = tensor("op_3169_cast")]; + tensor var_3170_begin_0 = const()[name = tensor("op_3170_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_3170_end_0 = const()[name = tensor("op_3170_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_3170_end_mask_0 = const()[name = tensor("op_3170_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3170_cast = slice_by_index(begin = var_3170_begin_0, end = var_3170_end_0, end_mask = var_3170_end_mask_0, x = var_3136_cast)[name = tensor("op_3170_cast")]; + tensor var_3171_begin_0 = const()[name = tensor("op_3171_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3171_end_0 = const()[name = tensor("op_3171_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_3171_end_mask_0 = const()[name = tensor("op_3171_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3171_cast = slice_by_index(begin = var_3171_begin_0, end = var_3171_end_0, end_mask = var_3171_end_mask_0, x = var_3140_cast)[name = tensor("op_3171_cast")]; + tensor var_3172_begin_0 = const()[name = tensor("op_3172_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_3172_end_0 = const()[name = tensor("op_3172_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_3172_end_mask_0 = const()[name = tensor("op_3172_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3172_cast = slice_by_index(begin = var_3172_begin_0, end = var_3172_end_0, end_mask = var_3172_end_mask_0, x = var_3140_cast)[name = tensor("op_3172_cast")]; + tensor var_3173_begin_0 = const()[name = tensor("op_3173_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3173_end_0 = const()[name = tensor("op_3173_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_3173_end_mask_0 = const()[name = tensor("op_3173_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3173_cast = slice_by_index(begin = var_3173_begin_0, end = var_3173_end_0, end_mask = var_3173_end_mask_0, x = var_3144_cast)[name = tensor("op_3173_cast")]; + tensor var_3174_begin_0 = const()[name = tensor("op_3174_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_3174_end_0 = const()[name = tensor("op_3174_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_3174_end_mask_0 = const()[name = tensor("op_3174_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3174_cast = slice_by_index(begin = var_3174_begin_0, end = var_3174_end_0, end_mask = var_3174_end_mask_0, x = var_3144_cast)[name = tensor("op_3174_cast")]; + tensor var_3175_begin_0 = const()[name = tensor("op_3175_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3175_end_0 = const()[name = tensor("op_3175_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_3175_end_mask_0 = const()[name = tensor("op_3175_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3175_cast = slice_by_index(begin = var_3175_begin_0, end = var_3175_end_0, end_mask = var_3175_end_mask_0, x = var_3148_cast)[name = tensor("op_3175_cast")]; + tensor var_3176_begin_0 = const()[name = tensor("op_3176_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_3176_end_0 = const()[name = tensor("op_3176_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_3176_end_mask_0 = const()[name = tensor("op_3176_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3176_cast = slice_by_index(begin = var_3176_begin_0, end = var_3176_end_0, end_mask = var_3176_end_mask_0, x = var_3148_cast)[name = tensor("op_3176_cast")]; + tensor var_3177_begin_0 = const()[name = tensor("op_3177_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3177_end_0 = const()[name = tensor("op_3177_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_3177_end_mask_0 = const()[name = tensor("op_3177_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3177_cast = slice_by_index(begin = var_3177_begin_0, end = var_3177_end_0, end_mask = var_3177_end_mask_0, x = var_3152_cast)[name = tensor("op_3177_cast")]; + tensor var_3178_begin_0 = const()[name = tensor("op_3178_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_3178_end_0 = const()[name = tensor("op_3178_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_3178_end_mask_0 = const()[name = tensor("op_3178_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3178_cast = slice_by_index(begin = var_3178_begin_0, end = var_3178_end_0, end_mask = var_3178_end_mask_0, x = var_3152_cast)[name = tensor("op_3178_cast")]; + tensor var_3179_begin_0 = const()[name = tensor("op_3179_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3179_end_0 = const()[name = tensor("op_3179_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_3179_end_mask_0 = const()[name = tensor("op_3179_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3179_cast = slice_by_index(begin = var_3179_begin_0, end = var_3179_end_0, end_mask = var_3179_end_mask_0, x = var_3156_cast)[name = tensor("op_3179_cast")]; + tensor var_3180_begin_0 = const()[name = tensor("op_3180_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_3180_end_0 = const()[name = tensor("op_3180_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_3180_end_mask_0 = const()[name = tensor("op_3180_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3180_cast = slice_by_index(begin = var_3180_begin_0, end = var_3180_end_0, end_mask = var_3180_end_mask_0, x = var_3156_cast)[name = tensor("op_3180_cast")]; + tensor var_3181_begin_0 = const()[name = tensor("op_3181_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3181_end_0 = const()[name = tensor("op_3181_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_3181_end_mask_0 = const()[name = tensor("op_3181_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3181_cast = slice_by_index(begin = var_3181_begin_0, end = var_3181_end_0, end_mask = var_3181_end_mask_0, x = var_3160_cast)[name = tensor("op_3181_cast")]; + tensor var_3182_begin_0 = const()[name = tensor("op_3182_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_3182_end_0 = const()[name = tensor("op_3182_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_3182_end_mask_0 = const()[name = tensor("op_3182_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3182_cast = slice_by_index(begin = var_3182_begin_0, end = var_3182_end_0, end_mask = var_3182_end_mask_0, x = var_3160_cast)[name = tensor("op_3182_cast")]; + tensor k_27_perm_0 = const()[name = tensor("k_27_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_3187_begin_0 = const()[name = tensor("op_3187_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3187_end_0 = const()[name = tensor("op_3187_end_0"), val = tensor([2, 1024, 1, 64])]; + tensor var_3187_end_mask_0 = const()[name = tensor("op_3187_end_mask_0"), val = tensor([true, true, true, false])]; + tensor transpose_25 = transpose(perm = k_27_perm_0, x = k_25_cast)[name = tensor("transpose_25")]; + tensor var_3187_cast = slice_by_index(begin = var_3187_begin_0, end = var_3187_end_0, end_mask = var_3187_end_mask_0, x = transpose_25)[name = tensor("op_3187_cast")]; + tensor var_3191_begin_0 = const()[name = tensor("op_3191_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_3191_end_0 = const()[name = tensor("op_3191_end_0"), val = tensor([2, 1024, 1, 128])]; + tensor var_3191_end_mask_0 = const()[name = tensor("op_3191_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3191_cast = slice_by_index(begin = var_3191_begin_0, end = var_3191_end_0, end_mask = var_3191_end_mask_0, x = transpose_25)[name = tensor("op_3191_cast")]; + tensor var_3195_begin_0 = const()[name = tensor("op_3195_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_3195_end_0 = const()[name = tensor("op_3195_end_0"), val = tensor([2, 1024, 1, 192])]; + tensor var_3195_end_mask_0 = const()[name = tensor("op_3195_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3195_cast = slice_by_index(begin = var_3195_begin_0, end = var_3195_end_0, end_mask = var_3195_end_mask_0, x = transpose_25)[name = tensor("op_3195_cast")]; + tensor var_3199_begin_0 = const()[name = tensor("op_3199_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_3199_end_0 = const()[name = tensor("op_3199_end_0"), val = tensor([2, 1024, 1, 256])]; + tensor var_3199_end_mask_0 = const()[name = tensor("op_3199_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3199_cast = slice_by_index(begin = var_3199_begin_0, end = var_3199_end_0, end_mask = var_3199_end_mask_0, x = transpose_25)[name = tensor("op_3199_cast")]; + tensor var_3203_begin_0 = const()[name = tensor("op_3203_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_3203_end_0 = const()[name = tensor("op_3203_end_0"), val = tensor([2, 1024, 1, 320])]; + tensor var_3203_end_mask_0 = const()[name = tensor("op_3203_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3203_cast = slice_by_index(begin = var_3203_begin_0, end = var_3203_end_0, end_mask = var_3203_end_mask_0, x = transpose_25)[name = tensor("op_3203_cast")]; + tensor var_3207_begin_0 = const()[name = tensor("op_3207_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_3207_end_0 = const()[name = tensor("op_3207_end_0"), val = tensor([2, 1024, 1, 384])]; + tensor var_3207_end_mask_0 = const()[name = tensor("op_3207_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3207_cast = slice_by_index(begin = var_3207_begin_0, end = var_3207_end_0, end_mask = var_3207_end_mask_0, x = transpose_25)[name = tensor("op_3207_cast")]; + tensor var_3211_begin_0 = const()[name = tensor("op_3211_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_3211_end_0 = const()[name = tensor("op_3211_end_0"), val = tensor([2, 1024, 1, 448])]; + tensor var_3211_end_mask_0 = const()[name = tensor("op_3211_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3211_cast = slice_by_index(begin = var_3211_begin_0, end = var_3211_end_0, end_mask = var_3211_end_mask_0, x = transpose_25)[name = tensor("op_3211_cast")]; + tensor var_3215_begin_0 = const()[name = tensor("op_3215_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_3215_end_0 = const()[name = tensor("op_3215_end_0"), val = tensor([2, 1024, 1, 512])]; + tensor var_3215_end_mask_0 = const()[name = tensor("op_3215_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3215_cast = slice_by_index(begin = var_3215_begin_0, end = var_3215_end_0, end_mask = var_3215_end_mask_0, x = transpose_25)[name = tensor("op_3215_cast")]; + tensor var_3219_begin_0 = const()[name = tensor("op_3219_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_3219_end_0 = const()[name = tensor("op_3219_end_0"), val = tensor([2, 1024, 1, 576])]; + tensor var_3219_end_mask_0 = const()[name = tensor("op_3219_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3219_cast = slice_by_index(begin = var_3219_begin_0, end = var_3219_end_0, end_mask = var_3219_end_mask_0, x = transpose_25)[name = tensor("op_3219_cast")]; + tensor var_3223_begin_0 = const()[name = tensor("op_3223_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_3223_end_0 = const()[name = tensor("op_3223_end_0"), val = tensor([2, 1024, 1, 640])]; + tensor var_3223_end_mask_0 = const()[name = tensor("op_3223_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3223_cast = slice_by_index(begin = var_3223_begin_0, end = var_3223_end_0, end_mask = var_3223_end_mask_0, x = transpose_25)[name = tensor("op_3223_cast")]; + tensor var_3225_begin_0 = const()[name = tensor("op_3225_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3225_end_0 = const()[name = tensor("op_3225_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_3225_end_mask_0 = const()[name = tensor("op_3225_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3225_cast = slice_by_index(begin = var_3225_begin_0, end = var_3225_end_0, end_mask = var_3225_end_mask_0, x = v_13_cast)[name = tensor("op_3225_cast")]; + tensor var_3229_begin_0 = const()[name = tensor("op_3229_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_3229_end_0 = const()[name = tensor("op_3229_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_3229_end_mask_0 = const()[name = tensor("op_3229_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3229_cast = slice_by_index(begin = var_3229_begin_0, end = var_3229_end_0, end_mask = var_3229_end_mask_0, x = v_13_cast)[name = tensor("op_3229_cast")]; + tensor var_3233_begin_0 = const()[name = tensor("op_3233_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_3233_end_0 = const()[name = tensor("op_3233_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_3233_end_mask_0 = const()[name = tensor("op_3233_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3233_cast = slice_by_index(begin = var_3233_begin_0, end = var_3233_end_0, end_mask = var_3233_end_mask_0, x = v_13_cast)[name = tensor("op_3233_cast")]; + tensor var_3237_begin_0 = const()[name = tensor("op_3237_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_3237_end_0 = const()[name = tensor("op_3237_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_3237_end_mask_0 = const()[name = tensor("op_3237_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3237_cast = slice_by_index(begin = var_3237_begin_0, end = var_3237_end_0, end_mask = var_3237_end_mask_0, x = v_13_cast)[name = tensor("op_3237_cast")]; + tensor var_3241_begin_0 = const()[name = tensor("op_3241_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_3241_end_0 = const()[name = tensor("op_3241_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_3241_end_mask_0 = const()[name = tensor("op_3241_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3241_cast = slice_by_index(begin = var_3241_begin_0, end = var_3241_end_0, end_mask = var_3241_end_mask_0, x = v_13_cast)[name = tensor("op_3241_cast")]; + tensor var_3245_begin_0 = const()[name = tensor("op_3245_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_3245_end_0 = const()[name = tensor("op_3245_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_3245_end_mask_0 = const()[name = tensor("op_3245_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3245_cast = slice_by_index(begin = var_3245_begin_0, end = var_3245_end_0, end_mask = var_3245_end_mask_0, x = v_13_cast)[name = tensor("op_3245_cast")]; + tensor var_3249_begin_0 = const()[name = tensor("op_3249_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_3249_end_0 = const()[name = tensor("op_3249_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_3249_end_mask_0 = const()[name = tensor("op_3249_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3249_cast = slice_by_index(begin = var_3249_begin_0, end = var_3249_end_0, end_mask = var_3249_end_mask_0, x = v_13_cast)[name = tensor("op_3249_cast")]; + tensor var_3253_begin_0 = const()[name = tensor("op_3253_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_3253_end_0 = const()[name = tensor("op_3253_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_3253_end_mask_0 = const()[name = tensor("op_3253_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3253_cast = slice_by_index(begin = var_3253_begin_0, end = var_3253_end_0, end_mask = var_3253_end_mask_0, x = v_13_cast)[name = tensor("op_3253_cast")]; + tensor var_3257_begin_0 = const()[name = tensor("op_3257_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_3257_end_0 = const()[name = tensor("op_3257_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_3257_end_mask_0 = const()[name = tensor("op_3257_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3257_cast = slice_by_index(begin = var_3257_begin_0, end = var_3257_end_0, end_mask = var_3257_end_mask_0, x = v_13_cast)[name = tensor("op_3257_cast")]; + tensor var_3261_begin_0 = const()[name = tensor("op_3261_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_3261_end_0 = const()[name = tensor("op_3261_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_3261_end_mask_0 = const()[name = tensor("op_3261_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3261_cast = slice_by_index(begin = var_3261_begin_0, end = var_3261_end_0, end_mask = var_3261_end_mask_0, x = v_13_cast)[name = tensor("op_3261_cast")]; + tensor var_3265_equation_0 = const()[name = tensor("op_3265_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3265_cast = einsum(equation = var_3265_equation_0, values = (var_3187_cast, var_3163_cast))[name = tensor("op_3265_cast")]; + tensor var_3266_to_fp16 = const()[name = tensor("op_3266_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_401_cast = mul(x = var_3265_cast, y = var_3266_to_fp16)[name = tensor("aw_chunk_401_cast")]; + tensor var_3269_equation_0 = const()[name = tensor("op_3269_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3269_cast = einsum(equation = var_3269_equation_0, values = (var_3187_cast, var_3164_cast))[name = tensor("op_3269_cast")]; + tensor var_3270_to_fp16 = const()[name = tensor("op_3270_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_403_cast = mul(x = var_3269_cast, y = var_3270_to_fp16)[name = tensor("aw_chunk_403_cast")]; + tensor var_3273_equation_0 = const()[name = tensor("op_3273_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3273_cast = einsum(equation = var_3273_equation_0, values = (var_3191_cast, var_3165_cast))[name = tensor("op_3273_cast")]; + tensor var_3274_to_fp16 = const()[name = tensor("op_3274_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_405_cast = mul(x = var_3273_cast, y = var_3274_to_fp16)[name = tensor("aw_chunk_405_cast")]; + tensor var_3277_equation_0 = const()[name = tensor("op_3277_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3277_cast = einsum(equation = var_3277_equation_0, values = (var_3191_cast, var_3166_cast))[name = tensor("op_3277_cast")]; + tensor var_3278_to_fp16 = const()[name = tensor("op_3278_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_407_cast = mul(x = var_3277_cast, y = var_3278_to_fp16)[name = tensor("aw_chunk_407_cast")]; + tensor var_3281_equation_0 = const()[name = tensor("op_3281_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3281_cast = einsum(equation = var_3281_equation_0, values = (var_3195_cast, var_3167_cast))[name = tensor("op_3281_cast")]; + tensor var_3282_to_fp16 = const()[name = tensor("op_3282_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_409_cast = mul(x = var_3281_cast, y = var_3282_to_fp16)[name = tensor("aw_chunk_409_cast")]; + tensor var_3285_equation_0 = const()[name = tensor("op_3285_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3285_cast = einsum(equation = var_3285_equation_0, values = (var_3195_cast, var_3168_cast))[name = tensor("op_3285_cast")]; + tensor var_3286_to_fp16 = const()[name = tensor("op_3286_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_411_cast = mul(x = var_3285_cast, y = var_3286_to_fp16)[name = tensor("aw_chunk_411_cast")]; + tensor var_3289_equation_0 = const()[name = tensor("op_3289_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3289_cast = einsum(equation = var_3289_equation_0, values = (var_3199_cast, var_3169_cast))[name = tensor("op_3289_cast")]; + tensor var_3290_to_fp16 = const()[name = tensor("op_3290_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_413_cast = mul(x = var_3289_cast, y = var_3290_to_fp16)[name = tensor("aw_chunk_413_cast")]; + tensor var_3293_equation_0 = const()[name = tensor("op_3293_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3293_cast = einsum(equation = var_3293_equation_0, values = (var_3199_cast, var_3170_cast))[name = tensor("op_3293_cast")]; + tensor var_3294_to_fp16 = const()[name = tensor("op_3294_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_415_cast = mul(x = var_3293_cast, y = var_3294_to_fp16)[name = tensor("aw_chunk_415_cast")]; + tensor var_3297_equation_0 = const()[name = tensor("op_3297_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3297_cast = einsum(equation = var_3297_equation_0, values = (var_3203_cast, var_3171_cast))[name = tensor("op_3297_cast")]; + tensor var_3298_to_fp16 = const()[name = tensor("op_3298_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_417_cast = mul(x = var_3297_cast, y = var_3298_to_fp16)[name = tensor("aw_chunk_417_cast")]; + tensor var_3301_equation_0 = const()[name = tensor("op_3301_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3301_cast = einsum(equation = var_3301_equation_0, values = (var_3203_cast, var_3172_cast))[name = tensor("op_3301_cast")]; + tensor var_3302_to_fp16 = const()[name = tensor("op_3302_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_419_cast = mul(x = var_3301_cast, y = var_3302_to_fp16)[name = tensor("aw_chunk_419_cast")]; + tensor var_3305_equation_0 = const()[name = tensor("op_3305_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3305_cast = einsum(equation = var_3305_equation_0, values = (var_3207_cast, var_3173_cast))[name = tensor("op_3305_cast")]; + tensor var_3306_to_fp16 = const()[name = tensor("op_3306_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_421_cast = mul(x = var_3305_cast, y = var_3306_to_fp16)[name = tensor("aw_chunk_421_cast")]; + tensor var_3309_equation_0 = const()[name = tensor("op_3309_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3309_cast = einsum(equation = var_3309_equation_0, values = (var_3207_cast, var_3174_cast))[name = tensor("op_3309_cast")]; + tensor var_3310_to_fp16 = const()[name = tensor("op_3310_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_423_cast = mul(x = var_3309_cast, y = var_3310_to_fp16)[name = tensor("aw_chunk_423_cast")]; + tensor var_3313_equation_0 = const()[name = tensor("op_3313_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3313_cast = einsum(equation = var_3313_equation_0, values = (var_3211_cast, var_3175_cast))[name = tensor("op_3313_cast")]; + tensor var_3314_to_fp16 = const()[name = tensor("op_3314_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_425_cast = mul(x = var_3313_cast, y = var_3314_to_fp16)[name = tensor("aw_chunk_425_cast")]; + tensor var_3317_equation_0 = const()[name = tensor("op_3317_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3317_cast = einsum(equation = var_3317_equation_0, values = (var_3211_cast, var_3176_cast))[name = tensor("op_3317_cast")]; + tensor var_3318_to_fp16 = const()[name = tensor("op_3318_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_427_cast = mul(x = var_3317_cast, y = var_3318_to_fp16)[name = tensor("aw_chunk_427_cast")]; + tensor var_3321_equation_0 = const()[name = tensor("op_3321_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3321_cast = einsum(equation = var_3321_equation_0, values = (var_3215_cast, var_3177_cast))[name = tensor("op_3321_cast")]; + tensor var_3322_to_fp16 = const()[name = tensor("op_3322_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_429_cast = mul(x = var_3321_cast, y = var_3322_to_fp16)[name = tensor("aw_chunk_429_cast")]; + tensor var_3325_equation_0 = const()[name = tensor("op_3325_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3325_cast = einsum(equation = var_3325_equation_0, values = (var_3215_cast, var_3178_cast))[name = tensor("op_3325_cast")]; + tensor var_3326_to_fp16 = const()[name = tensor("op_3326_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_431_cast = mul(x = var_3325_cast, y = var_3326_to_fp16)[name = tensor("aw_chunk_431_cast")]; + tensor var_3329_equation_0 = const()[name = tensor("op_3329_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3329_cast = einsum(equation = var_3329_equation_0, values = (var_3219_cast, var_3179_cast))[name = tensor("op_3329_cast")]; + tensor var_3330_to_fp16 = const()[name = tensor("op_3330_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_433_cast = mul(x = var_3329_cast, y = var_3330_to_fp16)[name = tensor("aw_chunk_433_cast")]; + tensor var_3333_equation_0 = const()[name = tensor("op_3333_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3333_cast = einsum(equation = var_3333_equation_0, values = (var_3219_cast, var_3180_cast))[name = tensor("op_3333_cast")]; + tensor var_3334_to_fp16 = const()[name = tensor("op_3334_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_435_cast = mul(x = var_3333_cast, y = var_3334_to_fp16)[name = tensor("aw_chunk_435_cast")]; + tensor var_3337_equation_0 = const()[name = tensor("op_3337_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3337_cast = einsum(equation = var_3337_equation_0, values = (var_3223_cast, var_3181_cast))[name = tensor("op_3337_cast")]; + tensor var_3338_to_fp16 = const()[name = tensor("op_3338_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_437_cast = mul(x = var_3337_cast, y = var_3338_to_fp16)[name = tensor("aw_chunk_437_cast")]; + tensor var_3341_equation_0 = const()[name = tensor("op_3341_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3341_cast = einsum(equation = var_3341_equation_0, values = (var_3223_cast, var_3182_cast))[name = tensor("op_3341_cast")]; + tensor var_3342_to_fp16 = const()[name = tensor("op_3342_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_439_cast = mul(x = var_3341_cast, y = var_3342_to_fp16)[name = tensor("aw_chunk_439_cast")]; + tensor var_3344_cast = softmax(axis = var_2173, x = aw_chunk_401_cast)[name = tensor("op_3344_cast")]; + tensor var_3345_cast = softmax(axis = var_2173, x = aw_chunk_403_cast)[name = tensor("op_3345_cast")]; + tensor var_3346_cast = softmax(axis = var_2173, x = aw_chunk_405_cast)[name = tensor("op_3346_cast")]; + tensor var_3347_cast = softmax(axis = var_2173, x = aw_chunk_407_cast)[name = tensor("op_3347_cast")]; + tensor var_3348_cast = softmax(axis = var_2173, x = aw_chunk_409_cast)[name = tensor("op_3348_cast")]; + tensor var_3349_cast = softmax(axis = var_2173, x = aw_chunk_411_cast)[name = tensor("op_3349_cast")]; + tensor var_3350_cast = softmax(axis = var_2173, x = aw_chunk_413_cast)[name = tensor("op_3350_cast")]; + tensor var_3351_cast = softmax(axis = var_2173, x = aw_chunk_415_cast)[name = tensor("op_3351_cast")]; + tensor var_3352_cast = softmax(axis = var_2173, x = aw_chunk_417_cast)[name = tensor("op_3352_cast")]; + tensor var_3353_cast = softmax(axis = var_2173, x = aw_chunk_419_cast)[name = tensor("op_3353_cast")]; + tensor var_3354_cast = softmax(axis = var_2173, x = aw_chunk_421_cast)[name = tensor("op_3354_cast")]; + tensor var_3355_cast = softmax(axis = var_2173, x = aw_chunk_423_cast)[name = tensor("op_3355_cast")]; + tensor var_3356_cast = softmax(axis = var_2173, x = aw_chunk_425_cast)[name = tensor("op_3356_cast")]; + tensor var_3357_cast = softmax(axis = var_2173, x = aw_chunk_427_cast)[name = tensor("op_3357_cast")]; + tensor var_3358_cast = softmax(axis = var_2173, x = aw_chunk_429_cast)[name = tensor("op_3358_cast")]; + tensor var_3359_cast = softmax(axis = var_2173, x = aw_chunk_431_cast)[name = tensor("op_3359_cast")]; + tensor var_3360_cast = softmax(axis = var_2173, x = aw_chunk_433_cast)[name = tensor("op_3360_cast")]; + tensor var_3361_cast = softmax(axis = var_2173, x = aw_chunk_435_cast)[name = tensor("op_3361_cast")]; + tensor var_3362_cast = softmax(axis = var_2173, x = aw_chunk_437_cast)[name = tensor("op_3362_cast")]; + tensor var_3363_cast = softmax(axis = var_2173, x = aw_chunk_439_cast)[name = tensor("op_3363_cast")]; + tensor var_3365_equation_0 = const()[name = tensor("op_3365_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3365_cast = einsum(equation = var_3365_equation_0, values = (var_3225_cast, var_3344_cast))[name = tensor("op_3365_cast")]; + tensor var_3367_equation_0 = const()[name = tensor("op_3367_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3367_cast = einsum(equation = var_3367_equation_0, values = (var_3225_cast, var_3345_cast))[name = tensor("op_3367_cast")]; + tensor var_3369_equation_0 = const()[name = tensor("op_3369_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3369_cast = einsum(equation = var_3369_equation_0, values = (var_3229_cast, var_3346_cast))[name = tensor("op_3369_cast")]; + tensor var_3371_equation_0 = const()[name = tensor("op_3371_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3371_cast = einsum(equation = var_3371_equation_0, values = (var_3229_cast, var_3347_cast))[name = tensor("op_3371_cast")]; + tensor var_3373_equation_0 = const()[name = tensor("op_3373_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3373_cast = einsum(equation = var_3373_equation_0, values = (var_3233_cast, var_3348_cast))[name = tensor("op_3373_cast")]; + tensor var_3375_equation_0 = const()[name = tensor("op_3375_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3375_cast = einsum(equation = var_3375_equation_0, values = (var_3233_cast, var_3349_cast))[name = tensor("op_3375_cast")]; + tensor var_3377_equation_0 = const()[name = tensor("op_3377_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3377_cast = einsum(equation = var_3377_equation_0, values = (var_3237_cast, var_3350_cast))[name = tensor("op_3377_cast")]; + tensor var_3379_equation_0 = const()[name = tensor("op_3379_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3379_cast = einsum(equation = var_3379_equation_0, values = (var_3237_cast, var_3351_cast))[name = tensor("op_3379_cast")]; + tensor var_3381_equation_0 = const()[name = tensor("op_3381_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3381_cast = einsum(equation = var_3381_equation_0, values = (var_3241_cast, var_3352_cast))[name = tensor("op_3381_cast")]; + tensor var_3383_equation_0 = const()[name = tensor("op_3383_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3383_cast = einsum(equation = var_3383_equation_0, values = (var_3241_cast, var_3353_cast))[name = tensor("op_3383_cast")]; + tensor var_3385_equation_0 = const()[name = tensor("op_3385_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3385_cast = einsum(equation = var_3385_equation_0, values = (var_3245_cast, var_3354_cast))[name = tensor("op_3385_cast")]; + tensor var_3387_equation_0 = const()[name = tensor("op_3387_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3387_cast = einsum(equation = var_3387_equation_0, values = (var_3245_cast, var_3355_cast))[name = tensor("op_3387_cast")]; + tensor var_3389_equation_0 = const()[name = tensor("op_3389_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3389_cast = einsum(equation = var_3389_equation_0, values = (var_3249_cast, var_3356_cast))[name = tensor("op_3389_cast")]; + tensor var_3391_equation_0 = const()[name = tensor("op_3391_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3391_cast = einsum(equation = var_3391_equation_0, values = (var_3249_cast, var_3357_cast))[name = tensor("op_3391_cast")]; + tensor var_3393_equation_0 = const()[name = tensor("op_3393_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3393_cast = einsum(equation = var_3393_equation_0, values = (var_3253_cast, var_3358_cast))[name = tensor("op_3393_cast")]; + tensor var_3395_equation_0 = const()[name = tensor("op_3395_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3395_cast = einsum(equation = var_3395_equation_0, values = (var_3253_cast, var_3359_cast))[name = tensor("op_3395_cast")]; + tensor var_3397_equation_0 = const()[name = tensor("op_3397_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3397_cast = einsum(equation = var_3397_equation_0, values = (var_3257_cast, var_3360_cast))[name = tensor("op_3397_cast")]; + tensor var_3399_equation_0 = const()[name = tensor("op_3399_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3399_cast = einsum(equation = var_3399_equation_0, values = (var_3257_cast, var_3361_cast))[name = tensor("op_3399_cast")]; + tensor var_3401_equation_0 = const()[name = tensor("op_3401_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3401_cast = einsum(equation = var_3401_equation_0, values = (var_3261_cast, var_3362_cast))[name = tensor("op_3401_cast")]; + tensor var_3403_equation_0 = const()[name = tensor("op_3403_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3403_cast = einsum(equation = var_3403_equation_0, values = (var_3261_cast, var_3363_cast))[name = tensor("op_3403_cast")]; + tensor var_3405_interleave_0 = const()[name = tensor("op_3405_interleave_0"), val = tensor(false)]; + tensor var_3405_cast = concat(axis = var_2150, interleave = var_3405_interleave_0, values = (var_3365_cast, var_3367_cast))[name = tensor("op_3405_cast")]; + tensor var_3407_interleave_0 = const()[name = tensor("op_3407_interleave_0"), val = tensor(false)]; + tensor var_3407_cast = concat(axis = var_2150, interleave = var_3407_interleave_0, values = (var_3369_cast, var_3371_cast))[name = tensor("op_3407_cast")]; + tensor var_3409_interleave_0 = const()[name = tensor("op_3409_interleave_0"), val = tensor(false)]; + tensor var_3409_cast = concat(axis = var_2150, interleave = var_3409_interleave_0, values = (var_3373_cast, var_3375_cast))[name = tensor("op_3409_cast")]; + tensor var_3411_interleave_0 = const()[name = tensor("op_3411_interleave_0"), val = tensor(false)]; + tensor var_3411_cast = concat(axis = var_2150, interleave = var_3411_interleave_0, values = (var_3377_cast, var_3379_cast))[name = tensor("op_3411_cast")]; + tensor var_3413_interleave_0 = const()[name = tensor("op_3413_interleave_0"), val = tensor(false)]; + tensor var_3413_cast = concat(axis = var_2150, interleave = var_3413_interleave_0, values = (var_3381_cast, var_3383_cast))[name = tensor("op_3413_cast")]; + tensor var_3415_interleave_0 = const()[name = tensor("op_3415_interleave_0"), val = tensor(false)]; + tensor var_3415_cast = concat(axis = var_2150, interleave = var_3415_interleave_0, values = (var_3385_cast, var_3387_cast))[name = tensor("op_3415_cast")]; + tensor var_3417_interleave_0 = const()[name = tensor("op_3417_interleave_0"), val = tensor(false)]; + tensor var_3417_cast = concat(axis = var_2150, interleave = var_3417_interleave_0, values = (var_3389_cast, var_3391_cast))[name = tensor("op_3417_cast")]; + tensor var_3419_interleave_0 = const()[name = tensor("op_3419_interleave_0"), val = tensor(false)]; + tensor var_3419_cast = concat(axis = var_2150, interleave = var_3419_interleave_0, values = (var_3393_cast, var_3395_cast))[name = tensor("op_3419_cast")]; + tensor var_3421_interleave_0 = const()[name = tensor("op_3421_interleave_0"), val = tensor(false)]; + tensor var_3421_cast = concat(axis = var_2150, interleave = var_3421_interleave_0, values = (var_3397_cast, var_3399_cast))[name = tensor("op_3421_cast")]; + tensor var_3423_interleave_0 = const()[name = tensor("op_3423_interleave_0"), val = tensor(false)]; + tensor var_3423_cast = concat(axis = var_2150, interleave = var_3423_interleave_0, values = (var_3401_cast, var_3403_cast))[name = tensor("op_3423_cast")]; + tensor input_105_interleave_0 = const()[name = tensor("input_105_interleave_0"), val = tensor(false)]; + tensor input_105_cast = concat(axis = var_2173, interleave = input_105_interleave_0, values = (var_3405_cast, var_3407_cast, var_3409_cast, var_3411_cast, var_3413_cast, var_3415_cast, var_3417_cast, var_3419_cast, var_3421_cast, var_3423_cast))[name = tensor("input_105_cast")]; + tensor var_3429 = const()[name = tensor("op_3429"), val = tensor([1, 1])]; + tensor var_3431 = const()[name = tensor("op_3431"), val = tensor([1, 1])]; + tensor var_3433_pad_type_0 = const()[name = tensor("op_3433_pad_type_0"), val = tensor("custom")]; + tensor var_3433_pad_0 = const()[name = tensor("op_3433_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_1_attentions_1_transformer_blocks_0_attn1_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(29190592))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(29497856))), name = tensor("down_blocks_1_attentions_1_transformer_blocks_0_attn1_to_out_0_weight_to_fp16_palettized"), shape = tensor([640, 640, 1, 1])]; + tensor down_blocks_1_attentions_1_transformer_blocks_0_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("down_blocks_1_attentions_1_transformer_blocks_0_attn1_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(29498048)))]; + tensor var_3433_cast = conv(bias = down_blocks_1_attentions_1_transformer_blocks_0_attn1_to_out_0_bias_to_fp16, dilations = var_3431, groups = var_2173, pad = var_3433_pad_0, pad_type = var_3433_pad_type_0, strides = var_3429, weight = down_blocks_1_attentions_1_transformer_blocks_0_attn1_to_out_0_weight_to_fp16_palettized, x = input_105_cast)[name = tensor("op_3433_cast")]; + tensor inputs_21_cast = add(x = var_3433_cast, y = inputs_19_cast)[name = tensor("inputs_21_cast")]; + tensor var_3437 = const()[name = tensor("op_3437"), val = tensor([1])]; + tensor channels_mean_21_cast = reduce_mean(axes = var_3437, keep_dims = var_2168, x = inputs_21_cast)[name = tensor("channels_mean_21_cast")]; + tensor zero_mean_21_cast = sub(x = inputs_21_cast, y = channels_mean_21_cast)[name = tensor("zero_mean_21_cast")]; + tensor zero_mean_sq_21_cast = mul(x = zero_mean_21_cast, y = zero_mean_21_cast)[name = tensor("zero_mean_sq_21_cast")]; + tensor var_3441 = const()[name = tensor("op_3441"), val = tensor([1])]; + tensor var_3442_cast = reduce_mean(axes = var_3441, keep_dims = var_2168, x = zero_mean_sq_21_cast)[name = tensor("op_3442_cast")]; + tensor var_3443_to_fp16 = const()[name = tensor("op_3443_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_3444_cast = add(x = var_3442_cast, y = var_3443_to_fp16)[name = tensor("op_3444_cast")]; + tensor denom_21_epsilon_0_to_fp16 = const()[name = tensor("denom_21_epsilon_0_to_fp16"), val = tensor(0x1p-24)]; + tensor denom_21_cast = rsqrt(epsilon = denom_21_epsilon_0_to_fp16, x = var_3444_cast)[name = tensor("denom_21_cast")]; + tensor out_21_cast = mul(x = zero_mean_21_cast, y = denom_21_cast)[name = tensor("out_21_cast")]; + tensor var_3448_to_fp16 = const()[name = tensor("op_3448_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(29499392)))]; + tensor var_3449_cast = add(x = out_21_cast, y = var_3448_to_fp16)[name = tensor("op_3449_cast")]; + tensor var_3451_to_fp16 = const()[name = tensor("op_3451_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(29500736)))]; + tensor hidden_states_65_cast = mul(x = var_3449_cast, y = var_3451_to_fp16)[name = tensor("hidden_states_65_cast")]; + tensor var_3458 = const()[name = tensor("op_3458"), val = tensor([1, 1])]; + tensor var_3460 = const()[name = tensor("op_3460"), val = tensor([1, 1])]; + tensor q_15_pad_type_0 = const()[name = tensor("q_15_pad_type_0"), val = tensor("custom")]; + tensor q_15_pad_0 = const()[name = tensor("q_15_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_1_attentions_1_transformer_blocks_0_attn2_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(29502080))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(29809344))), name = tensor("down_blocks_1_attentions_1_transformer_blocks_0_attn2_to_q_weight_to_fp16_palettized"), shape = tensor([640, 640, 1, 1])]; + tensor q_15_cast = conv(dilations = var_3460, groups = var_2173, pad = q_15_pad_0, pad_type = q_15_pad_type_0, strides = var_3458, weight = down_blocks_1_attentions_1_transformer_blocks_0_attn2_to_q_weight_to_fp16_palettized, x = hidden_states_65_cast)[name = tensor("q_15_cast")]; + tensor var_3464 = const()[name = tensor("op_3464"), val = tensor([1, 1])]; + tensor var_3466 = const()[name = tensor("op_3466"), val = tensor([1, 1])]; + tensor k_29_pad_type_0 = const()[name = tensor("k_29_pad_type_0"), val = tensor("custom")]; + tensor k_29_pad_0 = const()[name = tensor("k_29_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_1_attentions_1_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(29809536))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(30301120))), name = tensor("down_blocks_1_attentions_1_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized"), shape = tensor([640, 1024, 1, 1])]; + tensor k_29_cast = conv(dilations = var_3466, groups = var_2173, pad = k_29_pad_0, pad_type = k_29_pad_type_0, strides = var_3464, weight = down_blocks_1_attentions_1_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor("k_29_cast")]; + tensor var_3470 = const()[name = tensor("op_3470"), val = tensor([1, 1])]; + tensor var_3472 = const()[name = tensor("op_3472"), val = tensor([1, 1])]; + tensor v_15_pad_type_0 = const()[name = tensor("v_15_pad_type_0"), val = tensor("custom")]; + tensor v_15_pad_0 = const()[name = tensor("v_15_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_1_attentions_1_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(30301312))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(30792896))), name = tensor("down_blocks_1_attentions_1_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized"), shape = tensor([640, 1024, 1, 1])]; + tensor v_15_cast = conv(dilations = var_3472, groups = var_2173, pad = v_15_pad_0, pad_type = v_15_pad_type_0, strides = var_3470, weight = down_blocks_1_attentions_1_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor("v_15_cast")]; + tensor var_3476_begin_0 = const()[name = tensor("op_3476_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3476_end_0 = const()[name = tensor("op_3476_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_3476_end_mask_0 = const()[name = tensor("op_3476_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3476_cast = slice_by_index(begin = var_3476_begin_0, end = var_3476_end_0, end_mask = var_3476_end_mask_0, x = q_15_cast)[name = tensor("op_3476_cast")]; + tensor var_3480_begin_0 = const()[name = tensor("op_3480_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_3480_end_0 = const()[name = tensor("op_3480_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_3480_end_mask_0 = const()[name = tensor("op_3480_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3480_cast = slice_by_index(begin = var_3480_begin_0, end = var_3480_end_0, end_mask = var_3480_end_mask_0, x = q_15_cast)[name = tensor("op_3480_cast")]; + tensor var_3484_begin_0 = const()[name = tensor("op_3484_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_3484_end_0 = const()[name = tensor("op_3484_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_3484_end_mask_0 = const()[name = tensor("op_3484_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3484_cast = slice_by_index(begin = var_3484_begin_0, end = var_3484_end_0, end_mask = var_3484_end_mask_0, x = q_15_cast)[name = tensor("op_3484_cast")]; + tensor var_3488_begin_0 = const()[name = tensor("op_3488_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_3488_end_0 = const()[name = tensor("op_3488_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_3488_end_mask_0 = const()[name = tensor("op_3488_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3488_cast = slice_by_index(begin = var_3488_begin_0, end = var_3488_end_0, end_mask = var_3488_end_mask_0, x = q_15_cast)[name = tensor("op_3488_cast")]; + tensor var_3492_begin_0 = const()[name = tensor("op_3492_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_3492_end_0 = const()[name = tensor("op_3492_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_3492_end_mask_0 = const()[name = tensor("op_3492_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3492_cast = slice_by_index(begin = var_3492_begin_0, end = var_3492_end_0, end_mask = var_3492_end_mask_0, x = q_15_cast)[name = tensor("op_3492_cast")]; + tensor var_3496_begin_0 = const()[name = tensor("op_3496_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_3496_end_0 = const()[name = tensor("op_3496_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_3496_end_mask_0 = const()[name = tensor("op_3496_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3496_cast = slice_by_index(begin = var_3496_begin_0, end = var_3496_end_0, end_mask = var_3496_end_mask_0, x = q_15_cast)[name = tensor("op_3496_cast")]; + tensor var_3500_begin_0 = const()[name = tensor("op_3500_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_3500_end_0 = const()[name = tensor("op_3500_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_3500_end_mask_0 = const()[name = tensor("op_3500_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3500_cast = slice_by_index(begin = var_3500_begin_0, end = var_3500_end_0, end_mask = var_3500_end_mask_0, x = q_15_cast)[name = tensor("op_3500_cast")]; + tensor var_3504_begin_0 = const()[name = tensor("op_3504_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_3504_end_0 = const()[name = tensor("op_3504_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_3504_end_mask_0 = const()[name = tensor("op_3504_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3504_cast = slice_by_index(begin = var_3504_begin_0, end = var_3504_end_0, end_mask = var_3504_end_mask_0, x = q_15_cast)[name = tensor("op_3504_cast")]; + tensor var_3508_begin_0 = const()[name = tensor("op_3508_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_3508_end_0 = const()[name = tensor("op_3508_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_3508_end_mask_0 = const()[name = tensor("op_3508_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3508_cast = slice_by_index(begin = var_3508_begin_0, end = var_3508_end_0, end_mask = var_3508_end_mask_0, x = q_15_cast)[name = tensor("op_3508_cast")]; + tensor var_3512_begin_0 = const()[name = tensor("op_3512_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_3512_end_0 = const()[name = tensor("op_3512_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_3512_end_mask_0 = const()[name = tensor("op_3512_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3512_cast = slice_by_index(begin = var_3512_begin_0, end = var_3512_end_0, end_mask = var_3512_end_mask_0, x = q_15_cast)[name = tensor("op_3512_cast")]; + tensor var_3515_begin_0 = const()[name = tensor("op_3515_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3515_end_0 = const()[name = tensor("op_3515_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_3515_end_mask_0 = const()[name = tensor("op_3515_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3515_cast = slice_by_index(begin = var_3515_begin_0, end = var_3515_end_0, end_mask = var_3515_end_mask_0, x = var_3476_cast)[name = tensor("op_3515_cast")]; + tensor var_3516_begin_0 = const()[name = tensor("op_3516_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_3516_end_0 = const()[name = tensor("op_3516_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_3516_end_mask_0 = const()[name = tensor("op_3516_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3516_cast = slice_by_index(begin = var_3516_begin_0, end = var_3516_end_0, end_mask = var_3516_end_mask_0, x = var_3476_cast)[name = tensor("op_3516_cast")]; + tensor var_3517_begin_0 = const()[name = tensor("op_3517_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3517_end_0 = const()[name = tensor("op_3517_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_3517_end_mask_0 = const()[name = tensor("op_3517_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3517_cast = slice_by_index(begin = var_3517_begin_0, end = var_3517_end_0, end_mask = var_3517_end_mask_0, x = var_3480_cast)[name = tensor("op_3517_cast")]; + tensor var_3518_begin_0 = const()[name = tensor("op_3518_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_3518_end_0 = const()[name = tensor("op_3518_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_3518_end_mask_0 = const()[name = tensor("op_3518_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3518_cast = slice_by_index(begin = var_3518_begin_0, end = var_3518_end_0, end_mask = var_3518_end_mask_0, x = var_3480_cast)[name = tensor("op_3518_cast")]; + tensor var_3519_begin_0 = const()[name = tensor("op_3519_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3519_end_0 = const()[name = tensor("op_3519_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_3519_end_mask_0 = const()[name = tensor("op_3519_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3519_cast = slice_by_index(begin = var_3519_begin_0, end = var_3519_end_0, end_mask = var_3519_end_mask_0, x = var_3484_cast)[name = tensor("op_3519_cast")]; + tensor var_3520_begin_0 = const()[name = tensor("op_3520_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_3520_end_0 = const()[name = tensor("op_3520_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_3520_end_mask_0 = const()[name = tensor("op_3520_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3520_cast = slice_by_index(begin = var_3520_begin_0, end = var_3520_end_0, end_mask = var_3520_end_mask_0, x = var_3484_cast)[name = tensor("op_3520_cast")]; + tensor var_3521_begin_0 = const()[name = tensor("op_3521_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3521_end_0 = const()[name = tensor("op_3521_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_3521_end_mask_0 = const()[name = tensor("op_3521_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3521_cast = slice_by_index(begin = var_3521_begin_0, end = var_3521_end_0, end_mask = var_3521_end_mask_0, x = var_3488_cast)[name = tensor("op_3521_cast")]; + tensor var_3522_begin_0 = const()[name = tensor("op_3522_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_3522_end_0 = const()[name = tensor("op_3522_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_3522_end_mask_0 = const()[name = tensor("op_3522_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3522_cast = slice_by_index(begin = var_3522_begin_0, end = var_3522_end_0, end_mask = var_3522_end_mask_0, x = var_3488_cast)[name = tensor("op_3522_cast")]; + tensor var_3523_begin_0 = const()[name = tensor("op_3523_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3523_end_0 = const()[name = tensor("op_3523_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_3523_end_mask_0 = const()[name = tensor("op_3523_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3523_cast = slice_by_index(begin = var_3523_begin_0, end = var_3523_end_0, end_mask = var_3523_end_mask_0, x = var_3492_cast)[name = tensor("op_3523_cast")]; + tensor var_3524_begin_0 = const()[name = tensor("op_3524_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_3524_end_0 = const()[name = tensor("op_3524_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_3524_end_mask_0 = const()[name = tensor("op_3524_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3524_cast = slice_by_index(begin = var_3524_begin_0, end = var_3524_end_0, end_mask = var_3524_end_mask_0, x = var_3492_cast)[name = tensor("op_3524_cast")]; + tensor var_3525_begin_0 = const()[name = tensor("op_3525_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3525_end_0 = const()[name = tensor("op_3525_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_3525_end_mask_0 = const()[name = tensor("op_3525_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3525_cast = slice_by_index(begin = var_3525_begin_0, end = var_3525_end_0, end_mask = var_3525_end_mask_0, x = var_3496_cast)[name = tensor("op_3525_cast")]; + tensor var_3526_begin_0 = const()[name = tensor("op_3526_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_3526_end_0 = const()[name = tensor("op_3526_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_3526_end_mask_0 = const()[name = tensor("op_3526_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3526_cast = slice_by_index(begin = var_3526_begin_0, end = var_3526_end_0, end_mask = var_3526_end_mask_0, x = var_3496_cast)[name = tensor("op_3526_cast")]; + tensor var_3527_begin_0 = const()[name = tensor("op_3527_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3527_end_0 = const()[name = tensor("op_3527_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_3527_end_mask_0 = const()[name = tensor("op_3527_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3527_cast = slice_by_index(begin = var_3527_begin_0, end = var_3527_end_0, end_mask = var_3527_end_mask_0, x = var_3500_cast)[name = tensor("op_3527_cast")]; + tensor var_3528_begin_0 = const()[name = tensor("op_3528_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_3528_end_0 = const()[name = tensor("op_3528_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_3528_end_mask_0 = const()[name = tensor("op_3528_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3528_cast = slice_by_index(begin = var_3528_begin_0, end = var_3528_end_0, end_mask = var_3528_end_mask_0, x = var_3500_cast)[name = tensor("op_3528_cast")]; + tensor var_3529_begin_0 = const()[name = tensor("op_3529_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3529_end_0 = const()[name = tensor("op_3529_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_3529_end_mask_0 = const()[name = tensor("op_3529_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3529_cast = slice_by_index(begin = var_3529_begin_0, end = var_3529_end_0, end_mask = var_3529_end_mask_0, x = var_3504_cast)[name = tensor("op_3529_cast")]; + tensor var_3530_begin_0 = const()[name = tensor("op_3530_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_3530_end_0 = const()[name = tensor("op_3530_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_3530_end_mask_0 = const()[name = tensor("op_3530_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3530_cast = slice_by_index(begin = var_3530_begin_0, end = var_3530_end_0, end_mask = var_3530_end_mask_0, x = var_3504_cast)[name = tensor("op_3530_cast")]; + tensor var_3531_begin_0 = const()[name = tensor("op_3531_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3531_end_0 = const()[name = tensor("op_3531_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_3531_end_mask_0 = const()[name = tensor("op_3531_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3531_cast = slice_by_index(begin = var_3531_begin_0, end = var_3531_end_0, end_mask = var_3531_end_mask_0, x = var_3508_cast)[name = tensor("op_3531_cast")]; + tensor var_3532_begin_0 = const()[name = tensor("op_3532_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_3532_end_0 = const()[name = tensor("op_3532_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_3532_end_mask_0 = const()[name = tensor("op_3532_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3532_cast = slice_by_index(begin = var_3532_begin_0, end = var_3532_end_0, end_mask = var_3532_end_mask_0, x = var_3508_cast)[name = tensor("op_3532_cast")]; + tensor var_3533_begin_0 = const()[name = tensor("op_3533_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3533_end_0 = const()[name = tensor("op_3533_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_3533_end_mask_0 = const()[name = tensor("op_3533_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3533_cast = slice_by_index(begin = var_3533_begin_0, end = var_3533_end_0, end_mask = var_3533_end_mask_0, x = var_3512_cast)[name = tensor("op_3533_cast")]; + tensor var_3534_begin_0 = const()[name = tensor("op_3534_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_3534_end_0 = const()[name = tensor("op_3534_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_3534_end_mask_0 = const()[name = tensor("op_3534_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3534_cast = slice_by_index(begin = var_3534_begin_0, end = var_3534_end_0, end_mask = var_3534_end_mask_0, x = var_3512_cast)[name = tensor("op_3534_cast")]; + tensor k_31_perm_0 = const()[name = tensor("k_31_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_3539_begin_0 = const()[name = tensor("op_3539_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3539_end_0 = const()[name = tensor("op_3539_end_0"), val = tensor([2, 77, 1, 64])]; + tensor var_3539_end_mask_0 = const()[name = tensor("op_3539_end_mask_0"), val = tensor([true, true, true, false])]; + tensor transpose_24 = transpose(perm = k_31_perm_0, x = k_29_cast)[name = tensor("transpose_24")]; + tensor var_3539_cast = slice_by_index(begin = var_3539_begin_0, end = var_3539_end_0, end_mask = var_3539_end_mask_0, x = transpose_24)[name = tensor("op_3539_cast")]; + tensor var_3543_begin_0 = const()[name = tensor("op_3543_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_3543_end_0 = const()[name = tensor("op_3543_end_0"), val = tensor([2, 77, 1, 128])]; + tensor var_3543_end_mask_0 = const()[name = tensor("op_3543_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3543_cast = slice_by_index(begin = var_3543_begin_0, end = var_3543_end_0, end_mask = var_3543_end_mask_0, x = transpose_24)[name = tensor("op_3543_cast")]; + tensor var_3547_begin_0 = const()[name = tensor("op_3547_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_3547_end_0 = const()[name = tensor("op_3547_end_0"), val = tensor([2, 77, 1, 192])]; + tensor var_3547_end_mask_0 = const()[name = tensor("op_3547_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3547_cast = slice_by_index(begin = var_3547_begin_0, end = var_3547_end_0, end_mask = var_3547_end_mask_0, x = transpose_24)[name = tensor("op_3547_cast")]; + tensor var_3551_begin_0 = const()[name = tensor("op_3551_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_3551_end_0 = const()[name = tensor("op_3551_end_0"), val = tensor([2, 77, 1, 256])]; + tensor var_3551_end_mask_0 = const()[name = tensor("op_3551_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3551_cast = slice_by_index(begin = var_3551_begin_0, end = var_3551_end_0, end_mask = var_3551_end_mask_0, x = transpose_24)[name = tensor("op_3551_cast")]; + tensor var_3555_begin_0 = const()[name = tensor("op_3555_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_3555_end_0 = const()[name = tensor("op_3555_end_0"), val = tensor([2, 77, 1, 320])]; + tensor var_3555_end_mask_0 = const()[name = tensor("op_3555_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3555_cast = slice_by_index(begin = var_3555_begin_0, end = var_3555_end_0, end_mask = var_3555_end_mask_0, x = transpose_24)[name = tensor("op_3555_cast")]; + tensor var_3559_begin_0 = const()[name = tensor("op_3559_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_3559_end_0 = const()[name = tensor("op_3559_end_0"), val = tensor([2, 77, 1, 384])]; + tensor var_3559_end_mask_0 = const()[name = tensor("op_3559_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3559_cast = slice_by_index(begin = var_3559_begin_0, end = var_3559_end_0, end_mask = var_3559_end_mask_0, x = transpose_24)[name = tensor("op_3559_cast")]; + tensor var_3563_begin_0 = const()[name = tensor("op_3563_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_3563_end_0 = const()[name = tensor("op_3563_end_0"), val = tensor([2, 77, 1, 448])]; + tensor var_3563_end_mask_0 = const()[name = tensor("op_3563_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3563_cast = slice_by_index(begin = var_3563_begin_0, end = var_3563_end_0, end_mask = var_3563_end_mask_0, x = transpose_24)[name = tensor("op_3563_cast")]; + tensor var_3567_begin_0 = const()[name = tensor("op_3567_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_3567_end_0 = const()[name = tensor("op_3567_end_0"), val = tensor([2, 77, 1, 512])]; + tensor var_3567_end_mask_0 = const()[name = tensor("op_3567_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3567_cast = slice_by_index(begin = var_3567_begin_0, end = var_3567_end_0, end_mask = var_3567_end_mask_0, x = transpose_24)[name = tensor("op_3567_cast")]; + tensor var_3571_begin_0 = const()[name = tensor("op_3571_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_3571_end_0 = const()[name = tensor("op_3571_end_0"), val = tensor([2, 77, 1, 576])]; + tensor var_3571_end_mask_0 = const()[name = tensor("op_3571_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3571_cast = slice_by_index(begin = var_3571_begin_0, end = var_3571_end_0, end_mask = var_3571_end_mask_0, x = transpose_24)[name = tensor("op_3571_cast")]; + tensor var_3575_begin_0 = const()[name = tensor("op_3575_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_3575_end_0 = const()[name = tensor("op_3575_end_0"), val = tensor([2, 77, 1, 640])]; + tensor var_3575_end_mask_0 = const()[name = tensor("op_3575_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3575_cast = slice_by_index(begin = var_3575_begin_0, end = var_3575_end_0, end_mask = var_3575_end_mask_0, x = transpose_24)[name = tensor("op_3575_cast")]; + tensor var_3577_begin_0 = const()[name = tensor("op_3577_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3577_end_0 = const()[name = tensor("op_3577_end_0"), val = tensor([2, 64, 1, 77])]; + tensor var_3577_end_mask_0 = const()[name = tensor("op_3577_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3577_cast = slice_by_index(begin = var_3577_begin_0, end = var_3577_end_0, end_mask = var_3577_end_mask_0, x = v_15_cast)[name = tensor("op_3577_cast")]; + tensor var_3581_begin_0 = const()[name = tensor("op_3581_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_3581_end_0 = const()[name = tensor("op_3581_end_0"), val = tensor([2, 128, 1, 77])]; + tensor var_3581_end_mask_0 = const()[name = tensor("op_3581_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3581_cast = slice_by_index(begin = var_3581_begin_0, end = var_3581_end_0, end_mask = var_3581_end_mask_0, x = v_15_cast)[name = tensor("op_3581_cast")]; + tensor var_3585_begin_0 = const()[name = tensor("op_3585_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_3585_end_0 = const()[name = tensor("op_3585_end_0"), val = tensor([2, 192, 1, 77])]; + tensor var_3585_end_mask_0 = const()[name = tensor("op_3585_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3585_cast = slice_by_index(begin = var_3585_begin_0, end = var_3585_end_0, end_mask = var_3585_end_mask_0, x = v_15_cast)[name = tensor("op_3585_cast")]; + tensor var_3589_begin_0 = const()[name = tensor("op_3589_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_3589_end_0 = const()[name = tensor("op_3589_end_0"), val = tensor([2, 256, 1, 77])]; + tensor var_3589_end_mask_0 = const()[name = tensor("op_3589_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3589_cast = slice_by_index(begin = var_3589_begin_0, end = var_3589_end_0, end_mask = var_3589_end_mask_0, x = v_15_cast)[name = tensor("op_3589_cast")]; + tensor var_3593_begin_0 = const()[name = tensor("op_3593_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_3593_end_0 = const()[name = tensor("op_3593_end_0"), val = tensor([2, 320, 1, 77])]; + tensor var_3593_end_mask_0 = const()[name = tensor("op_3593_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3593_cast = slice_by_index(begin = var_3593_begin_0, end = var_3593_end_0, end_mask = var_3593_end_mask_0, x = v_15_cast)[name = tensor("op_3593_cast")]; + tensor var_3597_begin_0 = const()[name = tensor("op_3597_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_3597_end_0 = const()[name = tensor("op_3597_end_0"), val = tensor([2, 384, 1, 77])]; + tensor var_3597_end_mask_0 = const()[name = tensor("op_3597_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3597_cast = slice_by_index(begin = var_3597_begin_0, end = var_3597_end_0, end_mask = var_3597_end_mask_0, x = v_15_cast)[name = tensor("op_3597_cast")]; + tensor var_3601_begin_0 = const()[name = tensor("op_3601_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_3601_end_0 = const()[name = tensor("op_3601_end_0"), val = tensor([2, 448, 1, 77])]; + tensor var_3601_end_mask_0 = const()[name = tensor("op_3601_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3601_cast = slice_by_index(begin = var_3601_begin_0, end = var_3601_end_0, end_mask = var_3601_end_mask_0, x = v_15_cast)[name = tensor("op_3601_cast")]; + tensor var_3605_begin_0 = const()[name = tensor("op_3605_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_3605_end_0 = const()[name = tensor("op_3605_end_0"), val = tensor([2, 512, 1, 77])]; + tensor var_3605_end_mask_0 = const()[name = tensor("op_3605_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3605_cast = slice_by_index(begin = var_3605_begin_0, end = var_3605_end_0, end_mask = var_3605_end_mask_0, x = v_15_cast)[name = tensor("op_3605_cast")]; + tensor var_3609_begin_0 = const()[name = tensor("op_3609_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_3609_end_0 = const()[name = tensor("op_3609_end_0"), val = tensor([2, 576, 1, 77])]; + tensor var_3609_end_mask_0 = const()[name = tensor("op_3609_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3609_cast = slice_by_index(begin = var_3609_begin_0, end = var_3609_end_0, end_mask = var_3609_end_mask_0, x = v_15_cast)[name = tensor("op_3609_cast")]; + tensor var_3613_begin_0 = const()[name = tensor("op_3613_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_3613_end_0 = const()[name = tensor("op_3613_end_0"), val = tensor([2, 640, 1, 77])]; + tensor var_3613_end_mask_0 = const()[name = tensor("op_3613_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3613_cast = slice_by_index(begin = var_3613_begin_0, end = var_3613_end_0, end_mask = var_3613_end_mask_0, x = v_15_cast)[name = tensor("op_3613_cast")]; + tensor var_3617_equation_0 = const()[name = tensor("op_3617_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3617_cast = einsum(equation = var_3617_equation_0, values = (var_3539_cast, var_3515_cast))[name = tensor("op_3617_cast")]; + tensor var_3618_to_fp16 = const()[name = tensor("op_3618_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_441_cast = mul(x = var_3617_cast, y = var_3618_to_fp16)[name = tensor("aw_chunk_441_cast")]; + tensor var_3621_equation_0 = const()[name = tensor("op_3621_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3621_cast = einsum(equation = var_3621_equation_0, values = (var_3539_cast, var_3516_cast))[name = tensor("op_3621_cast")]; + tensor var_3622_to_fp16 = const()[name = tensor("op_3622_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_443_cast = mul(x = var_3621_cast, y = var_3622_to_fp16)[name = tensor("aw_chunk_443_cast")]; + tensor var_3625_equation_0 = const()[name = tensor("op_3625_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3625_cast = einsum(equation = var_3625_equation_0, values = (var_3543_cast, var_3517_cast))[name = tensor("op_3625_cast")]; + tensor var_3626_to_fp16 = const()[name = tensor("op_3626_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_445_cast = mul(x = var_3625_cast, y = var_3626_to_fp16)[name = tensor("aw_chunk_445_cast")]; + tensor var_3629_equation_0 = const()[name = tensor("op_3629_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3629_cast = einsum(equation = var_3629_equation_0, values = (var_3543_cast, var_3518_cast))[name = tensor("op_3629_cast")]; + tensor var_3630_to_fp16 = const()[name = tensor("op_3630_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_447_cast = mul(x = var_3629_cast, y = var_3630_to_fp16)[name = tensor("aw_chunk_447_cast")]; + tensor var_3633_equation_0 = const()[name = tensor("op_3633_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3633_cast = einsum(equation = var_3633_equation_0, values = (var_3547_cast, var_3519_cast))[name = tensor("op_3633_cast")]; + tensor var_3634_to_fp16 = const()[name = tensor("op_3634_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_449_cast = mul(x = var_3633_cast, y = var_3634_to_fp16)[name = tensor("aw_chunk_449_cast")]; + tensor var_3637_equation_0 = const()[name = tensor("op_3637_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3637_cast = einsum(equation = var_3637_equation_0, values = (var_3547_cast, var_3520_cast))[name = tensor("op_3637_cast")]; + tensor var_3638_to_fp16 = const()[name = tensor("op_3638_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_451_cast = mul(x = var_3637_cast, y = var_3638_to_fp16)[name = tensor("aw_chunk_451_cast")]; + tensor var_3641_equation_0 = const()[name = tensor("op_3641_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3641_cast = einsum(equation = var_3641_equation_0, values = (var_3551_cast, var_3521_cast))[name = tensor("op_3641_cast")]; + tensor var_3642_to_fp16 = const()[name = tensor("op_3642_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_453_cast = mul(x = var_3641_cast, y = var_3642_to_fp16)[name = tensor("aw_chunk_453_cast")]; + tensor var_3645_equation_0 = const()[name = tensor("op_3645_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3645_cast = einsum(equation = var_3645_equation_0, values = (var_3551_cast, var_3522_cast))[name = tensor("op_3645_cast")]; + tensor var_3646_to_fp16 = const()[name = tensor("op_3646_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_455_cast = mul(x = var_3645_cast, y = var_3646_to_fp16)[name = tensor("aw_chunk_455_cast")]; + tensor var_3649_equation_0 = const()[name = tensor("op_3649_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3649_cast = einsum(equation = var_3649_equation_0, values = (var_3555_cast, var_3523_cast))[name = tensor("op_3649_cast")]; + tensor var_3650_to_fp16 = const()[name = tensor("op_3650_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_457_cast = mul(x = var_3649_cast, y = var_3650_to_fp16)[name = tensor("aw_chunk_457_cast")]; + tensor var_3653_equation_0 = const()[name = tensor("op_3653_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3653_cast = einsum(equation = var_3653_equation_0, values = (var_3555_cast, var_3524_cast))[name = tensor("op_3653_cast")]; + tensor var_3654_to_fp16 = const()[name = tensor("op_3654_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_459_cast = mul(x = var_3653_cast, y = var_3654_to_fp16)[name = tensor("aw_chunk_459_cast")]; + tensor var_3657_equation_0 = const()[name = tensor("op_3657_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3657_cast = einsum(equation = var_3657_equation_0, values = (var_3559_cast, var_3525_cast))[name = tensor("op_3657_cast")]; + tensor var_3658_to_fp16 = const()[name = tensor("op_3658_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_461_cast = mul(x = var_3657_cast, y = var_3658_to_fp16)[name = tensor("aw_chunk_461_cast")]; + tensor var_3661_equation_0 = const()[name = tensor("op_3661_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3661_cast = einsum(equation = var_3661_equation_0, values = (var_3559_cast, var_3526_cast))[name = tensor("op_3661_cast")]; + tensor var_3662_to_fp16 = const()[name = tensor("op_3662_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_463_cast = mul(x = var_3661_cast, y = var_3662_to_fp16)[name = tensor("aw_chunk_463_cast")]; + tensor var_3665_equation_0 = const()[name = tensor("op_3665_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3665_cast = einsum(equation = var_3665_equation_0, values = (var_3563_cast, var_3527_cast))[name = tensor("op_3665_cast")]; + tensor var_3666_to_fp16 = const()[name = tensor("op_3666_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_465_cast = mul(x = var_3665_cast, y = var_3666_to_fp16)[name = tensor("aw_chunk_465_cast")]; + tensor var_3669_equation_0 = const()[name = tensor("op_3669_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3669_cast = einsum(equation = var_3669_equation_0, values = (var_3563_cast, var_3528_cast))[name = tensor("op_3669_cast")]; + tensor var_3670_to_fp16 = const()[name = tensor("op_3670_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_467_cast = mul(x = var_3669_cast, y = var_3670_to_fp16)[name = tensor("aw_chunk_467_cast")]; + tensor var_3673_equation_0 = const()[name = tensor("op_3673_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3673_cast = einsum(equation = var_3673_equation_0, values = (var_3567_cast, var_3529_cast))[name = tensor("op_3673_cast")]; + tensor var_3674_to_fp16 = const()[name = tensor("op_3674_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_469_cast = mul(x = var_3673_cast, y = var_3674_to_fp16)[name = tensor("aw_chunk_469_cast")]; + tensor var_3677_equation_0 = const()[name = tensor("op_3677_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3677_cast = einsum(equation = var_3677_equation_0, values = (var_3567_cast, var_3530_cast))[name = tensor("op_3677_cast")]; + tensor var_3678_to_fp16 = const()[name = tensor("op_3678_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_471_cast = mul(x = var_3677_cast, y = var_3678_to_fp16)[name = tensor("aw_chunk_471_cast")]; + tensor var_3681_equation_0 = const()[name = tensor("op_3681_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3681_cast = einsum(equation = var_3681_equation_0, values = (var_3571_cast, var_3531_cast))[name = tensor("op_3681_cast")]; + tensor var_3682_to_fp16 = const()[name = tensor("op_3682_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_473_cast = mul(x = var_3681_cast, y = var_3682_to_fp16)[name = tensor("aw_chunk_473_cast")]; + tensor var_3685_equation_0 = const()[name = tensor("op_3685_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3685_cast = einsum(equation = var_3685_equation_0, values = (var_3571_cast, var_3532_cast))[name = tensor("op_3685_cast")]; + tensor var_3686_to_fp16 = const()[name = tensor("op_3686_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_475_cast = mul(x = var_3685_cast, y = var_3686_to_fp16)[name = tensor("aw_chunk_475_cast")]; + tensor var_3689_equation_0 = const()[name = tensor("op_3689_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3689_cast = einsum(equation = var_3689_equation_0, values = (var_3575_cast, var_3533_cast))[name = tensor("op_3689_cast")]; + tensor var_3690_to_fp16 = const()[name = tensor("op_3690_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_477_cast = mul(x = var_3689_cast, y = var_3690_to_fp16)[name = tensor("aw_chunk_477_cast")]; + tensor var_3693_equation_0 = const()[name = tensor("op_3693_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3693_cast = einsum(equation = var_3693_equation_0, values = (var_3575_cast, var_3534_cast))[name = tensor("op_3693_cast")]; + tensor var_3694_to_fp16 = const()[name = tensor("op_3694_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_479_cast = mul(x = var_3693_cast, y = var_3694_to_fp16)[name = tensor("aw_chunk_479_cast")]; + tensor var_3696_cast = softmax(axis = var_2173, x = aw_chunk_441_cast)[name = tensor("op_3696_cast")]; + tensor var_3697_cast = softmax(axis = var_2173, x = aw_chunk_443_cast)[name = tensor("op_3697_cast")]; + tensor var_3698_cast = softmax(axis = var_2173, x = aw_chunk_445_cast)[name = tensor("op_3698_cast")]; + tensor var_3699_cast = softmax(axis = var_2173, x = aw_chunk_447_cast)[name = tensor("op_3699_cast")]; + tensor var_3700_cast = softmax(axis = var_2173, x = aw_chunk_449_cast)[name = tensor("op_3700_cast")]; + tensor var_3701_cast = softmax(axis = var_2173, x = aw_chunk_451_cast)[name = tensor("op_3701_cast")]; + tensor var_3702_cast = softmax(axis = var_2173, x = aw_chunk_453_cast)[name = tensor("op_3702_cast")]; + tensor var_3703_cast = softmax(axis = var_2173, x = aw_chunk_455_cast)[name = tensor("op_3703_cast")]; + tensor var_3704_cast = softmax(axis = var_2173, x = aw_chunk_457_cast)[name = tensor("op_3704_cast")]; + tensor var_3705_cast = softmax(axis = var_2173, x = aw_chunk_459_cast)[name = tensor("op_3705_cast")]; + tensor var_3706_cast = softmax(axis = var_2173, x = aw_chunk_461_cast)[name = tensor("op_3706_cast")]; + tensor var_3707_cast = softmax(axis = var_2173, x = aw_chunk_463_cast)[name = tensor("op_3707_cast")]; + tensor var_3708_cast = softmax(axis = var_2173, x = aw_chunk_465_cast)[name = tensor("op_3708_cast")]; + tensor var_3709_cast = softmax(axis = var_2173, x = aw_chunk_467_cast)[name = tensor("op_3709_cast")]; + tensor var_3710_cast = softmax(axis = var_2173, x = aw_chunk_469_cast)[name = tensor("op_3710_cast")]; + tensor var_3711_cast = softmax(axis = var_2173, x = aw_chunk_471_cast)[name = tensor("op_3711_cast")]; + tensor var_3712_cast = softmax(axis = var_2173, x = aw_chunk_473_cast)[name = tensor("op_3712_cast")]; + tensor var_3713_cast = softmax(axis = var_2173, x = aw_chunk_475_cast)[name = tensor("op_3713_cast")]; + tensor var_3714_cast = softmax(axis = var_2173, x = aw_chunk_477_cast)[name = tensor("op_3714_cast")]; + tensor var_3715_cast = softmax(axis = var_2173, x = aw_chunk_479_cast)[name = tensor("op_3715_cast")]; + tensor var_3717_equation_0 = const()[name = tensor("op_3717_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3717_cast = einsum(equation = var_3717_equation_0, values = (var_3577_cast, var_3696_cast))[name = tensor("op_3717_cast")]; + tensor var_3719_equation_0 = const()[name = tensor("op_3719_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3719_cast = einsum(equation = var_3719_equation_0, values = (var_3577_cast, var_3697_cast))[name = tensor("op_3719_cast")]; + tensor var_3721_equation_0 = const()[name = tensor("op_3721_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3721_cast = einsum(equation = var_3721_equation_0, values = (var_3581_cast, var_3698_cast))[name = tensor("op_3721_cast")]; + tensor var_3723_equation_0 = const()[name = tensor("op_3723_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3723_cast = einsum(equation = var_3723_equation_0, values = (var_3581_cast, var_3699_cast))[name = tensor("op_3723_cast")]; + tensor var_3725_equation_0 = const()[name = tensor("op_3725_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3725_cast = einsum(equation = var_3725_equation_0, values = (var_3585_cast, var_3700_cast))[name = tensor("op_3725_cast")]; + tensor var_3727_equation_0 = const()[name = tensor("op_3727_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3727_cast = einsum(equation = var_3727_equation_0, values = (var_3585_cast, var_3701_cast))[name = tensor("op_3727_cast")]; + tensor var_3729_equation_0 = const()[name = tensor("op_3729_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3729_cast = einsum(equation = var_3729_equation_0, values = (var_3589_cast, var_3702_cast))[name = tensor("op_3729_cast")]; + tensor var_3731_equation_0 = const()[name = tensor("op_3731_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3731_cast = einsum(equation = var_3731_equation_0, values = (var_3589_cast, var_3703_cast))[name = tensor("op_3731_cast")]; + tensor var_3733_equation_0 = const()[name = tensor("op_3733_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3733_cast = einsum(equation = var_3733_equation_0, values = (var_3593_cast, var_3704_cast))[name = tensor("op_3733_cast")]; + tensor var_3735_equation_0 = const()[name = tensor("op_3735_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3735_cast = einsum(equation = var_3735_equation_0, values = (var_3593_cast, var_3705_cast))[name = tensor("op_3735_cast")]; + tensor var_3737_equation_0 = const()[name = tensor("op_3737_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3737_cast = einsum(equation = var_3737_equation_0, values = (var_3597_cast, var_3706_cast))[name = tensor("op_3737_cast")]; + tensor var_3739_equation_0 = const()[name = tensor("op_3739_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3739_cast = einsum(equation = var_3739_equation_0, values = (var_3597_cast, var_3707_cast))[name = tensor("op_3739_cast")]; + tensor var_3741_equation_0 = const()[name = tensor("op_3741_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3741_cast = einsum(equation = var_3741_equation_0, values = (var_3601_cast, var_3708_cast))[name = tensor("op_3741_cast")]; + tensor var_3743_equation_0 = const()[name = tensor("op_3743_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3743_cast = einsum(equation = var_3743_equation_0, values = (var_3601_cast, var_3709_cast))[name = tensor("op_3743_cast")]; + tensor var_3745_equation_0 = const()[name = tensor("op_3745_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3745_cast = einsum(equation = var_3745_equation_0, values = (var_3605_cast, var_3710_cast))[name = tensor("op_3745_cast")]; + tensor var_3747_equation_0 = const()[name = tensor("op_3747_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3747_cast = einsum(equation = var_3747_equation_0, values = (var_3605_cast, var_3711_cast))[name = tensor("op_3747_cast")]; + tensor var_3749_equation_0 = const()[name = tensor("op_3749_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3749_cast = einsum(equation = var_3749_equation_0, values = (var_3609_cast, var_3712_cast))[name = tensor("op_3749_cast")]; + tensor var_3751_equation_0 = const()[name = tensor("op_3751_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3751_cast = einsum(equation = var_3751_equation_0, values = (var_3609_cast, var_3713_cast))[name = tensor("op_3751_cast")]; + tensor var_3753_equation_0 = const()[name = tensor("op_3753_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3753_cast = einsum(equation = var_3753_equation_0, values = (var_3613_cast, var_3714_cast))[name = tensor("op_3753_cast")]; + tensor var_3755_equation_0 = const()[name = tensor("op_3755_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3755_cast = einsum(equation = var_3755_equation_0, values = (var_3613_cast, var_3715_cast))[name = tensor("op_3755_cast")]; + tensor var_3757_interleave_0 = const()[name = tensor("op_3757_interleave_0"), val = tensor(false)]; + tensor var_3757_cast = concat(axis = var_2150, interleave = var_3757_interleave_0, values = (var_3717_cast, var_3719_cast))[name = tensor("op_3757_cast")]; + tensor var_3759_interleave_0 = const()[name = tensor("op_3759_interleave_0"), val = tensor(false)]; + tensor var_3759_cast = concat(axis = var_2150, interleave = var_3759_interleave_0, values = (var_3721_cast, var_3723_cast))[name = tensor("op_3759_cast")]; + tensor var_3761_interleave_0 = const()[name = tensor("op_3761_interleave_0"), val = tensor(false)]; + tensor var_3761_cast = concat(axis = var_2150, interleave = var_3761_interleave_0, values = (var_3725_cast, var_3727_cast))[name = tensor("op_3761_cast")]; + tensor var_3763_interleave_0 = const()[name = tensor("op_3763_interleave_0"), val = tensor(false)]; + tensor var_3763_cast = concat(axis = var_2150, interleave = var_3763_interleave_0, values = (var_3729_cast, var_3731_cast))[name = tensor("op_3763_cast")]; + tensor var_3765_interleave_0 = const()[name = tensor("op_3765_interleave_0"), val = tensor(false)]; + tensor var_3765_cast = concat(axis = var_2150, interleave = var_3765_interleave_0, values = (var_3733_cast, var_3735_cast))[name = tensor("op_3765_cast")]; + tensor var_3767_interleave_0 = const()[name = tensor("op_3767_interleave_0"), val = tensor(false)]; + tensor var_3767_cast = concat(axis = var_2150, interleave = var_3767_interleave_0, values = (var_3737_cast, var_3739_cast))[name = tensor("op_3767_cast")]; + tensor var_3769_interleave_0 = const()[name = tensor("op_3769_interleave_0"), val = tensor(false)]; + tensor var_3769_cast = concat(axis = var_2150, interleave = var_3769_interleave_0, values = (var_3741_cast, var_3743_cast))[name = tensor("op_3769_cast")]; + tensor var_3771_interleave_0 = const()[name = tensor("op_3771_interleave_0"), val = tensor(false)]; + tensor var_3771_cast = concat(axis = var_2150, interleave = var_3771_interleave_0, values = (var_3745_cast, var_3747_cast))[name = tensor("op_3771_cast")]; + tensor var_3773_interleave_0 = const()[name = tensor("op_3773_interleave_0"), val = tensor(false)]; + tensor var_3773_cast = concat(axis = var_2150, interleave = var_3773_interleave_0, values = (var_3749_cast, var_3751_cast))[name = tensor("op_3773_cast")]; + tensor var_3775_interleave_0 = const()[name = tensor("op_3775_interleave_0"), val = tensor(false)]; + tensor var_3775_cast = concat(axis = var_2150, interleave = var_3775_interleave_0, values = (var_3753_cast, var_3755_cast))[name = tensor("op_3775_cast")]; + tensor input_107_interleave_0 = const()[name = tensor("input_107_interleave_0"), val = tensor(false)]; + tensor input_107_cast = concat(axis = var_2173, interleave = input_107_interleave_0, values = (var_3757_cast, var_3759_cast, var_3761_cast, var_3763_cast, var_3765_cast, var_3767_cast, var_3769_cast, var_3771_cast, var_3773_cast, var_3775_cast))[name = tensor("input_107_cast")]; + tensor var_3781 = const()[name = tensor("op_3781"), val = tensor([1, 1])]; + tensor var_3783 = const()[name = tensor("op_3783"), val = tensor([1, 1])]; + tensor var_3785_pad_type_0 = const()[name = tensor("op_3785_pad_type_0"), val = tensor("custom")]; + tensor var_3785_pad_0 = const()[name = tensor("op_3785_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_1_attentions_1_transformer_blocks_0_attn2_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(30793088))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(31100352))), name = tensor("down_blocks_1_attentions_1_transformer_blocks_0_attn2_to_out_0_weight_to_fp16_palettized"), shape = tensor([640, 640, 1, 1])]; + tensor down_blocks_1_attentions_1_transformer_blocks_0_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("down_blocks_1_attentions_1_transformer_blocks_0_attn2_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(31100544)))]; + tensor var_3785_cast = conv(bias = down_blocks_1_attentions_1_transformer_blocks_0_attn2_to_out_0_bias_to_fp16, dilations = var_3783, groups = var_2173, pad = var_3785_pad_0, pad_type = var_3785_pad_type_0, strides = var_3781, weight = down_blocks_1_attentions_1_transformer_blocks_0_attn2_to_out_0_weight_to_fp16_palettized, x = input_107_cast)[name = tensor("op_3785_cast")]; + tensor inputs_23_cast = add(x = var_3785_cast, y = inputs_21_cast)[name = tensor("inputs_23_cast")]; + tensor var_3789 = const()[name = tensor("op_3789"), val = tensor([1])]; + tensor channels_mean_23_cast = reduce_mean(axes = var_3789, keep_dims = var_2168, x = inputs_23_cast)[name = tensor("channels_mean_23_cast")]; + tensor zero_mean_23_cast = sub(x = inputs_23_cast, y = channels_mean_23_cast)[name = tensor("zero_mean_23_cast")]; + tensor zero_mean_sq_23_cast = mul(x = zero_mean_23_cast, y = zero_mean_23_cast)[name = tensor("zero_mean_sq_23_cast")]; + tensor var_3793 = const()[name = tensor("op_3793"), val = tensor([1])]; + tensor var_3794_cast = reduce_mean(axes = var_3793, keep_dims = var_2168, x = zero_mean_sq_23_cast)[name = tensor("op_3794_cast")]; + tensor var_3795_to_fp16 = const()[name = tensor("op_3795_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_3796_cast = add(x = var_3794_cast, y = var_3795_to_fp16)[name = tensor("op_3796_cast")]; + tensor denom_23_epsilon_0_to_fp16 = const()[name = tensor("denom_23_epsilon_0_to_fp16"), val = tensor(0x1p-24)]; + tensor denom_23_cast = rsqrt(epsilon = denom_23_epsilon_0_to_fp16, x = var_3796_cast)[name = tensor("denom_23_cast")]; + tensor out_23_cast = mul(x = zero_mean_23_cast, y = denom_23_cast)[name = tensor("out_23_cast")]; + tensor var_3800_to_fp16 = const()[name = tensor("op_3800_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(31101888)))]; + tensor var_3801_cast = add(x = out_23_cast, y = var_3800_to_fp16)[name = tensor("op_3801_cast")]; + tensor var_3803_to_fp16 = const()[name = tensor("op_3803_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(31103232)))]; + tensor input_109_cast = mul(x = var_3801_cast, y = var_3803_to_fp16)[name = tensor("input_109_cast")]; + tensor var_3811 = const()[name = tensor("op_3811"), val = tensor([1, 1])]; + tensor var_3813 = const()[name = tensor("op_3813"), val = tensor([1, 1])]; + tensor var_3815_pad_type_0 = const()[name = tensor("op_3815_pad_type_0"), val = tensor("custom")]; + tensor var_3815_pad_0 = const()[name = tensor("op_3815_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_1_attentions_1_transformer_blocks_0_ff_net_0_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(31104576))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(33562240))), name = tensor("down_blocks_1_attentions_1_transformer_blocks_0_ff_net_0_proj_weight_to_fp16_palettized"), shape = tensor([5120, 640, 1, 1])]; + tensor down_blocks_1_attentions_1_transformer_blocks_0_ff_net_0_proj_bias_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(33562432))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(33566336))), name = tensor("down_blocks_1_attentions_1_transformer_blocks_0_ff_net_0_proj_bias_to_fp16_palettized"), shape = tensor([5120])]; + tensor var_3815_cast = conv(bias = down_blocks_1_attentions_1_transformer_blocks_0_ff_net_0_proj_bias_to_fp16_palettized, dilations = var_3813, groups = var_2173, pad = var_3815_pad_0, pad_type = var_3815_pad_type_0, strides = var_3811, weight = down_blocks_1_attentions_1_transformer_blocks_0_ff_net_0_proj_weight_to_fp16_palettized, x = input_109_cast)[name = tensor("op_3815_cast")]; + tensor var_3816_split_sizes_0 = const()[name = tensor("op_3816_split_sizes_0"), val = tensor([2560, 2560])]; + tensor var_3816_axis_0 = const()[name = tensor("op_3816_axis_0"), val = tensor(1)]; + tensor var_3816_cast_0, tensor var_3816_cast_1 = split(axis = var_3816_axis_0, split_sizes = var_3816_split_sizes_0, x = var_3815_cast)[name = tensor("op_3816_cast")]; + tensor var_3818_mode_0 = const()[name = tensor("op_3818_mode_0"), val = tensor("EXACT")]; + tensor var_3818_cast = gelu(mode = var_3818_mode_0, x = var_3816_cast_1)[name = tensor("op_3818_cast")]; + tensor input_111_cast = mul(x = var_3816_cast_0, y = var_3818_cast)[name = tensor("input_111_cast")]; + tensor var_3822 = const()[name = tensor("op_3822"), val = tensor([1, 1])]; + tensor var_3824 = const()[name = tensor("op_3824"), val = tensor([1, 1])]; + tensor var_3826_pad_type_0 = const()[name = tensor("op_3826_pad_type_0"), val = tensor("custom")]; + tensor var_3826_pad_0 = const()[name = tensor("op_3826_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_1_attentions_1_transformer_blocks_0_ff_net_2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(33566528))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(34795392))), name = tensor("down_blocks_1_attentions_1_transformer_blocks_0_ff_net_2_weight_to_fp16_palettized"), shape = tensor([640, 2560, 1, 1])]; + tensor down_blocks_1_attentions_1_transformer_blocks_0_ff_net_2_bias_to_fp16 = const()[name = tensor("down_blocks_1_attentions_1_transformer_blocks_0_ff_net_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(34795584)))]; + tensor var_3826_cast = conv(bias = down_blocks_1_attentions_1_transformer_blocks_0_ff_net_2_bias_to_fp16, dilations = var_3824, groups = var_2173, pad = var_3826_pad_0, pad_type = var_3826_pad_type_0, strides = var_3822, weight = down_blocks_1_attentions_1_transformer_blocks_0_ff_net_2_weight_to_fp16_palettized, x = input_111_cast)[name = tensor("op_3826_cast")]; + tensor hidden_states_69_cast = add(x = var_3826_cast, y = inputs_23_cast)[name = tensor("hidden_states_69_cast")]; + tensor var_3828 = const()[name = tensor("op_3828"), val = tensor([2, 640, 32, 32])]; + tensor input_113_cast = reshape(shape = var_3828, x = hidden_states_69_cast)[name = tensor("input_113_cast")]; + tensor var_3832 = const()[name = tensor("op_3832"), val = tensor([1, 1])]; + tensor var_3834 = const()[name = tensor("op_3834"), val = tensor([1, 1])]; + tensor hidden_states_71_pad_type_0 = const()[name = tensor("hidden_states_71_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_71_pad_0 = const()[name = tensor("hidden_states_71_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_1_attentions_1_proj_out_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(34796928))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(35104192))), name = tensor("down_blocks_1_attentions_1_proj_out_weight_to_fp16_palettized"), shape = tensor([640, 640, 1, 1])]; + tensor down_blocks_1_attentions_1_proj_out_bias_to_fp16 = const()[name = tensor("down_blocks_1_attentions_1_proj_out_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(35104384)))]; + tensor hidden_states_71_cast = conv(bias = down_blocks_1_attentions_1_proj_out_bias_to_fp16, dilations = var_3834, groups = var_2173, pad = hidden_states_71_pad_0, pad_type = hidden_states_71_pad_type_0, strides = var_3832, weight = down_blocks_1_attentions_1_proj_out_weight_to_fp16_palettized, x = input_113_cast)[name = tensor("hidden_states_71_cast")]; + tensor input_115_cast = add(x = hidden_states_71_cast, y = hidden_states_59_cast)[name = tensor("input_115_cast")]; + tensor var_3841 = const()[name = tensor("op_3841"), val = tensor([2, 2])]; + tensor var_3843 = const()[name = tensor("op_3843"), val = tensor([1, 1])]; + tensor input_117_pad_type_0 = const()[name = tensor("input_117_pad_type_0"), val = tensor("custom")]; + tensor input_117_pad_0 = const()[name = tensor("input_117_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor down_blocks_1_downsamplers_0_conv_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(35105728))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(37870592))), name = tensor("down_blocks_1_downsamplers_0_conv_weight_to_fp16_palettized"), shape = tensor([640, 640, 3, 3])]; + tensor down_blocks_1_downsamplers_0_conv_bias_to_fp16 = const()[name = tensor("down_blocks_1_downsamplers_0_conv_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(37870784)))]; + tensor input_117_cast = conv(bias = down_blocks_1_downsamplers_0_conv_bias_to_fp16, dilations = var_3843, groups = var_2173, pad = input_117_pad_0, pad_type = input_117_pad_type_0, strides = var_3841, weight = down_blocks_1_downsamplers_0_conv_weight_to_fp16_palettized, x = input_115_cast)[name = tensor("input_117_cast")]; + tensor var_3878 = const()[name = tensor("op_3878"), val = tensor(true)]; + tensor var_3883 = const()[name = tensor("op_3883"), val = tensor(1)]; + tensor reshape_48_shape_0 = const()[name = tensor("reshape_48_shape_0"), val = tensor([2, 32, 20, 16, 16])]; + tensor reshape_48_cast = reshape(shape = reshape_48_shape_0, x = input_117_cast)[name = tensor("reshape_48_cast")]; + tensor reduce_mean_36_axes_0 = const()[name = tensor("reduce_mean_36_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_36_keep_dims_0 = const()[name = tensor("reduce_mean_36_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_36_cast = reduce_mean(axes = reduce_mean_36_axes_0, keep_dims = reduce_mean_36_keep_dims_0, x = reshape_48_cast)[name = tensor("reduce_mean_36_cast")]; + tensor sub_24_cast = sub(x = reshape_48_cast, y = reduce_mean_36_cast)[name = tensor("sub_24_cast")]; + tensor square_12_cast = square(x = sub_24_cast)[name = tensor("square_12_cast")]; + tensor reduce_mean_38_axes_0 = const()[name = tensor("reduce_mean_38_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_38_keep_dims_0 = const()[name = tensor("reduce_mean_38_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_38_cast = reduce_mean(axes = reduce_mean_38_axes_0, keep_dims = reduce_mean_38_keep_dims_0, x = square_12_cast)[name = tensor("reduce_mean_38_cast")]; + tensor add_24_y_0_to_fp16 = const()[name = tensor("add_24_y_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_24_cast = add(x = reduce_mean_38_cast, y = add_24_y_0_to_fp16)[name = tensor("add_24_cast")]; + tensor sqrt_12_cast = sqrt(x = add_24_cast)[name = tensor("sqrt_12_cast")]; + tensor real_div_12_cast = real_div(x = sub_24_cast, y = sqrt_12_cast)[name = tensor("real_div_12_cast")]; + tensor reshape_49_shape_0 = const()[name = tensor("reshape_49_shape_0"), val = tensor([2, 640, 16, 16])]; + tensor reshape_49_cast = reshape(shape = reshape_49_shape_0, x = real_div_12_cast)[name = tensor("reshape_49_cast")]; + tensor add_25_gamma_0_to_fp16 = const()[name = tensor("add_25_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(37872128)))]; + tensor add_25_beta_0_to_fp16 = const()[name = tensor("add_25_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(37873472)))]; + tensor add_25_epsilon_0_to_fp16 = const()[name = tensor("add_25_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_25_cast = batch_norm(beta = add_25_beta_0_to_fp16, epsilon = add_25_epsilon_0_to_fp16, gamma = add_25_gamma_0_to_fp16, mean = add_15_mean_0_to_fp16, variance = add_15_variance_0_to_fp16, x = reshape_49_cast)[name = tensor("add_25_cast")]; + tensor input_121_cast = silu(x = add_25_cast)[name = tensor("input_121_cast")]; + tensor var_3906 = const()[name = tensor("op_3906"), val = tensor([1, 1])]; + tensor var_3908 = const()[name = tensor("op_3908"), val = tensor([1, 1])]; + tensor hidden_states_73_pad_type_0 = const()[name = tensor("hidden_states_73_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_73_pad_0 = const()[name = tensor("hidden_states_73_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor down_blocks_2_resnets_0_conv1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(37874816))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(43404480))), name = tensor("down_blocks_2_resnets_0_conv1_weight_to_fp16_palettized"), shape = tensor([1280, 640, 3, 3])]; + tensor down_blocks_2_resnets_0_conv1_bias_to_fp16 = const()[name = tensor("down_blocks_2_resnets_0_conv1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(43404672)))]; + tensor hidden_states_73_cast = conv(bias = down_blocks_2_resnets_0_conv1_bias_to_fp16, dilations = var_3908, groups = var_3883, pad = hidden_states_73_pad_0, pad_type = hidden_states_73_pad_type_0, strides = var_3906, weight = down_blocks_2_resnets_0_conv1_weight_to_fp16_palettized, x = input_121_cast)[name = tensor("hidden_states_73_cast")]; + tensor var_3914 = const()[name = tensor("op_3914"), val = tensor([1, 1])]; + tensor var_3916 = const()[name = tensor("op_3916"), val = tensor([1, 1])]; + tensor temb_9_pad_type_0 = const()[name = tensor("temb_9_pad_type_0"), val = tensor("custom")]; + tensor temb_9_pad_0 = const()[name = tensor("temb_9_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_resnets_0_time_emb_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(43407296))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(44636160))), name = tensor("down_blocks_2_resnets_0_time_emb_proj_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor down_blocks_2_resnets_0_time_emb_proj_bias_to_fp16 = const()[name = tensor("down_blocks_2_resnets_0_time_emb_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(44636352)))]; + tensor temb_9_cast = conv(bias = down_blocks_2_resnets_0_time_emb_proj_bias_to_fp16, dilations = var_3916, groups = var_3883, pad = temb_9_pad_0, pad_type = temb_9_pad_type_0, strides = var_3914, weight = down_blocks_2_resnets_0_time_emb_proj_weight_to_fp16_palettized, x = input_15_cast)[name = tensor("temb_9_cast")]; + tensor input_125_cast = add(x = hidden_states_73_cast, y = temb_9_cast)[name = tensor("input_125_cast")]; + tensor reshape_52_shape_0 = const()[name = tensor("reshape_52_shape_0"), val = tensor([2, 32, 40, 16, 16])]; + tensor reshape_52_cast = reshape(shape = reshape_52_shape_0, x = input_125_cast)[name = tensor("reshape_52_cast")]; + tensor reduce_mean_39_axes_0 = const()[name = tensor("reduce_mean_39_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_39_keep_dims_0 = const()[name = tensor("reduce_mean_39_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_39_cast = reduce_mean(axes = reduce_mean_39_axes_0, keep_dims = reduce_mean_39_keep_dims_0, x = reshape_52_cast)[name = tensor("reduce_mean_39_cast")]; + tensor sub_26_cast = sub(x = reshape_52_cast, y = reduce_mean_39_cast)[name = tensor("sub_26_cast")]; + tensor square_13_cast = square(x = sub_26_cast)[name = tensor("square_13_cast")]; + tensor reduce_mean_41_axes_0 = const()[name = tensor("reduce_mean_41_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_41_keep_dims_0 = const()[name = tensor("reduce_mean_41_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_41_cast = reduce_mean(axes = reduce_mean_41_axes_0, keep_dims = reduce_mean_41_keep_dims_0, x = square_13_cast)[name = tensor("reduce_mean_41_cast")]; + tensor add_26_y_0_to_fp16 = const()[name = tensor("add_26_y_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_26_cast = add(x = reduce_mean_41_cast, y = add_26_y_0_to_fp16)[name = tensor("add_26_cast")]; + tensor sqrt_13_cast = sqrt(x = add_26_cast)[name = tensor("sqrt_13_cast")]; + tensor real_div_13_cast = real_div(x = sub_26_cast, y = sqrt_13_cast)[name = tensor("real_div_13_cast")]; + tensor reshape_53_shape_0 = const()[name = tensor("reshape_53_shape_0"), val = tensor([2, 1280, 16, 16])]; + tensor reshape_53_cast = reshape(shape = reshape_53_shape_0, x = real_div_13_cast)[name = tensor("reshape_53_cast")]; + tensor add_27_mean_0_to_fp16 = const()[name = tensor("add_27_mean_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(44638976)))]; + tensor add_27_variance_0_to_fp16 = const()[name = tensor("add_27_variance_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(44641600)))]; + tensor add_27_gamma_0_to_fp16 = const()[name = tensor("add_27_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(44644224)))]; + tensor add_27_beta_0_to_fp16 = const()[name = tensor("add_27_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(44646848)))]; + tensor add_27_epsilon_0_to_fp16 = const()[name = tensor("add_27_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_27_cast = batch_norm(beta = add_27_beta_0_to_fp16, epsilon = add_27_epsilon_0_to_fp16, gamma = add_27_gamma_0_to_fp16, mean = add_27_mean_0_to_fp16, variance = add_27_variance_0_to_fp16, x = reshape_53_cast)[name = tensor("add_27_cast")]; + tensor input_129_cast = silu(x = add_27_cast)[name = tensor("input_129_cast")]; + tensor var_3926 = const()[name = tensor("op_3926"), val = tensor([1, 1])]; + tensor var_3928 = const()[name = tensor("op_3928"), val = tensor([1, 1])]; + tensor hidden_states_75_pad_type_0 = const()[name = tensor("hidden_states_75_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_75_pad_0 = const()[name = tensor("hidden_states_75_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor down_blocks_2_resnets_0_conv2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(44649472))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(55708736))), name = tensor("down_blocks_2_resnets_0_conv2_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 3, 3])]; + tensor down_blocks_2_resnets_0_conv2_bias_to_fp16 = const()[name = tensor("down_blocks_2_resnets_0_conv2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(55708928)))]; + tensor hidden_states_75_cast = conv(bias = down_blocks_2_resnets_0_conv2_bias_to_fp16, dilations = var_3928, groups = var_3883, pad = hidden_states_75_pad_0, pad_type = hidden_states_75_pad_type_0, strides = var_3926, weight = down_blocks_2_resnets_0_conv2_weight_to_fp16_palettized, x = input_129_cast)[name = tensor("hidden_states_75_cast")]; + tensor var_3933 = const()[name = tensor("op_3933"), val = tensor([1, 1])]; + tensor var_3935 = const()[name = tensor("op_3935"), val = tensor([1, 1])]; + tensor x_3_pad_type_0 = const()[name = tensor("x_3_pad_type_0"), val = tensor("custom")]; + tensor x_3_pad_0 = const()[name = tensor("x_3_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_resnets_0_conv_shortcut_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(55711552))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(56326016))), name = tensor("down_blocks_2_resnets_0_conv_shortcut_weight_to_fp16_palettized"), shape = tensor([1280, 640, 1, 1])]; + tensor down_blocks_2_resnets_0_conv_shortcut_bias_to_fp16 = const()[name = tensor("down_blocks_2_resnets_0_conv_shortcut_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(56326208)))]; + tensor x_3_cast = conv(bias = down_blocks_2_resnets_0_conv_shortcut_bias_to_fp16, dilations = var_3935, groups = var_3883, pad = x_3_pad_0, pad_type = x_3_pad_type_0, strides = var_3933, weight = down_blocks_2_resnets_0_conv_shortcut_weight_to_fp16_palettized, x = input_117_cast)[name = tensor("x_3_cast")]; + tensor hidden_states_77_cast = add(x = x_3_cast, y = hidden_states_75_cast)[name = tensor("hidden_states_77_cast")]; + tensor reshape_56_shape_0 = const()[name = tensor("reshape_56_shape_0"), val = tensor([2, 32, 40, 16, 16])]; + tensor reshape_56_cast = reshape(shape = reshape_56_shape_0, x = hidden_states_77_cast)[name = tensor("reshape_56_cast")]; + tensor reduce_mean_42_axes_0 = const()[name = tensor("reduce_mean_42_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_42_keep_dims_0 = const()[name = tensor("reduce_mean_42_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_42_cast = reduce_mean(axes = reduce_mean_42_axes_0, keep_dims = reduce_mean_42_keep_dims_0, x = reshape_56_cast)[name = tensor("reduce_mean_42_cast")]; + tensor sub_28_cast = sub(x = reshape_56_cast, y = reduce_mean_42_cast)[name = tensor("sub_28_cast")]; + tensor square_14_cast = square(x = sub_28_cast)[name = tensor("square_14_cast")]; + tensor reduce_mean_44_axes_0 = const()[name = tensor("reduce_mean_44_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_44_keep_dims_0 = const()[name = tensor("reduce_mean_44_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_44_cast = reduce_mean(axes = reduce_mean_44_axes_0, keep_dims = reduce_mean_44_keep_dims_0, x = square_14_cast)[name = tensor("reduce_mean_44_cast")]; + tensor add_28_y_0_to_fp16 = const()[name = tensor("add_28_y_0_to_fp16"), val = tensor(0x1.1p-20)]; + tensor add_28_cast = add(x = reduce_mean_44_cast, y = add_28_y_0_to_fp16)[name = tensor("add_28_cast")]; + tensor sqrt_14_cast = sqrt(x = add_28_cast)[name = tensor("sqrt_14_cast")]; + tensor real_div_14_cast = real_div(x = sub_28_cast, y = sqrt_14_cast)[name = tensor("real_div_14_cast")]; + tensor reshape_57_shape_0 = const()[name = tensor("reshape_57_shape_0"), val = tensor([2, 1280, 16, 16])]; + tensor reshape_57_cast = reshape(shape = reshape_57_shape_0, x = real_div_14_cast)[name = tensor("reshape_57_cast")]; + tensor add_29_gamma_0_to_fp16 = const()[name = tensor("add_29_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(56328832)))]; + tensor add_29_beta_0_to_fp16 = const()[name = tensor("add_29_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(56331456)))]; + tensor add_29_epsilon_0_to_fp16 = const()[name = tensor("add_29_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_29_cast = batch_norm(beta = add_29_beta_0_to_fp16, epsilon = add_29_epsilon_0_to_fp16, gamma = add_29_gamma_0_to_fp16, mean = add_27_mean_0_to_fp16, variance = add_27_variance_0_to_fp16, x = reshape_57_cast)[name = tensor("add_29_cast")]; + tensor var_3955 = const()[name = tensor("op_3955"), val = tensor([1, 1])]; + tensor var_3957 = const()[name = tensor("op_3957"), val = tensor([1, 1])]; + tensor hidden_states_79_pad_type_0 = const()[name = tensor("hidden_states_79_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_79_pad_0 = const()[name = tensor("hidden_states_79_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_0_proj_in_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(56334080))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(57562944))), name = tensor("down_blocks_2_attentions_0_proj_in_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor down_blocks_2_attentions_0_proj_in_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_proj_in_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(57563136)))]; + tensor hidden_states_79_cast = conv(bias = down_blocks_2_attentions_0_proj_in_bias_to_fp16, dilations = var_3957, groups = var_3883, pad = hidden_states_79_pad_0, pad_type = hidden_states_79_pad_type_0, strides = var_3955, weight = down_blocks_2_attentions_0_proj_in_weight_to_fp16_palettized, x = add_29_cast)[name = tensor("hidden_states_79_cast")]; + tensor var_3962 = const()[name = tensor("op_3962"), val = tensor([2, 1280, 1, 256])]; + tensor inputs_25_cast = reshape(shape = var_3962, x = hidden_states_79_cast)[name = tensor("inputs_25_cast")]; + tensor var_3972 = const()[name = tensor("op_3972"), val = tensor([1])]; + tensor channels_mean_25_cast = reduce_mean(axes = var_3972, keep_dims = var_3878, x = inputs_25_cast)[name = tensor("channels_mean_25_cast")]; + tensor zero_mean_25_cast = sub(x = inputs_25_cast, y = channels_mean_25_cast)[name = tensor("zero_mean_25_cast")]; + tensor zero_mean_sq_25_cast = mul(x = zero_mean_25_cast, y = zero_mean_25_cast)[name = tensor("zero_mean_sq_25_cast")]; + tensor var_3976 = const()[name = tensor("op_3976"), val = tensor([1])]; + tensor var_3977_cast = reduce_mean(axes = var_3976, keep_dims = var_3878, x = zero_mean_sq_25_cast)[name = tensor("op_3977_cast")]; + tensor var_3978_to_fp16 = const()[name = tensor("op_3978_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_3979_cast = add(x = var_3977_cast, y = var_3978_to_fp16)[name = tensor("op_3979_cast")]; + tensor denom_25_epsilon_0_to_fp16 = const()[name = tensor("denom_25_epsilon_0_to_fp16"), val = tensor(0x1p-24)]; + tensor denom_25_cast = rsqrt(epsilon = denom_25_epsilon_0_to_fp16, x = var_3979_cast)[name = tensor("denom_25_cast")]; + tensor out_25_cast = mul(x = zero_mean_25_cast, y = denom_25_cast)[name = tensor("out_25_cast")]; + tensor var_3983_to_fp16 = const()[name = tensor("op_3983_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(57565760)))]; + tensor var_3984_cast = add(x = out_25_cast, y = var_3983_to_fp16)[name = tensor("op_3984_cast")]; + tensor var_3986_to_fp16 = const()[name = tensor("op_3986_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(57568384)))]; + tensor hidden_states_81_cast = mul(x = var_3984_cast, y = var_3986_to_fp16)[name = tensor("hidden_states_81_cast")]; + tensor var_3993 = const()[name = tensor("op_3993"), val = tensor([1, 1])]; + tensor var_3995 = const()[name = tensor("op_3995"), val = tensor([1, 1])]; + tensor q_17_pad_type_0 = const()[name = tensor("q_17_pad_type_0"), val = tensor("custom")]; + tensor q_17_pad_0 = const()[name = tensor("q_17_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(57571008))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(58799872))), name = tensor("down_blocks_2_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor q_17_cast = conv(dilations = var_3995, groups = var_3883, pad = q_17_pad_0, pad_type = q_17_pad_type_0, strides = var_3993, weight = down_blocks_2_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16_palettized, x = hidden_states_81_cast)[name = tensor("q_17_cast")]; + tensor var_3999 = const()[name = tensor("op_3999"), val = tensor([1, 1])]; + tensor var_4001 = const()[name = tensor("op_4001"), val = tensor([1, 1])]; + tensor k_33_pad_type_0 = const()[name = tensor("k_33_pad_type_0"), val = tensor("custom")]; + tensor k_33_pad_0 = const()[name = tensor("k_33_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(58800064))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(60028928))), name = tensor("down_blocks_2_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor k_33_cast = conv(dilations = var_4001, groups = var_3883, pad = k_33_pad_0, pad_type = k_33_pad_type_0, strides = var_3999, weight = down_blocks_2_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16_palettized, x = hidden_states_81_cast)[name = tensor("k_33_cast")]; + tensor var_4005 = const()[name = tensor("op_4005"), val = tensor([1, 1])]; + tensor var_4007 = const()[name = tensor("op_4007"), val = tensor([1, 1])]; + tensor v_17_pad_type_0 = const()[name = tensor("v_17_pad_type_0"), val = tensor("custom")]; + tensor v_17_pad_0 = const()[name = tensor("v_17_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(60029120))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(61257984))), name = tensor("down_blocks_2_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor v_17_cast = conv(dilations = var_4007, groups = var_3883, pad = v_17_pad_0, pad_type = v_17_pad_type_0, strides = var_4005, weight = down_blocks_2_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16_palettized, x = hidden_states_81_cast)[name = tensor("v_17_cast")]; + tensor var_4011_begin_0 = const()[name = tensor("op_4011_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4011_end_0 = const()[name = tensor("op_4011_end_0"), val = tensor([2, 64, 1, 256])]; + tensor var_4011_end_mask_0 = const()[name = tensor("op_4011_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4011_cast = slice_by_index(begin = var_4011_begin_0, end = var_4011_end_0, end_mask = var_4011_end_mask_0, x = q_17_cast)[name = tensor("op_4011_cast")]; + tensor var_4015_begin_0 = const()[name = tensor("op_4015_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_4015_end_0 = const()[name = tensor("op_4015_end_0"), val = tensor([2, 128, 1, 256])]; + tensor var_4015_end_mask_0 = const()[name = tensor("op_4015_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4015_cast = slice_by_index(begin = var_4015_begin_0, end = var_4015_end_0, end_mask = var_4015_end_mask_0, x = q_17_cast)[name = tensor("op_4015_cast")]; + tensor var_4019_begin_0 = const()[name = tensor("op_4019_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_4019_end_0 = const()[name = tensor("op_4019_end_0"), val = tensor([2, 192, 1, 256])]; + tensor var_4019_end_mask_0 = const()[name = tensor("op_4019_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4019_cast = slice_by_index(begin = var_4019_begin_0, end = var_4019_end_0, end_mask = var_4019_end_mask_0, x = q_17_cast)[name = tensor("op_4019_cast")]; + tensor var_4023_begin_0 = const()[name = tensor("op_4023_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_4023_end_0 = const()[name = tensor("op_4023_end_0"), val = tensor([2, 256, 1, 256])]; + tensor var_4023_end_mask_0 = const()[name = tensor("op_4023_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4023_cast = slice_by_index(begin = var_4023_begin_0, end = var_4023_end_0, end_mask = var_4023_end_mask_0, x = q_17_cast)[name = tensor("op_4023_cast")]; + tensor var_4027_begin_0 = const()[name = tensor("op_4027_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_4027_end_0 = const()[name = tensor("op_4027_end_0"), val = tensor([2, 320, 1, 256])]; + tensor var_4027_end_mask_0 = const()[name = tensor("op_4027_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4027_cast = slice_by_index(begin = var_4027_begin_0, end = var_4027_end_0, end_mask = var_4027_end_mask_0, x = q_17_cast)[name = tensor("op_4027_cast")]; + tensor var_4031_begin_0 = const()[name = tensor("op_4031_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_4031_end_0 = const()[name = tensor("op_4031_end_0"), val = tensor([2, 384, 1, 256])]; + tensor var_4031_end_mask_0 = const()[name = tensor("op_4031_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4031_cast = slice_by_index(begin = var_4031_begin_0, end = var_4031_end_0, end_mask = var_4031_end_mask_0, x = q_17_cast)[name = tensor("op_4031_cast")]; + tensor var_4035_begin_0 = const()[name = tensor("op_4035_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_4035_end_0 = const()[name = tensor("op_4035_end_0"), val = tensor([2, 448, 1, 256])]; + tensor var_4035_end_mask_0 = const()[name = tensor("op_4035_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4035_cast = slice_by_index(begin = var_4035_begin_0, end = var_4035_end_0, end_mask = var_4035_end_mask_0, x = q_17_cast)[name = tensor("op_4035_cast")]; + tensor var_4039_begin_0 = const()[name = tensor("op_4039_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_4039_end_0 = const()[name = tensor("op_4039_end_0"), val = tensor([2, 512, 1, 256])]; + tensor var_4039_end_mask_0 = const()[name = tensor("op_4039_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4039_cast = slice_by_index(begin = var_4039_begin_0, end = var_4039_end_0, end_mask = var_4039_end_mask_0, x = q_17_cast)[name = tensor("op_4039_cast")]; + tensor var_4043_begin_0 = const()[name = tensor("op_4043_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_4043_end_0 = const()[name = tensor("op_4043_end_0"), val = tensor([2, 576, 1, 256])]; + tensor var_4043_end_mask_0 = const()[name = tensor("op_4043_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4043_cast = slice_by_index(begin = var_4043_begin_0, end = var_4043_end_0, end_mask = var_4043_end_mask_0, x = q_17_cast)[name = tensor("op_4043_cast")]; + tensor var_4047_begin_0 = const()[name = tensor("op_4047_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_4047_end_0 = const()[name = tensor("op_4047_end_0"), val = tensor([2, 640, 1, 256])]; + tensor var_4047_end_mask_0 = const()[name = tensor("op_4047_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4047_cast = slice_by_index(begin = var_4047_begin_0, end = var_4047_end_0, end_mask = var_4047_end_mask_0, x = q_17_cast)[name = tensor("op_4047_cast")]; + tensor var_4051_begin_0 = const()[name = tensor("op_4051_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_4051_end_0 = const()[name = tensor("op_4051_end_0"), val = tensor([2, 704, 1, 256])]; + tensor var_4051_end_mask_0 = const()[name = tensor("op_4051_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4051_cast = slice_by_index(begin = var_4051_begin_0, end = var_4051_end_0, end_mask = var_4051_end_mask_0, x = q_17_cast)[name = tensor("op_4051_cast")]; + tensor var_4055_begin_0 = const()[name = tensor("op_4055_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_4055_end_0 = const()[name = tensor("op_4055_end_0"), val = tensor([2, 768, 1, 256])]; + tensor var_4055_end_mask_0 = const()[name = tensor("op_4055_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4055_cast = slice_by_index(begin = var_4055_begin_0, end = var_4055_end_0, end_mask = var_4055_end_mask_0, x = q_17_cast)[name = tensor("op_4055_cast")]; + tensor var_4059_begin_0 = const()[name = tensor("op_4059_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_4059_end_0 = const()[name = tensor("op_4059_end_0"), val = tensor([2, 832, 1, 256])]; + tensor var_4059_end_mask_0 = const()[name = tensor("op_4059_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4059_cast = slice_by_index(begin = var_4059_begin_0, end = var_4059_end_0, end_mask = var_4059_end_mask_0, x = q_17_cast)[name = tensor("op_4059_cast")]; + tensor var_4063_begin_0 = const()[name = tensor("op_4063_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_4063_end_0 = const()[name = tensor("op_4063_end_0"), val = tensor([2, 896, 1, 256])]; + tensor var_4063_end_mask_0 = const()[name = tensor("op_4063_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4063_cast = slice_by_index(begin = var_4063_begin_0, end = var_4063_end_0, end_mask = var_4063_end_mask_0, x = q_17_cast)[name = tensor("op_4063_cast")]; + tensor var_4067_begin_0 = const()[name = tensor("op_4067_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_4067_end_0 = const()[name = tensor("op_4067_end_0"), val = tensor([2, 960, 1, 256])]; + tensor var_4067_end_mask_0 = const()[name = tensor("op_4067_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4067_cast = slice_by_index(begin = var_4067_begin_0, end = var_4067_end_0, end_mask = var_4067_end_mask_0, x = q_17_cast)[name = tensor("op_4067_cast")]; + tensor var_4071_begin_0 = const()[name = tensor("op_4071_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_4071_end_0 = const()[name = tensor("op_4071_end_0"), val = tensor([2, 1024, 1, 256])]; + tensor var_4071_end_mask_0 = const()[name = tensor("op_4071_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4071_cast = slice_by_index(begin = var_4071_begin_0, end = var_4071_end_0, end_mask = var_4071_end_mask_0, x = q_17_cast)[name = tensor("op_4071_cast")]; + tensor var_4075_begin_0 = const()[name = tensor("op_4075_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_4075_end_0 = const()[name = tensor("op_4075_end_0"), val = tensor([2, 1088, 1, 256])]; + tensor var_4075_end_mask_0 = const()[name = tensor("op_4075_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4075_cast = slice_by_index(begin = var_4075_begin_0, end = var_4075_end_0, end_mask = var_4075_end_mask_0, x = q_17_cast)[name = tensor("op_4075_cast")]; + tensor var_4079_begin_0 = const()[name = tensor("op_4079_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_4079_end_0 = const()[name = tensor("op_4079_end_0"), val = tensor([2, 1152, 1, 256])]; + tensor var_4079_end_mask_0 = const()[name = tensor("op_4079_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4079_cast = slice_by_index(begin = var_4079_begin_0, end = var_4079_end_0, end_mask = var_4079_end_mask_0, x = q_17_cast)[name = tensor("op_4079_cast")]; + tensor var_4083_begin_0 = const()[name = tensor("op_4083_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_4083_end_0 = const()[name = tensor("op_4083_end_0"), val = tensor([2, 1216, 1, 256])]; + tensor var_4083_end_mask_0 = const()[name = tensor("op_4083_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4083_cast = slice_by_index(begin = var_4083_begin_0, end = var_4083_end_0, end_mask = var_4083_end_mask_0, x = q_17_cast)[name = tensor("op_4083_cast")]; + tensor var_4087_begin_0 = const()[name = tensor("op_4087_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_4087_end_0 = const()[name = tensor("op_4087_end_0"), val = tensor([2, 1280, 1, 256])]; + tensor var_4087_end_mask_0 = const()[name = tensor("op_4087_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4087_cast = slice_by_index(begin = var_4087_begin_0, end = var_4087_end_0, end_mask = var_4087_end_mask_0, x = q_17_cast)[name = tensor("op_4087_cast")]; + tensor k_35_perm_0 = const()[name = tensor("k_35_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_4094_begin_0 = const()[name = tensor("op_4094_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4094_end_0 = const()[name = tensor("op_4094_end_0"), val = tensor([2, 256, 1, 64])]; + tensor var_4094_end_mask_0 = const()[name = tensor("op_4094_end_mask_0"), val = tensor([true, true, true, false])]; + tensor transpose_23 = transpose(perm = k_35_perm_0, x = k_33_cast)[name = tensor("transpose_23")]; + tensor var_4094_cast = slice_by_index(begin = var_4094_begin_0, end = var_4094_end_0, end_mask = var_4094_end_mask_0, x = transpose_23)[name = tensor("op_4094_cast")]; + tensor var_4098_begin_0 = const()[name = tensor("op_4098_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_4098_end_0 = const()[name = tensor("op_4098_end_0"), val = tensor([2, 256, 1, 128])]; + tensor var_4098_end_mask_0 = const()[name = tensor("op_4098_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4098_cast = slice_by_index(begin = var_4098_begin_0, end = var_4098_end_0, end_mask = var_4098_end_mask_0, x = transpose_23)[name = tensor("op_4098_cast")]; + tensor var_4102_begin_0 = const()[name = tensor("op_4102_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_4102_end_0 = const()[name = tensor("op_4102_end_0"), val = tensor([2, 256, 1, 192])]; + tensor var_4102_end_mask_0 = const()[name = tensor("op_4102_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4102_cast = slice_by_index(begin = var_4102_begin_0, end = var_4102_end_0, end_mask = var_4102_end_mask_0, x = transpose_23)[name = tensor("op_4102_cast")]; + tensor var_4106_begin_0 = const()[name = tensor("op_4106_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_4106_end_0 = const()[name = tensor("op_4106_end_0"), val = tensor([2, 256, 1, 256])]; + tensor var_4106_end_mask_0 = const()[name = tensor("op_4106_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4106_cast = slice_by_index(begin = var_4106_begin_0, end = var_4106_end_0, end_mask = var_4106_end_mask_0, x = transpose_23)[name = tensor("op_4106_cast")]; + tensor var_4110_begin_0 = const()[name = tensor("op_4110_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_4110_end_0 = const()[name = tensor("op_4110_end_0"), val = tensor([2, 256, 1, 320])]; + tensor var_4110_end_mask_0 = const()[name = tensor("op_4110_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4110_cast = slice_by_index(begin = var_4110_begin_0, end = var_4110_end_0, end_mask = var_4110_end_mask_0, x = transpose_23)[name = tensor("op_4110_cast")]; + tensor var_4114_begin_0 = const()[name = tensor("op_4114_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_4114_end_0 = const()[name = tensor("op_4114_end_0"), val = tensor([2, 256, 1, 384])]; + tensor var_4114_end_mask_0 = const()[name = tensor("op_4114_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4114_cast = slice_by_index(begin = var_4114_begin_0, end = var_4114_end_0, end_mask = var_4114_end_mask_0, x = transpose_23)[name = tensor("op_4114_cast")]; + tensor var_4118_begin_0 = const()[name = tensor("op_4118_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_4118_end_0 = const()[name = tensor("op_4118_end_0"), val = tensor([2, 256, 1, 448])]; + tensor var_4118_end_mask_0 = const()[name = tensor("op_4118_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4118_cast = slice_by_index(begin = var_4118_begin_0, end = var_4118_end_0, end_mask = var_4118_end_mask_0, x = transpose_23)[name = tensor("op_4118_cast")]; + tensor var_4122_begin_0 = const()[name = tensor("op_4122_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_4122_end_0 = const()[name = tensor("op_4122_end_0"), val = tensor([2, 256, 1, 512])]; + tensor var_4122_end_mask_0 = const()[name = tensor("op_4122_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4122_cast = slice_by_index(begin = var_4122_begin_0, end = var_4122_end_0, end_mask = var_4122_end_mask_0, x = transpose_23)[name = tensor("op_4122_cast")]; + tensor var_4126_begin_0 = const()[name = tensor("op_4126_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_4126_end_0 = const()[name = tensor("op_4126_end_0"), val = tensor([2, 256, 1, 576])]; + tensor var_4126_end_mask_0 = const()[name = tensor("op_4126_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4126_cast = slice_by_index(begin = var_4126_begin_0, end = var_4126_end_0, end_mask = var_4126_end_mask_0, x = transpose_23)[name = tensor("op_4126_cast")]; + tensor var_4130_begin_0 = const()[name = tensor("op_4130_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_4130_end_0 = const()[name = tensor("op_4130_end_0"), val = tensor([2, 256, 1, 640])]; + tensor var_4130_end_mask_0 = const()[name = tensor("op_4130_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4130_cast = slice_by_index(begin = var_4130_begin_0, end = var_4130_end_0, end_mask = var_4130_end_mask_0, x = transpose_23)[name = tensor("op_4130_cast")]; + tensor var_4134_begin_0 = const()[name = tensor("op_4134_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_4134_end_0 = const()[name = tensor("op_4134_end_0"), val = tensor([2, 256, 1, 704])]; + tensor var_4134_end_mask_0 = const()[name = tensor("op_4134_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4134_cast = slice_by_index(begin = var_4134_begin_0, end = var_4134_end_0, end_mask = var_4134_end_mask_0, x = transpose_23)[name = tensor("op_4134_cast")]; + tensor var_4138_begin_0 = const()[name = tensor("op_4138_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_4138_end_0 = const()[name = tensor("op_4138_end_0"), val = tensor([2, 256, 1, 768])]; + tensor var_4138_end_mask_0 = const()[name = tensor("op_4138_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4138_cast = slice_by_index(begin = var_4138_begin_0, end = var_4138_end_0, end_mask = var_4138_end_mask_0, x = transpose_23)[name = tensor("op_4138_cast")]; + tensor var_4142_begin_0 = const()[name = tensor("op_4142_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_4142_end_0 = const()[name = tensor("op_4142_end_0"), val = tensor([2, 256, 1, 832])]; + tensor var_4142_end_mask_0 = const()[name = tensor("op_4142_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4142_cast = slice_by_index(begin = var_4142_begin_0, end = var_4142_end_0, end_mask = var_4142_end_mask_0, x = transpose_23)[name = tensor("op_4142_cast")]; + tensor var_4146_begin_0 = const()[name = tensor("op_4146_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_4146_end_0 = const()[name = tensor("op_4146_end_0"), val = tensor([2, 256, 1, 896])]; + tensor var_4146_end_mask_0 = const()[name = tensor("op_4146_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4146_cast = slice_by_index(begin = var_4146_begin_0, end = var_4146_end_0, end_mask = var_4146_end_mask_0, x = transpose_23)[name = tensor("op_4146_cast")]; + tensor var_4150_begin_0 = const()[name = tensor("op_4150_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_4150_end_0 = const()[name = tensor("op_4150_end_0"), val = tensor([2, 256, 1, 960])]; + tensor var_4150_end_mask_0 = const()[name = tensor("op_4150_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4150_cast = slice_by_index(begin = var_4150_begin_0, end = var_4150_end_0, end_mask = var_4150_end_mask_0, x = transpose_23)[name = tensor("op_4150_cast")]; + tensor var_4154_begin_0 = const()[name = tensor("op_4154_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_4154_end_0 = const()[name = tensor("op_4154_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_4154_end_mask_0 = const()[name = tensor("op_4154_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4154_cast = slice_by_index(begin = var_4154_begin_0, end = var_4154_end_0, end_mask = var_4154_end_mask_0, x = transpose_23)[name = tensor("op_4154_cast")]; + tensor var_4158_begin_0 = const()[name = tensor("op_4158_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_4158_end_0 = const()[name = tensor("op_4158_end_0"), val = tensor([2, 256, 1, 1088])]; + tensor var_4158_end_mask_0 = const()[name = tensor("op_4158_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4158_cast = slice_by_index(begin = var_4158_begin_0, end = var_4158_end_0, end_mask = var_4158_end_mask_0, x = transpose_23)[name = tensor("op_4158_cast")]; + tensor var_4162_begin_0 = const()[name = tensor("op_4162_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_4162_end_0 = const()[name = tensor("op_4162_end_0"), val = tensor([2, 256, 1, 1152])]; + tensor var_4162_end_mask_0 = const()[name = tensor("op_4162_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4162_cast = slice_by_index(begin = var_4162_begin_0, end = var_4162_end_0, end_mask = var_4162_end_mask_0, x = transpose_23)[name = tensor("op_4162_cast")]; + tensor var_4166_begin_0 = const()[name = tensor("op_4166_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_4166_end_0 = const()[name = tensor("op_4166_end_0"), val = tensor([2, 256, 1, 1216])]; + tensor var_4166_end_mask_0 = const()[name = tensor("op_4166_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4166_cast = slice_by_index(begin = var_4166_begin_0, end = var_4166_end_0, end_mask = var_4166_end_mask_0, x = transpose_23)[name = tensor("op_4166_cast")]; + tensor var_4170_begin_0 = const()[name = tensor("op_4170_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_4170_end_0 = const()[name = tensor("op_4170_end_0"), val = tensor([2, 256, 1, 1280])]; + tensor var_4170_end_mask_0 = const()[name = tensor("op_4170_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4170_cast = slice_by_index(begin = var_4170_begin_0, end = var_4170_end_0, end_mask = var_4170_end_mask_0, x = transpose_23)[name = tensor("op_4170_cast")]; + tensor var_4172_begin_0 = const()[name = tensor("op_4172_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4172_end_0 = const()[name = tensor("op_4172_end_0"), val = tensor([2, 64, 1, 256])]; + tensor var_4172_end_mask_0 = const()[name = tensor("op_4172_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4172_cast = slice_by_index(begin = var_4172_begin_0, end = var_4172_end_0, end_mask = var_4172_end_mask_0, x = v_17_cast)[name = tensor("op_4172_cast")]; + tensor var_4176_begin_0 = const()[name = tensor("op_4176_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_4176_end_0 = const()[name = tensor("op_4176_end_0"), val = tensor([2, 128, 1, 256])]; + tensor var_4176_end_mask_0 = const()[name = tensor("op_4176_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4176_cast = slice_by_index(begin = var_4176_begin_0, end = var_4176_end_0, end_mask = var_4176_end_mask_0, x = v_17_cast)[name = tensor("op_4176_cast")]; + tensor var_4180_begin_0 = const()[name = tensor("op_4180_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_4180_end_0 = const()[name = tensor("op_4180_end_0"), val = tensor([2, 192, 1, 256])]; + tensor var_4180_end_mask_0 = const()[name = tensor("op_4180_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4180_cast = slice_by_index(begin = var_4180_begin_0, end = var_4180_end_0, end_mask = var_4180_end_mask_0, x = v_17_cast)[name = tensor("op_4180_cast")]; + tensor var_4184_begin_0 = const()[name = tensor("op_4184_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_4184_end_0 = const()[name = tensor("op_4184_end_0"), val = tensor([2, 256, 1, 256])]; + tensor var_4184_end_mask_0 = const()[name = tensor("op_4184_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4184_cast = slice_by_index(begin = var_4184_begin_0, end = var_4184_end_0, end_mask = var_4184_end_mask_0, x = v_17_cast)[name = tensor("op_4184_cast")]; + tensor var_4188_begin_0 = const()[name = tensor("op_4188_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_4188_end_0 = const()[name = tensor("op_4188_end_0"), val = tensor([2, 320, 1, 256])]; + tensor var_4188_end_mask_0 = const()[name = tensor("op_4188_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4188_cast = slice_by_index(begin = var_4188_begin_0, end = var_4188_end_0, end_mask = var_4188_end_mask_0, x = v_17_cast)[name = tensor("op_4188_cast")]; + tensor var_4192_begin_0 = const()[name = tensor("op_4192_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_4192_end_0 = const()[name = tensor("op_4192_end_0"), val = tensor([2, 384, 1, 256])]; + tensor var_4192_end_mask_0 = const()[name = tensor("op_4192_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4192_cast = slice_by_index(begin = var_4192_begin_0, end = var_4192_end_0, end_mask = var_4192_end_mask_0, x = v_17_cast)[name = tensor("op_4192_cast")]; + tensor var_4196_begin_0 = const()[name = tensor("op_4196_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_4196_end_0 = const()[name = tensor("op_4196_end_0"), val = tensor([2, 448, 1, 256])]; + tensor var_4196_end_mask_0 = const()[name = tensor("op_4196_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4196_cast = slice_by_index(begin = var_4196_begin_0, end = var_4196_end_0, end_mask = var_4196_end_mask_0, x = v_17_cast)[name = tensor("op_4196_cast")]; + tensor var_4200_begin_0 = const()[name = tensor("op_4200_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_4200_end_0 = const()[name = tensor("op_4200_end_0"), val = tensor([2, 512, 1, 256])]; + tensor var_4200_end_mask_0 = const()[name = tensor("op_4200_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4200_cast = slice_by_index(begin = var_4200_begin_0, end = var_4200_end_0, end_mask = var_4200_end_mask_0, x = v_17_cast)[name = tensor("op_4200_cast")]; + tensor var_4204_begin_0 = const()[name = tensor("op_4204_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_4204_end_0 = const()[name = tensor("op_4204_end_0"), val = tensor([2, 576, 1, 256])]; + tensor var_4204_end_mask_0 = const()[name = tensor("op_4204_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4204_cast = slice_by_index(begin = var_4204_begin_0, end = var_4204_end_0, end_mask = var_4204_end_mask_0, x = v_17_cast)[name = tensor("op_4204_cast")]; + tensor var_4208_begin_0 = const()[name = tensor("op_4208_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_4208_end_0 = const()[name = tensor("op_4208_end_0"), val = tensor([2, 640, 1, 256])]; + tensor var_4208_end_mask_0 = const()[name = tensor("op_4208_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4208_cast = slice_by_index(begin = var_4208_begin_0, end = var_4208_end_0, end_mask = var_4208_end_mask_0, x = v_17_cast)[name = tensor("op_4208_cast")]; + tensor var_4212_begin_0 = const()[name = tensor("op_4212_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_4212_end_0 = const()[name = tensor("op_4212_end_0"), val = tensor([2, 704, 1, 256])]; + tensor var_4212_end_mask_0 = const()[name = tensor("op_4212_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4212_cast = slice_by_index(begin = var_4212_begin_0, end = var_4212_end_0, end_mask = var_4212_end_mask_0, x = v_17_cast)[name = tensor("op_4212_cast")]; + tensor var_4216_begin_0 = const()[name = tensor("op_4216_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_4216_end_0 = const()[name = tensor("op_4216_end_0"), val = tensor([2, 768, 1, 256])]; + tensor var_4216_end_mask_0 = const()[name = tensor("op_4216_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4216_cast = slice_by_index(begin = var_4216_begin_0, end = var_4216_end_0, end_mask = var_4216_end_mask_0, x = v_17_cast)[name = tensor("op_4216_cast")]; + tensor var_4220_begin_0 = const()[name = tensor("op_4220_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_4220_end_0 = const()[name = tensor("op_4220_end_0"), val = tensor([2, 832, 1, 256])]; + tensor var_4220_end_mask_0 = const()[name = tensor("op_4220_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4220_cast = slice_by_index(begin = var_4220_begin_0, end = var_4220_end_0, end_mask = var_4220_end_mask_0, x = v_17_cast)[name = tensor("op_4220_cast")]; + tensor var_4224_begin_0 = const()[name = tensor("op_4224_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_4224_end_0 = const()[name = tensor("op_4224_end_0"), val = tensor([2, 896, 1, 256])]; + tensor var_4224_end_mask_0 = const()[name = tensor("op_4224_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4224_cast = slice_by_index(begin = var_4224_begin_0, end = var_4224_end_0, end_mask = var_4224_end_mask_0, x = v_17_cast)[name = tensor("op_4224_cast")]; + tensor var_4228_begin_0 = const()[name = tensor("op_4228_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_4228_end_0 = const()[name = tensor("op_4228_end_0"), val = tensor([2, 960, 1, 256])]; + tensor var_4228_end_mask_0 = const()[name = tensor("op_4228_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4228_cast = slice_by_index(begin = var_4228_begin_0, end = var_4228_end_0, end_mask = var_4228_end_mask_0, x = v_17_cast)[name = tensor("op_4228_cast")]; + tensor var_4232_begin_0 = const()[name = tensor("op_4232_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_4232_end_0 = const()[name = tensor("op_4232_end_0"), val = tensor([2, 1024, 1, 256])]; + tensor var_4232_end_mask_0 = const()[name = tensor("op_4232_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4232_cast = slice_by_index(begin = var_4232_begin_0, end = var_4232_end_0, end_mask = var_4232_end_mask_0, x = v_17_cast)[name = tensor("op_4232_cast")]; + tensor var_4236_begin_0 = const()[name = tensor("op_4236_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_4236_end_0 = const()[name = tensor("op_4236_end_0"), val = tensor([2, 1088, 1, 256])]; + tensor var_4236_end_mask_0 = const()[name = tensor("op_4236_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4236_cast = slice_by_index(begin = var_4236_begin_0, end = var_4236_end_0, end_mask = var_4236_end_mask_0, x = v_17_cast)[name = tensor("op_4236_cast")]; + tensor var_4240_begin_0 = const()[name = tensor("op_4240_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_4240_end_0 = const()[name = tensor("op_4240_end_0"), val = tensor([2, 1152, 1, 256])]; + tensor var_4240_end_mask_0 = const()[name = tensor("op_4240_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4240_cast = slice_by_index(begin = var_4240_begin_0, end = var_4240_end_0, end_mask = var_4240_end_mask_0, x = v_17_cast)[name = tensor("op_4240_cast")]; + tensor var_4244_begin_0 = const()[name = tensor("op_4244_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_4244_end_0 = const()[name = tensor("op_4244_end_0"), val = tensor([2, 1216, 1, 256])]; + tensor var_4244_end_mask_0 = const()[name = tensor("op_4244_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4244_cast = slice_by_index(begin = var_4244_begin_0, end = var_4244_end_0, end_mask = var_4244_end_mask_0, x = v_17_cast)[name = tensor("op_4244_cast")]; + tensor var_4248_begin_0 = const()[name = tensor("op_4248_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_4248_end_0 = const()[name = tensor("op_4248_end_0"), val = tensor([2, 1280, 1, 256])]; + tensor var_4248_end_mask_0 = const()[name = tensor("op_4248_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4248_cast = slice_by_index(begin = var_4248_begin_0, end = var_4248_end_0, end_mask = var_4248_end_mask_0, x = v_17_cast)[name = tensor("op_4248_cast")]; + tensor var_4252_equation_0 = const()[name = tensor("op_4252_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4252_cast = einsum(equation = var_4252_equation_0, values = (var_4094_cast, var_4011_cast))[name = tensor("op_4252_cast")]; + tensor var_4253_to_fp16 = const()[name = tensor("op_4253_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_1_cast = mul(x = var_4252_cast, y = var_4253_to_fp16)[name = tensor("aw_1_cast")]; + tensor var_4256_equation_0 = const()[name = tensor("op_4256_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4256_cast = einsum(equation = var_4256_equation_0, values = (var_4098_cast, var_4015_cast))[name = tensor("op_4256_cast")]; + tensor var_4257_to_fp16 = const()[name = tensor("op_4257_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_3_cast = mul(x = var_4256_cast, y = var_4257_to_fp16)[name = tensor("aw_3_cast")]; + tensor var_4260_equation_0 = const()[name = tensor("op_4260_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4260_cast = einsum(equation = var_4260_equation_0, values = (var_4102_cast, var_4019_cast))[name = tensor("op_4260_cast")]; + tensor var_4261_to_fp16 = const()[name = tensor("op_4261_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_5_cast = mul(x = var_4260_cast, y = var_4261_to_fp16)[name = tensor("aw_5_cast")]; + tensor var_4264_equation_0 = const()[name = tensor("op_4264_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4264_cast = einsum(equation = var_4264_equation_0, values = (var_4106_cast, var_4023_cast))[name = tensor("op_4264_cast")]; + tensor var_4265_to_fp16 = const()[name = tensor("op_4265_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_7_cast = mul(x = var_4264_cast, y = var_4265_to_fp16)[name = tensor("aw_7_cast")]; + tensor var_4268_equation_0 = const()[name = tensor("op_4268_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4268_cast = einsum(equation = var_4268_equation_0, values = (var_4110_cast, var_4027_cast))[name = tensor("op_4268_cast")]; + tensor var_4269_to_fp16 = const()[name = tensor("op_4269_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_9_cast = mul(x = var_4268_cast, y = var_4269_to_fp16)[name = tensor("aw_9_cast")]; + tensor var_4272_equation_0 = const()[name = tensor("op_4272_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4272_cast = einsum(equation = var_4272_equation_0, values = (var_4114_cast, var_4031_cast))[name = tensor("op_4272_cast")]; + tensor var_4273_to_fp16 = const()[name = tensor("op_4273_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_11_cast = mul(x = var_4272_cast, y = var_4273_to_fp16)[name = tensor("aw_11_cast")]; + tensor var_4276_equation_0 = const()[name = tensor("op_4276_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4276_cast = einsum(equation = var_4276_equation_0, values = (var_4118_cast, var_4035_cast))[name = tensor("op_4276_cast")]; + tensor var_4277_to_fp16 = const()[name = tensor("op_4277_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_13_cast = mul(x = var_4276_cast, y = var_4277_to_fp16)[name = tensor("aw_13_cast")]; + tensor var_4280_equation_0 = const()[name = tensor("op_4280_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4280_cast = einsum(equation = var_4280_equation_0, values = (var_4122_cast, var_4039_cast))[name = tensor("op_4280_cast")]; + tensor var_4281_to_fp16 = const()[name = tensor("op_4281_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_15_cast = mul(x = var_4280_cast, y = var_4281_to_fp16)[name = tensor("aw_15_cast")]; + tensor var_4284_equation_0 = const()[name = tensor("op_4284_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4284_cast = einsum(equation = var_4284_equation_0, values = (var_4126_cast, var_4043_cast))[name = tensor("op_4284_cast")]; + tensor var_4285_to_fp16 = const()[name = tensor("op_4285_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_17_cast = mul(x = var_4284_cast, y = var_4285_to_fp16)[name = tensor("aw_17_cast")]; + tensor var_4288_equation_0 = const()[name = tensor("op_4288_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4288_cast = einsum(equation = var_4288_equation_0, values = (var_4130_cast, var_4047_cast))[name = tensor("op_4288_cast")]; + tensor var_4289_to_fp16 = const()[name = tensor("op_4289_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_19_cast = mul(x = var_4288_cast, y = var_4289_to_fp16)[name = tensor("aw_19_cast")]; + tensor var_4292_equation_0 = const()[name = tensor("op_4292_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4292_cast = einsum(equation = var_4292_equation_0, values = (var_4134_cast, var_4051_cast))[name = tensor("op_4292_cast")]; + tensor var_4293_to_fp16 = const()[name = tensor("op_4293_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_21_cast = mul(x = var_4292_cast, y = var_4293_to_fp16)[name = tensor("aw_21_cast")]; + tensor var_4296_equation_0 = const()[name = tensor("op_4296_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4296_cast = einsum(equation = var_4296_equation_0, values = (var_4138_cast, var_4055_cast))[name = tensor("op_4296_cast")]; + tensor var_4297_to_fp16 = const()[name = tensor("op_4297_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_23_cast = mul(x = var_4296_cast, y = var_4297_to_fp16)[name = tensor("aw_23_cast")]; + tensor var_4300_equation_0 = const()[name = tensor("op_4300_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4300_cast = einsum(equation = var_4300_equation_0, values = (var_4142_cast, var_4059_cast))[name = tensor("op_4300_cast")]; + tensor var_4301_to_fp16 = const()[name = tensor("op_4301_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_25_cast = mul(x = var_4300_cast, y = var_4301_to_fp16)[name = tensor("aw_25_cast")]; + tensor var_4304_equation_0 = const()[name = tensor("op_4304_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4304_cast = einsum(equation = var_4304_equation_0, values = (var_4146_cast, var_4063_cast))[name = tensor("op_4304_cast")]; + tensor var_4305_to_fp16 = const()[name = tensor("op_4305_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_27_cast = mul(x = var_4304_cast, y = var_4305_to_fp16)[name = tensor("aw_27_cast")]; + tensor var_4308_equation_0 = const()[name = tensor("op_4308_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4308_cast = einsum(equation = var_4308_equation_0, values = (var_4150_cast, var_4067_cast))[name = tensor("op_4308_cast")]; + tensor var_4309_to_fp16 = const()[name = tensor("op_4309_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_29_cast = mul(x = var_4308_cast, y = var_4309_to_fp16)[name = tensor("aw_29_cast")]; + tensor var_4312_equation_0 = const()[name = tensor("op_4312_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4312_cast = einsum(equation = var_4312_equation_0, values = (var_4154_cast, var_4071_cast))[name = tensor("op_4312_cast")]; + tensor var_4313_to_fp16 = const()[name = tensor("op_4313_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_31_cast = mul(x = var_4312_cast, y = var_4313_to_fp16)[name = tensor("aw_31_cast")]; + tensor var_4316_equation_0 = const()[name = tensor("op_4316_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4316_cast = einsum(equation = var_4316_equation_0, values = (var_4158_cast, var_4075_cast))[name = tensor("op_4316_cast")]; + tensor var_4317_to_fp16 = const()[name = tensor("op_4317_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_33_cast = mul(x = var_4316_cast, y = var_4317_to_fp16)[name = tensor("aw_33_cast")]; + tensor var_4320_equation_0 = const()[name = tensor("op_4320_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4320_cast = einsum(equation = var_4320_equation_0, values = (var_4162_cast, var_4079_cast))[name = tensor("op_4320_cast")]; + tensor var_4321_to_fp16 = const()[name = tensor("op_4321_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_35_cast = mul(x = var_4320_cast, y = var_4321_to_fp16)[name = tensor("aw_35_cast")]; + tensor var_4324_equation_0 = const()[name = tensor("op_4324_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4324_cast = einsum(equation = var_4324_equation_0, values = (var_4166_cast, var_4083_cast))[name = tensor("op_4324_cast")]; + tensor var_4325_to_fp16 = const()[name = tensor("op_4325_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_37_cast = mul(x = var_4324_cast, y = var_4325_to_fp16)[name = tensor("aw_37_cast")]; + tensor var_4328_equation_0 = const()[name = tensor("op_4328_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4328_cast = einsum(equation = var_4328_equation_0, values = (var_4170_cast, var_4087_cast))[name = tensor("op_4328_cast")]; + tensor var_4329_to_fp16 = const()[name = tensor("op_4329_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_39_cast = mul(x = var_4328_cast, y = var_4329_to_fp16)[name = tensor("aw_39_cast")]; + tensor var_4331_cast = softmax(axis = var_3883, x = aw_1_cast)[name = tensor("op_4331_cast")]; + tensor var_4332_cast = softmax(axis = var_3883, x = aw_3_cast)[name = tensor("op_4332_cast")]; + tensor var_4333_cast = softmax(axis = var_3883, x = aw_5_cast)[name = tensor("op_4333_cast")]; + tensor var_4334_cast = softmax(axis = var_3883, x = aw_7_cast)[name = tensor("op_4334_cast")]; + tensor var_4335_cast = softmax(axis = var_3883, x = aw_9_cast)[name = tensor("op_4335_cast")]; + tensor var_4336_cast = softmax(axis = var_3883, x = aw_11_cast)[name = tensor("op_4336_cast")]; + tensor var_4337_cast = softmax(axis = var_3883, x = aw_13_cast)[name = tensor("op_4337_cast")]; + tensor var_4338_cast = softmax(axis = var_3883, x = aw_15_cast)[name = tensor("op_4338_cast")]; + tensor var_4339_cast = softmax(axis = var_3883, x = aw_17_cast)[name = tensor("op_4339_cast")]; + tensor var_4340_cast = softmax(axis = var_3883, x = aw_19_cast)[name = tensor("op_4340_cast")]; + tensor var_4341_cast = softmax(axis = var_3883, x = aw_21_cast)[name = tensor("op_4341_cast")]; + tensor var_4342_cast = softmax(axis = var_3883, x = aw_23_cast)[name = tensor("op_4342_cast")]; + tensor var_4343_cast = softmax(axis = var_3883, x = aw_25_cast)[name = tensor("op_4343_cast")]; + tensor var_4344_cast = softmax(axis = var_3883, x = aw_27_cast)[name = tensor("op_4344_cast")]; + tensor var_4345_cast = softmax(axis = var_3883, x = aw_29_cast)[name = tensor("op_4345_cast")]; + tensor var_4346_cast = softmax(axis = var_3883, x = aw_31_cast)[name = tensor("op_4346_cast")]; + tensor var_4347_cast = softmax(axis = var_3883, x = aw_33_cast)[name = tensor("op_4347_cast")]; + tensor var_4348_cast = softmax(axis = var_3883, x = aw_35_cast)[name = tensor("op_4348_cast")]; + tensor var_4349_cast = softmax(axis = var_3883, x = aw_37_cast)[name = tensor("op_4349_cast")]; + tensor var_4350_cast = softmax(axis = var_3883, x = aw_39_cast)[name = tensor("op_4350_cast")]; + tensor var_4352_equation_0 = const()[name = tensor("op_4352_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4352_cast = einsum(equation = var_4352_equation_0, values = (var_4172_cast, var_4331_cast))[name = tensor("op_4352_cast")]; + tensor var_4354_equation_0 = const()[name = tensor("op_4354_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4354_cast = einsum(equation = var_4354_equation_0, values = (var_4176_cast, var_4332_cast))[name = tensor("op_4354_cast")]; + tensor var_4356_equation_0 = const()[name = tensor("op_4356_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4356_cast = einsum(equation = var_4356_equation_0, values = (var_4180_cast, var_4333_cast))[name = tensor("op_4356_cast")]; + tensor var_4358_equation_0 = const()[name = tensor("op_4358_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4358_cast = einsum(equation = var_4358_equation_0, values = (var_4184_cast, var_4334_cast))[name = tensor("op_4358_cast")]; + tensor var_4360_equation_0 = const()[name = tensor("op_4360_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4360_cast = einsum(equation = var_4360_equation_0, values = (var_4188_cast, var_4335_cast))[name = tensor("op_4360_cast")]; + tensor var_4362_equation_0 = const()[name = tensor("op_4362_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4362_cast = einsum(equation = var_4362_equation_0, values = (var_4192_cast, var_4336_cast))[name = tensor("op_4362_cast")]; + tensor var_4364_equation_0 = const()[name = tensor("op_4364_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4364_cast = einsum(equation = var_4364_equation_0, values = (var_4196_cast, var_4337_cast))[name = tensor("op_4364_cast")]; + tensor var_4366_equation_0 = const()[name = tensor("op_4366_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4366_cast = einsum(equation = var_4366_equation_0, values = (var_4200_cast, var_4338_cast))[name = tensor("op_4366_cast")]; + tensor var_4368_equation_0 = const()[name = tensor("op_4368_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4368_cast = einsum(equation = var_4368_equation_0, values = (var_4204_cast, var_4339_cast))[name = tensor("op_4368_cast")]; + tensor var_4370_equation_0 = const()[name = tensor("op_4370_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4370_cast = einsum(equation = var_4370_equation_0, values = (var_4208_cast, var_4340_cast))[name = tensor("op_4370_cast")]; + tensor var_4372_equation_0 = const()[name = tensor("op_4372_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4372_cast = einsum(equation = var_4372_equation_0, values = (var_4212_cast, var_4341_cast))[name = tensor("op_4372_cast")]; + tensor var_4374_equation_0 = const()[name = tensor("op_4374_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4374_cast = einsum(equation = var_4374_equation_0, values = (var_4216_cast, var_4342_cast))[name = tensor("op_4374_cast")]; + tensor var_4376_equation_0 = const()[name = tensor("op_4376_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4376_cast = einsum(equation = var_4376_equation_0, values = (var_4220_cast, var_4343_cast))[name = tensor("op_4376_cast")]; + tensor var_4378_equation_0 = const()[name = tensor("op_4378_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4378_cast = einsum(equation = var_4378_equation_0, values = (var_4224_cast, var_4344_cast))[name = tensor("op_4378_cast")]; + tensor var_4380_equation_0 = const()[name = tensor("op_4380_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4380_cast = einsum(equation = var_4380_equation_0, values = (var_4228_cast, var_4345_cast))[name = tensor("op_4380_cast")]; + tensor var_4382_equation_0 = const()[name = tensor("op_4382_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4382_cast = einsum(equation = var_4382_equation_0, values = (var_4232_cast, var_4346_cast))[name = tensor("op_4382_cast")]; + tensor var_4384_equation_0 = const()[name = tensor("op_4384_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4384_cast = einsum(equation = var_4384_equation_0, values = (var_4236_cast, var_4347_cast))[name = tensor("op_4384_cast")]; + tensor var_4386_equation_0 = const()[name = tensor("op_4386_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4386_cast = einsum(equation = var_4386_equation_0, values = (var_4240_cast, var_4348_cast))[name = tensor("op_4386_cast")]; + tensor var_4388_equation_0 = const()[name = tensor("op_4388_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4388_cast = einsum(equation = var_4388_equation_0, values = (var_4244_cast, var_4349_cast))[name = tensor("op_4388_cast")]; + tensor var_4390_equation_0 = const()[name = tensor("op_4390_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4390_cast = einsum(equation = var_4390_equation_0, values = (var_4248_cast, var_4350_cast))[name = tensor("op_4390_cast")]; + tensor input_133_interleave_0 = const()[name = tensor("input_133_interleave_0"), val = tensor(false)]; + tensor input_133_cast = concat(axis = var_3883, interleave = input_133_interleave_0, values = (var_4352_cast, var_4354_cast, var_4356_cast, var_4358_cast, var_4360_cast, var_4362_cast, var_4364_cast, var_4366_cast, var_4368_cast, var_4370_cast, var_4372_cast, var_4374_cast, var_4376_cast, var_4378_cast, var_4380_cast, var_4382_cast, var_4384_cast, var_4386_cast, var_4388_cast, var_4390_cast))[name = tensor("input_133_cast")]; + tensor var_4396 = const()[name = tensor("op_4396"), val = tensor([1, 1])]; + tensor var_4398 = const()[name = tensor("op_4398"), val = tensor([1, 1])]; + tensor var_4400_pad_type_0 = const()[name = tensor("op_4400_pad_type_0"), val = tensor("custom")]; + tensor var_4400_pad_0 = const()[name = tensor("op_4400_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(61258176))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(62487040))), name = tensor("down_blocks_2_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor down_blocks_2_attentions_0_transformer_blocks_0_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_0_attn1_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(62487232)))]; + tensor var_4400_cast = conv(bias = down_blocks_2_attentions_0_transformer_blocks_0_attn1_to_out_0_bias_to_fp16, dilations = var_4398, groups = var_3883, pad = var_4400_pad_0, pad_type = var_4400_pad_type_0, strides = var_4396, weight = down_blocks_2_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16_palettized, x = input_133_cast)[name = tensor("op_4400_cast")]; + tensor inputs_27_cast = add(x = var_4400_cast, y = inputs_25_cast)[name = tensor("inputs_27_cast")]; + tensor var_4404 = const()[name = tensor("op_4404"), val = tensor([1])]; + tensor channels_mean_27_cast = reduce_mean(axes = var_4404, keep_dims = var_3878, x = inputs_27_cast)[name = tensor("channels_mean_27_cast")]; + tensor zero_mean_27_cast = sub(x = inputs_27_cast, y = channels_mean_27_cast)[name = tensor("zero_mean_27_cast")]; + tensor zero_mean_sq_27_cast = mul(x = zero_mean_27_cast, y = zero_mean_27_cast)[name = tensor("zero_mean_sq_27_cast")]; + tensor var_4408 = const()[name = tensor("op_4408"), val = tensor([1])]; + tensor var_4409_cast = reduce_mean(axes = var_4408, keep_dims = var_3878, x = zero_mean_sq_27_cast)[name = tensor("op_4409_cast")]; + tensor var_4410_to_fp16 = const()[name = tensor("op_4410_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_4411_cast = add(x = var_4409_cast, y = var_4410_to_fp16)[name = tensor("op_4411_cast")]; + tensor denom_27_epsilon_0_to_fp16 = const()[name = tensor("denom_27_epsilon_0_to_fp16"), val = tensor(0x1p-24)]; + tensor denom_27_cast = rsqrt(epsilon = denom_27_epsilon_0_to_fp16, x = var_4411_cast)[name = tensor("denom_27_cast")]; + tensor out_27_cast = mul(x = zero_mean_27_cast, y = denom_27_cast)[name = tensor("out_27_cast")]; + tensor var_4415_to_fp16 = const()[name = tensor("op_4415_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(62489856)))]; + tensor var_4416_cast = add(x = out_27_cast, y = var_4415_to_fp16)[name = tensor("op_4416_cast")]; + tensor var_4418_to_fp16 = const()[name = tensor("op_4418_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(62492480)))]; + tensor hidden_states_83_cast = mul(x = var_4416_cast, y = var_4418_to_fp16)[name = tensor("hidden_states_83_cast")]; + tensor var_4425 = const()[name = tensor("op_4425"), val = tensor([1, 1])]; + tensor var_4427 = const()[name = tensor("op_4427"), val = tensor([1, 1])]; + tensor q_19_pad_type_0 = const()[name = tensor("q_19_pad_type_0"), val = tensor("custom")]; + tensor q_19_pad_0 = const()[name = tensor("q_19_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(62495104))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(63723968))), name = tensor("down_blocks_2_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor q_19_cast = conv(dilations = var_4427, groups = var_3883, pad = q_19_pad_0, pad_type = q_19_pad_type_0, strides = var_4425, weight = down_blocks_2_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16_palettized, x = hidden_states_83_cast)[name = tensor("q_19_cast")]; + tensor var_4431 = const()[name = tensor("op_4431"), val = tensor([1, 1])]; + tensor var_4433 = const()[name = tensor("op_4433"), val = tensor([1, 1])]; + tensor k_37_pad_type_0 = const()[name = tensor("k_37_pad_type_0"), val = tensor("custom")]; + tensor k_37_pad_0 = const()[name = tensor("k_37_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(63724160))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(64707264))), name = tensor("down_blocks_2_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized"), shape = tensor([1280, 1024, 1, 1])]; + tensor k_37_cast = conv(dilations = var_4433, groups = var_3883, pad = k_37_pad_0, pad_type = k_37_pad_type_0, strides = var_4431, weight = down_blocks_2_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor("k_37_cast")]; + tensor var_4437 = const()[name = tensor("op_4437"), val = tensor([1, 1])]; + tensor var_4439 = const()[name = tensor("op_4439"), val = tensor([1, 1])]; + tensor v_19_pad_type_0 = const()[name = tensor("v_19_pad_type_0"), val = tensor("custom")]; + tensor v_19_pad_0 = const()[name = tensor("v_19_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(64707456))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(65690560))), name = tensor("down_blocks_2_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized"), shape = tensor([1280, 1024, 1, 1])]; + tensor v_19_cast = conv(dilations = var_4439, groups = var_3883, pad = v_19_pad_0, pad_type = v_19_pad_type_0, strides = var_4437, weight = down_blocks_2_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor("v_19_cast")]; + tensor var_4443_begin_0 = const()[name = tensor("op_4443_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4443_end_0 = const()[name = tensor("op_4443_end_0"), val = tensor([2, 64, 1, 256])]; + tensor var_4443_end_mask_0 = const()[name = tensor("op_4443_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4443_cast = slice_by_index(begin = var_4443_begin_0, end = var_4443_end_0, end_mask = var_4443_end_mask_0, x = q_19_cast)[name = tensor("op_4443_cast")]; + tensor var_4447_begin_0 = const()[name = tensor("op_4447_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_4447_end_0 = const()[name = tensor("op_4447_end_0"), val = tensor([2, 128, 1, 256])]; + tensor var_4447_end_mask_0 = const()[name = tensor("op_4447_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4447_cast = slice_by_index(begin = var_4447_begin_0, end = var_4447_end_0, end_mask = var_4447_end_mask_0, x = q_19_cast)[name = tensor("op_4447_cast")]; + tensor var_4451_begin_0 = const()[name = tensor("op_4451_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_4451_end_0 = const()[name = tensor("op_4451_end_0"), val = tensor([2, 192, 1, 256])]; + tensor var_4451_end_mask_0 = const()[name = tensor("op_4451_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4451_cast = slice_by_index(begin = var_4451_begin_0, end = var_4451_end_0, end_mask = var_4451_end_mask_0, x = q_19_cast)[name = tensor("op_4451_cast")]; + tensor var_4455_begin_0 = const()[name = tensor("op_4455_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_4455_end_0 = const()[name = tensor("op_4455_end_0"), val = tensor([2, 256, 1, 256])]; + tensor var_4455_end_mask_0 = const()[name = tensor("op_4455_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4455_cast = slice_by_index(begin = var_4455_begin_0, end = var_4455_end_0, end_mask = var_4455_end_mask_0, x = q_19_cast)[name = tensor("op_4455_cast")]; + tensor var_4459_begin_0 = const()[name = tensor("op_4459_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_4459_end_0 = const()[name = tensor("op_4459_end_0"), val = tensor([2, 320, 1, 256])]; + tensor var_4459_end_mask_0 = const()[name = tensor("op_4459_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4459_cast = slice_by_index(begin = var_4459_begin_0, end = var_4459_end_0, end_mask = var_4459_end_mask_0, x = q_19_cast)[name = tensor("op_4459_cast")]; + tensor var_4463_begin_0 = const()[name = tensor("op_4463_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_4463_end_0 = const()[name = tensor("op_4463_end_0"), val = tensor([2, 384, 1, 256])]; + tensor var_4463_end_mask_0 = const()[name = tensor("op_4463_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4463_cast = slice_by_index(begin = var_4463_begin_0, end = var_4463_end_0, end_mask = var_4463_end_mask_0, x = q_19_cast)[name = tensor("op_4463_cast")]; + tensor var_4467_begin_0 = const()[name = tensor("op_4467_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_4467_end_0 = const()[name = tensor("op_4467_end_0"), val = tensor([2, 448, 1, 256])]; + tensor var_4467_end_mask_0 = const()[name = tensor("op_4467_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4467_cast = slice_by_index(begin = var_4467_begin_0, end = var_4467_end_0, end_mask = var_4467_end_mask_0, x = q_19_cast)[name = tensor("op_4467_cast")]; + tensor var_4471_begin_0 = const()[name = tensor("op_4471_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_4471_end_0 = const()[name = tensor("op_4471_end_0"), val = tensor([2, 512, 1, 256])]; + tensor var_4471_end_mask_0 = const()[name = tensor("op_4471_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4471_cast = slice_by_index(begin = var_4471_begin_0, end = var_4471_end_0, end_mask = var_4471_end_mask_0, x = q_19_cast)[name = tensor("op_4471_cast")]; + tensor var_4475_begin_0 = const()[name = tensor("op_4475_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_4475_end_0 = const()[name = tensor("op_4475_end_0"), val = tensor([2, 576, 1, 256])]; + tensor var_4475_end_mask_0 = const()[name = tensor("op_4475_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4475_cast = slice_by_index(begin = var_4475_begin_0, end = var_4475_end_0, end_mask = var_4475_end_mask_0, x = q_19_cast)[name = tensor("op_4475_cast")]; + tensor var_4479_begin_0 = const()[name = tensor("op_4479_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_4479_end_0 = const()[name = tensor("op_4479_end_0"), val = tensor([2, 640, 1, 256])]; + tensor var_4479_end_mask_0 = const()[name = tensor("op_4479_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4479_cast = slice_by_index(begin = var_4479_begin_0, end = var_4479_end_0, end_mask = var_4479_end_mask_0, x = q_19_cast)[name = tensor("op_4479_cast")]; + tensor var_4483_begin_0 = const()[name = tensor("op_4483_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_4483_end_0 = const()[name = tensor("op_4483_end_0"), val = tensor([2, 704, 1, 256])]; + tensor var_4483_end_mask_0 = const()[name = tensor("op_4483_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4483_cast = slice_by_index(begin = var_4483_begin_0, end = var_4483_end_0, end_mask = var_4483_end_mask_0, x = q_19_cast)[name = tensor("op_4483_cast")]; + tensor var_4487_begin_0 = const()[name = tensor("op_4487_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_4487_end_0 = const()[name = tensor("op_4487_end_0"), val = tensor([2, 768, 1, 256])]; + tensor var_4487_end_mask_0 = const()[name = tensor("op_4487_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4487_cast = slice_by_index(begin = var_4487_begin_0, end = var_4487_end_0, end_mask = var_4487_end_mask_0, x = q_19_cast)[name = tensor("op_4487_cast")]; + tensor var_4491_begin_0 = const()[name = tensor("op_4491_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_4491_end_0 = const()[name = tensor("op_4491_end_0"), val = tensor([2, 832, 1, 256])]; + tensor var_4491_end_mask_0 = const()[name = tensor("op_4491_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4491_cast = slice_by_index(begin = var_4491_begin_0, end = var_4491_end_0, end_mask = var_4491_end_mask_0, x = q_19_cast)[name = tensor("op_4491_cast")]; + tensor var_4495_begin_0 = const()[name = tensor("op_4495_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_4495_end_0 = const()[name = tensor("op_4495_end_0"), val = tensor([2, 896, 1, 256])]; + tensor var_4495_end_mask_0 = const()[name = tensor("op_4495_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4495_cast = slice_by_index(begin = var_4495_begin_0, end = var_4495_end_0, end_mask = var_4495_end_mask_0, x = q_19_cast)[name = tensor("op_4495_cast")]; + tensor var_4499_begin_0 = const()[name = tensor("op_4499_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_4499_end_0 = const()[name = tensor("op_4499_end_0"), val = tensor([2, 960, 1, 256])]; + tensor var_4499_end_mask_0 = const()[name = tensor("op_4499_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4499_cast = slice_by_index(begin = var_4499_begin_0, end = var_4499_end_0, end_mask = var_4499_end_mask_0, x = q_19_cast)[name = tensor("op_4499_cast")]; + tensor var_4503_begin_0 = const()[name = tensor("op_4503_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_4503_end_0 = const()[name = tensor("op_4503_end_0"), val = tensor([2, 1024, 1, 256])]; + tensor var_4503_end_mask_0 = const()[name = tensor("op_4503_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4503_cast = slice_by_index(begin = var_4503_begin_0, end = var_4503_end_0, end_mask = var_4503_end_mask_0, x = q_19_cast)[name = tensor("op_4503_cast")]; + tensor var_4507_begin_0 = const()[name = tensor("op_4507_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_4507_end_0 = const()[name = tensor("op_4507_end_0"), val = tensor([2, 1088, 1, 256])]; + tensor var_4507_end_mask_0 = const()[name = tensor("op_4507_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4507_cast = slice_by_index(begin = var_4507_begin_0, end = var_4507_end_0, end_mask = var_4507_end_mask_0, x = q_19_cast)[name = tensor("op_4507_cast")]; + tensor var_4511_begin_0 = const()[name = tensor("op_4511_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_4511_end_0 = const()[name = tensor("op_4511_end_0"), val = tensor([2, 1152, 1, 256])]; + tensor var_4511_end_mask_0 = const()[name = tensor("op_4511_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4511_cast = slice_by_index(begin = var_4511_begin_0, end = var_4511_end_0, end_mask = var_4511_end_mask_0, x = q_19_cast)[name = tensor("op_4511_cast")]; + tensor var_4515_begin_0 = const()[name = tensor("op_4515_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_4515_end_0 = const()[name = tensor("op_4515_end_0"), val = tensor([2, 1216, 1, 256])]; + tensor var_4515_end_mask_0 = const()[name = tensor("op_4515_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4515_cast = slice_by_index(begin = var_4515_begin_0, end = var_4515_end_0, end_mask = var_4515_end_mask_0, x = q_19_cast)[name = tensor("op_4515_cast")]; + tensor var_4519_begin_0 = const()[name = tensor("op_4519_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_4519_end_0 = const()[name = tensor("op_4519_end_0"), val = tensor([2, 1280, 1, 256])]; + tensor var_4519_end_mask_0 = const()[name = tensor("op_4519_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4519_cast = slice_by_index(begin = var_4519_begin_0, end = var_4519_end_0, end_mask = var_4519_end_mask_0, x = q_19_cast)[name = tensor("op_4519_cast")]; + tensor k_39_perm_0 = const()[name = tensor("k_39_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_4526_begin_0 = const()[name = tensor("op_4526_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4526_end_0 = const()[name = tensor("op_4526_end_0"), val = tensor([2, 77, 1, 64])]; + tensor var_4526_end_mask_0 = const()[name = tensor("op_4526_end_mask_0"), val = tensor([true, true, true, false])]; + tensor transpose_22 = transpose(perm = k_39_perm_0, x = k_37_cast)[name = tensor("transpose_22")]; + tensor var_4526_cast = slice_by_index(begin = var_4526_begin_0, end = var_4526_end_0, end_mask = var_4526_end_mask_0, x = transpose_22)[name = tensor("op_4526_cast")]; + tensor var_4530_begin_0 = const()[name = tensor("op_4530_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_4530_end_0 = const()[name = tensor("op_4530_end_0"), val = tensor([2, 77, 1, 128])]; + tensor var_4530_end_mask_0 = const()[name = tensor("op_4530_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4530_cast = slice_by_index(begin = var_4530_begin_0, end = var_4530_end_0, end_mask = var_4530_end_mask_0, x = transpose_22)[name = tensor("op_4530_cast")]; + tensor var_4534_begin_0 = const()[name = tensor("op_4534_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_4534_end_0 = const()[name = tensor("op_4534_end_0"), val = tensor([2, 77, 1, 192])]; + tensor var_4534_end_mask_0 = const()[name = tensor("op_4534_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4534_cast = slice_by_index(begin = var_4534_begin_0, end = var_4534_end_0, end_mask = var_4534_end_mask_0, x = transpose_22)[name = tensor("op_4534_cast")]; + tensor var_4538_begin_0 = const()[name = tensor("op_4538_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_4538_end_0 = const()[name = tensor("op_4538_end_0"), val = tensor([2, 77, 1, 256])]; + tensor var_4538_end_mask_0 = const()[name = tensor("op_4538_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4538_cast = slice_by_index(begin = var_4538_begin_0, end = var_4538_end_0, end_mask = var_4538_end_mask_0, x = transpose_22)[name = tensor("op_4538_cast")]; + tensor var_4542_begin_0 = const()[name = tensor("op_4542_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_4542_end_0 = const()[name = tensor("op_4542_end_0"), val = tensor([2, 77, 1, 320])]; + tensor var_4542_end_mask_0 = const()[name = tensor("op_4542_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4542_cast = slice_by_index(begin = var_4542_begin_0, end = var_4542_end_0, end_mask = var_4542_end_mask_0, x = transpose_22)[name = tensor("op_4542_cast")]; + tensor var_4546_begin_0 = const()[name = tensor("op_4546_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_4546_end_0 = const()[name = tensor("op_4546_end_0"), val = tensor([2, 77, 1, 384])]; + tensor var_4546_end_mask_0 = const()[name = tensor("op_4546_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4546_cast = slice_by_index(begin = var_4546_begin_0, end = var_4546_end_0, end_mask = var_4546_end_mask_0, x = transpose_22)[name = tensor("op_4546_cast")]; + tensor var_4550_begin_0 = const()[name = tensor("op_4550_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_4550_end_0 = const()[name = tensor("op_4550_end_0"), val = tensor([2, 77, 1, 448])]; + tensor var_4550_end_mask_0 = const()[name = tensor("op_4550_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4550_cast = slice_by_index(begin = var_4550_begin_0, end = var_4550_end_0, end_mask = var_4550_end_mask_0, x = transpose_22)[name = tensor("op_4550_cast")]; + tensor var_4554_begin_0 = const()[name = tensor("op_4554_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_4554_end_0 = const()[name = tensor("op_4554_end_0"), val = tensor([2, 77, 1, 512])]; + tensor var_4554_end_mask_0 = const()[name = tensor("op_4554_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4554_cast = slice_by_index(begin = var_4554_begin_0, end = var_4554_end_0, end_mask = var_4554_end_mask_0, x = transpose_22)[name = tensor("op_4554_cast")]; + tensor var_4558_begin_0 = const()[name = tensor("op_4558_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_4558_end_0 = const()[name = tensor("op_4558_end_0"), val = tensor([2, 77, 1, 576])]; + tensor var_4558_end_mask_0 = const()[name = tensor("op_4558_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4558_cast = slice_by_index(begin = var_4558_begin_0, end = var_4558_end_0, end_mask = var_4558_end_mask_0, x = transpose_22)[name = tensor("op_4558_cast")]; + tensor var_4562_begin_0 = const()[name = tensor("op_4562_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_4562_end_0 = const()[name = tensor("op_4562_end_0"), val = tensor([2, 77, 1, 640])]; + tensor var_4562_end_mask_0 = const()[name = tensor("op_4562_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4562_cast = slice_by_index(begin = var_4562_begin_0, end = var_4562_end_0, end_mask = var_4562_end_mask_0, x = transpose_22)[name = tensor("op_4562_cast")]; + tensor var_4566_begin_0 = const()[name = tensor("op_4566_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_4566_end_0 = const()[name = tensor("op_4566_end_0"), val = tensor([2, 77, 1, 704])]; + tensor var_4566_end_mask_0 = const()[name = tensor("op_4566_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4566_cast = slice_by_index(begin = var_4566_begin_0, end = var_4566_end_0, end_mask = var_4566_end_mask_0, x = transpose_22)[name = tensor("op_4566_cast")]; + tensor var_4570_begin_0 = const()[name = tensor("op_4570_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_4570_end_0 = const()[name = tensor("op_4570_end_0"), val = tensor([2, 77, 1, 768])]; + tensor var_4570_end_mask_0 = const()[name = tensor("op_4570_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4570_cast = slice_by_index(begin = var_4570_begin_0, end = var_4570_end_0, end_mask = var_4570_end_mask_0, x = transpose_22)[name = tensor("op_4570_cast")]; + tensor var_4574_begin_0 = const()[name = tensor("op_4574_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_4574_end_0 = const()[name = tensor("op_4574_end_0"), val = tensor([2, 77, 1, 832])]; + tensor var_4574_end_mask_0 = const()[name = tensor("op_4574_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4574_cast = slice_by_index(begin = var_4574_begin_0, end = var_4574_end_0, end_mask = var_4574_end_mask_0, x = transpose_22)[name = tensor("op_4574_cast")]; + tensor var_4578_begin_0 = const()[name = tensor("op_4578_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_4578_end_0 = const()[name = tensor("op_4578_end_0"), val = tensor([2, 77, 1, 896])]; + tensor var_4578_end_mask_0 = const()[name = tensor("op_4578_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4578_cast = slice_by_index(begin = var_4578_begin_0, end = var_4578_end_0, end_mask = var_4578_end_mask_0, x = transpose_22)[name = tensor("op_4578_cast")]; + tensor var_4582_begin_0 = const()[name = tensor("op_4582_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_4582_end_0 = const()[name = tensor("op_4582_end_0"), val = tensor([2, 77, 1, 960])]; + tensor var_4582_end_mask_0 = const()[name = tensor("op_4582_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4582_cast = slice_by_index(begin = var_4582_begin_0, end = var_4582_end_0, end_mask = var_4582_end_mask_0, x = transpose_22)[name = tensor("op_4582_cast")]; + tensor var_4586_begin_0 = const()[name = tensor("op_4586_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_4586_end_0 = const()[name = tensor("op_4586_end_0"), val = tensor([2, 77, 1, 1024])]; + tensor var_4586_end_mask_0 = const()[name = tensor("op_4586_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4586_cast = slice_by_index(begin = var_4586_begin_0, end = var_4586_end_0, end_mask = var_4586_end_mask_0, x = transpose_22)[name = tensor("op_4586_cast")]; + tensor var_4590_begin_0 = const()[name = tensor("op_4590_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_4590_end_0 = const()[name = tensor("op_4590_end_0"), val = tensor([2, 77, 1, 1088])]; + tensor var_4590_end_mask_0 = const()[name = tensor("op_4590_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4590_cast = slice_by_index(begin = var_4590_begin_0, end = var_4590_end_0, end_mask = var_4590_end_mask_0, x = transpose_22)[name = tensor("op_4590_cast")]; + tensor var_4594_begin_0 = const()[name = tensor("op_4594_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_4594_end_0 = const()[name = tensor("op_4594_end_0"), val = tensor([2, 77, 1, 1152])]; + tensor var_4594_end_mask_0 = const()[name = tensor("op_4594_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4594_cast = slice_by_index(begin = var_4594_begin_0, end = var_4594_end_0, end_mask = var_4594_end_mask_0, x = transpose_22)[name = tensor("op_4594_cast")]; + tensor var_4598_begin_0 = const()[name = tensor("op_4598_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_4598_end_0 = const()[name = tensor("op_4598_end_0"), val = tensor([2, 77, 1, 1216])]; + tensor var_4598_end_mask_0 = const()[name = tensor("op_4598_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4598_cast = slice_by_index(begin = var_4598_begin_0, end = var_4598_end_0, end_mask = var_4598_end_mask_0, x = transpose_22)[name = tensor("op_4598_cast")]; + tensor var_4602_begin_0 = const()[name = tensor("op_4602_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_4602_end_0 = const()[name = tensor("op_4602_end_0"), val = tensor([2, 77, 1, 1280])]; + tensor var_4602_end_mask_0 = const()[name = tensor("op_4602_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4602_cast = slice_by_index(begin = var_4602_begin_0, end = var_4602_end_0, end_mask = var_4602_end_mask_0, x = transpose_22)[name = tensor("op_4602_cast")]; + tensor var_4604_begin_0 = const()[name = tensor("op_4604_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4604_end_0 = const()[name = tensor("op_4604_end_0"), val = tensor([2, 64, 1, 77])]; + tensor var_4604_end_mask_0 = const()[name = tensor("op_4604_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4604_cast = slice_by_index(begin = var_4604_begin_0, end = var_4604_end_0, end_mask = var_4604_end_mask_0, x = v_19_cast)[name = tensor("op_4604_cast")]; + tensor var_4608_begin_0 = const()[name = tensor("op_4608_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_4608_end_0 = const()[name = tensor("op_4608_end_0"), val = tensor([2, 128, 1, 77])]; + tensor var_4608_end_mask_0 = const()[name = tensor("op_4608_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4608_cast = slice_by_index(begin = var_4608_begin_0, end = var_4608_end_0, end_mask = var_4608_end_mask_0, x = v_19_cast)[name = tensor("op_4608_cast")]; + tensor var_4612_begin_0 = const()[name = tensor("op_4612_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_4612_end_0 = const()[name = tensor("op_4612_end_0"), val = tensor([2, 192, 1, 77])]; + tensor var_4612_end_mask_0 = const()[name = tensor("op_4612_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4612_cast = slice_by_index(begin = var_4612_begin_0, end = var_4612_end_0, end_mask = var_4612_end_mask_0, x = v_19_cast)[name = tensor("op_4612_cast")]; + tensor var_4616_begin_0 = const()[name = tensor("op_4616_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_4616_end_0 = const()[name = tensor("op_4616_end_0"), val = tensor([2, 256, 1, 77])]; + tensor var_4616_end_mask_0 = const()[name = tensor("op_4616_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4616_cast = slice_by_index(begin = var_4616_begin_0, end = var_4616_end_0, end_mask = var_4616_end_mask_0, x = v_19_cast)[name = tensor("op_4616_cast")]; + tensor var_4620_begin_0 = const()[name = tensor("op_4620_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_4620_end_0 = const()[name = tensor("op_4620_end_0"), val = tensor([2, 320, 1, 77])]; + tensor var_4620_end_mask_0 = const()[name = tensor("op_4620_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4620_cast = slice_by_index(begin = var_4620_begin_0, end = var_4620_end_0, end_mask = var_4620_end_mask_0, x = v_19_cast)[name = tensor("op_4620_cast")]; + tensor var_4624_begin_0 = const()[name = tensor("op_4624_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_4624_end_0 = const()[name = tensor("op_4624_end_0"), val = tensor([2, 384, 1, 77])]; + tensor var_4624_end_mask_0 = const()[name = tensor("op_4624_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4624_cast = slice_by_index(begin = var_4624_begin_0, end = var_4624_end_0, end_mask = var_4624_end_mask_0, x = v_19_cast)[name = tensor("op_4624_cast")]; + tensor var_4628_begin_0 = const()[name = tensor("op_4628_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_4628_end_0 = const()[name = tensor("op_4628_end_0"), val = tensor([2, 448, 1, 77])]; + tensor var_4628_end_mask_0 = const()[name = tensor("op_4628_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4628_cast = slice_by_index(begin = var_4628_begin_0, end = var_4628_end_0, end_mask = var_4628_end_mask_0, x = v_19_cast)[name = tensor("op_4628_cast")]; + tensor var_4632_begin_0 = const()[name = tensor("op_4632_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_4632_end_0 = const()[name = tensor("op_4632_end_0"), val = tensor([2, 512, 1, 77])]; + tensor var_4632_end_mask_0 = const()[name = tensor("op_4632_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4632_cast = slice_by_index(begin = var_4632_begin_0, end = var_4632_end_0, end_mask = var_4632_end_mask_0, x = v_19_cast)[name = tensor("op_4632_cast")]; + tensor var_4636_begin_0 = const()[name = tensor("op_4636_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_4636_end_0 = const()[name = tensor("op_4636_end_0"), val = tensor([2, 576, 1, 77])]; + tensor var_4636_end_mask_0 = const()[name = tensor("op_4636_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4636_cast = slice_by_index(begin = var_4636_begin_0, end = var_4636_end_0, end_mask = var_4636_end_mask_0, x = v_19_cast)[name = tensor("op_4636_cast")]; + tensor var_4640_begin_0 = const()[name = tensor("op_4640_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_4640_end_0 = const()[name = tensor("op_4640_end_0"), val = tensor([2, 640, 1, 77])]; + tensor var_4640_end_mask_0 = const()[name = tensor("op_4640_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4640_cast = slice_by_index(begin = var_4640_begin_0, end = var_4640_end_0, end_mask = var_4640_end_mask_0, x = v_19_cast)[name = tensor("op_4640_cast")]; + tensor var_4644_begin_0 = const()[name = tensor("op_4644_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_4644_end_0 = const()[name = tensor("op_4644_end_0"), val = tensor([2, 704, 1, 77])]; + tensor var_4644_end_mask_0 = const()[name = tensor("op_4644_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4644_cast = slice_by_index(begin = var_4644_begin_0, end = var_4644_end_0, end_mask = var_4644_end_mask_0, x = v_19_cast)[name = tensor("op_4644_cast")]; + tensor var_4648_begin_0 = const()[name = tensor("op_4648_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_4648_end_0 = const()[name = tensor("op_4648_end_0"), val = tensor([2, 768, 1, 77])]; + tensor var_4648_end_mask_0 = const()[name = tensor("op_4648_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4648_cast = slice_by_index(begin = var_4648_begin_0, end = var_4648_end_0, end_mask = var_4648_end_mask_0, x = v_19_cast)[name = tensor("op_4648_cast")]; + tensor var_4652_begin_0 = const()[name = tensor("op_4652_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_4652_end_0 = const()[name = tensor("op_4652_end_0"), val = tensor([2, 832, 1, 77])]; + tensor var_4652_end_mask_0 = const()[name = tensor("op_4652_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4652_cast = slice_by_index(begin = var_4652_begin_0, end = var_4652_end_0, end_mask = var_4652_end_mask_0, x = v_19_cast)[name = tensor("op_4652_cast")]; + tensor var_4656_begin_0 = const()[name = tensor("op_4656_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_4656_end_0 = const()[name = tensor("op_4656_end_0"), val = tensor([2, 896, 1, 77])]; + tensor var_4656_end_mask_0 = const()[name = tensor("op_4656_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4656_cast = slice_by_index(begin = var_4656_begin_0, end = var_4656_end_0, end_mask = var_4656_end_mask_0, x = v_19_cast)[name = tensor("op_4656_cast")]; + tensor var_4660_begin_0 = const()[name = tensor("op_4660_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_4660_end_0 = const()[name = tensor("op_4660_end_0"), val = tensor([2, 960, 1, 77])]; + tensor var_4660_end_mask_0 = const()[name = tensor("op_4660_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4660_cast = slice_by_index(begin = var_4660_begin_0, end = var_4660_end_0, end_mask = var_4660_end_mask_0, x = v_19_cast)[name = tensor("op_4660_cast")]; + tensor var_4664_begin_0 = const()[name = tensor("op_4664_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_4664_end_0 = const()[name = tensor("op_4664_end_0"), val = tensor([2, 1024, 1, 77])]; + tensor var_4664_end_mask_0 = const()[name = tensor("op_4664_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4664_cast = slice_by_index(begin = var_4664_begin_0, end = var_4664_end_0, end_mask = var_4664_end_mask_0, x = v_19_cast)[name = tensor("op_4664_cast")]; + tensor var_4668_begin_0 = const()[name = tensor("op_4668_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_4668_end_0 = const()[name = tensor("op_4668_end_0"), val = tensor([2, 1088, 1, 77])]; + tensor var_4668_end_mask_0 = const()[name = tensor("op_4668_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4668_cast = slice_by_index(begin = var_4668_begin_0, end = var_4668_end_0, end_mask = var_4668_end_mask_0, x = v_19_cast)[name = tensor("op_4668_cast")]; + tensor var_4672_begin_0 = const()[name = tensor("op_4672_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_4672_end_0 = const()[name = tensor("op_4672_end_0"), val = tensor([2, 1152, 1, 77])]; + tensor var_4672_end_mask_0 = const()[name = tensor("op_4672_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4672_cast = slice_by_index(begin = var_4672_begin_0, end = var_4672_end_0, end_mask = var_4672_end_mask_0, x = v_19_cast)[name = tensor("op_4672_cast")]; + tensor var_4676_begin_0 = const()[name = tensor("op_4676_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_4676_end_0 = const()[name = tensor("op_4676_end_0"), val = tensor([2, 1216, 1, 77])]; + tensor var_4676_end_mask_0 = const()[name = tensor("op_4676_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4676_cast = slice_by_index(begin = var_4676_begin_0, end = var_4676_end_0, end_mask = var_4676_end_mask_0, x = v_19_cast)[name = tensor("op_4676_cast")]; + tensor var_4680_begin_0 = const()[name = tensor("op_4680_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_4680_end_0 = const()[name = tensor("op_4680_end_0"), val = tensor([2, 1280, 1, 77])]; + tensor var_4680_end_mask_0 = const()[name = tensor("op_4680_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4680_cast = slice_by_index(begin = var_4680_begin_0, end = var_4680_end_0, end_mask = var_4680_end_mask_0, x = v_19_cast)[name = tensor("op_4680_cast")]; + tensor var_4684_equation_0 = const()[name = tensor("op_4684_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4684_cast = einsum(equation = var_4684_equation_0, values = (var_4526_cast, var_4443_cast))[name = tensor("op_4684_cast")]; + tensor var_4685_to_fp16 = const()[name = tensor("op_4685_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_41_cast = mul(x = var_4684_cast, y = var_4685_to_fp16)[name = tensor("aw_41_cast")]; + tensor var_4688_equation_0 = const()[name = tensor("op_4688_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4688_cast = einsum(equation = var_4688_equation_0, values = (var_4530_cast, var_4447_cast))[name = tensor("op_4688_cast")]; + tensor var_4689_to_fp16 = const()[name = tensor("op_4689_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_43_cast = mul(x = var_4688_cast, y = var_4689_to_fp16)[name = tensor("aw_43_cast")]; + tensor var_4692_equation_0 = const()[name = tensor("op_4692_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4692_cast = einsum(equation = var_4692_equation_0, values = (var_4534_cast, var_4451_cast))[name = tensor("op_4692_cast")]; + tensor var_4693_to_fp16 = const()[name = tensor("op_4693_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_45_cast = mul(x = var_4692_cast, y = var_4693_to_fp16)[name = tensor("aw_45_cast")]; + tensor var_4696_equation_0 = const()[name = tensor("op_4696_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4696_cast = einsum(equation = var_4696_equation_0, values = (var_4538_cast, var_4455_cast))[name = tensor("op_4696_cast")]; + tensor var_4697_to_fp16 = const()[name = tensor("op_4697_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_47_cast = mul(x = var_4696_cast, y = var_4697_to_fp16)[name = tensor("aw_47_cast")]; + tensor var_4700_equation_0 = const()[name = tensor("op_4700_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4700_cast = einsum(equation = var_4700_equation_0, values = (var_4542_cast, var_4459_cast))[name = tensor("op_4700_cast")]; + tensor var_4701_to_fp16 = const()[name = tensor("op_4701_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_49_cast = mul(x = var_4700_cast, y = var_4701_to_fp16)[name = tensor("aw_49_cast")]; + tensor var_4704_equation_0 = const()[name = tensor("op_4704_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4704_cast = einsum(equation = var_4704_equation_0, values = (var_4546_cast, var_4463_cast))[name = tensor("op_4704_cast")]; + tensor var_4705_to_fp16 = const()[name = tensor("op_4705_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_51_cast = mul(x = var_4704_cast, y = var_4705_to_fp16)[name = tensor("aw_51_cast")]; + tensor var_4708_equation_0 = const()[name = tensor("op_4708_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4708_cast = einsum(equation = var_4708_equation_0, values = (var_4550_cast, var_4467_cast))[name = tensor("op_4708_cast")]; + tensor var_4709_to_fp16 = const()[name = tensor("op_4709_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_53_cast = mul(x = var_4708_cast, y = var_4709_to_fp16)[name = tensor("aw_53_cast")]; + tensor var_4712_equation_0 = const()[name = tensor("op_4712_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4712_cast = einsum(equation = var_4712_equation_0, values = (var_4554_cast, var_4471_cast))[name = tensor("op_4712_cast")]; + tensor var_4713_to_fp16 = const()[name = tensor("op_4713_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_55_cast = mul(x = var_4712_cast, y = var_4713_to_fp16)[name = tensor("aw_55_cast")]; + tensor var_4716_equation_0 = const()[name = tensor("op_4716_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4716_cast = einsum(equation = var_4716_equation_0, values = (var_4558_cast, var_4475_cast))[name = tensor("op_4716_cast")]; + tensor var_4717_to_fp16 = const()[name = tensor("op_4717_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_57_cast = mul(x = var_4716_cast, y = var_4717_to_fp16)[name = tensor("aw_57_cast")]; + tensor var_4720_equation_0 = const()[name = tensor("op_4720_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4720_cast = einsum(equation = var_4720_equation_0, values = (var_4562_cast, var_4479_cast))[name = tensor("op_4720_cast")]; + tensor var_4721_to_fp16 = const()[name = tensor("op_4721_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_59_cast = mul(x = var_4720_cast, y = var_4721_to_fp16)[name = tensor("aw_59_cast")]; + tensor var_4724_equation_0 = const()[name = tensor("op_4724_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4724_cast = einsum(equation = var_4724_equation_0, values = (var_4566_cast, var_4483_cast))[name = tensor("op_4724_cast")]; + tensor var_4725_to_fp16 = const()[name = tensor("op_4725_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_61_cast = mul(x = var_4724_cast, y = var_4725_to_fp16)[name = tensor("aw_61_cast")]; + tensor var_4728_equation_0 = const()[name = tensor("op_4728_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4728_cast = einsum(equation = var_4728_equation_0, values = (var_4570_cast, var_4487_cast))[name = tensor("op_4728_cast")]; + tensor var_4729_to_fp16 = const()[name = tensor("op_4729_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_63_cast = mul(x = var_4728_cast, y = var_4729_to_fp16)[name = tensor("aw_63_cast")]; + tensor var_4732_equation_0 = const()[name = tensor("op_4732_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4732_cast = einsum(equation = var_4732_equation_0, values = (var_4574_cast, var_4491_cast))[name = tensor("op_4732_cast")]; + tensor var_4733_to_fp16 = const()[name = tensor("op_4733_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_65_cast = mul(x = var_4732_cast, y = var_4733_to_fp16)[name = tensor("aw_65_cast")]; + tensor var_4736_equation_0 = const()[name = tensor("op_4736_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4736_cast = einsum(equation = var_4736_equation_0, values = (var_4578_cast, var_4495_cast))[name = tensor("op_4736_cast")]; + tensor var_4737_to_fp16 = const()[name = tensor("op_4737_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_67_cast = mul(x = var_4736_cast, y = var_4737_to_fp16)[name = tensor("aw_67_cast")]; + tensor var_4740_equation_0 = const()[name = tensor("op_4740_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4740_cast = einsum(equation = var_4740_equation_0, values = (var_4582_cast, var_4499_cast))[name = tensor("op_4740_cast")]; + tensor var_4741_to_fp16 = const()[name = tensor("op_4741_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_69_cast = mul(x = var_4740_cast, y = var_4741_to_fp16)[name = tensor("aw_69_cast")]; + tensor var_4744_equation_0 = const()[name = tensor("op_4744_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4744_cast = einsum(equation = var_4744_equation_0, values = (var_4586_cast, var_4503_cast))[name = tensor("op_4744_cast")]; + tensor var_4745_to_fp16 = const()[name = tensor("op_4745_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_71_cast = mul(x = var_4744_cast, y = var_4745_to_fp16)[name = tensor("aw_71_cast")]; + tensor var_4748_equation_0 = const()[name = tensor("op_4748_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4748_cast = einsum(equation = var_4748_equation_0, values = (var_4590_cast, var_4507_cast))[name = tensor("op_4748_cast")]; + tensor var_4749_to_fp16 = const()[name = tensor("op_4749_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_73_cast = mul(x = var_4748_cast, y = var_4749_to_fp16)[name = tensor("aw_73_cast")]; + tensor var_4752_equation_0 = const()[name = tensor("op_4752_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4752_cast = einsum(equation = var_4752_equation_0, values = (var_4594_cast, var_4511_cast))[name = tensor("op_4752_cast")]; + tensor var_4753_to_fp16 = const()[name = tensor("op_4753_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_75_cast = mul(x = var_4752_cast, y = var_4753_to_fp16)[name = tensor("aw_75_cast")]; + tensor var_4756_equation_0 = const()[name = tensor("op_4756_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4756_cast = einsum(equation = var_4756_equation_0, values = (var_4598_cast, var_4515_cast))[name = tensor("op_4756_cast")]; + tensor var_4757_to_fp16 = const()[name = tensor("op_4757_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_77_cast = mul(x = var_4756_cast, y = var_4757_to_fp16)[name = tensor("aw_77_cast")]; + tensor var_4760_equation_0 = const()[name = tensor("op_4760_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4760_cast = einsum(equation = var_4760_equation_0, values = (var_4602_cast, var_4519_cast))[name = tensor("op_4760_cast")]; + tensor var_4761_to_fp16 = const()[name = tensor("op_4761_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_79_cast = mul(x = var_4760_cast, y = var_4761_to_fp16)[name = tensor("aw_79_cast")]; + tensor var_4763_cast = softmax(axis = var_3883, x = aw_41_cast)[name = tensor("op_4763_cast")]; + tensor var_4764_cast = softmax(axis = var_3883, x = aw_43_cast)[name = tensor("op_4764_cast")]; + tensor var_4765_cast = softmax(axis = var_3883, x = aw_45_cast)[name = tensor("op_4765_cast")]; + tensor var_4766_cast = softmax(axis = var_3883, x = aw_47_cast)[name = tensor("op_4766_cast")]; + tensor var_4767_cast = softmax(axis = var_3883, x = aw_49_cast)[name = tensor("op_4767_cast")]; + tensor var_4768_cast = softmax(axis = var_3883, x = aw_51_cast)[name = tensor("op_4768_cast")]; + tensor var_4769_cast = softmax(axis = var_3883, x = aw_53_cast)[name = tensor("op_4769_cast")]; + tensor var_4770_cast = softmax(axis = var_3883, x = aw_55_cast)[name = tensor("op_4770_cast")]; + tensor var_4771_cast = softmax(axis = var_3883, x = aw_57_cast)[name = tensor("op_4771_cast")]; + tensor var_4772_cast = softmax(axis = var_3883, x = aw_59_cast)[name = tensor("op_4772_cast")]; + tensor var_4773_cast = softmax(axis = var_3883, x = aw_61_cast)[name = tensor("op_4773_cast")]; + tensor var_4774_cast = softmax(axis = var_3883, x = aw_63_cast)[name = tensor("op_4774_cast")]; + tensor var_4775_cast = softmax(axis = var_3883, x = aw_65_cast)[name = tensor("op_4775_cast")]; + tensor var_4776_cast = softmax(axis = var_3883, x = aw_67_cast)[name = tensor("op_4776_cast")]; + tensor var_4777_cast = softmax(axis = var_3883, x = aw_69_cast)[name = tensor("op_4777_cast")]; + tensor var_4778_cast = softmax(axis = var_3883, x = aw_71_cast)[name = tensor("op_4778_cast")]; + tensor var_4779_cast = softmax(axis = var_3883, x = aw_73_cast)[name = tensor("op_4779_cast")]; + tensor var_4780_cast = softmax(axis = var_3883, x = aw_75_cast)[name = tensor("op_4780_cast")]; + tensor var_4781_cast = softmax(axis = var_3883, x = aw_77_cast)[name = tensor("op_4781_cast")]; + tensor var_4782_cast = softmax(axis = var_3883, x = aw_79_cast)[name = tensor("op_4782_cast")]; + tensor var_4784_equation_0 = const()[name = tensor("op_4784_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4784_cast = einsum(equation = var_4784_equation_0, values = (var_4604_cast, var_4763_cast))[name = tensor("op_4784_cast")]; + tensor var_4786_equation_0 = const()[name = tensor("op_4786_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4786_cast = einsum(equation = var_4786_equation_0, values = (var_4608_cast, var_4764_cast))[name = tensor("op_4786_cast")]; + tensor var_4788_equation_0 = const()[name = tensor("op_4788_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4788_cast = einsum(equation = var_4788_equation_0, values = (var_4612_cast, var_4765_cast))[name = tensor("op_4788_cast")]; + tensor var_4790_equation_0 = const()[name = tensor("op_4790_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4790_cast = einsum(equation = var_4790_equation_0, values = (var_4616_cast, var_4766_cast))[name = tensor("op_4790_cast")]; + tensor var_4792_equation_0 = const()[name = tensor("op_4792_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4792_cast = einsum(equation = var_4792_equation_0, values = (var_4620_cast, var_4767_cast))[name = tensor("op_4792_cast")]; + tensor var_4794_equation_0 = const()[name = tensor("op_4794_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4794_cast = einsum(equation = var_4794_equation_0, values = (var_4624_cast, var_4768_cast))[name = tensor("op_4794_cast")]; + tensor var_4796_equation_0 = const()[name = tensor("op_4796_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4796_cast = einsum(equation = var_4796_equation_0, values = (var_4628_cast, var_4769_cast))[name = tensor("op_4796_cast")]; + tensor var_4798_equation_0 = const()[name = tensor("op_4798_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4798_cast = einsum(equation = var_4798_equation_0, values = (var_4632_cast, var_4770_cast))[name = tensor("op_4798_cast")]; + tensor var_4800_equation_0 = const()[name = tensor("op_4800_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4800_cast = einsum(equation = var_4800_equation_0, values = (var_4636_cast, var_4771_cast))[name = tensor("op_4800_cast")]; + tensor var_4802_equation_0 = const()[name = tensor("op_4802_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4802_cast = einsum(equation = var_4802_equation_0, values = (var_4640_cast, var_4772_cast))[name = tensor("op_4802_cast")]; + tensor var_4804_equation_0 = const()[name = tensor("op_4804_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4804_cast = einsum(equation = var_4804_equation_0, values = (var_4644_cast, var_4773_cast))[name = tensor("op_4804_cast")]; + tensor var_4806_equation_0 = const()[name = tensor("op_4806_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4806_cast = einsum(equation = var_4806_equation_0, values = (var_4648_cast, var_4774_cast))[name = tensor("op_4806_cast")]; + tensor var_4808_equation_0 = const()[name = tensor("op_4808_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4808_cast = einsum(equation = var_4808_equation_0, values = (var_4652_cast, var_4775_cast))[name = tensor("op_4808_cast")]; + tensor var_4810_equation_0 = const()[name = tensor("op_4810_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4810_cast = einsum(equation = var_4810_equation_0, values = (var_4656_cast, var_4776_cast))[name = tensor("op_4810_cast")]; + tensor var_4812_equation_0 = const()[name = tensor("op_4812_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4812_cast = einsum(equation = var_4812_equation_0, values = (var_4660_cast, var_4777_cast))[name = tensor("op_4812_cast")]; + tensor var_4814_equation_0 = const()[name = tensor("op_4814_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4814_cast = einsum(equation = var_4814_equation_0, values = (var_4664_cast, var_4778_cast))[name = tensor("op_4814_cast")]; + tensor var_4816_equation_0 = const()[name = tensor("op_4816_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4816_cast = einsum(equation = var_4816_equation_0, values = (var_4668_cast, var_4779_cast))[name = tensor("op_4816_cast")]; + tensor var_4818_equation_0 = const()[name = tensor("op_4818_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4818_cast = einsum(equation = var_4818_equation_0, values = (var_4672_cast, var_4780_cast))[name = tensor("op_4818_cast")]; + tensor var_4820_equation_0 = const()[name = tensor("op_4820_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4820_cast = einsum(equation = var_4820_equation_0, values = (var_4676_cast, var_4781_cast))[name = tensor("op_4820_cast")]; + tensor var_4822_equation_0 = const()[name = tensor("op_4822_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4822_cast = einsum(equation = var_4822_equation_0, values = (var_4680_cast, var_4782_cast))[name = tensor("op_4822_cast")]; + tensor input_135_interleave_0 = const()[name = tensor("input_135_interleave_0"), val = tensor(false)]; + tensor input_135_cast = concat(axis = var_3883, interleave = input_135_interleave_0, values = (var_4784_cast, var_4786_cast, var_4788_cast, var_4790_cast, var_4792_cast, var_4794_cast, var_4796_cast, var_4798_cast, var_4800_cast, var_4802_cast, var_4804_cast, var_4806_cast, var_4808_cast, var_4810_cast, var_4812_cast, var_4814_cast, var_4816_cast, var_4818_cast, var_4820_cast, var_4822_cast))[name = tensor("input_135_cast")]; + tensor var_4828 = const()[name = tensor("op_4828"), val = tensor([1, 1])]; + tensor var_4830 = const()[name = tensor("op_4830"), val = tensor([1, 1])]; + tensor var_4832_pad_type_0 = const()[name = tensor("op_4832_pad_type_0"), val = tensor("custom")]; + tensor var_4832_pad_0 = const()[name = tensor("op_4832_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(65690752))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(66919616))), name = tensor("down_blocks_2_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor down_blocks_2_attentions_0_transformer_blocks_0_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_0_attn2_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(66919808)))]; + tensor var_4832_cast = conv(bias = down_blocks_2_attentions_0_transformer_blocks_0_attn2_to_out_0_bias_to_fp16, dilations = var_4830, groups = var_3883, pad = var_4832_pad_0, pad_type = var_4832_pad_type_0, strides = var_4828, weight = down_blocks_2_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16_palettized, x = input_135_cast)[name = tensor("op_4832_cast")]; + tensor inputs_29_cast = add(x = var_4832_cast, y = inputs_27_cast)[name = tensor("inputs_29_cast")]; + tensor var_4836 = const()[name = tensor("op_4836"), val = tensor([1])]; + tensor channels_mean_29_cast = reduce_mean(axes = var_4836, keep_dims = var_3878, x = inputs_29_cast)[name = tensor("channels_mean_29_cast")]; + tensor zero_mean_29_cast = sub(x = inputs_29_cast, y = channels_mean_29_cast)[name = tensor("zero_mean_29_cast")]; + tensor zero_mean_sq_29_cast = mul(x = zero_mean_29_cast, y = zero_mean_29_cast)[name = tensor("zero_mean_sq_29_cast")]; + tensor var_4840 = const()[name = tensor("op_4840"), val = tensor([1])]; + tensor var_4841_cast = reduce_mean(axes = var_4840, keep_dims = var_3878, x = zero_mean_sq_29_cast)[name = tensor("op_4841_cast")]; + tensor var_4842_to_fp16 = const()[name = tensor("op_4842_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_4843_cast = add(x = var_4841_cast, y = var_4842_to_fp16)[name = tensor("op_4843_cast")]; + tensor denom_29_epsilon_0_to_fp16 = const()[name = tensor("denom_29_epsilon_0_to_fp16"), val = tensor(0x1p-24)]; + tensor denom_29_cast = rsqrt(epsilon = denom_29_epsilon_0_to_fp16, x = var_4843_cast)[name = tensor("denom_29_cast")]; + tensor out_29_cast = mul(x = zero_mean_29_cast, y = denom_29_cast)[name = tensor("out_29_cast")]; + tensor var_4847_to_fp16 = const()[name = tensor("op_4847_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(66922432)))]; + tensor var_4848_cast = add(x = out_29_cast, y = var_4847_to_fp16)[name = tensor("op_4848_cast")]; + tensor var_4850_to_fp16 = const()[name = tensor("op_4850_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(66925056)))]; + tensor input_137_cast = mul(x = var_4848_cast, y = var_4850_to_fp16)[name = tensor("input_137_cast")]; + tensor var_4858 = const()[name = tensor("op_4858"), val = tensor([1, 1])]; + tensor var_4860 = const()[name = tensor("op_4860"), val = tensor([1, 1])]; + tensor var_4862_pad_type_0 = const()[name = tensor("op_4862_pad_type_0"), val = tensor("custom")]; + tensor var_4862_pad_0 = const()[name = tensor("op_4862_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(66927680))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(76758144))), name = tensor("down_blocks_2_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16_palettized"), shape = tensor([10240, 1280, 1, 1])]; + tensor down_blocks_2_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(76758336))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(76766080))), name = tensor("down_blocks_2_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16_palettized"), shape = tensor([10240])]; + tensor var_4862_cast = conv(bias = down_blocks_2_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16_palettized, dilations = var_4860, groups = var_3883, pad = var_4862_pad_0, pad_type = var_4862_pad_type_0, strides = var_4858, weight = down_blocks_2_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16_palettized, x = input_137_cast)[name = tensor("op_4862_cast")]; + tensor var_4863_split_sizes_0 = const()[name = tensor("op_4863_split_sizes_0"), val = tensor([5120, 5120])]; + tensor var_4863_axis_0 = const()[name = tensor("op_4863_axis_0"), val = tensor(1)]; + tensor var_4863_cast_0, tensor var_4863_cast_1 = split(axis = var_4863_axis_0, split_sizes = var_4863_split_sizes_0, x = var_4862_cast)[name = tensor("op_4863_cast")]; + tensor var_4865_mode_0 = const()[name = tensor("op_4865_mode_0"), val = tensor("EXACT")]; + tensor var_4865_cast = gelu(mode = var_4865_mode_0, x = var_4863_cast_1)[name = tensor("op_4865_cast")]; + tensor input_139_cast = mul(x = var_4863_cast_0, y = var_4865_cast)[name = tensor("input_139_cast")]; + tensor var_4869 = const()[name = tensor("op_4869"), val = tensor([1, 1])]; + tensor var_4871 = const()[name = tensor("op_4871"), val = tensor([1, 1])]; + tensor var_4873_pad_type_0 = const()[name = tensor("op_4873_pad_type_0"), val = tensor("custom")]; + tensor var_4873_pad_0 = const()[name = tensor("op_4873_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(76766272))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(81681536))), name = tensor("down_blocks_2_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16_palettized"), shape = tensor([1280, 5120, 1, 1])]; + tensor down_blocks_2_attentions_0_transformer_blocks_0_ff_net_2_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_transformer_blocks_0_ff_net_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(81681728)))]; + tensor var_4873_cast = conv(bias = down_blocks_2_attentions_0_transformer_blocks_0_ff_net_2_bias_to_fp16, dilations = var_4871, groups = var_3883, pad = var_4873_pad_0, pad_type = var_4873_pad_type_0, strides = var_4869, weight = down_blocks_2_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16_palettized, x = input_139_cast)[name = tensor("op_4873_cast")]; + tensor hidden_states_87_cast = add(x = var_4873_cast, y = inputs_29_cast)[name = tensor("hidden_states_87_cast")]; + tensor var_4875 = const()[name = tensor("op_4875"), val = tensor([2, 1280, 16, 16])]; + tensor input_141_cast = reshape(shape = var_4875, x = hidden_states_87_cast)[name = tensor("input_141_cast")]; + tensor var_4879 = const()[name = tensor("op_4879"), val = tensor([1, 1])]; + tensor var_4881 = const()[name = tensor("op_4881"), val = tensor([1, 1])]; + tensor hidden_states_89_pad_type_0 = const()[name = tensor("hidden_states_89_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_89_pad_0 = const()[name = tensor("hidden_states_89_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_0_proj_out_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(81684352))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(82913216))), name = tensor("down_blocks_2_attentions_0_proj_out_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor down_blocks_2_attentions_0_proj_out_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_0_proj_out_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(82913408)))]; + tensor hidden_states_89_cast = conv(bias = down_blocks_2_attentions_0_proj_out_bias_to_fp16, dilations = var_4881, groups = var_3883, pad = hidden_states_89_pad_0, pad_type = hidden_states_89_pad_type_0, strides = var_4879, weight = down_blocks_2_attentions_0_proj_out_weight_to_fp16_palettized, x = input_141_cast)[name = tensor("hidden_states_89_cast")]; + tensor input_143_cast = add(x = hidden_states_89_cast, y = hidden_states_77_cast)[name = tensor("input_143_cast")]; + tensor reshape_60_shape_0 = const()[name = tensor("reshape_60_shape_0"), val = tensor([2, 32, 40, 16, 16])]; + tensor reshape_60_cast = reshape(shape = reshape_60_shape_0, x = input_143_cast)[name = tensor("reshape_60_cast")]; + tensor reduce_mean_45_axes_0 = const()[name = tensor("reduce_mean_45_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_45_keep_dims_0 = const()[name = tensor("reduce_mean_45_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_45_cast = reduce_mean(axes = reduce_mean_45_axes_0, keep_dims = reduce_mean_45_keep_dims_0, x = reshape_60_cast)[name = tensor("reduce_mean_45_cast")]; + tensor sub_30_cast = sub(x = reshape_60_cast, y = reduce_mean_45_cast)[name = tensor("sub_30_cast")]; + tensor square_15_cast = square(x = sub_30_cast)[name = tensor("square_15_cast")]; + tensor reduce_mean_47_axes_0 = const()[name = tensor("reduce_mean_47_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_47_keep_dims_0 = const()[name = tensor("reduce_mean_47_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_47_cast = reduce_mean(axes = reduce_mean_47_axes_0, keep_dims = reduce_mean_47_keep_dims_0, x = square_15_cast)[name = tensor("reduce_mean_47_cast")]; + tensor add_30_y_0_to_fp16 = const()[name = tensor("add_30_y_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_30_cast = add(x = reduce_mean_47_cast, y = add_30_y_0_to_fp16)[name = tensor("add_30_cast")]; + tensor sqrt_15_cast = sqrt(x = add_30_cast)[name = tensor("sqrt_15_cast")]; + tensor real_div_15_cast = real_div(x = sub_30_cast, y = sqrt_15_cast)[name = tensor("real_div_15_cast")]; + tensor reshape_61_shape_0 = const()[name = tensor("reshape_61_shape_0"), val = tensor([2, 1280, 16, 16])]; + tensor reshape_61_cast = reshape(shape = reshape_61_shape_0, x = real_div_15_cast)[name = tensor("reshape_61_cast")]; + tensor add_31_gamma_0_to_fp16 = const()[name = tensor("add_31_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(82916032)))]; + tensor add_31_beta_0_to_fp16 = const()[name = tensor("add_31_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(82918656)))]; + tensor add_31_epsilon_0_to_fp16 = const()[name = tensor("add_31_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_31_cast = batch_norm(beta = add_31_beta_0_to_fp16, epsilon = add_31_epsilon_0_to_fp16, gamma = add_31_gamma_0_to_fp16, mean = add_27_mean_0_to_fp16, variance = add_27_variance_0_to_fp16, x = reshape_61_cast)[name = tensor("add_31_cast")]; + tensor input_147_cast = silu(x = add_31_cast)[name = tensor("input_147_cast")]; + tensor var_4896 = const()[name = tensor("op_4896"), val = tensor([1, 1])]; + tensor var_4898 = const()[name = tensor("op_4898"), val = tensor([1, 1])]; + tensor hidden_states_91_pad_type_0 = const()[name = tensor("hidden_states_91_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_91_pad_0 = const()[name = tensor("hidden_states_91_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor down_blocks_2_resnets_1_conv1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(82921280))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(93980544))), name = tensor("down_blocks_2_resnets_1_conv1_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 3, 3])]; + tensor down_blocks_2_resnets_1_conv1_bias_to_fp16 = const()[name = tensor("down_blocks_2_resnets_1_conv1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(93980736)))]; + tensor hidden_states_91_cast = conv(bias = down_blocks_2_resnets_1_conv1_bias_to_fp16, dilations = var_4898, groups = var_3883, pad = hidden_states_91_pad_0, pad_type = hidden_states_91_pad_type_0, strides = var_4896, weight = down_blocks_2_resnets_1_conv1_weight_to_fp16_palettized, x = input_147_cast)[name = tensor("hidden_states_91_cast")]; + tensor var_4904 = const()[name = tensor("op_4904"), val = tensor([1, 1])]; + tensor var_4906 = const()[name = tensor("op_4906"), val = tensor([1, 1])]; + tensor temb_11_pad_type_0 = const()[name = tensor("temb_11_pad_type_0"), val = tensor("custom")]; + tensor temb_11_pad_0 = const()[name = tensor("temb_11_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_resnets_1_time_emb_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(93983360))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(95212224))), name = tensor("down_blocks_2_resnets_1_time_emb_proj_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor down_blocks_2_resnets_1_time_emb_proj_bias_to_fp16 = const()[name = tensor("down_blocks_2_resnets_1_time_emb_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(95212416)))]; + tensor temb_11_cast = conv(bias = down_blocks_2_resnets_1_time_emb_proj_bias_to_fp16, dilations = var_4906, groups = var_3883, pad = temb_11_pad_0, pad_type = temb_11_pad_type_0, strides = var_4904, weight = down_blocks_2_resnets_1_time_emb_proj_weight_to_fp16_palettized, x = input_15_cast)[name = tensor("temb_11_cast")]; + tensor input_151_cast = add(x = hidden_states_91_cast, y = temb_11_cast)[name = tensor("input_151_cast")]; + tensor reshape_64_shape_0 = const()[name = tensor("reshape_64_shape_0"), val = tensor([2, 32, 40, 16, 16])]; + tensor reshape_64_cast = reshape(shape = reshape_64_shape_0, x = input_151_cast)[name = tensor("reshape_64_cast")]; + tensor reduce_mean_48_axes_0 = const()[name = tensor("reduce_mean_48_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_48_keep_dims_0 = const()[name = tensor("reduce_mean_48_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_48_cast = reduce_mean(axes = reduce_mean_48_axes_0, keep_dims = reduce_mean_48_keep_dims_0, x = reshape_64_cast)[name = tensor("reduce_mean_48_cast")]; + tensor sub_32_cast = sub(x = reshape_64_cast, y = reduce_mean_48_cast)[name = tensor("sub_32_cast")]; + tensor square_16_cast = square(x = sub_32_cast)[name = tensor("square_16_cast")]; + tensor reduce_mean_50_axes_0 = const()[name = tensor("reduce_mean_50_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_50_keep_dims_0 = const()[name = tensor("reduce_mean_50_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_50_cast = reduce_mean(axes = reduce_mean_50_axes_0, keep_dims = reduce_mean_50_keep_dims_0, x = square_16_cast)[name = tensor("reduce_mean_50_cast")]; + tensor add_32_y_0_to_fp16 = const()[name = tensor("add_32_y_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_32_cast = add(x = reduce_mean_50_cast, y = add_32_y_0_to_fp16)[name = tensor("add_32_cast")]; + tensor sqrt_16_cast = sqrt(x = add_32_cast)[name = tensor("sqrt_16_cast")]; + tensor real_div_16_cast = real_div(x = sub_32_cast, y = sqrt_16_cast)[name = tensor("real_div_16_cast")]; + tensor reshape_65_shape_0 = const()[name = tensor("reshape_65_shape_0"), val = tensor([2, 1280, 16, 16])]; + tensor reshape_65_cast = reshape(shape = reshape_65_shape_0, x = real_div_16_cast)[name = tensor("reshape_65_cast")]; + tensor add_33_gamma_0_to_fp16 = const()[name = tensor("add_33_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(95215040)))]; + tensor add_33_beta_0_to_fp16 = const()[name = tensor("add_33_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(95217664)))]; + tensor add_33_epsilon_0_to_fp16 = const()[name = tensor("add_33_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_33_cast = batch_norm(beta = add_33_beta_0_to_fp16, epsilon = add_33_epsilon_0_to_fp16, gamma = add_33_gamma_0_to_fp16, mean = add_27_mean_0_to_fp16, variance = add_27_variance_0_to_fp16, x = reshape_65_cast)[name = tensor("add_33_cast")]; + tensor input_155_cast = silu(x = add_33_cast)[name = tensor("input_155_cast")]; + tensor var_4916 = const()[name = tensor("op_4916"), val = tensor([1, 1])]; + tensor var_4918 = const()[name = tensor("op_4918"), val = tensor([1, 1])]; + tensor hidden_states_93_pad_type_0 = const()[name = tensor("hidden_states_93_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_93_pad_0 = const()[name = tensor("hidden_states_93_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor down_blocks_2_resnets_1_conv2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(95220288))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(106279552))), name = tensor("down_blocks_2_resnets_1_conv2_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 3, 3])]; + tensor down_blocks_2_resnets_1_conv2_bias_to_fp16 = const()[name = tensor("down_blocks_2_resnets_1_conv2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(106279744)))]; + tensor hidden_states_93_cast = conv(bias = down_blocks_2_resnets_1_conv2_bias_to_fp16, dilations = var_4918, groups = var_3883, pad = hidden_states_93_pad_0, pad_type = hidden_states_93_pad_type_0, strides = var_4916, weight = down_blocks_2_resnets_1_conv2_weight_to_fp16_palettized, x = input_155_cast)[name = tensor("hidden_states_93_cast")]; + tensor hidden_states_95_cast = add(x = input_143_cast, y = hidden_states_93_cast)[name = tensor("hidden_states_95_cast")]; + tensor reshape_68_shape_0 = const()[name = tensor("reshape_68_shape_0"), val = tensor([2, 32, 40, 16, 16])]; + tensor reshape_68_cast = reshape(shape = reshape_68_shape_0, x = hidden_states_95_cast)[name = tensor("reshape_68_cast")]; + tensor reduce_mean_51_axes_0 = const()[name = tensor("reduce_mean_51_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_51_keep_dims_0 = const()[name = tensor("reduce_mean_51_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_51_cast = reduce_mean(axes = reduce_mean_51_axes_0, keep_dims = reduce_mean_51_keep_dims_0, x = reshape_68_cast)[name = tensor("reduce_mean_51_cast")]; + tensor sub_34_cast = sub(x = reshape_68_cast, y = reduce_mean_51_cast)[name = tensor("sub_34_cast")]; + tensor square_17_cast = square(x = sub_34_cast)[name = tensor("square_17_cast")]; + tensor reduce_mean_53_axes_0 = const()[name = tensor("reduce_mean_53_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_53_keep_dims_0 = const()[name = tensor("reduce_mean_53_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_53_cast = reduce_mean(axes = reduce_mean_53_axes_0, keep_dims = reduce_mean_53_keep_dims_0, x = square_17_cast)[name = tensor("reduce_mean_53_cast")]; + tensor add_34_y_0_to_fp16 = const()[name = tensor("add_34_y_0_to_fp16"), val = tensor(0x1.1p-20)]; + tensor add_34_cast = add(x = reduce_mean_53_cast, y = add_34_y_0_to_fp16)[name = tensor("add_34_cast")]; + tensor sqrt_17_cast = sqrt(x = add_34_cast)[name = tensor("sqrt_17_cast")]; + tensor real_div_17_cast = real_div(x = sub_34_cast, y = sqrt_17_cast)[name = tensor("real_div_17_cast")]; + tensor reshape_69_shape_0 = const()[name = tensor("reshape_69_shape_0"), val = tensor([2, 1280, 16, 16])]; + tensor reshape_69_cast = reshape(shape = reshape_69_shape_0, x = real_div_17_cast)[name = tensor("reshape_69_cast")]; + tensor add_35_gamma_0_to_fp16 = const()[name = tensor("add_35_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(106282368)))]; + tensor add_35_beta_0_to_fp16 = const()[name = tensor("add_35_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(106284992)))]; + tensor add_35_epsilon_0_to_fp16 = const()[name = tensor("add_35_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_35_cast = batch_norm(beta = add_35_beta_0_to_fp16, epsilon = add_35_epsilon_0_to_fp16, gamma = add_35_gamma_0_to_fp16, mean = add_27_mean_0_to_fp16, variance = add_27_variance_0_to_fp16, x = reshape_69_cast)[name = tensor("add_35_cast")]; + tensor var_4938 = const()[name = tensor("op_4938"), val = tensor([1, 1])]; + tensor var_4940 = const()[name = tensor("op_4940"), val = tensor([1, 1])]; + tensor hidden_states_97_pad_type_0 = const()[name = tensor("hidden_states_97_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_97_pad_0 = const()[name = tensor("hidden_states_97_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_1_proj_in_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(106287616))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(107516480))), name = tensor("down_blocks_2_attentions_1_proj_in_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor down_blocks_2_attentions_1_proj_in_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_proj_in_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(107516672)))]; + tensor hidden_states_97_cast = conv(bias = down_blocks_2_attentions_1_proj_in_bias_to_fp16, dilations = var_4940, groups = var_3883, pad = hidden_states_97_pad_0, pad_type = hidden_states_97_pad_type_0, strides = var_4938, weight = down_blocks_2_attentions_1_proj_in_weight_to_fp16_palettized, x = add_35_cast)[name = tensor("hidden_states_97_cast")]; + tensor var_4945 = const()[name = tensor("op_4945"), val = tensor([2, 1280, 1, 256])]; + tensor inputs_31_cast = reshape(shape = var_4945, x = hidden_states_97_cast)[name = tensor("inputs_31_cast")]; + tensor var_4955 = const()[name = tensor("op_4955"), val = tensor([1])]; + tensor channels_mean_31_cast = reduce_mean(axes = var_4955, keep_dims = var_3878, x = inputs_31_cast)[name = tensor("channels_mean_31_cast")]; + tensor zero_mean_31_cast = sub(x = inputs_31_cast, y = channels_mean_31_cast)[name = tensor("zero_mean_31_cast")]; + tensor zero_mean_sq_31_cast = mul(x = zero_mean_31_cast, y = zero_mean_31_cast)[name = tensor("zero_mean_sq_31_cast")]; + tensor var_4959 = const()[name = tensor("op_4959"), val = tensor([1])]; + tensor var_4960_cast = reduce_mean(axes = var_4959, keep_dims = var_3878, x = zero_mean_sq_31_cast)[name = tensor("op_4960_cast")]; + tensor var_4961_to_fp16 = const()[name = tensor("op_4961_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_4962_cast = add(x = var_4960_cast, y = var_4961_to_fp16)[name = tensor("op_4962_cast")]; + tensor denom_31_epsilon_0_to_fp16 = const()[name = tensor("denom_31_epsilon_0_to_fp16"), val = tensor(0x1p-24)]; + tensor denom_31_cast = rsqrt(epsilon = denom_31_epsilon_0_to_fp16, x = var_4962_cast)[name = tensor("denom_31_cast")]; + tensor out_31_cast = mul(x = zero_mean_31_cast, y = denom_31_cast)[name = tensor("out_31_cast")]; + tensor var_4966_to_fp16 = const()[name = tensor("op_4966_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(107519296)))]; + tensor var_4967_cast = add(x = out_31_cast, y = var_4966_to_fp16)[name = tensor("op_4967_cast")]; + tensor var_4969_to_fp16 = const()[name = tensor("op_4969_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(107521920)))]; + tensor hidden_states_99_cast = mul(x = var_4967_cast, y = var_4969_to_fp16)[name = tensor("hidden_states_99_cast")]; + tensor var_4976 = const()[name = tensor("op_4976"), val = tensor([1, 1])]; + tensor var_4978 = const()[name = tensor("op_4978"), val = tensor([1, 1])]; + tensor q_21_pad_type_0 = const()[name = tensor("q_21_pad_type_0"), val = tensor("custom")]; + tensor q_21_pad_0 = const()[name = tensor("q_21_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_1_transformer_blocks_0_attn1_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(107524544))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(108753408))), name = tensor("down_blocks_2_attentions_1_transformer_blocks_0_attn1_to_q_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor q_21_cast = conv(dilations = var_4978, groups = var_3883, pad = q_21_pad_0, pad_type = q_21_pad_type_0, strides = var_4976, weight = down_blocks_2_attentions_1_transformer_blocks_0_attn1_to_q_weight_to_fp16_palettized, x = hidden_states_99_cast)[name = tensor("q_21_cast")]; + tensor var_4982 = const()[name = tensor("op_4982"), val = tensor([1, 1])]; + tensor var_4984 = const()[name = tensor("op_4984"), val = tensor([1, 1])]; + tensor k_41_pad_type_0 = const()[name = tensor("k_41_pad_type_0"), val = tensor("custom")]; + tensor k_41_pad_0 = const()[name = tensor("k_41_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_1_transformer_blocks_0_attn1_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(108753600))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(109982464))), name = tensor("down_blocks_2_attentions_1_transformer_blocks_0_attn1_to_k_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor k_41_cast = conv(dilations = var_4984, groups = var_3883, pad = k_41_pad_0, pad_type = k_41_pad_type_0, strides = var_4982, weight = down_blocks_2_attentions_1_transformer_blocks_0_attn1_to_k_weight_to_fp16_palettized, x = hidden_states_99_cast)[name = tensor("k_41_cast")]; + tensor var_4988 = const()[name = tensor("op_4988"), val = tensor([1, 1])]; + tensor var_4990 = const()[name = tensor("op_4990"), val = tensor([1, 1])]; + tensor v_21_pad_type_0 = const()[name = tensor("v_21_pad_type_0"), val = tensor("custom")]; + tensor v_21_pad_0 = const()[name = tensor("v_21_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_1_transformer_blocks_0_attn1_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(109982656))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(111211520))), name = tensor("down_blocks_2_attentions_1_transformer_blocks_0_attn1_to_v_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor v_21_cast = conv(dilations = var_4990, groups = var_3883, pad = v_21_pad_0, pad_type = v_21_pad_type_0, strides = var_4988, weight = down_blocks_2_attentions_1_transformer_blocks_0_attn1_to_v_weight_to_fp16_palettized, x = hidden_states_99_cast)[name = tensor("v_21_cast")]; + tensor var_4994_begin_0 = const()[name = tensor("op_4994_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4994_end_0 = const()[name = tensor("op_4994_end_0"), val = tensor([2, 64, 1, 256])]; + tensor var_4994_end_mask_0 = const()[name = tensor("op_4994_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4994_cast = slice_by_index(begin = var_4994_begin_0, end = var_4994_end_0, end_mask = var_4994_end_mask_0, x = q_21_cast)[name = tensor("op_4994_cast")]; + tensor var_4998_begin_0 = const()[name = tensor("op_4998_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_4998_end_0 = const()[name = tensor("op_4998_end_0"), val = tensor([2, 128, 1, 256])]; + tensor var_4998_end_mask_0 = const()[name = tensor("op_4998_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4998_cast = slice_by_index(begin = var_4998_begin_0, end = var_4998_end_0, end_mask = var_4998_end_mask_0, x = q_21_cast)[name = tensor("op_4998_cast")]; + tensor var_5002_begin_0 = const()[name = tensor("op_5002_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_5002_end_0 = const()[name = tensor("op_5002_end_0"), val = tensor([2, 192, 1, 256])]; + tensor var_5002_end_mask_0 = const()[name = tensor("op_5002_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5002_cast = slice_by_index(begin = var_5002_begin_0, end = var_5002_end_0, end_mask = var_5002_end_mask_0, x = q_21_cast)[name = tensor("op_5002_cast")]; + tensor var_5006_begin_0 = const()[name = tensor("op_5006_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_5006_end_0 = const()[name = tensor("op_5006_end_0"), val = tensor([2, 256, 1, 256])]; + tensor var_5006_end_mask_0 = const()[name = tensor("op_5006_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5006_cast = slice_by_index(begin = var_5006_begin_0, end = var_5006_end_0, end_mask = var_5006_end_mask_0, x = q_21_cast)[name = tensor("op_5006_cast")]; + tensor var_5010_begin_0 = const()[name = tensor("op_5010_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_5010_end_0 = const()[name = tensor("op_5010_end_0"), val = tensor([2, 320, 1, 256])]; + tensor var_5010_end_mask_0 = const()[name = tensor("op_5010_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5010_cast = slice_by_index(begin = var_5010_begin_0, end = var_5010_end_0, end_mask = var_5010_end_mask_0, x = q_21_cast)[name = tensor("op_5010_cast")]; + tensor var_5014_begin_0 = const()[name = tensor("op_5014_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_5014_end_0 = const()[name = tensor("op_5014_end_0"), val = tensor([2, 384, 1, 256])]; + tensor var_5014_end_mask_0 = const()[name = tensor("op_5014_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5014_cast = slice_by_index(begin = var_5014_begin_0, end = var_5014_end_0, end_mask = var_5014_end_mask_0, x = q_21_cast)[name = tensor("op_5014_cast")]; + tensor var_5018_begin_0 = const()[name = tensor("op_5018_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_5018_end_0 = const()[name = tensor("op_5018_end_0"), val = tensor([2, 448, 1, 256])]; + tensor var_5018_end_mask_0 = const()[name = tensor("op_5018_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5018_cast = slice_by_index(begin = var_5018_begin_0, end = var_5018_end_0, end_mask = var_5018_end_mask_0, x = q_21_cast)[name = tensor("op_5018_cast")]; + tensor var_5022_begin_0 = const()[name = tensor("op_5022_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_5022_end_0 = const()[name = tensor("op_5022_end_0"), val = tensor([2, 512, 1, 256])]; + tensor var_5022_end_mask_0 = const()[name = tensor("op_5022_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5022_cast = slice_by_index(begin = var_5022_begin_0, end = var_5022_end_0, end_mask = var_5022_end_mask_0, x = q_21_cast)[name = tensor("op_5022_cast")]; + tensor var_5026_begin_0 = const()[name = tensor("op_5026_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_5026_end_0 = const()[name = tensor("op_5026_end_0"), val = tensor([2, 576, 1, 256])]; + tensor var_5026_end_mask_0 = const()[name = tensor("op_5026_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5026_cast = slice_by_index(begin = var_5026_begin_0, end = var_5026_end_0, end_mask = var_5026_end_mask_0, x = q_21_cast)[name = tensor("op_5026_cast")]; + tensor var_5030_begin_0 = const()[name = tensor("op_5030_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_5030_end_0 = const()[name = tensor("op_5030_end_0"), val = tensor([2, 640, 1, 256])]; + tensor var_5030_end_mask_0 = const()[name = tensor("op_5030_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5030_cast = slice_by_index(begin = var_5030_begin_0, end = var_5030_end_0, end_mask = var_5030_end_mask_0, x = q_21_cast)[name = tensor("op_5030_cast")]; + tensor var_5034_begin_0 = const()[name = tensor("op_5034_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_5034_end_0 = const()[name = tensor("op_5034_end_0"), val = tensor([2, 704, 1, 256])]; + tensor var_5034_end_mask_0 = const()[name = tensor("op_5034_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5034_cast = slice_by_index(begin = var_5034_begin_0, end = var_5034_end_0, end_mask = var_5034_end_mask_0, x = q_21_cast)[name = tensor("op_5034_cast")]; + tensor var_5038_begin_0 = const()[name = tensor("op_5038_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_5038_end_0 = const()[name = tensor("op_5038_end_0"), val = tensor([2, 768, 1, 256])]; + tensor var_5038_end_mask_0 = const()[name = tensor("op_5038_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5038_cast = slice_by_index(begin = var_5038_begin_0, end = var_5038_end_0, end_mask = var_5038_end_mask_0, x = q_21_cast)[name = tensor("op_5038_cast")]; + tensor var_5042_begin_0 = const()[name = tensor("op_5042_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_5042_end_0 = const()[name = tensor("op_5042_end_0"), val = tensor([2, 832, 1, 256])]; + tensor var_5042_end_mask_0 = const()[name = tensor("op_5042_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5042_cast = slice_by_index(begin = var_5042_begin_0, end = var_5042_end_0, end_mask = var_5042_end_mask_0, x = q_21_cast)[name = tensor("op_5042_cast")]; + tensor var_5046_begin_0 = const()[name = tensor("op_5046_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_5046_end_0 = const()[name = tensor("op_5046_end_0"), val = tensor([2, 896, 1, 256])]; + tensor var_5046_end_mask_0 = const()[name = tensor("op_5046_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5046_cast = slice_by_index(begin = var_5046_begin_0, end = var_5046_end_0, end_mask = var_5046_end_mask_0, x = q_21_cast)[name = tensor("op_5046_cast")]; + tensor var_5050_begin_0 = const()[name = tensor("op_5050_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_5050_end_0 = const()[name = tensor("op_5050_end_0"), val = tensor([2, 960, 1, 256])]; + tensor var_5050_end_mask_0 = const()[name = tensor("op_5050_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5050_cast = slice_by_index(begin = var_5050_begin_0, end = var_5050_end_0, end_mask = var_5050_end_mask_0, x = q_21_cast)[name = tensor("op_5050_cast")]; + tensor var_5054_begin_0 = const()[name = tensor("op_5054_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_5054_end_0 = const()[name = tensor("op_5054_end_0"), val = tensor([2, 1024, 1, 256])]; + tensor var_5054_end_mask_0 = const()[name = tensor("op_5054_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5054_cast = slice_by_index(begin = var_5054_begin_0, end = var_5054_end_0, end_mask = var_5054_end_mask_0, x = q_21_cast)[name = tensor("op_5054_cast")]; + tensor var_5058_begin_0 = const()[name = tensor("op_5058_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_5058_end_0 = const()[name = tensor("op_5058_end_0"), val = tensor([2, 1088, 1, 256])]; + tensor var_5058_end_mask_0 = const()[name = tensor("op_5058_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5058_cast = slice_by_index(begin = var_5058_begin_0, end = var_5058_end_0, end_mask = var_5058_end_mask_0, x = q_21_cast)[name = tensor("op_5058_cast")]; + tensor var_5062_begin_0 = const()[name = tensor("op_5062_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_5062_end_0 = const()[name = tensor("op_5062_end_0"), val = tensor([2, 1152, 1, 256])]; + tensor var_5062_end_mask_0 = const()[name = tensor("op_5062_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5062_cast = slice_by_index(begin = var_5062_begin_0, end = var_5062_end_0, end_mask = var_5062_end_mask_0, x = q_21_cast)[name = tensor("op_5062_cast")]; + tensor var_5066_begin_0 = const()[name = tensor("op_5066_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_5066_end_0 = const()[name = tensor("op_5066_end_0"), val = tensor([2, 1216, 1, 256])]; + tensor var_5066_end_mask_0 = const()[name = tensor("op_5066_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5066_cast = slice_by_index(begin = var_5066_begin_0, end = var_5066_end_0, end_mask = var_5066_end_mask_0, x = q_21_cast)[name = tensor("op_5066_cast")]; + tensor var_5070_begin_0 = const()[name = tensor("op_5070_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_5070_end_0 = const()[name = tensor("op_5070_end_0"), val = tensor([2, 1280, 1, 256])]; + tensor var_5070_end_mask_0 = const()[name = tensor("op_5070_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5070_cast = slice_by_index(begin = var_5070_begin_0, end = var_5070_end_0, end_mask = var_5070_end_mask_0, x = q_21_cast)[name = tensor("op_5070_cast")]; + tensor k_43_perm_0 = const()[name = tensor("k_43_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_5077_begin_0 = const()[name = tensor("op_5077_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5077_end_0 = const()[name = tensor("op_5077_end_0"), val = tensor([2, 256, 1, 64])]; + tensor var_5077_end_mask_0 = const()[name = tensor("op_5077_end_mask_0"), val = tensor([true, true, true, false])]; + tensor transpose_21 = transpose(perm = k_43_perm_0, x = k_41_cast)[name = tensor("transpose_21")]; + tensor var_5077_cast = slice_by_index(begin = var_5077_begin_0, end = var_5077_end_0, end_mask = var_5077_end_mask_0, x = transpose_21)[name = tensor("op_5077_cast")]; + tensor var_5081_begin_0 = const()[name = tensor("op_5081_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_5081_end_0 = const()[name = tensor("op_5081_end_0"), val = tensor([2, 256, 1, 128])]; + tensor var_5081_end_mask_0 = const()[name = tensor("op_5081_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5081_cast = slice_by_index(begin = var_5081_begin_0, end = var_5081_end_0, end_mask = var_5081_end_mask_0, x = transpose_21)[name = tensor("op_5081_cast")]; + tensor var_5085_begin_0 = const()[name = tensor("op_5085_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_5085_end_0 = const()[name = tensor("op_5085_end_0"), val = tensor([2, 256, 1, 192])]; + tensor var_5085_end_mask_0 = const()[name = tensor("op_5085_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5085_cast = slice_by_index(begin = var_5085_begin_0, end = var_5085_end_0, end_mask = var_5085_end_mask_0, x = transpose_21)[name = tensor("op_5085_cast")]; + tensor var_5089_begin_0 = const()[name = tensor("op_5089_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_5089_end_0 = const()[name = tensor("op_5089_end_0"), val = tensor([2, 256, 1, 256])]; + tensor var_5089_end_mask_0 = const()[name = tensor("op_5089_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5089_cast = slice_by_index(begin = var_5089_begin_0, end = var_5089_end_0, end_mask = var_5089_end_mask_0, x = transpose_21)[name = tensor("op_5089_cast")]; + tensor var_5093_begin_0 = const()[name = tensor("op_5093_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_5093_end_0 = const()[name = tensor("op_5093_end_0"), val = tensor([2, 256, 1, 320])]; + tensor var_5093_end_mask_0 = const()[name = tensor("op_5093_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5093_cast = slice_by_index(begin = var_5093_begin_0, end = var_5093_end_0, end_mask = var_5093_end_mask_0, x = transpose_21)[name = tensor("op_5093_cast")]; + tensor var_5097_begin_0 = const()[name = tensor("op_5097_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_5097_end_0 = const()[name = tensor("op_5097_end_0"), val = tensor([2, 256, 1, 384])]; + tensor var_5097_end_mask_0 = const()[name = tensor("op_5097_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5097_cast = slice_by_index(begin = var_5097_begin_0, end = var_5097_end_0, end_mask = var_5097_end_mask_0, x = transpose_21)[name = tensor("op_5097_cast")]; + tensor var_5101_begin_0 = const()[name = tensor("op_5101_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_5101_end_0 = const()[name = tensor("op_5101_end_0"), val = tensor([2, 256, 1, 448])]; + tensor var_5101_end_mask_0 = const()[name = tensor("op_5101_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5101_cast = slice_by_index(begin = var_5101_begin_0, end = var_5101_end_0, end_mask = var_5101_end_mask_0, x = transpose_21)[name = tensor("op_5101_cast")]; + tensor var_5105_begin_0 = const()[name = tensor("op_5105_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_5105_end_0 = const()[name = tensor("op_5105_end_0"), val = tensor([2, 256, 1, 512])]; + tensor var_5105_end_mask_0 = const()[name = tensor("op_5105_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5105_cast = slice_by_index(begin = var_5105_begin_0, end = var_5105_end_0, end_mask = var_5105_end_mask_0, x = transpose_21)[name = tensor("op_5105_cast")]; + tensor var_5109_begin_0 = const()[name = tensor("op_5109_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_5109_end_0 = const()[name = tensor("op_5109_end_0"), val = tensor([2, 256, 1, 576])]; + tensor var_5109_end_mask_0 = const()[name = tensor("op_5109_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5109_cast = slice_by_index(begin = var_5109_begin_0, end = var_5109_end_0, end_mask = var_5109_end_mask_0, x = transpose_21)[name = tensor("op_5109_cast")]; + tensor var_5113_begin_0 = const()[name = tensor("op_5113_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_5113_end_0 = const()[name = tensor("op_5113_end_0"), val = tensor([2, 256, 1, 640])]; + tensor var_5113_end_mask_0 = const()[name = tensor("op_5113_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5113_cast = slice_by_index(begin = var_5113_begin_0, end = var_5113_end_0, end_mask = var_5113_end_mask_0, x = transpose_21)[name = tensor("op_5113_cast")]; + tensor var_5117_begin_0 = const()[name = tensor("op_5117_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_5117_end_0 = const()[name = tensor("op_5117_end_0"), val = tensor([2, 256, 1, 704])]; + tensor var_5117_end_mask_0 = const()[name = tensor("op_5117_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5117_cast = slice_by_index(begin = var_5117_begin_0, end = var_5117_end_0, end_mask = var_5117_end_mask_0, x = transpose_21)[name = tensor("op_5117_cast")]; + tensor var_5121_begin_0 = const()[name = tensor("op_5121_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_5121_end_0 = const()[name = tensor("op_5121_end_0"), val = tensor([2, 256, 1, 768])]; + tensor var_5121_end_mask_0 = const()[name = tensor("op_5121_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5121_cast = slice_by_index(begin = var_5121_begin_0, end = var_5121_end_0, end_mask = var_5121_end_mask_0, x = transpose_21)[name = tensor("op_5121_cast")]; + tensor var_5125_begin_0 = const()[name = tensor("op_5125_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_5125_end_0 = const()[name = tensor("op_5125_end_0"), val = tensor([2, 256, 1, 832])]; + tensor var_5125_end_mask_0 = const()[name = tensor("op_5125_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5125_cast = slice_by_index(begin = var_5125_begin_0, end = var_5125_end_0, end_mask = var_5125_end_mask_0, x = transpose_21)[name = tensor("op_5125_cast")]; + tensor var_5129_begin_0 = const()[name = tensor("op_5129_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_5129_end_0 = const()[name = tensor("op_5129_end_0"), val = tensor([2, 256, 1, 896])]; + tensor var_5129_end_mask_0 = const()[name = tensor("op_5129_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5129_cast = slice_by_index(begin = var_5129_begin_0, end = var_5129_end_0, end_mask = var_5129_end_mask_0, x = transpose_21)[name = tensor("op_5129_cast")]; + tensor var_5133_begin_0 = const()[name = tensor("op_5133_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_5133_end_0 = const()[name = tensor("op_5133_end_0"), val = tensor([2, 256, 1, 960])]; + tensor var_5133_end_mask_0 = const()[name = tensor("op_5133_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5133_cast = slice_by_index(begin = var_5133_begin_0, end = var_5133_end_0, end_mask = var_5133_end_mask_0, x = transpose_21)[name = tensor("op_5133_cast")]; + tensor var_5137_begin_0 = const()[name = tensor("op_5137_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_5137_end_0 = const()[name = tensor("op_5137_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_5137_end_mask_0 = const()[name = tensor("op_5137_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5137_cast = slice_by_index(begin = var_5137_begin_0, end = var_5137_end_0, end_mask = var_5137_end_mask_0, x = transpose_21)[name = tensor("op_5137_cast")]; + tensor var_5141_begin_0 = const()[name = tensor("op_5141_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_5141_end_0 = const()[name = tensor("op_5141_end_0"), val = tensor([2, 256, 1, 1088])]; + tensor var_5141_end_mask_0 = const()[name = tensor("op_5141_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5141_cast = slice_by_index(begin = var_5141_begin_0, end = var_5141_end_0, end_mask = var_5141_end_mask_0, x = transpose_21)[name = tensor("op_5141_cast")]; + tensor var_5145_begin_0 = const()[name = tensor("op_5145_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_5145_end_0 = const()[name = tensor("op_5145_end_0"), val = tensor([2, 256, 1, 1152])]; + tensor var_5145_end_mask_0 = const()[name = tensor("op_5145_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5145_cast = slice_by_index(begin = var_5145_begin_0, end = var_5145_end_0, end_mask = var_5145_end_mask_0, x = transpose_21)[name = tensor("op_5145_cast")]; + tensor var_5149_begin_0 = const()[name = tensor("op_5149_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_5149_end_0 = const()[name = tensor("op_5149_end_0"), val = tensor([2, 256, 1, 1216])]; + tensor var_5149_end_mask_0 = const()[name = tensor("op_5149_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5149_cast = slice_by_index(begin = var_5149_begin_0, end = var_5149_end_0, end_mask = var_5149_end_mask_0, x = transpose_21)[name = tensor("op_5149_cast")]; + tensor var_5153_begin_0 = const()[name = tensor("op_5153_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_5153_end_0 = const()[name = tensor("op_5153_end_0"), val = tensor([2, 256, 1, 1280])]; + tensor var_5153_end_mask_0 = const()[name = tensor("op_5153_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5153_cast = slice_by_index(begin = var_5153_begin_0, end = var_5153_end_0, end_mask = var_5153_end_mask_0, x = transpose_21)[name = tensor("op_5153_cast")]; + tensor var_5155_begin_0 = const()[name = tensor("op_5155_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5155_end_0 = const()[name = tensor("op_5155_end_0"), val = tensor([2, 64, 1, 256])]; + tensor var_5155_end_mask_0 = const()[name = tensor("op_5155_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5155_cast = slice_by_index(begin = var_5155_begin_0, end = var_5155_end_0, end_mask = var_5155_end_mask_0, x = v_21_cast)[name = tensor("op_5155_cast")]; + tensor var_5159_begin_0 = const()[name = tensor("op_5159_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_5159_end_0 = const()[name = tensor("op_5159_end_0"), val = tensor([2, 128, 1, 256])]; + tensor var_5159_end_mask_0 = const()[name = tensor("op_5159_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5159_cast = slice_by_index(begin = var_5159_begin_0, end = var_5159_end_0, end_mask = var_5159_end_mask_0, x = v_21_cast)[name = tensor("op_5159_cast")]; + tensor var_5163_begin_0 = const()[name = tensor("op_5163_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_5163_end_0 = const()[name = tensor("op_5163_end_0"), val = tensor([2, 192, 1, 256])]; + tensor var_5163_end_mask_0 = const()[name = tensor("op_5163_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5163_cast = slice_by_index(begin = var_5163_begin_0, end = var_5163_end_0, end_mask = var_5163_end_mask_0, x = v_21_cast)[name = tensor("op_5163_cast")]; + tensor var_5167_begin_0 = const()[name = tensor("op_5167_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_5167_end_0 = const()[name = tensor("op_5167_end_0"), val = tensor([2, 256, 1, 256])]; + tensor var_5167_end_mask_0 = const()[name = tensor("op_5167_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5167_cast = slice_by_index(begin = var_5167_begin_0, end = var_5167_end_0, end_mask = var_5167_end_mask_0, x = v_21_cast)[name = tensor("op_5167_cast")]; + tensor var_5171_begin_0 = const()[name = tensor("op_5171_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_5171_end_0 = const()[name = tensor("op_5171_end_0"), val = tensor([2, 320, 1, 256])]; + tensor var_5171_end_mask_0 = const()[name = tensor("op_5171_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5171_cast = slice_by_index(begin = var_5171_begin_0, end = var_5171_end_0, end_mask = var_5171_end_mask_0, x = v_21_cast)[name = tensor("op_5171_cast")]; + tensor var_5175_begin_0 = const()[name = tensor("op_5175_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_5175_end_0 = const()[name = tensor("op_5175_end_0"), val = tensor([2, 384, 1, 256])]; + tensor var_5175_end_mask_0 = const()[name = tensor("op_5175_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5175_cast = slice_by_index(begin = var_5175_begin_0, end = var_5175_end_0, end_mask = var_5175_end_mask_0, x = v_21_cast)[name = tensor("op_5175_cast")]; + tensor var_5179_begin_0 = const()[name = tensor("op_5179_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_5179_end_0 = const()[name = tensor("op_5179_end_0"), val = tensor([2, 448, 1, 256])]; + tensor var_5179_end_mask_0 = const()[name = tensor("op_5179_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5179_cast = slice_by_index(begin = var_5179_begin_0, end = var_5179_end_0, end_mask = var_5179_end_mask_0, x = v_21_cast)[name = tensor("op_5179_cast")]; + tensor var_5183_begin_0 = const()[name = tensor("op_5183_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_5183_end_0 = const()[name = tensor("op_5183_end_0"), val = tensor([2, 512, 1, 256])]; + tensor var_5183_end_mask_0 = const()[name = tensor("op_5183_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5183_cast = slice_by_index(begin = var_5183_begin_0, end = var_5183_end_0, end_mask = var_5183_end_mask_0, x = v_21_cast)[name = tensor("op_5183_cast")]; + tensor var_5187_begin_0 = const()[name = tensor("op_5187_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_5187_end_0 = const()[name = tensor("op_5187_end_0"), val = tensor([2, 576, 1, 256])]; + tensor var_5187_end_mask_0 = const()[name = tensor("op_5187_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5187_cast = slice_by_index(begin = var_5187_begin_0, end = var_5187_end_0, end_mask = var_5187_end_mask_0, x = v_21_cast)[name = tensor("op_5187_cast")]; + tensor var_5191_begin_0 = const()[name = tensor("op_5191_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_5191_end_0 = const()[name = tensor("op_5191_end_0"), val = tensor([2, 640, 1, 256])]; + tensor var_5191_end_mask_0 = const()[name = tensor("op_5191_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5191_cast = slice_by_index(begin = var_5191_begin_0, end = var_5191_end_0, end_mask = var_5191_end_mask_0, x = v_21_cast)[name = tensor("op_5191_cast")]; + tensor var_5195_begin_0 = const()[name = tensor("op_5195_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_5195_end_0 = const()[name = tensor("op_5195_end_0"), val = tensor([2, 704, 1, 256])]; + tensor var_5195_end_mask_0 = const()[name = tensor("op_5195_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5195_cast = slice_by_index(begin = var_5195_begin_0, end = var_5195_end_0, end_mask = var_5195_end_mask_0, x = v_21_cast)[name = tensor("op_5195_cast")]; + tensor var_5199_begin_0 = const()[name = tensor("op_5199_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_5199_end_0 = const()[name = tensor("op_5199_end_0"), val = tensor([2, 768, 1, 256])]; + tensor var_5199_end_mask_0 = const()[name = tensor("op_5199_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5199_cast = slice_by_index(begin = var_5199_begin_0, end = var_5199_end_0, end_mask = var_5199_end_mask_0, x = v_21_cast)[name = tensor("op_5199_cast")]; + tensor var_5203_begin_0 = const()[name = tensor("op_5203_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_5203_end_0 = const()[name = tensor("op_5203_end_0"), val = tensor([2, 832, 1, 256])]; + tensor var_5203_end_mask_0 = const()[name = tensor("op_5203_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5203_cast = slice_by_index(begin = var_5203_begin_0, end = var_5203_end_0, end_mask = var_5203_end_mask_0, x = v_21_cast)[name = tensor("op_5203_cast")]; + tensor var_5207_begin_0 = const()[name = tensor("op_5207_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_5207_end_0 = const()[name = tensor("op_5207_end_0"), val = tensor([2, 896, 1, 256])]; + tensor var_5207_end_mask_0 = const()[name = tensor("op_5207_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5207_cast = slice_by_index(begin = var_5207_begin_0, end = var_5207_end_0, end_mask = var_5207_end_mask_0, x = v_21_cast)[name = tensor("op_5207_cast")]; + tensor var_5211_begin_0 = const()[name = tensor("op_5211_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_5211_end_0 = const()[name = tensor("op_5211_end_0"), val = tensor([2, 960, 1, 256])]; + tensor var_5211_end_mask_0 = const()[name = tensor("op_5211_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5211_cast = slice_by_index(begin = var_5211_begin_0, end = var_5211_end_0, end_mask = var_5211_end_mask_0, x = v_21_cast)[name = tensor("op_5211_cast")]; + tensor var_5215_begin_0 = const()[name = tensor("op_5215_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_5215_end_0 = const()[name = tensor("op_5215_end_0"), val = tensor([2, 1024, 1, 256])]; + tensor var_5215_end_mask_0 = const()[name = tensor("op_5215_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5215_cast = slice_by_index(begin = var_5215_begin_0, end = var_5215_end_0, end_mask = var_5215_end_mask_0, x = v_21_cast)[name = tensor("op_5215_cast")]; + tensor var_5219_begin_0 = const()[name = tensor("op_5219_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_5219_end_0 = const()[name = tensor("op_5219_end_0"), val = tensor([2, 1088, 1, 256])]; + tensor var_5219_end_mask_0 = const()[name = tensor("op_5219_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5219_cast = slice_by_index(begin = var_5219_begin_0, end = var_5219_end_0, end_mask = var_5219_end_mask_0, x = v_21_cast)[name = tensor("op_5219_cast")]; + tensor var_5223_begin_0 = const()[name = tensor("op_5223_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_5223_end_0 = const()[name = tensor("op_5223_end_0"), val = tensor([2, 1152, 1, 256])]; + tensor var_5223_end_mask_0 = const()[name = tensor("op_5223_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5223_cast = slice_by_index(begin = var_5223_begin_0, end = var_5223_end_0, end_mask = var_5223_end_mask_0, x = v_21_cast)[name = tensor("op_5223_cast")]; + tensor var_5227_begin_0 = const()[name = tensor("op_5227_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_5227_end_0 = const()[name = tensor("op_5227_end_0"), val = tensor([2, 1216, 1, 256])]; + tensor var_5227_end_mask_0 = const()[name = tensor("op_5227_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5227_cast = slice_by_index(begin = var_5227_begin_0, end = var_5227_end_0, end_mask = var_5227_end_mask_0, x = v_21_cast)[name = tensor("op_5227_cast")]; + tensor var_5231_begin_0 = const()[name = tensor("op_5231_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_5231_end_0 = const()[name = tensor("op_5231_end_0"), val = tensor([2, 1280, 1, 256])]; + tensor var_5231_end_mask_0 = const()[name = tensor("op_5231_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5231_cast = slice_by_index(begin = var_5231_begin_0, end = var_5231_end_0, end_mask = var_5231_end_mask_0, x = v_21_cast)[name = tensor("op_5231_cast")]; + tensor var_5235_equation_0 = const()[name = tensor("op_5235_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5235_cast = einsum(equation = var_5235_equation_0, values = (var_5077_cast, var_4994_cast))[name = tensor("op_5235_cast")]; + tensor var_5236_to_fp16 = const()[name = tensor("op_5236_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_81_cast = mul(x = var_5235_cast, y = var_5236_to_fp16)[name = tensor("aw_81_cast")]; + tensor var_5239_equation_0 = const()[name = tensor("op_5239_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5239_cast = einsum(equation = var_5239_equation_0, values = (var_5081_cast, var_4998_cast))[name = tensor("op_5239_cast")]; + tensor var_5240_to_fp16 = const()[name = tensor("op_5240_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_83_cast = mul(x = var_5239_cast, y = var_5240_to_fp16)[name = tensor("aw_83_cast")]; + tensor var_5243_equation_0 = const()[name = tensor("op_5243_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5243_cast = einsum(equation = var_5243_equation_0, values = (var_5085_cast, var_5002_cast))[name = tensor("op_5243_cast")]; + tensor var_5244_to_fp16 = const()[name = tensor("op_5244_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_85_cast = mul(x = var_5243_cast, y = var_5244_to_fp16)[name = tensor("aw_85_cast")]; + tensor var_5247_equation_0 = const()[name = tensor("op_5247_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5247_cast = einsum(equation = var_5247_equation_0, values = (var_5089_cast, var_5006_cast))[name = tensor("op_5247_cast")]; + tensor var_5248_to_fp16 = const()[name = tensor("op_5248_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_87_cast = mul(x = var_5247_cast, y = var_5248_to_fp16)[name = tensor("aw_87_cast")]; + tensor var_5251_equation_0 = const()[name = tensor("op_5251_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5251_cast = einsum(equation = var_5251_equation_0, values = (var_5093_cast, var_5010_cast))[name = tensor("op_5251_cast")]; + tensor var_5252_to_fp16 = const()[name = tensor("op_5252_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_89_cast = mul(x = var_5251_cast, y = var_5252_to_fp16)[name = tensor("aw_89_cast")]; + tensor var_5255_equation_0 = const()[name = tensor("op_5255_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5255_cast = einsum(equation = var_5255_equation_0, values = (var_5097_cast, var_5014_cast))[name = tensor("op_5255_cast")]; + tensor var_5256_to_fp16 = const()[name = tensor("op_5256_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_91_cast = mul(x = var_5255_cast, y = var_5256_to_fp16)[name = tensor("aw_91_cast")]; + tensor var_5259_equation_0 = const()[name = tensor("op_5259_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5259_cast = einsum(equation = var_5259_equation_0, values = (var_5101_cast, var_5018_cast))[name = tensor("op_5259_cast")]; + tensor var_5260_to_fp16 = const()[name = tensor("op_5260_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_93_cast = mul(x = var_5259_cast, y = var_5260_to_fp16)[name = tensor("aw_93_cast")]; + tensor var_5263_equation_0 = const()[name = tensor("op_5263_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5263_cast = einsum(equation = var_5263_equation_0, values = (var_5105_cast, var_5022_cast))[name = tensor("op_5263_cast")]; + tensor var_5264_to_fp16 = const()[name = tensor("op_5264_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_95_cast = mul(x = var_5263_cast, y = var_5264_to_fp16)[name = tensor("aw_95_cast")]; + tensor var_5267_equation_0 = const()[name = tensor("op_5267_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5267_cast = einsum(equation = var_5267_equation_0, values = (var_5109_cast, var_5026_cast))[name = tensor("op_5267_cast")]; + tensor var_5268_to_fp16 = const()[name = tensor("op_5268_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_97_cast = mul(x = var_5267_cast, y = var_5268_to_fp16)[name = tensor("aw_97_cast")]; + tensor var_5271_equation_0 = const()[name = tensor("op_5271_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5271_cast = einsum(equation = var_5271_equation_0, values = (var_5113_cast, var_5030_cast))[name = tensor("op_5271_cast")]; + tensor var_5272_to_fp16 = const()[name = tensor("op_5272_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_99_cast = mul(x = var_5271_cast, y = var_5272_to_fp16)[name = tensor("aw_99_cast")]; + tensor var_5275_equation_0 = const()[name = tensor("op_5275_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5275_cast = einsum(equation = var_5275_equation_0, values = (var_5117_cast, var_5034_cast))[name = tensor("op_5275_cast")]; + tensor var_5276_to_fp16 = const()[name = tensor("op_5276_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_101_cast = mul(x = var_5275_cast, y = var_5276_to_fp16)[name = tensor("aw_101_cast")]; + tensor var_5279_equation_0 = const()[name = tensor("op_5279_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5279_cast = einsum(equation = var_5279_equation_0, values = (var_5121_cast, var_5038_cast))[name = tensor("op_5279_cast")]; + tensor var_5280_to_fp16 = const()[name = tensor("op_5280_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_103_cast = mul(x = var_5279_cast, y = var_5280_to_fp16)[name = tensor("aw_103_cast")]; + tensor var_5283_equation_0 = const()[name = tensor("op_5283_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5283_cast = einsum(equation = var_5283_equation_0, values = (var_5125_cast, var_5042_cast))[name = tensor("op_5283_cast")]; + tensor var_5284_to_fp16 = const()[name = tensor("op_5284_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_105_cast = mul(x = var_5283_cast, y = var_5284_to_fp16)[name = tensor("aw_105_cast")]; + tensor var_5287_equation_0 = const()[name = tensor("op_5287_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5287_cast = einsum(equation = var_5287_equation_0, values = (var_5129_cast, var_5046_cast))[name = tensor("op_5287_cast")]; + tensor var_5288_to_fp16 = const()[name = tensor("op_5288_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_107_cast = mul(x = var_5287_cast, y = var_5288_to_fp16)[name = tensor("aw_107_cast")]; + tensor var_5291_equation_0 = const()[name = tensor("op_5291_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5291_cast = einsum(equation = var_5291_equation_0, values = (var_5133_cast, var_5050_cast))[name = tensor("op_5291_cast")]; + tensor var_5292_to_fp16 = const()[name = tensor("op_5292_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_109_cast = mul(x = var_5291_cast, y = var_5292_to_fp16)[name = tensor("aw_109_cast")]; + tensor var_5295_equation_0 = const()[name = tensor("op_5295_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5295_cast = einsum(equation = var_5295_equation_0, values = (var_5137_cast, var_5054_cast))[name = tensor("op_5295_cast")]; + tensor var_5296_to_fp16 = const()[name = tensor("op_5296_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_111_cast = mul(x = var_5295_cast, y = var_5296_to_fp16)[name = tensor("aw_111_cast")]; + tensor var_5299_equation_0 = const()[name = tensor("op_5299_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5299_cast = einsum(equation = var_5299_equation_0, values = (var_5141_cast, var_5058_cast))[name = tensor("op_5299_cast")]; + tensor var_5300_to_fp16 = const()[name = tensor("op_5300_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_113_cast = mul(x = var_5299_cast, y = var_5300_to_fp16)[name = tensor("aw_113_cast")]; + tensor var_5303_equation_0 = const()[name = tensor("op_5303_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5303_cast = einsum(equation = var_5303_equation_0, values = (var_5145_cast, var_5062_cast))[name = tensor("op_5303_cast")]; + tensor var_5304_to_fp16 = const()[name = tensor("op_5304_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_115_cast = mul(x = var_5303_cast, y = var_5304_to_fp16)[name = tensor("aw_115_cast")]; + tensor var_5307_equation_0 = const()[name = tensor("op_5307_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5307_cast = einsum(equation = var_5307_equation_0, values = (var_5149_cast, var_5066_cast))[name = tensor("op_5307_cast")]; + tensor var_5308_to_fp16 = const()[name = tensor("op_5308_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_117_cast = mul(x = var_5307_cast, y = var_5308_to_fp16)[name = tensor("aw_117_cast")]; + tensor var_5311_equation_0 = const()[name = tensor("op_5311_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5311_cast = einsum(equation = var_5311_equation_0, values = (var_5153_cast, var_5070_cast))[name = tensor("op_5311_cast")]; + tensor var_5312_to_fp16 = const()[name = tensor("op_5312_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_119_cast = mul(x = var_5311_cast, y = var_5312_to_fp16)[name = tensor("aw_119_cast")]; + tensor var_5314_cast = softmax(axis = var_3883, x = aw_81_cast)[name = tensor("op_5314_cast")]; + tensor var_5315_cast = softmax(axis = var_3883, x = aw_83_cast)[name = tensor("op_5315_cast")]; + tensor var_5316_cast = softmax(axis = var_3883, x = aw_85_cast)[name = tensor("op_5316_cast")]; + tensor var_5317_cast = softmax(axis = var_3883, x = aw_87_cast)[name = tensor("op_5317_cast")]; + tensor var_5318_cast = softmax(axis = var_3883, x = aw_89_cast)[name = tensor("op_5318_cast")]; + tensor var_5319_cast = softmax(axis = var_3883, x = aw_91_cast)[name = tensor("op_5319_cast")]; + tensor var_5320_cast = softmax(axis = var_3883, x = aw_93_cast)[name = tensor("op_5320_cast")]; + tensor var_5321_cast = softmax(axis = var_3883, x = aw_95_cast)[name = tensor("op_5321_cast")]; + tensor var_5322_cast = softmax(axis = var_3883, x = aw_97_cast)[name = tensor("op_5322_cast")]; + tensor var_5323_cast = softmax(axis = var_3883, x = aw_99_cast)[name = tensor("op_5323_cast")]; + tensor var_5324_cast = softmax(axis = var_3883, x = aw_101_cast)[name = tensor("op_5324_cast")]; + tensor var_5325_cast = softmax(axis = var_3883, x = aw_103_cast)[name = tensor("op_5325_cast")]; + tensor var_5326_cast = softmax(axis = var_3883, x = aw_105_cast)[name = tensor("op_5326_cast")]; + tensor var_5327_cast = softmax(axis = var_3883, x = aw_107_cast)[name = tensor("op_5327_cast")]; + tensor var_5328_cast = softmax(axis = var_3883, x = aw_109_cast)[name = tensor("op_5328_cast")]; + tensor var_5329_cast = softmax(axis = var_3883, x = aw_111_cast)[name = tensor("op_5329_cast")]; + tensor var_5330_cast = softmax(axis = var_3883, x = aw_113_cast)[name = tensor("op_5330_cast")]; + tensor var_5331_cast = softmax(axis = var_3883, x = aw_115_cast)[name = tensor("op_5331_cast")]; + tensor var_5332_cast = softmax(axis = var_3883, x = aw_117_cast)[name = tensor("op_5332_cast")]; + tensor var_5333_cast = softmax(axis = var_3883, x = aw_119_cast)[name = tensor("op_5333_cast")]; + tensor var_5335_equation_0 = const()[name = tensor("op_5335_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_5335_cast = einsum(equation = var_5335_equation_0, values = (var_5155_cast, var_5314_cast))[name = tensor("op_5335_cast")]; + tensor var_5337_equation_0 = const()[name = tensor("op_5337_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_5337_cast = einsum(equation = var_5337_equation_0, values = (var_5159_cast, var_5315_cast))[name = tensor("op_5337_cast")]; + tensor var_5339_equation_0 = const()[name = tensor("op_5339_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_5339_cast = einsum(equation = var_5339_equation_0, values = (var_5163_cast, var_5316_cast))[name = tensor("op_5339_cast")]; + tensor var_5341_equation_0 = const()[name = tensor("op_5341_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_5341_cast = einsum(equation = var_5341_equation_0, values = (var_5167_cast, var_5317_cast))[name = tensor("op_5341_cast")]; + tensor var_5343_equation_0 = const()[name = tensor("op_5343_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_5343_cast = einsum(equation = var_5343_equation_0, values = (var_5171_cast, var_5318_cast))[name = tensor("op_5343_cast")]; + tensor var_5345_equation_0 = const()[name = tensor("op_5345_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_5345_cast = einsum(equation = var_5345_equation_0, values = (var_5175_cast, var_5319_cast))[name = tensor("op_5345_cast")]; + tensor var_5347_equation_0 = const()[name = tensor("op_5347_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_5347_cast = einsum(equation = var_5347_equation_0, values = (var_5179_cast, var_5320_cast))[name = tensor("op_5347_cast")]; + tensor var_5349_equation_0 = const()[name = tensor("op_5349_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_5349_cast = einsum(equation = var_5349_equation_0, values = (var_5183_cast, var_5321_cast))[name = tensor("op_5349_cast")]; + tensor var_5351_equation_0 = const()[name = tensor("op_5351_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_5351_cast = einsum(equation = var_5351_equation_0, values = (var_5187_cast, var_5322_cast))[name = tensor("op_5351_cast")]; + tensor var_5353_equation_0 = const()[name = tensor("op_5353_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_5353_cast = einsum(equation = var_5353_equation_0, values = (var_5191_cast, var_5323_cast))[name = tensor("op_5353_cast")]; + tensor var_5355_equation_0 = const()[name = tensor("op_5355_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_5355_cast = einsum(equation = var_5355_equation_0, values = (var_5195_cast, var_5324_cast))[name = tensor("op_5355_cast")]; + tensor var_5357_equation_0 = const()[name = tensor("op_5357_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_5357_cast = einsum(equation = var_5357_equation_0, values = (var_5199_cast, var_5325_cast))[name = tensor("op_5357_cast")]; + tensor var_5359_equation_0 = const()[name = tensor("op_5359_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_5359_cast = einsum(equation = var_5359_equation_0, values = (var_5203_cast, var_5326_cast))[name = tensor("op_5359_cast")]; + tensor var_5361_equation_0 = const()[name = tensor("op_5361_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_5361_cast = einsum(equation = var_5361_equation_0, values = (var_5207_cast, var_5327_cast))[name = tensor("op_5361_cast")]; + tensor var_5363_equation_0 = const()[name = tensor("op_5363_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_5363_cast = einsum(equation = var_5363_equation_0, values = (var_5211_cast, var_5328_cast))[name = tensor("op_5363_cast")]; + tensor var_5365_equation_0 = const()[name = tensor("op_5365_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_5365_cast = einsum(equation = var_5365_equation_0, values = (var_5215_cast, var_5329_cast))[name = tensor("op_5365_cast")]; + tensor var_5367_equation_0 = const()[name = tensor("op_5367_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_5367_cast = einsum(equation = var_5367_equation_0, values = (var_5219_cast, var_5330_cast))[name = tensor("op_5367_cast")]; + tensor var_5369_equation_0 = const()[name = tensor("op_5369_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_5369_cast = einsum(equation = var_5369_equation_0, values = (var_5223_cast, var_5331_cast))[name = tensor("op_5369_cast")]; + tensor var_5371_equation_0 = const()[name = tensor("op_5371_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_5371_cast = einsum(equation = var_5371_equation_0, values = (var_5227_cast, var_5332_cast))[name = tensor("op_5371_cast")]; + tensor var_5373_equation_0 = const()[name = tensor("op_5373_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_5373_cast = einsum(equation = var_5373_equation_0, values = (var_5231_cast, var_5333_cast))[name = tensor("op_5373_cast")]; + tensor input_159_interleave_0 = const()[name = tensor("input_159_interleave_0"), val = tensor(false)]; + tensor input_159_cast = concat(axis = var_3883, interleave = input_159_interleave_0, values = (var_5335_cast, var_5337_cast, var_5339_cast, var_5341_cast, var_5343_cast, var_5345_cast, var_5347_cast, var_5349_cast, var_5351_cast, var_5353_cast, var_5355_cast, var_5357_cast, var_5359_cast, var_5361_cast, var_5363_cast, var_5365_cast, var_5367_cast, var_5369_cast, var_5371_cast, var_5373_cast))[name = tensor("input_159_cast")]; + tensor var_5379 = const()[name = tensor("op_5379"), val = tensor([1, 1])]; + tensor var_5381 = const()[name = tensor("op_5381"), val = tensor([1, 1])]; + tensor var_5383_pad_type_0 = const()[name = tensor("op_5383_pad_type_0"), val = tensor("custom")]; + tensor var_5383_pad_0 = const()[name = tensor("op_5383_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_1_transformer_blocks_0_attn1_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(111211712))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(112440576))), name = tensor("down_blocks_2_attentions_1_transformer_blocks_0_attn1_to_out_0_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor down_blocks_2_attentions_1_transformer_blocks_0_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_0_attn1_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(112440768)))]; + tensor var_5383_cast = conv(bias = down_blocks_2_attentions_1_transformer_blocks_0_attn1_to_out_0_bias_to_fp16, dilations = var_5381, groups = var_3883, pad = var_5383_pad_0, pad_type = var_5383_pad_type_0, strides = var_5379, weight = down_blocks_2_attentions_1_transformer_blocks_0_attn1_to_out_0_weight_to_fp16_palettized, x = input_159_cast)[name = tensor("op_5383_cast")]; + tensor inputs_33_cast = add(x = var_5383_cast, y = inputs_31_cast)[name = tensor("inputs_33_cast")]; + tensor var_5387 = const()[name = tensor("op_5387"), val = tensor([1])]; + tensor channels_mean_33_cast = reduce_mean(axes = var_5387, keep_dims = var_3878, x = inputs_33_cast)[name = tensor("channels_mean_33_cast")]; + tensor zero_mean_33_cast = sub(x = inputs_33_cast, y = channels_mean_33_cast)[name = tensor("zero_mean_33_cast")]; + tensor zero_mean_sq_33_cast = mul(x = zero_mean_33_cast, y = zero_mean_33_cast)[name = tensor("zero_mean_sq_33_cast")]; + tensor var_5391 = const()[name = tensor("op_5391"), val = tensor([1])]; + tensor var_5392_cast = reduce_mean(axes = var_5391, keep_dims = var_3878, x = zero_mean_sq_33_cast)[name = tensor("op_5392_cast")]; + tensor var_5393_to_fp16 = const()[name = tensor("op_5393_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_5394_cast = add(x = var_5392_cast, y = var_5393_to_fp16)[name = tensor("op_5394_cast")]; + tensor denom_33_epsilon_0_to_fp16 = const()[name = tensor("denom_33_epsilon_0_to_fp16"), val = tensor(0x1p-24)]; + tensor denom_33_cast = rsqrt(epsilon = denom_33_epsilon_0_to_fp16, x = var_5394_cast)[name = tensor("denom_33_cast")]; + tensor out_33_cast = mul(x = zero_mean_33_cast, y = denom_33_cast)[name = tensor("out_33_cast")]; + tensor var_5398_to_fp16 = const()[name = tensor("op_5398_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(112443392)))]; + tensor var_5399_cast = add(x = out_33_cast, y = var_5398_to_fp16)[name = tensor("op_5399_cast")]; + tensor var_5401_to_fp16 = const()[name = tensor("op_5401_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(112446016)))]; + tensor hidden_states_101_cast = mul(x = var_5399_cast, y = var_5401_to_fp16)[name = tensor("hidden_states_101_cast")]; + tensor var_5408 = const()[name = tensor("op_5408"), val = tensor([1, 1])]; + tensor var_5410 = const()[name = tensor("op_5410"), val = tensor([1, 1])]; + tensor q_23_pad_type_0 = const()[name = tensor("q_23_pad_type_0"), val = tensor("custom")]; + tensor q_23_pad_0 = const()[name = tensor("q_23_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_1_transformer_blocks_0_attn2_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(112448640))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(113677504))), name = tensor("down_blocks_2_attentions_1_transformer_blocks_0_attn2_to_q_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor q_23_cast = conv(dilations = var_5410, groups = var_3883, pad = q_23_pad_0, pad_type = q_23_pad_type_0, strides = var_5408, weight = down_blocks_2_attentions_1_transformer_blocks_0_attn2_to_q_weight_to_fp16_palettized, x = hidden_states_101_cast)[name = tensor("q_23_cast")]; + tensor var_5414 = const()[name = tensor("op_5414"), val = tensor([1, 1])]; + tensor var_5416 = const()[name = tensor("op_5416"), val = tensor([1, 1])]; + tensor k_45_pad_type_0 = const()[name = tensor("k_45_pad_type_0"), val = tensor("custom")]; + tensor k_45_pad_0 = const()[name = tensor("k_45_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_1_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(113677696))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(114660800))), name = tensor("down_blocks_2_attentions_1_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized"), shape = tensor([1280, 1024, 1, 1])]; + tensor k_45_cast = conv(dilations = var_5416, groups = var_3883, pad = k_45_pad_0, pad_type = k_45_pad_type_0, strides = var_5414, weight = down_blocks_2_attentions_1_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor("k_45_cast")]; + tensor var_5420 = const()[name = tensor("op_5420"), val = tensor([1, 1])]; + tensor var_5422 = const()[name = tensor("op_5422"), val = tensor([1, 1])]; + tensor v_23_pad_type_0 = const()[name = tensor("v_23_pad_type_0"), val = tensor("custom")]; + tensor v_23_pad_0 = const()[name = tensor("v_23_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_1_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(114660992))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(115644096))), name = tensor("down_blocks_2_attentions_1_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized"), shape = tensor([1280, 1024, 1, 1])]; + tensor v_23_cast = conv(dilations = var_5422, groups = var_3883, pad = v_23_pad_0, pad_type = v_23_pad_type_0, strides = var_5420, weight = down_blocks_2_attentions_1_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor("v_23_cast")]; + tensor var_5426_begin_0 = const()[name = tensor("op_5426_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5426_end_0 = const()[name = tensor("op_5426_end_0"), val = tensor([2, 64, 1, 256])]; + tensor var_5426_end_mask_0 = const()[name = tensor("op_5426_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5426_cast = slice_by_index(begin = var_5426_begin_0, end = var_5426_end_0, end_mask = var_5426_end_mask_0, x = q_23_cast)[name = tensor("op_5426_cast")]; + tensor var_5430_begin_0 = const()[name = tensor("op_5430_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_5430_end_0 = const()[name = tensor("op_5430_end_0"), val = tensor([2, 128, 1, 256])]; + tensor var_5430_end_mask_0 = const()[name = tensor("op_5430_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5430_cast = slice_by_index(begin = var_5430_begin_0, end = var_5430_end_0, end_mask = var_5430_end_mask_0, x = q_23_cast)[name = tensor("op_5430_cast")]; + tensor var_5434_begin_0 = const()[name = tensor("op_5434_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_5434_end_0 = const()[name = tensor("op_5434_end_0"), val = tensor([2, 192, 1, 256])]; + tensor var_5434_end_mask_0 = const()[name = tensor("op_5434_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5434_cast = slice_by_index(begin = var_5434_begin_0, end = var_5434_end_0, end_mask = var_5434_end_mask_0, x = q_23_cast)[name = tensor("op_5434_cast")]; + tensor var_5438_begin_0 = const()[name = tensor("op_5438_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_5438_end_0 = const()[name = tensor("op_5438_end_0"), val = tensor([2, 256, 1, 256])]; + tensor var_5438_end_mask_0 = const()[name = tensor("op_5438_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5438_cast = slice_by_index(begin = var_5438_begin_0, end = var_5438_end_0, end_mask = var_5438_end_mask_0, x = q_23_cast)[name = tensor("op_5438_cast")]; + tensor var_5442_begin_0 = const()[name = tensor("op_5442_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_5442_end_0 = const()[name = tensor("op_5442_end_0"), val = tensor([2, 320, 1, 256])]; + tensor var_5442_end_mask_0 = const()[name = tensor("op_5442_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5442_cast = slice_by_index(begin = var_5442_begin_0, end = var_5442_end_0, end_mask = var_5442_end_mask_0, x = q_23_cast)[name = tensor("op_5442_cast")]; + tensor var_5446_begin_0 = const()[name = tensor("op_5446_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_5446_end_0 = const()[name = tensor("op_5446_end_0"), val = tensor([2, 384, 1, 256])]; + tensor var_5446_end_mask_0 = const()[name = tensor("op_5446_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5446_cast = slice_by_index(begin = var_5446_begin_0, end = var_5446_end_0, end_mask = var_5446_end_mask_0, x = q_23_cast)[name = tensor("op_5446_cast")]; + tensor var_5450_begin_0 = const()[name = tensor("op_5450_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_5450_end_0 = const()[name = tensor("op_5450_end_0"), val = tensor([2, 448, 1, 256])]; + tensor var_5450_end_mask_0 = const()[name = tensor("op_5450_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5450_cast = slice_by_index(begin = var_5450_begin_0, end = var_5450_end_0, end_mask = var_5450_end_mask_0, x = q_23_cast)[name = tensor("op_5450_cast")]; + tensor var_5454_begin_0 = const()[name = tensor("op_5454_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_5454_end_0 = const()[name = tensor("op_5454_end_0"), val = tensor([2, 512, 1, 256])]; + tensor var_5454_end_mask_0 = const()[name = tensor("op_5454_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5454_cast = slice_by_index(begin = var_5454_begin_0, end = var_5454_end_0, end_mask = var_5454_end_mask_0, x = q_23_cast)[name = tensor("op_5454_cast")]; + tensor var_5458_begin_0 = const()[name = tensor("op_5458_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_5458_end_0 = const()[name = tensor("op_5458_end_0"), val = tensor([2, 576, 1, 256])]; + tensor var_5458_end_mask_0 = const()[name = tensor("op_5458_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5458_cast = slice_by_index(begin = var_5458_begin_0, end = var_5458_end_0, end_mask = var_5458_end_mask_0, x = q_23_cast)[name = tensor("op_5458_cast")]; + tensor var_5462_begin_0 = const()[name = tensor("op_5462_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_5462_end_0 = const()[name = tensor("op_5462_end_0"), val = tensor([2, 640, 1, 256])]; + tensor var_5462_end_mask_0 = const()[name = tensor("op_5462_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5462_cast = slice_by_index(begin = var_5462_begin_0, end = var_5462_end_0, end_mask = var_5462_end_mask_0, x = q_23_cast)[name = tensor("op_5462_cast")]; + tensor var_5466_begin_0 = const()[name = tensor("op_5466_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_5466_end_0 = const()[name = tensor("op_5466_end_0"), val = tensor([2, 704, 1, 256])]; + tensor var_5466_end_mask_0 = const()[name = tensor("op_5466_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5466_cast = slice_by_index(begin = var_5466_begin_0, end = var_5466_end_0, end_mask = var_5466_end_mask_0, x = q_23_cast)[name = tensor("op_5466_cast")]; + tensor var_5470_begin_0 = const()[name = tensor("op_5470_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_5470_end_0 = const()[name = tensor("op_5470_end_0"), val = tensor([2, 768, 1, 256])]; + tensor var_5470_end_mask_0 = const()[name = tensor("op_5470_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5470_cast = slice_by_index(begin = var_5470_begin_0, end = var_5470_end_0, end_mask = var_5470_end_mask_0, x = q_23_cast)[name = tensor("op_5470_cast")]; + tensor var_5474_begin_0 = const()[name = tensor("op_5474_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_5474_end_0 = const()[name = tensor("op_5474_end_0"), val = tensor([2, 832, 1, 256])]; + tensor var_5474_end_mask_0 = const()[name = tensor("op_5474_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5474_cast = slice_by_index(begin = var_5474_begin_0, end = var_5474_end_0, end_mask = var_5474_end_mask_0, x = q_23_cast)[name = tensor("op_5474_cast")]; + tensor var_5478_begin_0 = const()[name = tensor("op_5478_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_5478_end_0 = const()[name = tensor("op_5478_end_0"), val = tensor([2, 896, 1, 256])]; + tensor var_5478_end_mask_0 = const()[name = tensor("op_5478_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5478_cast = slice_by_index(begin = var_5478_begin_0, end = var_5478_end_0, end_mask = var_5478_end_mask_0, x = q_23_cast)[name = tensor("op_5478_cast")]; + tensor var_5482_begin_0 = const()[name = tensor("op_5482_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_5482_end_0 = const()[name = tensor("op_5482_end_0"), val = tensor([2, 960, 1, 256])]; + tensor var_5482_end_mask_0 = const()[name = tensor("op_5482_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5482_cast = slice_by_index(begin = var_5482_begin_0, end = var_5482_end_0, end_mask = var_5482_end_mask_0, x = q_23_cast)[name = tensor("op_5482_cast")]; + tensor var_5486_begin_0 = const()[name = tensor("op_5486_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_5486_end_0 = const()[name = tensor("op_5486_end_0"), val = tensor([2, 1024, 1, 256])]; + tensor var_5486_end_mask_0 = const()[name = tensor("op_5486_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5486_cast = slice_by_index(begin = var_5486_begin_0, end = var_5486_end_0, end_mask = var_5486_end_mask_0, x = q_23_cast)[name = tensor("op_5486_cast")]; + tensor var_5490_begin_0 = const()[name = tensor("op_5490_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_5490_end_0 = const()[name = tensor("op_5490_end_0"), val = tensor([2, 1088, 1, 256])]; + tensor var_5490_end_mask_0 = const()[name = tensor("op_5490_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5490_cast = slice_by_index(begin = var_5490_begin_0, end = var_5490_end_0, end_mask = var_5490_end_mask_0, x = q_23_cast)[name = tensor("op_5490_cast")]; + tensor var_5494_begin_0 = const()[name = tensor("op_5494_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_5494_end_0 = const()[name = tensor("op_5494_end_0"), val = tensor([2, 1152, 1, 256])]; + tensor var_5494_end_mask_0 = const()[name = tensor("op_5494_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5494_cast = slice_by_index(begin = var_5494_begin_0, end = var_5494_end_0, end_mask = var_5494_end_mask_0, x = q_23_cast)[name = tensor("op_5494_cast")]; + tensor var_5498_begin_0 = const()[name = tensor("op_5498_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_5498_end_0 = const()[name = tensor("op_5498_end_0"), val = tensor([2, 1216, 1, 256])]; + tensor var_5498_end_mask_0 = const()[name = tensor("op_5498_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5498_cast = slice_by_index(begin = var_5498_begin_0, end = var_5498_end_0, end_mask = var_5498_end_mask_0, x = q_23_cast)[name = tensor("op_5498_cast")]; + tensor var_5502_begin_0 = const()[name = tensor("op_5502_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_5502_end_0 = const()[name = tensor("op_5502_end_0"), val = tensor([2, 1280, 1, 256])]; + tensor var_5502_end_mask_0 = const()[name = tensor("op_5502_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5502_cast = slice_by_index(begin = var_5502_begin_0, end = var_5502_end_0, end_mask = var_5502_end_mask_0, x = q_23_cast)[name = tensor("op_5502_cast")]; + tensor k_47_perm_0 = const()[name = tensor("k_47_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_5509_begin_0 = const()[name = tensor("op_5509_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5509_end_0 = const()[name = tensor("op_5509_end_0"), val = tensor([2, 77, 1, 64])]; + tensor var_5509_end_mask_0 = const()[name = tensor("op_5509_end_mask_0"), val = tensor([true, true, true, false])]; + tensor transpose_20 = transpose(perm = k_47_perm_0, x = k_45_cast)[name = tensor("transpose_20")]; + tensor var_5509_cast = slice_by_index(begin = var_5509_begin_0, end = var_5509_end_0, end_mask = var_5509_end_mask_0, x = transpose_20)[name = tensor("op_5509_cast")]; + tensor var_5513_begin_0 = const()[name = tensor("op_5513_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_5513_end_0 = const()[name = tensor("op_5513_end_0"), val = tensor([2, 77, 1, 128])]; + tensor var_5513_end_mask_0 = const()[name = tensor("op_5513_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5513_cast = slice_by_index(begin = var_5513_begin_0, end = var_5513_end_0, end_mask = var_5513_end_mask_0, x = transpose_20)[name = tensor("op_5513_cast")]; + tensor var_5517_begin_0 = const()[name = tensor("op_5517_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_5517_end_0 = const()[name = tensor("op_5517_end_0"), val = tensor([2, 77, 1, 192])]; + tensor var_5517_end_mask_0 = const()[name = tensor("op_5517_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5517_cast = slice_by_index(begin = var_5517_begin_0, end = var_5517_end_0, end_mask = var_5517_end_mask_0, x = transpose_20)[name = tensor("op_5517_cast")]; + tensor var_5521_begin_0 = const()[name = tensor("op_5521_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_5521_end_0 = const()[name = tensor("op_5521_end_0"), val = tensor([2, 77, 1, 256])]; + tensor var_5521_end_mask_0 = const()[name = tensor("op_5521_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5521_cast = slice_by_index(begin = var_5521_begin_0, end = var_5521_end_0, end_mask = var_5521_end_mask_0, x = transpose_20)[name = tensor("op_5521_cast")]; + tensor var_5525_begin_0 = const()[name = tensor("op_5525_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_5525_end_0 = const()[name = tensor("op_5525_end_0"), val = tensor([2, 77, 1, 320])]; + tensor var_5525_end_mask_0 = const()[name = tensor("op_5525_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5525_cast = slice_by_index(begin = var_5525_begin_0, end = var_5525_end_0, end_mask = var_5525_end_mask_0, x = transpose_20)[name = tensor("op_5525_cast")]; + tensor var_5529_begin_0 = const()[name = tensor("op_5529_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_5529_end_0 = const()[name = tensor("op_5529_end_0"), val = tensor([2, 77, 1, 384])]; + tensor var_5529_end_mask_0 = const()[name = tensor("op_5529_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5529_cast = slice_by_index(begin = var_5529_begin_0, end = var_5529_end_0, end_mask = var_5529_end_mask_0, x = transpose_20)[name = tensor("op_5529_cast")]; + tensor var_5533_begin_0 = const()[name = tensor("op_5533_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_5533_end_0 = const()[name = tensor("op_5533_end_0"), val = tensor([2, 77, 1, 448])]; + tensor var_5533_end_mask_0 = const()[name = tensor("op_5533_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5533_cast = slice_by_index(begin = var_5533_begin_0, end = var_5533_end_0, end_mask = var_5533_end_mask_0, x = transpose_20)[name = tensor("op_5533_cast")]; + tensor var_5537_begin_0 = const()[name = tensor("op_5537_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_5537_end_0 = const()[name = tensor("op_5537_end_0"), val = tensor([2, 77, 1, 512])]; + tensor var_5537_end_mask_0 = const()[name = tensor("op_5537_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5537_cast = slice_by_index(begin = var_5537_begin_0, end = var_5537_end_0, end_mask = var_5537_end_mask_0, x = transpose_20)[name = tensor("op_5537_cast")]; + tensor var_5541_begin_0 = const()[name = tensor("op_5541_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_5541_end_0 = const()[name = tensor("op_5541_end_0"), val = tensor([2, 77, 1, 576])]; + tensor var_5541_end_mask_0 = const()[name = tensor("op_5541_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5541_cast = slice_by_index(begin = var_5541_begin_0, end = var_5541_end_0, end_mask = var_5541_end_mask_0, x = transpose_20)[name = tensor("op_5541_cast")]; + tensor var_5545_begin_0 = const()[name = tensor("op_5545_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_5545_end_0 = const()[name = tensor("op_5545_end_0"), val = tensor([2, 77, 1, 640])]; + tensor var_5545_end_mask_0 = const()[name = tensor("op_5545_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5545_cast = slice_by_index(begin = var_5545_begin_0, end = var_5545_end_0, end_mask = var_5545_end_mask_0, x = transpose_20)[name = tensor("op_5545_cast")]; + tensor var_5549_begin_0 = const()[name = tensor("op_5549_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_5549_end_0 = const()[name = tensor("op_5549_end_0"), val = tensor([2, 77, 1, 704])]; + tensor var_5549_end_mask_0 = const()[name = tensor("op_5549_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5549_cast = slice_by_index(begin = var_5549_begin_0, end = var_5549_end_0, end_mask = var_5549_end_mask_0, x = transpose_20)[name = tensor("op_5549_cast")]; + tensor var_5553_begin_0 = const()[name = tensor("op_5553_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_5553_end_0 = const()[name = tensor("op_5553_end_0"), val = tensor([2, 77, 1, 768])]; + tensor var_5553_end_mask_0 = const()[name = tensor("op_5553_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5553_cast = slice_by_index(begin = var_5553_begin_0, end = var_5553_end_0, end_mask = var_5553_end_mask_0, x = transpose_20)[name = tensor("op_5553_cast")]; + tensor var_5557_begin_0 = const()[name = tensor("op_5557_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_5557_end_0 = const()[name = tensor("op_5557_end_0"), val = tensor([2, 77, 1, 832])]; + tensor var_5557_end_mask_0 = const()[name = tensor("op_5557_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5557_cast = slice_by_index(begin = var_5557_begin_0, end = var_5557_end_0, end_mask = var_5557_end_mask_0, x = transpose_20)[name = tensor("op_5557_cast")]; + tensor var_5561_begin_0 = const()[name = tensor("op_5561_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_5561_end_0 = const()[name = tensor("op_5561_end_0"), val = tensor([2, 77, 1, 896])]; + tensor var_5561_end_mask_0 = const()[name = tensor("op_5561_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5561_cast = slice_by_index(begin = var_5561_begin_0, end = var_5561_end_0, end_mask = var_5561_end_mask_0, x = transpose_20)[name = tensor("op_5561_cast")]; + tensor var_5565_begin_0 = const()[name = tensor("op_5565_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_5565_end_0 = const()[name = tensor("op_5565_end_0"), val = tensor([2, 77, 1, 960])]; + tensor var_5565_end_mask_0 = const()[name = tensor("op_5565_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5565_cast = slice_by_index(begin = var_5565_begin_0, end = var_5565_end_0, end_mask = var_5565_end_mask_0, x = transpose_20)[name = tensor("op_5565_cast")]; + tensor var_5569_begin_0 = const()[name = tensor("op_5569_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_5569_end_0 = const()[name = tensor("op_5569_end_0"), val = tensor([2, 77, 1, 1024])]; + tensor var_5569_end_mask_0 = const()[name = tensor("op_5569_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5569_cast = slice_by_index(begin = var_5569_begin_0, end = var_5569_end_0, end_mask = var_5569_end_mask_0, x = transpose_20)[name = tensor("op_5569_cast")]; + tensor var_5573_begin_0 = const()[name = tensor("op_5573_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_5573_end_0 = const()[name = tensor("op_5573_end_0"), val = tensor([2, 77, 1, 1088])]; + tensor var_5573_end_mask_0 = const()[name = tensor("op_5573_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5573_cast = slice_by_index(begin = var_5573_begin_0, end = var_5573_end_0, end_mask = var_5573_end_mask_0, x = transpose_20)[name = tensor("op_5573_cast")]; + tensor var_5577_begin_0 = const()[name = tensor("op_5577_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_5577_end_0 = const()[name = tensor("op_5577_end_0"), val = tensor([2, 77, 1, 1152])]; + tensor var_5577_end_mask_0 = const()[name = tensor("op_5577_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5577_cast = slice_by_index(begin = var_5577_begin_0, end = var_5577_end_0, end_mask = var_5577_end_mask_0, x = transpose_20)[name = tensor("op_5577_cast")]; + tensor var_5581_begin_0 = const()[name = tensor("op_5581_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_5581_end_0 = const()[name = tensor("op_5581_end_0"), val = tensor([2, 77, 1, 1216])]; + tensor var_5581_end_mask_0 = const()[name = tensor("op_5581_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5581_cast = slice_by_index(begin = var_5581_begin_0, end = var_5581_end_0, end_mask = var_5581_end_mask_0, x = transpose_20)[name = tensor("op_5581_cast")]; + tensor var_5585_begin_0 = const()[name = tensor("op_5585_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_5585_end_0 = const()[name = tensor("op_5585_end_0"), val = tensor([2, 77, 1, 1280])]; + tensor var_5585_end_mask_0 = const()[name = tensor("op_5585_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5585_cast = slice_by_index(begin = var_5585_begin_0, end = var_5585_end_0, end_mask = var_5585_end_mask_0, x = transpose_20)[name = tensor("op_5585_cast")]; + tensor var_5587_begin_0 = const()[name = tensor("op_5587_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5587_end_0 = const()[name = tensor("op_5587_end_0"), val = tensor([2, 64, 1, 77])]; + tensor var_5587_end_mask_0 = const()[name = tensor("op_5587_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5587_cast = slice_by_index(begin = var_5587_begin_0, end = var_5587_end_0, end_mask = var_5587_end_mask_0, x = v_23_cast)[name = tensor("op_5587_cast")]; + tensor var_5591_begin_0 = const()[name = tensor("op_5591_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_5591_end_0 = const()[name = tensor("op_5591_end_0"), val = tensor([2, 128, 1, 77])]; + tensor var_5591_end_mask_0 = const()[name = tensor("op_5591_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5591_cast = slice_by_index(begin = var_5591_begin_0, end = var_5591_end_0, end_mask = var_5591_end_mask_0, x = v_23_cast)[name = tensor("op_5591_cast")]; + tensor var_5595_begin_0 = const()[name = tensor("op_5595_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_5595_end_0 = const()[name = tensor("op_5595_end_0"), val = tensor([2, 192, 1, 77])]; + tensor var_5595_end_mask_0 = const()[name = tensor("op_5595_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5595_cast = slice_by_index(begin = var_5595_begin_0, end = var_5595_end_0, end_mask = var_5595_end_mask_0, x = v_23_cast)[name = tensor("op_5595_cast")]; + tensor var_5599_begin_0 = const()[name = tensor("op_5599_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_5599_end_0 = const()[name = tensor("op_5599_end_0"), val = tensor([2, 256, 1, 77])]; + tensor var_5599_end_mask_0 = const()[name = tensor("op_5599_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5599_cast = slice_by_index(begin = var_5599_begin_0, end = var_5599_end_0, end_mask = var_5599_end_mask_0, x = v_23_cast)[name = tensor("op_5599_cast")]; + tensor var_5603_begin_0 = const()[name = tensor("op_5603_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_5603_end_0 = const()[name = tensor("op_5603_end_0"), val = tensor([2, 320, 1, 77])]; + tensor var_5603_end_mask_0 = const()[name = tensor("op_5603_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5603_cast = slice_by_index(begin = var_5603_begin_0, end = var_5603_end_0, end_mask = var_5603_end_mask_0, x = v_23_cast)[name = tensor("op_5603_cast")]; + tensor var_5607_begin_0 = const()[name = tensor("op_5607_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_5607_end_0 = const()[name = tensor("op_5607_end_0"), val = tensor([2, 384, 1, 77])]; + tensor var_5607_end_mask_0 = const()[name = tensor("op_5607_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5607_cast = slice_by_index(begin = var_5607_begin_0, end = var_5607_end_0, end_mask = var_5607_end_mask_0, x = v_23_cast)[name = tensor("op_5607_cast")]; + tensor var_5611_begin_0 = const()[name = tensor("op_5611_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_5611_end_0 = const()[name = tensor("op_5611_end_0"), val = tensor([2, 448, 1, 77])]; + tensor var_5611_end_mask_0 = const()[name = tensor("op_5611_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5611_cast = slice_by_index(begin = var_5611_begin_0, end = var_5611_end_0, end_mask = var_5611_end_mask_0, x = v_23_cast)[name = tensor("op_5611_cast")]; + tensor var_5615_begin_0 = const()[name = tensor("op_5615_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_5615_end_0 = const()[name = tensor("op_5615_end_0"), val = tensor([2, 512, 1, 77])]; + tensor var_5615_end_mask_0 = const()[name = tensor("op_5615_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5615_cast = slice_by_index(begin = var_5615_begin_0, end = var_5615_end_0, end_mask = var_5615_end_mask_0, x = v_23_cast)[name = tensor("op_5615_cast")]; + tensor var_5619_begin_0 = const()[name = tensor("op_5619_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_5619_end_0 = const()[name = tensor("op_5619_end_0"), val = tensor([2, 576, 1, 77])]; + tensor var_5619_end_mask_0 = const()[name = tensor("op_5619_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5619_cast = slice_by_index(begin = var_5619_begin_0, end = var_5619_end_0, end_mask = var_5619_end_mask_0, x = v_23_cast)[name = tensor("op_5619_cast")]; + tensor var_5623_begin_0 = const()[name = tensor("op_5623_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_5623_end_0 = const()[name = tensor("op_5623_end_0"), val = tensor([2, 640, 1, 77])]; + tensor var_5623_end_mask_0 = const()[name = tensor("op_5623_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5623_cast = slice_by_index(begin = var_5623_begin_0, end = var_5623_end_0, end_mask = var_5623_end_mask_0, x = v_23_cast)[name = tensor("op_5623_cast")]; + tensor var_5627_begin_0 = const()[name = tensor("op_5627_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_5627_end_0 = const()[name = tensor("op_5627_end_0"), val = tensor([2, 704, 1, 77])]; + tensor var_5627_end_mask_0 = const()[name = tensor("op_5627_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5627_cast = slice_by_index(begin = var_5627_begin_0, end = var_5627_end_0, end_mask = var_5627_end_mask_0, x = v_23_cast)[name = tensor("op_5627_cast")]; + tensor var_5631_begin_0 = const()[name = tensor("op_5631_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_5631_end_0 = const()[name = tensor("op_5631_end_0"), val = tensor([2, 768, 1, 77])]; + tensor var_5631_end_mask_0 = const()[name = tensor("op_5631_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5631_cast = slice_by_index(begin = var_5631_begin_0, end = var_5631_end_0, end_mask = var_5631_end_mask_0, x = v_23_cast)[name = tensor("op_5631_cast")]; + tensor var_5635_begin_0 = const()[name = tensor("op_5635_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_5635_end_0 = const()[name = tensor("op_5635_end_0"), val = tensor([2, 832, 1, 77])]; + tensor var_5635_end_mask_0 = const()[name = tensor("op_5635_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5635_cast = slice_by_index(begin = var_5635_begin_0, end = var_5635_end_0, end_mask = var_5635_end_mask_0, x = v_23_cast)[name = tensor("op_5635_cast")]; + tensor var_5639_begin_0 = const()[name = tensor("op_5639_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_5639_end_0 = const()[name = tensor("op_5639_end_0"), val = tensor([2, 896, 1, 77])]; + tensor var_5639_end_mask_0 = const()[name = tensor("op_5639_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5639_cast = slice_by_index(begin = var_5639_begin_0, end = var_5639_end_0, end_mask = var_5639_end_mask_0, x = v_23_cast)[name = tensor("op_5639_cast")]; + tensor var_5643_begin_0 = const()[name = tensor("op_5643_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_5643_end_0 = const()[name = tensor("op_5643_end_0"), val = tensor([2, 960, 1, 77])]; + tensor var_5643_end_mask_0 = const()[name = tensor("op_5643_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5643_cast = slice_by_index(begin = var_5643_begin_0, end = var_5643_end_0, end_mask = var_5643_end_mask_0, x = v_23_cast)[name = tensor("op_5643_cast")]; + tensor var_5647_begin_0 = const()[name = tensor("op_5647_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_5647_end_0 = const()[name = tensor("op_5647_end_0"), val = tensor([2, 1024, 1, 77])]; + tensor var_5647_end_mask_0 = const()[name = tensor("op_5647_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5647_cast = slice_by_index(begin = var_5647_begin_0, end = var_5647_end_0, end_mask = var_5647_end_mask_0, x = v_23_cast)[name = tensor("op_5647_cast")]; + tensor var_5651_begin_0 = const()[name = tensor("op_5651_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_5651_end_0 = const()[name = tensor("op_5651_end_0"), val = tensor([2, 1088, 1, 77])]; + tensor var_5651_end_mask_0 = const()[name = tensor("op_5651_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5651_cast = slice_by_index(begin = var_5651_begin_0, end = var_5651_end_0, end_mask = var_5651_end_mask_0, x = v_23_cast)[name = tensor("op_5651_cast")]; + tensor var_5655_begin_0 = const()[name = tensor("op_5655_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_5655_end_0 = const()[name = tensor("op_5655_end_0"), val = tensor([2, 1152, 1, 77])]; + tensor var_5655_end_mask_0 = const()[name = tensor("op_5655_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5655_cast = slice_by_index(begin = var_5655_begin_0, end = var_5655_end_0, end_mask = var_5655_end_mask_0, x = v_23_cast)[name = tensor("op_5655_cast")]; + tensor var_5659_begin_0 = const()[name = tensor("op_5659_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_5659_end_0 = const()[name = tensor("op_5659_end_0"), val = tensor([2, 1216, 1, 77])]; + tensor var_5659_end_mask_0 = const()[name = tensor("op_5659_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5659_cast = slice_by_index(begin = var_5659_begin_0, end = var_5659_end_0, end_mask = var_5659_end_mask_0, x = v_23_cast)[name = tensor("op_5659_cast")]; + tensor var_5663_begin_0 = const()[name = tensor("op_5663_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_5663_end_0 = const()[name = tensor("op_5663_end_0"), val = tensor([2, 1280, 1, 77])]; + tensor var_5663_end_mask_0 = const()[name = tensor("op_5663_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5663_cast = slice_by_index(begin = var_5663_begin_0, end = var_5663_end_0, end_mask = var_5663_end_mask_0, x = v_23_cast)[name = tensor("op_5663_cast")]; + tensor var_5667_equation_0 = const()[name = tensor("op_5667_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5667_cast = einsum(equation = var_5667_equation_0, values = (var_5509_cast, var_5426_cast))[name = tensor("op_5667_cast")]; + tensor var_5668_to_fp16 = const()[name = tensor("op_5668_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_121_cast = mul(x = var_5667_cast, y = var_5668_to_fp16)[name = tensor("aw_121_cast")]; + tensor var_5671_equation_0 = const()[name = tensor("op_5671_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5671_cast = einsum(equation = var_5671_equation_0, values = (var_5513_cast, var_5430_cast))[name = tensor("op_5671_cast")]; + tensor var_5672_to_fp16 = const()[name = tensor("op_5672_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_123_cast = mul(x = var_5671_cast, y = var_5672_to_fp16)[name = tensor("aw_123_cast")]; + tensor var_5675_equation_0 = const()[name = tensor("op_5675_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5675_cast = einsum(equation = var_5675_equation_0, values = (var_5517_cast, var_5434_cast))[name = tensor("op_5675_cast")]; + tensor var_5676_to_fp16 = const()[name = tensor("op_5676_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_125_cast = mul(x = var_5675_cast, y = var_5676_to_fp16)[name = tensor("aw_125_cast")]; + tensor var_5679_equation_0 = const()[name = tensor("op_5679_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5679_cast = einsum(equation = var_5679_equation_0, values = (var_5521_cast, var_5438_cast))[name = tensor("op_5679_cast")]; + tensor var_5680_to_fp16 = const()[name = tensor("op_5680_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_127_cast = mul(x = var_5679_cast, y = var_5680_to_fp16)[name = tensor("aw_127_cast")]; + tensor var_5683_equation_0 = const()[name = tensor("op_5683_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5683_cast = einsum(equation = var_5683_equation_0, values = (var_5525_cast, var_5442_cast))[name = tensor("op_5683_cast")]; + tensor var_5684_to_fp16 = const()[name = tensor("op_5684_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_129_cast = mul(x = var_5683_cast, y = var_5684_to_fp16)[name = tensor("aw_129_cast")]; + tensor var_5687_equation_0 = const()[name = tensor("op_5687_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5687_cast = einsum(equation = var_5687_equation_0, values = (var_5529_cast, var_5446_cast))[name = tensor("op_5687_cast")]; + tensor var_5688_to_fp16 = const()[name = tensor("op_5688_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_131_cast = mul(x = var_5687_cast, y = var_5688_to_fp16)[name = tensor("aw_131_cast")]; + tensor var_5691_equation_0 = const()[name = tensor("op_5691_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5691_cast = einsum(equation = var_5691_equation_0, values = (var_5533_cast, var_5450_cast))[name = tensor("op_5691_cast")]; + tensor var_5692_to_fp16 = const()[name = tensor("op_5692_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_133_cast = mul(x = var_5691_cast, y = var_5692_to_fp16)[name = tensor("aw_133_cast")]; + tensor var_5695_equation_0 = const()[name = tensor("op_5695_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5695_cast = einsum(equation = var_5695_equation_0, values = (var_5537_cast, var_5454_cast))[name = tensor("op_5695_cast")]; + tensor var_5696_to_fp16 = const()[name = tensor("op_5696_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_135_cast = mul(x = var_5695_cast, y = var_5696_to_fp16)[name = tensor("aw_135_cast")]; + tensor var_5699_equation_0 = const()[name = tensor("op_5699_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5699_cast = einsum(equation = var_5699_equation_0, values = (var_5541_cast, var_5458_cast))[name = tensor("op_5699_cast")]; + tensor var_5700_to_fp16 = const()[name = tensor("op_5700_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_137_cast = mul(x = var_5699_cast, y = var_5700_to_fp16)[name = tensor("aw_137_cast")]; + tensor var_5703_equation_0 = const()[name = tensor("op_5703_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5703_cast = einsum(equation = var_5703_equation_0, values = (var_5545_cast, var_5462_cast))[name = tensor("op_5703_cast")]; + tensor var_5704_to_fp16 = const()[name = tensor("op_5704_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_139_cast = mul(x = var_5703_cast, y = var_5704_to_fp16)[name = tensor("aw_139_cast")]; + tensor var_5707_equation_0 = const()[name = tensor("op_5707_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5707_cast = einsum(equation = var_5707_equation_0, values = (var_5549_cast, var_5466_cast))[name = tensor("op_5707_cast")]; + tensor var_5708_to_fp16 = const()[name = tensor("op_5708_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_141_cast = mul(x = var_5707_cast, y = var_5708_to_fp16)[name = tensor("aw_141_cast")]; + tensor var_5711_equation_0 = const()[name = tensor("op_5711_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5711_cast = einsum(equation = var_5711_equation_0, values = (var_5553_cast, var_5470_cast))[name = tensor("op_5711_cast")]; + tensor var_5712_to_fp16 = const()[name = tensor("op_5712_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_143_cast = mul(x = var_5711_cast, y = var_5712_to_fp16)[name = tensor("aw_143_cast")]; + tensor var_5715_equation_0 = const()[name = tensor("op_5715_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5715_cast = einsum(equation = var_5715_equation_0, values = (var_5557_cast, var_5474_cast))[name = tensor("op_5715_cast")]; + tensor var_5716_to_fp16 = const()[name = tensor("op_5716_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_145_cast = mul(x = var_5715_cast, y = var_5716_to_fp16)[name = tensor("aw_145_cast")]; + tensor var_5719_equation_0 = const()[name = tensor("op_5719_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5719_cast = einsum(equation = var_5719_equation_0, values = (var_5561_cast, var_5478_cast))[name = tensor("op_5719_cast")]; + tensor var_5720_to_fp16 = const()[name = tensor("op_5720_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_147_cast = mul(x = var_5719_cast, y = var_5720_to_fp16)[name = tensor("aw_147_cast")]; + tensor var_5723_equation_0 = const()[name = tensor("op_5723_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5723_cast = einsum(equation = var_5723_equation_0, values = (var_5565_cast, var_5482_cast))[name = tensor("op_5723_cast")]; + tensor var_5724_to_fp16 = const()[name = tensor("op_5724_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_149_cast = mul(x = var_5723_cast, y = var_5724_to_fp16)[name = tensor("aw_149_cast")]; + tensor var_5727_equation_0 = const()[name = tensor("op_5727_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5727_cast = einsum(equation = var_5727_equation_0, values = (var_5569_cast, var_5486_cast))[name = tensor("op_5727_cast")]; + tensor var_5728_to_fp16 = const()[name = tensor("op_5728_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_151_cast = mul(x = var_5727_cast, y = var_5728_to_fp16)[name = tensor("aw_151_cast")]; + tensor var_5731_equation_0 = const()[name = tensor("op_5731_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5731_cast = einsum(equation = var_5731_equation_0, values = (var_5573_cast, var_5490_cast))[name = tensor("op_5731_cast")]; + tensor var_5732_to_fp16 = const()[name = tensor("op_5732_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_153_cast = mul(x = var_5731_cast, y = var_5732_to_fp16)[name = tensor("aw_153_cast")]; + tensor var_5735_equation_0 = const()[name = tensor("op_5735_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5735_cast = einsum(equation = var_5735_equation_0, values = (var_5577_cast, var_5494_cast))[name = tensor("op_5735_cast")]; + tensor var_5736_to_fp16 = const()[name = tensor("op_5736_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_155_cast = mul(x = var_5735_cast, y = var_5736_to_fp16)[name = tensor("aw_155_cast")]; + tensor var_5739_equation_0 = const()[name = tensor("op_5739_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5739_cast = einsum(equation = var_5739_equation_0, values = (var_5581_cast, var_5498_cast))[name = tensor("op_5739_cast")]; + tensor var_5740_to_fp16 = const()[name = tensor("op_5740_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_157_cast = mul(x = var_5739_cast, y = var_5740_to_fp16)[name = tensor("aw_157_cast")]; + tensor var_5743_equation_0 = const()[name = tensor("op_5743_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_5743_cast = einsum(equation = var_5743_equation_0, values = (var_5585_cast, var_5502_cast))[name = tensor("op_5743_cast")]; + tensor var_5744_to_fp16 = const()[name = tensor("op_5744_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_159_cast = mul(x = var_5743_cast, y = var_5744_to_fp16)[name = tensor("aw_159_cast")]; + tensor var_5746_cast = softmax(axis = var_3883, x = aw_121_cast)[name = tensor("op_5746_cast")]; + tensor var_5747_cast = softmax(axis = var_3883, x = aw_123_cast)[name = tensor("op_5747_cast")]; + tensor var_5748_cast = softmax(axis = var_3883, x = aw_125_cast)[name = tensor("op_5748_cast")]; + tensor var_5749_cast = softmax(axis = var_3883, x = aw_127_cast)[name = tensor("op_5749_cast")]; + tensor var_5750_cast = softmax(axis = var_3883, x = aw_129_cast)[name = tensor("op_5750_cast")]; + tensor var_5751_cast = softmax(axis = var_3883, x = aw_131_cast)[name = tensor("op_5751_cast")]; + tensor var_5752_cast = softmax(axis = var_3883, x = aw_133_cast)[name = tensor("op_5752_cast")]; + tensor var_5753_cast = softmax(axis = var_3883, x = aw_135_cast)[name = tensor("op_5753_cast")]; + tensor var_5754_cast = softmax(axis = var_3883, x = aw_137_cast)[name = tensor("op_5754_cast")]; + tensor var_5755_cast = softmax(axis = var_3883, x = aw_139_cast)[name = tensor("op_5755_cast")]; + tensor var_5756_cast = softmax(axis = var_3883, x = aw_141_cast)[name = tensor("op_5756_cast")]; + tensor var_5757_cast = softmax(axis = var_3883, x = aw_143_cast)[name = tensor("op_5757_cast")]; + tensor var_5758_cast = softmax(axis = var_3883, x = aw_145_cast)[name = tensor("op_5758_cast")]; + tensor var_5759_cast = softmax(axis = var_3883, x = aw_147_cast)[name = tensor("op_5759_cast")]; + tensor var_5760_cast = softmax(axis = var_3883, x = aw_149_cast)[name = tensor("op_5760_cast")]; + tensor var_5761_cast = softmax(axis = var_3883, x = aw_151_cast)[name = tensor("op_5761_cast")]; + tensor var_5762_cast = softmax(axis = var_3883, x = aw_153_cast)[name = tensor("op_5762_cast")]; + tensor var_5763_cast = softmax(axis = var_3883, x = aw_155_cast)[name = tensor("op_5763_cast")]; + tensor var_5764_cast = softmax(axis = var_3883, x = aw_157_cast)[name = tensor("op_5764_cast")]; + tensor var_5765_cast = softmax(axis = var_3883, x = aw_159_cast)[name = tensor("op_5765_cast")]; + tensor var_5767_equation_0 = const()[name = tensor("op_5767_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_5767_cast = einsum(equation = var_5767_equation_0, values = (var_5587_cast, var_5746_cast))[name = tensor("op_5767_cast")]; + tensor var_5769_equation_0 = const()[name = tensor("op_5769_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_5769_cast = einsum(equation = var_5769_equation_0, values = (var_5591_cast, var_5747_cast))[name = tensor("op_5769_cast")]; + tensor var_5771_equation_0 = const()[name = tensor("op_5771_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_5771_cast = einsum(equation = var_5771_equation_0, values = (var_5595_cast, var_5748_cast))[name = tensor("op_5771_cast")]; + tensor var_5773_equation_0 = const()[name = tensor("op_5773_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_5773_cast = einsum(equation = var_5773_equation_0, values = (var_5599_cast, var_5749_cast))[name = tensor("op_5773_cast")]; + tensor var_5775_equation_0 = const()[name = tensor("op_5775_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_5775_cast = einsum(equation = var_5775_equation_0, values = (var_5603_cast, var_5750_cast))[name = tensor("op_5775_cast")]; + tensor var_5777_equation_0 = const()[name = tensor("op_5777_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_5777_cast = einsum(equation = var_5777_equation_0, values = (var_5607_cast, var_5751_cast))[name = tensor("op_5777_cast")]; + tensor var_5779_equation_0 = const()[name = tensor("op_5779_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_5779_cast = einsum(equation = var_5779_equation_0, values = (var_5611_cast, var_5752_cast))[name = tensor("op_5779_cast")]; + tensor var_5781_equation_0 = const()[name = tensor("op_5781_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_5781_cast = einsum(equation = var_5781_equation_0, values = (var_5615_cast, var_5753_cast))[name = tensor("op_5781_cast")]; + tensor var_5783_equation_0 = const()[name = tensor("op_5783_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_5783_cast = einsum(equation = var_5783_equation_0, values = (var_5619_cast, var_5754_cast))[name = tensor("op_5783_cast")]; + tensor var_5785_equation_0 = const()[name = tensor("op_5785_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_5785_cast = einsum(equation = var_5785_equation_0, values = (var_5623_cast, var_5755_cast))[name = tensor("op_5785_cast")]; + tensor var_5787_equation_0 = const()[name = tensor("op_5787_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_5787_cast = einsum(equation = var_5787_equation_0, values = (var_5627_cast, var_5756_cast))[name = tensor("op_5787_cast")]; + tensor var_5789_equation_0 = const()[name = tensor("op_5789_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_5789_cast = einsum(equation = var_5789_equation_0, values = (var_5631_cast, var_5757_cast))[name = tensor("op_5789_cast")]; + tensor var_5791_equation_0 = const()[name = tensor("op_5791_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_5791_cast = einsum(equation = var_5791_equation_0, values = (var_5635_cast, var_5758_cast))[name = tensor("op_5791_cast")]; + tensor var_5793_equation_0 = const()[name = tensor("op_5793_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_5793_cast = einsum(equation = var_5793_equation_0, values = (var_5639_cast, var_5759_cast))[name = tensor("op_5793_cast")]; + tensor var_5795_equation_0 = const()[name = tensor("op_5795_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_5795_cast = einsum(equation = var_5795_equation_0, values = (var_5643_cast, var_5760_cast))[name = tensor("op_5795_cast")]; + tensor var_5797_equation_0 = const()[name = tensor("op_5797_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_5797_cast = einsum(equation = var_5797_equation_0, values = (var_5647_cast, var_5761_cast))[name = tensor("op_5797_cast")]; + tensor var_5799_equation_0 = const()[name = tensor("op_5799_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_5799_cast = einsum(equation = var_5799_equation_0, values = (var_5651_cast, var_5762_cast))[name = tensor("op_5799_cast")]; + tensor var_5801_equation_0 = const()[name = tensor("op_5801_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_5801_cast = einsum(equation = var_5801_equation_0, values = (var_5655_cast, var_5763_cast))[name = tensor("op_5801_cast")]; + tensor var_5803_equation_0 = const()[name = tensor("op_5803_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_5803_cast = einsum(equation = var_5803_equation_0, values = (var_5659_cast, var_5764_cast))[name = tensor("op_5803_cast")]; + tensor var_5805_equation_0 = const()[name = tensor("op_5805_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_5805_cast = einsum(equation = var_5805_equation_0, values = (var_5663_cast, var_5765_cast))[name = tensor("op_5805_cast")]; + tensor input_161_interleave_0 = const()[name = tensor("input_161_interleave_0"), val = tensor(false)]; + tensor input_161_cast = concat(axis = var_3883, interleave = input_161_interleave_0, values = (var_5767_cast, var_5769_cast, var_5771_cast, var_5773_cast, var_5775_cast, var_5777_cast, var_5779_cast, var_5781_cast, var_5783_cast, var_5785_cast, var_5787_cast, var_5789_cast, var_5791_cast, var_5793_cast, var_5795_cast, var_5797_cast, var_5799_cast, var_5801_cast, var_5803_cast, var_5805_cast))[name = tensor("input_161_cast")]; + tensor var_5811 = const()[name = tensor("op_5811"), val = tensor([1, 1])]; + tensor var_5813 = const()[name = tensor("op_5813"), val = tensor([1, 1])]; + tensor var_5815_pad_type_0 = const()[name = tensor("op_5815_pad_type_0"), val = tensor("custom")]; + tensor var_5815_pad_0 = const()[name = tensor("op_5815_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_1_transformer_blocks_0_attn2_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(115644288))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(116873152))), name = tensor("down_blocks_2_attentions_1_transformer_blocks_0_attn2_to_out_0_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor down_blocks_2_attentions_1_transformer_blocks_0_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_0_attn2_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(116873344)))]; + tensor var_5815_cast = conv(bias = down_blocks_2_attentions_1_transformer_blocks_0_attn2_to_out_0_bias_to_fp16, dilations = var_5813, groups = var_3883, pad = var_5815_pad_0, pad_type = var_5815_pad_type_0, strides = var_5811, weight = down_blocks_2_attentions_1_transformer_blocks_0_attn2_to_out_0_weight_to_fp16_palettized, x = input_161_cast)[name = tensor("op_5815_cast")]; + tensor inputs_35_cast = add(x = var_5815_cast, y = inputs_33_cast)[name = tensor("inputs_35_cast")]; + tensor var_5819 = const()[name = tensor("op_5819"), val = tensor([1])]; + tensor channels_mean_35_cast = reduce_mean(axes = var_5819, keep_dims = var_3878, x = inputs_35_cast)[name = tensor("channels_mean_35_cast")]; + tensor zero_mean_35_cast = sub(x = inputs_35_cast, y = channels_mean_35_cast)[name = tensor("zero_mean_35_cast")]; + tensor zero_mean_sq_35_cast = mul(x = zero_mean_35_cast, y = zero_mean_35_cast)[name = tensor("zero_mean_sq_35_cast")]; + tensor var_5823 = const()[name = tensor("op_5823"), val = tensor([1])]; + tensor var_5824_cast = reduce_mean(axes = var_5823, keep_dims = var_3878, x = zero_mean_sq_35_cast)[name = tensor("op_5824_cast")]; + tensor var_5825_to_fp16 = const()[name = tensor("op_5825_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_5826_cast = add(x = var_5824_cast, y = var_5825_to_fp16)[name = tensor("op_5826_cast")]; + tensor denom_35_epsilon_0_to_fp16 = const()[name = tensor("denom_35_epsilon_0_to_fp16"), val = tensor(0x1p-24)]; + tensor denom_35_cast = rsqrt(epsilon = denom_35_epsilon_0_to_fp16, x = var_5826_cast)[name = tensor("denom_35_cast")]; + tensor out_35_cast = mul(x = zero_mean_35_cast, y = denom_35_cast)[name = tensor("out_35_cast")]; + tensor var_5830_to_fp16 = const()[name = tensor("op_5830_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(116875968)))]; + tensor var_5831_cast = add(x = out_35_cast, y = var_5830_to_fp16)[name = tensor("op_5831_cast")]; + tensor var_5833_to_fp16 = const()[name = tensor("op_5833_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(116878592)))]; + tensor input_163_cast = mul(x = var_5831_cast, y = var_5833_to_fp16)[name = tensor("input_163_cast")]; + tensor var_5841 = const()[name = tensor("op_5841"), val = tensor([1, 1])]; + tensor var_5843 = const()[name = tensor("op_5843"), val = tensor([1, 1])]; + tensor var_5845_pad_type_0 = const()[name = tensor("op_5845_pad_type_0"), val = tensor("custom")]; + tensor var_5845_pad_0 = const()[name = tensor("op_5845_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_1_transformer_blocks_0_ff_net_0_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(116881216))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(126711680))), name = tensor("down_blocks_2_attentions_1_transformer_blocks_0_ff_net_0_proj_weight_to_fp16_palettized"), shape = tensor([10240, 1280, 1, 1])]; + tensor down_blocks_2_attentions_1_transformer_blocks_0_ff_net_0_proj_bias_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(126711872))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(126719616))), name = tensor("down_blocks_2_attentions_1_transformer_blocks_0_ff_net_0_proj_bias_to_fp16_palettized"), shape = tensor([10240])]; + tensor var_5845_cast = conv(bias = down_blocks_2_attentions_1_transformer_blocks_0_ff_net_0_proj_bias_to_fp16_palettized, dilations = var_5843, groups = var_3883, pad = var_5845_pad_0, pad_type = var_5845_pad_type_0, strides = var_5841, weight = down_blocks_2_attentions_1_transformer_blocks_0_ff_net_0_proj_weight_to_fp16_palettized, x = input_163_cast)[name = tensor("op_5845_cast")]; + tensor var_5846_split_sizes_0 = const()[name = tensor("op_5846_split_sizes_0"), val = tensor([5120, 5120])]; + tensor var_5846_axis_0 = const()[name = tensor("op_5846_axis_0"), val = tensor(1)]; + tensor var_5846_cast_0, tensor var_5846_cast_1 = split(axis = var_5846_axis_0, split_sizes = var_5846_split_sizes_0, x = var_5845_cast)[name = tensor("op_5846_cast")]; + tensor var_5848_mode_0 = const()[name = tensor("op_5848_mode_0"), val = tensor("EXACT")]; + tensor var_5848_cast = gelu(mode = var_5848_mode_0, x = var_5846_cast_1)[name = tensor("op_5848_cast")]; + tensor input_165_cast = mul(x = var_5846_cast_0, y = var_5848_cast)[name = tensor("input_165_cast")]; + tensor var_5852 = const()[name = tensor("op_5852"), val = tensor([1, 1])]; + tensor var_5854 = const()[name = tensor("op_5854"), val = tensor([1, 1])]; + tensor var_5856_pad_type_0 = const()[name = tensor("op_5856_pad_type_0"), val = tensor("custom")]; + tensor var_5856_pad_0 = const()[name = tensor("op_5856_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_1_transformer_blocks_0_ff_net_2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(126719808))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(131635072))), name = tensor("down_blocks_2_attentions_1_transformer_blocks_0_ff_net_2_weight_to_fp16_palettized"), shape = tensor([1280, 5120, 1, 1])]; + tensor down_blocks_2_attentions_1_transformer_blocks_0_ff_net_2_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_transformer_blocks_0_ff_net_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(131635264)))]; + tensor var_5856_cast = conv(bias = down_blocks_2_attentions_1_transformer_blocks_0_ff_net_2_bias_to_fp16, dilations = var_5854, groups = var_3883, pad = var_5856_pad_0, pad_type = var_5856_pad_type_0, strides = var_5852, weight = down_blocks_2_attentions_1_transformer_blocks_0_ff_net_2_weight_to_fp16_palettized, x = input_165_cast)[name = tensor("op_5856_cast")]; + tensor hidden_states_105_cast = add(x = var_5856_cast, y = inputs_35_cast)[name = tensor("hidden_states_105_cast")]; + tensor var_5858 = const()[name = tensor("op_5858"), val = tensor([2, 1280, 16, 16])]; + tensor input_167_cast = reshape(shape = var_5858, x = hidden_states_105_cast)[name = tensor("input_167_cast")]; + tensor var_5862 = const()[name = tensor("op_5862"), val = tensor([1, 1])]; + tensor var_5864 = const()[name = tensor("op_5864"), val = tensor([1, 1])]; + tensor hidden_states_107_pad_type_0 = const()[name = tensor("hidden_states_107_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_107_pad_0 = const()[name = tensor("hidden_states_107_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_2_attentions_1_proj_out_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(131637888))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(132866752))), name = tensor("down_blocks_2_attentions_1_proj_out_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor down_blocks_2_attentions_1_proj_out_bias_to_fp16 = const()[name = tensor("down_blocks_2_attentions_1_proj_out_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(132866944)))]; + tensor hidden_states_107_cast = conv(bias = down_blocks_2_attentions_1_proj_out_bias_to_fp16, dilations = var_5864, groups = var_3883, pad = hidden_states_107_pad_0, pad_type = hidden_states_107_pad_type_0, strides = var_5862, weight = down_blocks_2_attentions_1_proj_out_weight_to_fp16_palettized, x = input_167_cast)[name = tensor("hidden_states_107_cast")]; + tensor input_169_cast = add(x = hidden_states_107_cast, y = hidden_states_95_cast)[name = tensor("input_169_cast")]; + tensor var_5871 = const()[name = tensor("op_5871"), val = tensor([2, 2])]; + tensor var_5873 = const()[name = tensor("op_5873"), val = tensor([1, 1])]; + tensor input_171_pad_type_0 = const()[name = tensor("input_171_pad_type_0"), val = tensor("custom")]; + tensor input_171_pad_0 = const()[name = tensor("input_171_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor down_blocks_2_downsamplers_0_conv_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(132869568))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(143928832))), name = tensor("down_blocks_2_downsamplers_0_conv_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 3, 3])]; + tensor down_blocks_2_downsamplers_0_conv_bias_to_fp16 = const()[name = tensor("down_blocks_2_downsamplers_0_conv_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(143929024)))]; + tensor input_171_cast = conv(bias = down_blocks_2_downsamplers_0_conv_bias_to_fp16, dilations = var_5873, groups = var_3883, pad = input_171_pad_0, pad_type = input_171_pad_type_0, strides = var_5871, weight = down_blocks_2_downsamplers_0_conv_weight_to_fp16_palettized, x = input_169_cast)[name = tensor("input_171_cast")]; + tensor var_5885 = const()[name = tensor("op_5885"), val = tensor(1)]; + tensor reshape_72_shape_0 = const()[name = tensor("reshape_72_shape_0"), val = tensor([2, 32, 40, 8, 8])]; + tensor reshape_72_cast = reshape(shape = reshape_72_shape_0, x = input_171_cast)[name = tensor("reshape_72_cast")]; + tensor reduce_mean_54_axes_0 = const()[name = tensor("reduce_mean_54_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_54_keep_dims_0 = const()[name = tensor("reduce_mean_54_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_54_cast = reduce_mean(axes = reduce_mean_54_axes_0, keep_dims = reduce_mean_54_keep_dims_0, x = reshape_72_cast)[name = tensor("reduce_mean_54_cast")]; + tensor sub_36_cast = sub(x = reshape_72_cast, y = reduce_mean_54_cast)[name = tensor("sub_36_cast")]; + tensor square_18_cast = square(x = sub_36_cast)[name = tensor("square_18_cast")]; + tensor reduce_mean_56_axes_0 = const()[name = tensor("reduce_mean_56_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_56_keep_dims_0 = const()[name = tensor("reduce_mean_56_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_56_cast = reduce_mean(axes = reduce_mean_56_axes_0, keep_dims = reduce_mean_56_keep_dims_0, x = square_18_cast)[name = tensor("reduce_mean_56_cast")]; + tensor add_36_y_0_to_fp16 = const()[name = tensor("add_36_y_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_36_cast = add(x = reduce_mean_56_cast, y = add_36_y_0_to_fp16)[name = tensor("add_36_cast")]; + tensor sqrt_18_cast = sqrt(x = add_36_cast)[name = tensor("sqrt_18_cast")]; + tensor real_div_18_cast = real_div(x = sub_36_cast, y = sqrt_18_cast)[name = tensor("real_div_18_cast")]; + tensor reshape_73_shape_0 = const()[name = tensor("reshape_73_shape_0"), val = tensor([2, 1280, 8, 8])]; + tensor reshape_73_cast = reshape(shape = reshape_73_shape_0, x = real_div_18_cast)[name = tensor("reshape_73_cast")]; + tensor add_37_gamma_0_to_fp16 = const()[name = tensor("add_37_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(143931648)))]; + tensor add_37_beta_0_to_fp16 = const()[name = tensor("add_37_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(143934272)))]; + tensor add_37_epsilon_0_to_fp16 = const()[name = tensor("add_37_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_37_cast = batch_norm(beta = add_37_beta_0_to_fp16, epsilon = add_37_epsilon_0_to_fp16, gamma = add_37_gamma_0_to_fp16, mean = add_27_mean_0_to_fp16, variance = add_27_variance_0_to_fp16, x = reshape_73_cast)[name = tensor("add_37_cast")]; + tensor input_175_cast = silu(x = add_37_cast)[name = tensor("input_175_cast")]; + tensor var_5901 = const()[name = tensor("op_5901"), val = tensor([1, 1])]; + tensor var_5903 = const()[name = tensor("op_5903"), val = tensor([1, 1])]; + tensor hidden_states_109_pad_type_0 = const()[name = tensor("hidden_states_109_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_109_pad_0 = const()[name = tensor("hidden_states_109_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor down_blocks_3_resnets_0_conv1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(143936896))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(154996160))), name = tensor("down_blocks_3_resnets_0_conv1_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 3, 3])]; + tensor down_blocks_3_resnets_0_conv1_bias_to_fp16 = const()[name = tensor("down_blocks_3_resnets_0_conv1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(154996352)))]; + tensor hidden_states_109_cast = conv(bias = down_blocks_3_resnets_0_conv1_bias_to_fp16, dilations = var_5903, groups = var_5885, pad = hidden_states_109_pad_0, pad_type = hidden_states_109_pad_type_0, strides = var_5901, weight = down_blocks_3_resnets_0_conv1_weight_to_fp16_palettized, x = input_175_cast)[name = tensor("hidden_states_109_cast")]; + tensor var_5909 = const()[name = tensor("op_5909"), val = tensor([1, 1])]; + tensor var_5911 = const()[name = tensor("op_5911"), val = tensor([1, 1])]; + tensor temb_13_pad_type_0 = const()[name = tensor("temb_13_pad_type_0"), val = tensor("custom")]; + tensor temb_13_pad_0 = const()[name = tensor("temb_13_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_3_resnets_0_time_emb_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(154998976))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(156227840))), name = tensor("down_blocks_3_resnets_0_time_emb_proj_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor down_blocks_3_resnets_0_time_emb_proj_bias_to_fp16 = const()[name = tensor("down_blocks_3_resnets_0_time_emb_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(156228032)))]; + tensor temb_13_cast = conv(bias = down_blocks_3_resnets_0_time_emb_proj_bias_to_fp16, dilations = var_5911, groups = var_5885, pad = temb_13_pad_0, pad_type = temb_13_pad_type_0, strides = var_5909, weight = down_blocks_3_resnets_0_time_emb_proj_weight_to_fp16_palettized, x = input_15_cast)[name = tensor("temb_13_cast")]; + tensor input_179_cast = add(x = hidden_states_109_cast, y = temb_13_cast)[name = tensor("input_179_cast")]; + tensor reshape_76_shape_0 = const()[name = tensor("reshape_76_shape_0"), val = tensor([2, 32, 40, 8, 8])]; + tensor reshape_76_cast = reshape(shape = reshape_76_shape_0, x = input_179_cast)[name = tensor("reshape_76_cast")]; + tensor reduce_mean_57_axes_0 = const()[name = tensor("reduce_mean_57_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_57_keep_dims_0 = const()[name = tensor("reduce_mean_57_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_57_cast = reduce_mean(axes = reduce_mean_57_axes_0, keep_dims = reduce_mean_57_keep_dims_0, x = reshape_76_cast)[name = tensor("reduce_mean_57_cast")]; + tensor sub_38_cast = sub(x = reshape_76_cast, y = reduce_mean_57_cast)[name = tensor("sub_38_cast")]; + tensor square_19_cast = square(x = sub_38_cast)[name = tensor("square_19_cast")]; + tensor reduce_mean_59_axes_0 = const()[name = tensor("reduce_mean_59_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_59_keep_dims_0 = const()[name = tensor("reduce_mean_59_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_59_cast = reduce_mean(axes = reduce_mean_59_axes_0, keep_dims = reduce_mean_59_keep_dims_0, x = square_19_cast)[name = tensor("reduce_mean_59_cast")]; + tensor add_38_y_0_to_fp16 = const()[name = tensor("add_38_y_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_38_cast = add(x = reduce_mean_59_cast, y = add_38_y_0_to_fp16)[name = tensor("add_38_cast")]; + tensor sqrt_19_cast = sqrt(x = add_38_cast)[name = tensor("sqrt_19_cast")]; + tensor real_div_19_cast = real_div(x = sub_38_cast, y = sqrt_19_cast)[name = tensor("real_div_19_cast")]; + tensor reshape_77_shape_0 = const()[name = tensor("reshape_77_shape_0"), val = tensor([2, 1280, 8, 8])]; + tensor reshape_77_cast = reshape(shape = reshape_77_shape_0, x = real_div_19_cast)[name = tensor("reshape_77_cast")]; + tensor add_39_gamma_0_to_fp16 = const()[name = tensor("add_39_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(156230656)))]; + tensor add_39_beta_0_to_fp16 = const()[name = tensor("add_39_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(156233280)))]; + tensor add_39_epsilon_0_to_fp16 = const()[name = tensor("add_39_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_39_cast = batch_norm(beta = add_39_beta_0_to_fp16, epsilon = add_39_epsilon_0_to_fp16, gamma = add_39_gamma_0_to_fp16, mean = add_27_mean_0_to_fp16, variance = add_27_variance_0_to_fp16, x = reshape_77_cast)[name = tensor("add_39_cast")]; + tensor input_183_cast = silu(x = add_39_cast)[name = tensor("input_183_cast")]; + tensor var_5921 = const()[name = tensor("op_5921"), val = tensor([1, 1])]; + tensor var_5923 = const()[name = tensor("op_5923"), val = tensor([1, 1])]; + tensor hidden_states_111_pad_type_0 = const()[name = tensor("hidden_states_111_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_111_pad_0 = const()[name = tensor("hidden_states_111_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor down_blocks_3_resnets_0_conv2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(156235904))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(167295168))), name = tensor("down_blocks_3_resnets_0_conv2_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 3, 3])]; + tensor down_blocks_3_resnets_0_conv2_bias_to_fp16 = const()[name = tensor("down_blocks_3_resnets_0_conv2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(167295360)))]; + tensor hidden_states_111_cast = conv(bias = down_blocks_3_resnets_0_conv2_bias_to_fp16, dilations = var_5923, groups = var_5885, pad = hidden_states_111_pad_0, pad_type = hidden_states_111_pad_type_0, strides = var_5921, weight = down_blocks_3_resnets_0_conv2_weight_to_fp16_palettized, x = input_183_cast)[name = tensor("hidden_states_111_cast")]; + tensor input_185_cast = add(x = input_171_cast, y = hidden_states_111_cast)[name = tensor("input_185_cast")]; + tensor reshape_80_shape_0 = const()[name = tensor("reshape_80_shape_0"), val = tensor([2, 32, 40, 8, 8])]; + tensor reshape_80_cast = reshape(shape = reshape_80_shape_0, x = input_185_cast)[name = tensor("reshape_80_cast")]; + tensor reduce_mean_60_axes_0 = const()[name = tensor("reduce_mean_60_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_60_keep_dims_0 = const()[name = tensor("reduce_mean_60_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_60_cast = reduce_mean(axes = reduce_mean_60_axes_0, keep_dims = reduce_mean_60_keep_dims_0, x = reshape_80_cast)[name = tensor("reduce_mean_60_cast")]; + tensor sub_40_cast = sub(x = reshape_80_cast, y = reduce_mean_60_cast)[name = tensor("sub_40_cast")]; + tensor square_20_cast = square(x = sub_40_cast)[name = tensor("square_20_cast")]; + tensor reduce_mean_62_axes_0 = const()[name = tensor("reduce_mean_62_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_62_keep_dims_0 = const()[name = tensor("reduce_mean_62_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_62_cast = reduce_mean(axes = reduce_mean_62_axes_0, keep_dims = reduce_mean_62_keep_dims_0, x = square_20_cast)[name = tensor("reduce_mean_62_cast")]; + tensor add_40_y_0_to_fp16 = const()[name = tensor("add_40_y_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_40_cast = add(x = reduce_mean_62_cast, y = add_40_y_0_to_fp16)[name = tensor("add_40_cast")]; + tensor sqrt_20_cast = sqrt(x = add_40_cast)[name = tensor("sqrt_20_cast")]; + tensor real_div_20_cast = real_div(x = sub_40_cast, y = sqrt_20_cast)[name = tensor("real_div_20_cast")]; + tensor reshape_81_shape_0 = const()[name = tensor("reshape_81_shape_0"), val = tensor([2, 1280, 8, 8])]; + tensor reshape_81_cast = reshape(shape = reshape_81_shape_0, x = real_div_20_cast)[name = tensor("reshape_81_cast")]; + tensor add_41_gamma_0_to_fp16 = const()[name = tensor("add_41_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(167297984)))]; + tensor add_41_beta_0_to_fp16 = const()[name = tensor("add_41_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(167300608)))]; + tensor add_41_epsilon_0_to_fp16 = const()[name = tensor("add_41_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_41_cast = batch_norm(beta = add_41_beta_0_to_fp16, epsilon = add_41_epsilon_0_to_fp16, gamma = add_41_gamma_0_to_fp16, mean = add_27_mean_0_to_fp16, variance = add_27_variance_0_to_fp16, x = reshape_81_cast)[name = tensor("add_41_cast")]; + tensor input_189_cast = silu(x = add_41_cast)[name = tensor("input_189_cast")]; + tensor var_5938 = const()[name = tensor("op_5938"), val = tensor([1, 1])]; + tensor var_5940 = const()[name = tensor("op_5940"), val = tensor([1, 1])]; + tensor hidden_states_113_pad_type_0 = const()[name = tensor("hidden_states_113_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_113_pad_0 = const()[name = tensor("hidden_states_113_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor down_blocks_3_resnets_1_conv1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(167303232))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(178362496))), name = tensor("down_blocks_3_resnets_1_conv1_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 3, 3])]; + tensor down_blocks_3_resnets_1_conv1_bias_to_fp16 = const()[name = tensor("down_blocks_3_resnets_1_conv1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(178362688)))]; + tensor hidden_states_113_cast = conv(bias = down_blocks_3_resnets_1_conv1_bias_to_fp16, dilations = var_5940, groups = var_5885, pad = hidden_states_113_pad_0, pad_type = hidden_states_113_pad_type_0, strides = var_5938, weight = down_blocks_3_resnets_1_conv1_weight_to_fp16_palettized, x = input_189_cast)[name = tensor("hidden_states_113_cast")]; + tensor var_5946 = const()[name = tensor("op_5946"), val = tensor([1, 1])]; + tensor var_5948 = const()[name = tensor("op_5948"), val = tensor([1, 1])]; + tensor temb_15_pad_type_0 = const()[name = tensor("temb_15_pad_type_0"), val = tensor("custom")]; + tensor temb_15_pad_0 = const()[name = tensor("temb_15_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor down_blocks_3_resnets_1_time_emb_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(178365312))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(179594176))), name = tensor("down_blocks_3_resnets_1_time_emb_proj_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor down_blocks_3_resnets_1_time_emb_proj_bias_to_fp16 = const()[name = tensor("down_blocks_3_resnets_1_time_emb_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(179594368)))]; + tensor temb_15_cast = conv(bias = down_blocks_3_resnets_1_time_emb_proj_bias_to_fp16, dilations = var_5948, groups = var_5885, pad = temb_15_pad_0, pad_type = temb_15_pad_type_0, strides = var_5946, weight = down_blocks_3_resnets_1_time_emb_proj_weight_to_fp16_palettized, x = input_15_cast)[name = tensor("temb_15_cast")]; + tensor input_193_cast = add(x = hidden_states_113_cast, y = temb_15_cast)[name = tensor("input_193_cast")]; + tensor reshape_84_shape_0 = const()[name = tensor("reshape_84_shape_0"), val = tensor([2, 32, 40, 8, 8])]; + tensor reshape_84_cast = reshape(shape = reshape_84_shape_0, x = input_193_cast)[name = tensor("reshape_84_cast")]; + tensor reduce_mean_63_axes_0 = const()[name = tensor("reduce_mean_63_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_63_keep_dims_0 = const()[name = tensor("reduce_mean_63_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_63_cast = reduce_mean(axes = reduce_mean_63_axes_0, keep_dims = reduce_mean_63_keep_dims_0, x = reshape_84_cast)[name = tensor("reduce_mean_63_cast")]; + tensor sub_42_cast = sub(x = reshape_84_cast, y = reduce_mean_63_cast)[name = tensor("sub_42_cast")]; + tensor square_21_cast = square(x = sub_42_cast)[name = tensor("square_21_cast")]; + tensor reduce_mean_65_axes_0 = const()[name = tensor("reduce_mean_65_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_65_keep_dims_0 = const()[name = tensor("reduce_mean_65_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_65_cast = reduce_mean(axes = reduce_mean_65_axes_0, keep_dims = reduce_mean_65_keep_dims_0, x = square_21_cast)[name = tensor("reduce_mean_65_cast")]; + tensor add_42_y_0_to_fp16 = const()[name = tensor("add_42_y_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_42_cast = add(x = reduce_mean_65_cast, y = add_42_y_0_to_fp16)[name = tensor("add_42_cast")]; + tensor sqrt_21_cast = sqrt(x = add_42_cast)[name = tensor("sqrt_21_cast")]; + tensor real_div_21_cast = real_div(x = sub_42_cast, y = sqrt_21_cast)[name = tensor("real_div_21_cast")]; + tensor reshape_85_shape_0 = const()[name = tensor("reshape_85_shape_0"), val = tensor([2, 1280, 8, 8])]; + tensor reshape_85_cast = reshape(shape = reshape_85_shape_0, x = real_div_21_cast)[name = tensor("reshape_85_cast")]; + tensor add_43_gamma_0_to_fp16 = const()[name = tensor("add_43_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(179596992)))]; + tensor add_43_beta_0_to_fp16 = const()[name = tensor("add_43_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(179599616)))]; + tensor add_43_epsilon_0_to_fp16 = const()[name = tensor("add_43_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_43_cast = batch_norm(beta = add_43_beta_0_to_fp16, epsilon = add_43_epsilon_0_to_fp16, gamma = add_43_gamma_0_to_fp16, mean = add_27_mean_0_to_fp16, variance = add_27_variance_0_to_fp16, x = reshape_85_cast)[name = tensor("add_43_cast")]; + tensor input_197_cast = silu(x = add_43_cast)[name = tensor("input_197_cast")]; + tensor var_5958 = const()[name = tensor("op_5958"), val = tensor([1, 1])]; + tensor var_5960 = const()[name = tensor("op_5960"), val = tensor([1, 1])]; + tensor hidden_states_115_pad_type_0 = const()[name = tensor("hidden_states_115_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_115_pad_0 = const()[name = tensor("hidden_states_115_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor down_blocks_3_resnets_1_conv2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(179602240))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(190661504))), name = tensor("down_blocks_3_resnets_1_conv2_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 3, 3])]; + tensor down_blocks_3_resnets_1_conv2_bias_to_fp16 = const()[name = tensor("down_blocks_3_resnets_1_conv2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(190661696)))]; + tensor hidden_states_115_cast = conv(bias = down_blocks_3_resnets_1_conv2_bias_to_fp16, dilations = var_5960, groups = var_5885, pad = hidden_states_115_pad_0, pad_type = hidden_states_115_pad_type_0, strides = var_5958, weight = down_blocks_3_resnets_1_conv2_weight_to_fp16_palettized, x = input_197_cast)[name = tensor("hidden_states_115_cast")]; + tensor input_199_cast = add(x = input_185_cast, y = hidden_states_115_cast)[name = tensor("input_199_cast")]; + tensor var_5995 = const()[name = tensor("op_5995"), val = tensor(true)]; + tensor var_6000 = const()[name = tensor("op_6000"), val = tensor(1)]; + tensor reshape_88_shape_0 = const()[name = tensor("reshape_88_shape_0"), val = tensor([2, 32, 40, 8, 8])]; + tensor reshape_88_cast = reshape(shape = reshape_88_shape_0, x = input_199_cast)[name = tensor("reshape_88_cast")]; + tensor reduce_mean_66_axes_0 = const()[name = tensor("reduce_mean_66_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_66_keep_dims_0 = const()[name = tensor("reduce_mean_66_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_66_cast = reduce_mean(axes = reduce_mean_66_axes_0, keep_dims = reduce_mean_66_keep_dims_0, x = reshape_88_cast)[name = tensor("reduce_mean_66_cast")]; + tensor sub_44_cast = sub(x = reshape_88_cast, y = reduce_mean_66_cast)[name = tensor("sub_44_cast")]; + tensor square_22_cast = square(x = sub_44_cast)[name = tensor("square_22_cast")]; + tensor reduce_mean_68_axes_0 = const()[name = tensor("reduce_mean_68_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_68_keep_dims_0 = const()[name = tensor("reduce_mean_68_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_68_cast = reduce_mean(axes = reduce_mean_68_axes_0, keep_dims = reduce_mean_68_keep_dims_0, x = square_22_cast)[name = tensor("reduce_mean_68_cast")]; + tensor add_44_y_0_to_fp16 = const()[name = tensor("add_44_y_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_44_cast = add(x = reduce_mean_68_cast, y = add_44_y_0_to_fp16)[name = tensor("add_44_cast")]; + tensor sqrt_22_cast = sqrt(x = add_44_cast)[name = tensor("sqrt_22_cast")]; + tensor real_div_22_cast = real_div(x = sub_44_cast, y = sqrt_22_cast)[name = tensor("real_div_22_cast")]; + tensor reshape_89_shape_0 = const()[name = tensor("reshape_89_shape_0"), val = tensor([2, 1280, 8, 8])]; + tensor reshape_89_cast = reshape(shape = reshape_89_shape_0, x = real_div_22_cast)[name = tensor("reshape_89_cast")]; + tensor add_45_gamma_0_to_fp16 = const()[name = tensor("add_45_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(190664320)))]; + tensor add_45_beta_0_to_fp16 = const()[name = tensor("add_45_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(190666944)))]; + tensor add_45_epsilon_0_to_fp16 = const()[name = tensor("add_45_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_45_cast = batch_norm(beta = add_45_beta_0_to_fp16, epsilon = add_45_epsilon_0_to_fp16, gamma = add_45_gamma_0_to_fp16, mean = add_27_mean_0_to_fp16, variance = add_27_variance_0_to_fp16, x = reshape_89_cast)[name = tensor("add_45_cast")]; + tensor input_203_cast = silu(x = add_45_cast)[name = tensor("input_203_cast")]; + tensor var_6018 = const()[name = tensor("op_6018"), val = tensor([1, 1])]; + tensor var_6020 = const()[name = tensor("op_6020"), val = tensor([1, 1])]; + tensor hidden_states_117_pad_type_0 = const()[name = tensor("hidden_states_117_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_117_pad_0 = const()[name = tensor("hidden_states_117_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor mid_block_resnets_0_conv1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(190669568))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(201728832))), name = tensor("mid_block_resnets_0_conv1_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 3, 3])]; + tensor mid_block_resnets_0_conv1_bias_to_fp16 = const()[name = tensor("mid_block_resnets_0_conv1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(201729024)))]; + tensor hidden_states_117_cast = conv(bias = mid_block_resnets_0_conv1_bias_to_fp16, dilations = var_6020, groups = var_6000, pad = hidden_states_117_pad_0, pad_type = hidden_states_117_pad_type_0, strides = var_6018, weight = mid_block_resnets_0_conv1_weight_to_fp16_palettized, x = input_203_cast)[name = tensor("hidden_states_117_cast")]; + tensor var_6026 = const()[name = tensor("op_6026"), val = tensor([1, 1])]; + tensor var_6028 = const()[name = tensor("op_6028"), val = tensor([1, 1])]; + tensor temb_17_pad_type_0 = const()[name = tensor("temb_17_pad_type_0"), val = tensor("custom")]; + tensor temb_17_pad_0 = const()[name = tensor("temb_17_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor mid_block_resnets_0_time_emb_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(201731648))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(202960512))), name = tensor("mid_block_resnets_0_time_emb_proj_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor mid_block_resnets_0_time_emb_proj_bias_to_fp16 = const()[name = tensor("mid_block_resnets_0_time_emb_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(202960704)))]; + tensor temb_17_cast = conv(bias = mid_block_resnets_0_time_emb_proj_bias_to_fp16, dilations = var_6028, groups = var_6000, pad = temb_17_pad_0, pad_type = temb_17_pad_type_0, strides = var_6026, weight = mid_block_resnets_0_time_emb_proj_weight_to_fp16_palettized, x = input_15_cast)[name = tensor("temb_17_cast")]; + tensor input_207_cast = add(x = hidden_states_117_cast, y = temb_17_cast)[name = tensor("input_207_cast")]; + tensor reshape_92_shape_0 = const()[name = tensor("reshape_92_shape_0"), val = tensor([2, 32, 40, 8, 8])]; + tensor reshape_92_cast = reshape(shape = reshape_92_shape_0, x = input_207_cast)[name = tensor("reshape_92_cast")]; + tensor reduce_mean_69_axes_0 = const()[name = tensor("reduce_mean_69_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_69_keep_dims_0 = const()[name = tensor("reduce_mean_69_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_69_cast = reduce_mean(axes = reduce_mean_69_axes_0, keep_dims = reduce_mean_69_keep_dims_0, x = reshape_92_cast)[name = tensor("reduce_mean_69_cast")]; + tensor sub_46_cast = sub(x = reshape_92_cast, y = reduce_mean_69_cast)[name = tensor("sub_46_cast")]; + tensor square_23_cast = square(x = sub_46_cast)[name = tensor("square_23_cast")]; + tensor reduce_mean_71_axes_0 = const()[name = tensor("reduce_mean_71_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_71_keep_dims_0 = const()[name = tensor("reduce_mean_71_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_71_cast = reduce_mean(axes = reduce_mean_71_axes_0, keep_dims = reduce_mean_71_keep_dims_0, x = square_23_cast)[name = tensor("reduce_mean_71_cast")]; + tensor add_46_y_0_to_fp16 = const()[name = tensor("add_46_y_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_46_cast = add(x = reduce_mean_71_cast, y = add_46_y_0_to_fp16)[name = tensor("add_46_cast")]; + tensor sqrt_23_cast = sqrt(x = add_46_cast)[name = tensor("sqrt_23_cast")]; + tensor real_div_23_cast = real_div(x = sub_46_cast, y = sqrt_23_cast)[name = tensor("real_div_23_cast")]; + tensor reshape_93_shape_0 = const()[name = tensor("reshape_93_shape_0"), val = tensor([2, 1280, 8, 8])]; + tensor reshape_93_cast = reshape(shape = reshape_93_shape_0, x = real_div_23_cast)[name = tensor("reshape_93_cast")]; + tensor add_47_gamma_0_to_fp16 = const()[name = tensor("add_47_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(202963328)))]; + tensor add_47_beta_0_to_fp16 = const()[name = tensor("add_47_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(202965952)))]; + tensor add_47_epsilon_0_to_fp16 = const()[name = tensor("add_47_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_47_cast = batch_norm(beta = add_47_beta_0_to_fp16, epsilon = add_47_epsilon_0_to_fp16, gamma = add_47_gamma_0_to_fp16, mean = add_27_mean_0_to_fp16, variance = add_27_variance_0_to_fp16, x = reshape_93_cast)[name = tensor("add_47_cast")]; + tensor input_211_cast = silu(x = add_47_cast)[name = tensor("input_211_cast")]; + tensor var_6038 = const()[name = tensor("op_6038"), val = tensor([1, 1])]; + tensor var_6040 = const()[name = tensor("op_6040"), val = tensor([1, 1])]; + tensor hidden_states_119_pad_type_0 = const()[name = tensor("hidden_states_119_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_119_pad_0 = const()[name = tensor("hidden_states_119_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor mid_block_resnets_0_conv2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(202968576))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(214027840))), name = tensor("mid_block_resnets_0_conv2_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 3, 3])]; + tensor mid_block_resnets_0_conv2_bias_to_fp16 = const()[name = tensor("mid_block_resnets_0_conv2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(214028032)))]; + tensor hidden_states_119_cast = conv(bias = mid_block_resnets_0_conv2_bias_to_fp16, dilations = var_6040, groups = var_6000, pad = hidden_states_119_pad_0, pad_type = hidden_states_119_pad_type_0, strides = var_6038, weight = mid_block_resnets_0_conv2_weight_to_fp16_palettized, x = input_211_cast)[name = tensor("hidden_states_119_cast")]; + tensor hidden_states_121_cast = add(x = input_199_cast, y = hidden_states_119_cast)[name = tensor("hidden_states_121_cast")]; + tensor reshape_96_shape_0 = const()[name = tensor("reshape_96_shape_0"), val = tensor([2, 32, 40, 8, 8])]; + tensor reshape_96_cast = reshape(shape = reshape_96_shape_0, x = hidden_states_121_cast)[name = tensor("reshape_96_cast")]; + tensor reduce_mean_72_axes_0 = const()[name = tensor("reduce_mean_72_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_72_keep_dims_0 = const()[name = tensor("reduce_mean_72_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_72_cast = reduce_mean(axes = reduce_mean_72_axes_0, keep_dims = reduce_mean_72_keep_dims_0, x = reshape_96_cast)[name = tensor("reduce_mean_72_cast")]; + tensor sub_48_cast = sub(x = reshape_96_cast, y = reduce_mean_72_cast)[name = tensor("sub_48_cast")]; + tensor square_24_cast = square(x = sub_48_cast)[name = tensor("square_24_cast")]; + tensor reduce_mean_74_axes_0 = const()[name = tensor("reduce_mean_74_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_74_keep_dims_0 = const()[name = tensor("reduce_mean_74_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_74_cast = reduce_mean(axes = reduce_mean_74_axes_0, keep_dims = reduce_mean_74_keep_dims_0, x = square_24_cast)[name = tensor("reduce_mean_74_cast")]; + tensor add_48_y_0_to_fp16 = const()[name = tensor("add_48_y_0_to_fp16"), val = tensor(0x1.1p-20)]; + tensor add_48_cast = add(x = reduce_mean_74_cast, y = add_48_y_0_to_fp16)[name = tensor("add_48_cast")]; + tensor sqrt_24_cast = sqrt(x = add_48_cast)[name = tensor("sqrt_24_cast")]; + tensor real_div_24_cast = real_div(x = sub_48_cast, y = sqrt_24_cast)[name = tensor("real_div_24_cast")]; + tensor reshape_97_shape_0 = const()[name = tensor("reshape_97_shape_0"), val = tensor([2, 1280, 8, 8])]; + tensor reshape_97_cast = reshape(shape = reshape_97_shape_0, x = real_div_24_cast)[name = tensor("reshape_97_cast")]; + tensor add_49_gamma_0_to_fp16 = const()[name = tensor("add_49_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(214030656)))]; + tensor add_49_beta_0_to_fp16 = const()[name = tensor("add_49_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(214033280)))]; + tensor add_49_epsilon_0_to_fp16 = const()[name = tensor("add_49_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_49_cast = batch_norm(beta = add_49_beta_0_to_fp16, epsilon = add_49_epsilon_0_to_fp16, gamma = add_49_gamma_0_to_fp16, mean = add_27_mean_0_to_fp16, variance = add_27_variance_0_to_fp16, x = reshape_97_cast)[name = tensor("add_49_cast")]; + tensor var_6060 = const()[name = tensor("op_6060"), val = tensor([1, 1])]; + tensor var_6062 = const()[name = tensor("op_6062"), val = tensor([1, 1])]; + tensor hidden_states_123_pad_type_0 = const()[name = tensor("hidden_states_123_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_123_pad_0 = const()[name = tensor("hidden_states_123_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor mid_block_attentions_0_proj_in_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(214035904))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(215264768))), name = tensor("mid_block_attentions_0_proj_in_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor mid_block_attentions_0_proj_in_bias_to_fp16 = const()[name = tensor("mid_block_attentions_0_proj_in_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(215264960)))]; + tensor hidden_states_123_cast = conv(bias = mid_block_attentions_0_proj_in_bias_to_fp16, dilations = var_6062, groups = var_6000, pad = hidden_states_123_pad_0, pad_type = hidden_states_123_pad_type_0, strides = var_6060, weight = mid_block_attentions_0_proj_in_weight_to_fp16_palettized, x = add_49_cast)[name = tensor("hidden_states_123_cast")]; + tensor var_6067 = const()[name = tensor("op_6067"), val = tensor([2, 1280, 1, 64])]; + tensor inputs_37_cast = reshape(shape = var_6067, x = hidden_states_123_cast)[name = tensor("inputs_37_cast")]; + tensor var_6077 = const()[name = tensor("op_6077"), val = tensor([1])]; + tensor channels_mean_37_cast = reduce_mean(axes = var_6077, keep_dims = var_5995, x = inputs_37_cast)[name = tensor("channels_mean_37_cast")]; + tensor zero_mean_37_cast = sub(x = inputs_37_cast, y = channels_mean_37_cast)[name = tensor("zero_mean_37_cast")]; + tensor zero_mean_sq_37_cast = mul(x = zero_mean_37_cast, y = zero_mean_37_cast)[name = tensor("zero_mean_sq_37_cast")]; + tensor var_6081 = const()[name = tensor("op_6081"), val = tensor([1])]; + tensor var_6082_cast = reduce_mean(axes = var_6081, keep_dims = var_5995, x = zero_mean_sq_37_cast)[name = tensor("op_6082_cast")]; + tensor var_6083_to_fp16 = const()[name = tensor("op_6083_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_6084_cast = add(x = var_6082_cast, y = var_6083_to_fp16)[name = tensor("op_6084_cast")]; + tensor denom_37_epsilon_0_to_fp16 = const()[name = tensor("denom_37_epsilon_0_to_fp16"), val = tensor(0x1p-24)]; + tensor denom_37_cast = rsqrt(epsilon = denom_37_epsilon_0_to_fp16, x = var_6084_cast)[name = tensor("denom_37_cast")]; + tensor out_37_cast = mul(x = zero_mean_37_cast, y = denom_37_cast)[name = tensor("out_37_cast")]; + tensor var_6088_to_fp16 = const()[name = tensor("op_6088_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(215267584)))]; + tensor var_6089_cast = add(x = out_37_cast, y = var_6088_to_fp16)[name = tensor("op_6089_cast")]; + tensor var_6091_to_fp16 = const()[name = tensor("op_6091_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(215270208)))]; + tensor hidden_states_125_cast = mul(x = var_6089_cast, y = var_6091_to_fp16)[name = tensor("hidden_states_125_cast")]; + tensor var_6098 = const()[name = tensor("op_6098"), val = tensor([1, 1])]; + tensor var_6100 = const()[name = tensor("op_6100"), val = tensor([1, 1])]; + tensor q_25_pad_type_0 = const()[name = tensor("q_25_pad_type_0"), val = tensor("custom")]; + tensor q_25_pad_0 = const()[name = tensor("q_25_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor mid_block_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(215272832))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(216501696))), name = tensor("mid_block_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor q_25_cast = conv(dilations = var_6100, groups = var_6000, pad = q_25_pad_0, pad_type = q_25_pad_type_0, strides = var_6098, weight = mid_block_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16_palettized, x = hidden_states_125_cast)[name = tensor("q_25_cast")]; + tensor var_6104 = const()[name = tensor("op_6104"), val = tensor([1, 1])]; + tensor var_6106 = const()[name = tensor("op_6106"), val = tensor([1, 1])]; + tensor k_49_pad_type_0 = const()[name = tensor("k_49_pad_type_0"), val = tensor("custom")]; + tensor k_49_pad_0 = const()[name = tensor("k_49_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor mid_block_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(216501888))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(217730752))), name = tensor("mid_block_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor k_49_cast = conv(dilations = var_6106, groups = var_6000, pad = k_49_pad_0, pad_type = k_49_pad_type_0, strides = var_6104, weight = mid_block_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16_palettized, x = hidden_states_125_cast)[name = tensor("k_49_cast")]; + tensor var_6110 = const()[name = tensor("op_6110"), val = tensor([1, 1])]; + tensor var_6112 = const()[name = tensor("op_6112"), val = tensor([1, 1])]; + tensor v_25_pad_type_0 = const()[name = tensor("v_25_pad_type_0"), val = tensor("custom")]; + tensor v_25_pad_0 = const()[name = tensor("v_25_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor mid_block_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(217730944))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(218959808))), name = tensor("mid_block_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor v_25_cast = conv(dilations = var_6112, groups = var_6000, pad = v_25_pad_0, pad_type = v_25_pad_type_0, strides = var_6110, weight = mid_block_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16_palettized, x = hidden_states_125_cast)[name = tensor("v_25_cast")]; + tensor var_6116_begin_0 = const()[name = tensor("op_6116_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6116_end_0 = const()[name = tensor("op_6116_end_0"), val = tensor([2, 64, 1, 64])]; + tensor var_6116_end_mask_0 = const()[name = tensor("op_6116_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6116_cast = slice_by_index(begin = var_6116_begin_0, end = var_6116_end_0, end_mask = var_6116_end_mask_0, x = q_25_cast)[name = tensor("op_6116_cast")]; + tensor var_6120_begin_0 = const()[name = tensor("op_6120_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_6120_end_0 = const()[name = tensor("op_6120_end_0"), val = tensor([2, 128, 1, 64])]; + tensor var_6120_end_mask_0 = const()[name = tensor("op_6120_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6120_cast = slice_by_index(begin = var_6120_begin_0, end = var_6120_end_0, end_mask = var_6120_end_mask_0, x = q_25_cast)[name = tensor("op_6120_cast")]; + tensor var_6124_begin_0 = const()[name = tensor("op_6124_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_6124_end_0 = const()[name = tensor("op_6124_end_0"), val = tensor([2, 192, 1, 64])]; + tensor var_6124_end_mask_0 = const()[name = tensor("op_6124_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6124_cast = slice_by_index(begin = var_6124_begin_0, end = var_6124_end_0, end_mask = var_6124_end_mask_0, x = q_25_cast)[name = tensor("op_6124_cast")]; + tensor var_6128_begin_0 = const()[name = tensor("op_6128_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_6128_end_0 = const()[name = tensor("op_6128_end_0"), val = tensor([2, 256, 1, 64])]; + tensor var_6128_end_mask_0 = const()[name = tensor("op_6128_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6128_cast = slice_by_index(begin = var_6128_begin_0, end = var_6128_end_0, end_mask = var_6128_end_mask_0, x = q_25_cast)[name = tensor("op_6128_cast")]; + tensor var_6132_begin_0 = const()[name = tensor("op_6132_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_6132_end_0 = const()[name = tensor("op_6132_end_0"), val = tensor([2, 320, 1, 64])]; + tensor var_6132_end_mask_0 = const()[name = tensor("op_6132_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6132_cast = slice_by_index(begin = var_6132_begin_0, end = var_6132_end_0, end_mask = var_6132_end_mask_0, x = q_25_cast)[name = tensor("op_6132_cast")]; + tensor var_6136_begin_0 = const()[name = tensor("op_6136_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_6136_end_0 = const()[name = tensor("op_6136_end_0"), val = tensor([2, 384, 1, 64])]; + tensor var_6136_end_mask_0 = const()[name = tensor("op_6136_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6136_cast = slice_by_index(begin = var_6136_begin_0, end = var_6136_end_0, end_mask = var_6136_end_mask_0, x = q_25_cast)[name = tensor("op_6136_cast")]; + tensor var_6140_begin_0 = const()[name = tensor("op_6140_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_6140_end_0 = const()[name = tensor("op_6140_end_0"), val = tensor([2, 448, 1, 64])]; + tensor var_6140_end_mask_0 = const()[name = tensor("op_6140_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6140_cast = slice_by_index(begin = var_6140_begin_0, end = var_6140_end_0, end_mask = var_6140_end_mask_0, x = q_25_cast)[name = tensor("op_6140_cast")]; + tensor var_6144_begin_0 = const()[name = tensor("op_6144_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_6144_end_0 = const()[name = tensor("op_6144_end_0"), val = tensor([2, 512, 1, 64])]; + tensor var_6144_end_mask_0 = const()[name = tensor("op_6144_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6144_cast = slice_by_index(begin = var_6144_begin_0, end = var_6144_end_0, end_mask = var_6144_end_mask_0, x = q_25_cast)[name = tensor("op_6144_cast")]; + tensor var_6148_begin_0 = const()[name = tensor("op_6148_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_6148_end_0 = const()[name = tensor("op_6148_end_0"), val = tensor([2, 576, 1, 64])]; + tensor var_6148_end_mask_0 = const()[name = tensor("op_6148_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6148_cast = slice_by_index(begin = var_6148_begin_0, end = var_6148_end_0, end_mask = var_6148_end_mask_0, x = q_25_cast)[name = tensor("op_6148_cast")]; + tensor var_6152_begin_0 = const()[name = tensor("op_6152_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_6152_end_0 = const()[name = tensor("op_6152_end_0"), val = tensor([2, 640, 1, 64])]; + tensor var_6152_end_mask_0 = const()[name = tensor("op_6152_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6152_cast = slice_by_index(begin = var_6152_begin_0, end = var_6152_end_0, end_mask = var_6152_end_mask_0, x = q_25_cast)[name = tensor("op_6152_cast")]; + tensor var_6156_begin_0 = const()[name = tensor("op_6156_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_6156_end_0 = const()[name = tensor("op_6156_end_0"), val = tensor([2, 704, 1, 64])]; + tensor var_6156_end_mask_0 = const()[name = tensor("op_6156_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6156_cast = slice_by_index(begin = var_6156_begin_0, end = var_6156_end_0, end_mask = var_6156_end_mask_0, x = q_25_cast)[name = tensor("op_6156_cast")]; + tensor var_6160_begin_0 = const()[name = tensor("op_6160_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_6160_end_0 = const()[name = tensor("op_6160_end_0"), val = tensor([2, 768, 1, 64])]; + tensor var_6160_end_mask_0 = const()[name = tensor("op_6160_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6160_cast = slice_by_index(begin = var_6160_begin_0, end = var_6160_end_0, end_mask = var_6160_end_mask_0, x = q_25_cast)[name = tensor("op_6160_cast")]; + tensor var_6164_begin_0 = const()[name = tensor("op_6164_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_6164_end_0 = const()[name = tensor("op_6164_end_0"), val = tensor([2, 832, 1, 64])]; + tensor var_6164_end_mask_0 = const()[name = tensor("op_6164_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6164_cast = slice_by_index(begin = var_6164_begin_0, end = var_6164_end_0, end_mask = var_6164_end_mask_0, x = q_25_cast)[name = tensor("op_6164_cast")]; + tensor var_6168_begin_0 = const()[name = tensor("op_6168_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_6168_end_0 = const()[name = tensor("op_6168_end_0"), val = tensor([2, 896, 1, 64])]; + tensor var_6168_end_mask_0 = const()[name = tensor("op_6168_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6168_cast = slice_by_index(begin = var_6168_begin_0, end = var_6168_end_0, end_mask = var_6168_end_mask_0, x = q_25_cast)[name = tensor("op_6168_cast")]; + tensor var_6172_begin_0 = const()[name = tensor("op_6172_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_6172_end_0 = const()[name = tensor("op_6172_end_0"), val = tensor([2, 960, 1, 64])]; + tensor var_6172_end_mask_0 = const()[name = tensor("op_6172_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6172_cast = slice_by_index(begin = var_6172_begin_0, end = var_6172_end_0, end_mask = var_6172_end_mask_0, x = q_25_cast)[name = tensor("op_6172_cast")]; + tensor var_6176_begin_0 = const()[name = tensor("op_6176_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_6176_end_0 = const()[name = tensor("op_6176_end_0"), val = tensor([2, 1024, 1, 64])]; + tensor var_6176_end_mask_0 = const()[name = tensor("op_6176_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6176_cast = slice_by_index(begin = var_6176_begin_0, end = var_6176_end_0, end_mask = var_6176_end_mask_0, x = q_25_cast)[name = tensor("op_6176_cast")]; + tensor var_6180_begin_0 = const()[name = tensor("op_6180_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_6180_end_0 = const()[name = tensor("op_6180_end_0"), val = tensor([2, 1088, 1, 64])]; + tensor var_6180_end_mask_0 = const()[name = tensor("op_6180_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6180_cast = slice_by_index(begin = var_6180_begin_0, end = var_6180_end_0, end_mask = var_6180_end_mask_0, x = q_25_cast)[name = tensor("op_6180_cast")]; + tensor var_6184_begin_0 = const()[name = tensor("op_6184_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_6184_end_0 = const()[name = tensor("op_6184_end_0"), val = tensor([2, 1152, 1, 64])]; + tensor var_6184_end_mask_0 = const()[name = tensor("op_6184_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6184_cast = slice_by_index(begin = var_6184_begin_0, end = var_6184_end_0, end_mask = var_6184_end_mask_0, x = q_25_cast)[name = tensor("op_6184_cast")]; + tensor var_6188_begin_0 = const()[name = tensor("op_6188_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_6188_end_0 = const()[name = tensor("op_6188_end_0"), val = tensor([2, 1216, 1, 64])]; + tensor var_6188_end_mask_0 = const()[name = tensor("op_6188_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6188_cast = slice_by_index(begin = var_6188_begin_0, end = var_6188_end_0, end_mask = var_6188_end_mask_0, x = q_25_cast)[name = tensor("op_6188_cast")]; + tensor var_6192_begin_0 = const()[name = tensor("op_6192_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_6192_end_0 = const()[name = tensor("op_6192_end_0"), val = tensor([2, 1280, 1, 64])]; + tensor var_6192_end_mask_0 = const()[name = tensor("op_6192_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6192_cast = slice_by_index(begin = var_6192_begin_0, end = var_6192_end_0, end_mask = var_6192_end_mask_0, x = q_25_cast)[name = tensor("op_6192_cast")]; + tensor k_51_perm_0 = const()[name = tensor("k_51_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_6199_begin_0 = const()[name = tensor("op_6199_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6199_end_0 = const()[name = tensor("op_6199_end_0"), val = tensor([2, 64, 1, 64])]; + tensor var_6199_end_mask_0 = const()[name = tensor("op_6199_end_mask_0"), val = tensor([true, true, true, false])]; + tensor transpose_19 = transpose(perm = k_51_perm_0, x = k_49_cast)[name = tensor("transpose_19")]; + tensor var_6199_cast = slice_by_index(begin = var_6199_begin_0, end = var_6199_end_0, end_mask = var_6199_end_mask_0, x = transpose_19)[name = tensor("op_6199_cast")]; + tensor var_6203_begin_0 = const()[name = tensor("op_6203_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_6203_end_0 = const()[name = tensor("op_6203_end_0"), val = tensor([2, 64, 1, 128])]; + tensor var_6203_end_mask_0 = const()[name = tensor("op_6203_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6203_cast = slice_by_index(begin = var_6203_begin_0, end = var_6203_end_0, end_mask = var_6203_end_mask_0, x = transpose_19)[name = tensor("op_6203_cast")]; + tensor var_6207_begin_0 = const()[name = tensor("op_6207_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_6207_end_0 = const()[name = tensor("op_6207_end_0"), val = tensor([2, 64, 1, 192])]; + tensor var_6207_end_mask_0 = const()[name = tensor("op_6207_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6207_cast = slice_by_index(begin = var_6207_begin_0, end = var_6207_end_0, end_mask = var_6207_end_mask_0, x = transpose_19)[name = tensor("op_6207_cast")]; + tensor var_6211_begin_0 = const()[name = tensor("op_6211_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_6211_end_0 = const()[name = tensor("op_6211_end_0"), val = tensor([2, 64, 1, 256])]; + tensor var_6211_end_mask_0 = const()[name = tensor("op_6211_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6211_cast = slice_by_index(begin = var_6211_begin_0, end = var_6211_end_0, end_mask = var_6211_end_mask_0, x = transpose_19)[name = tensor("op_6211_cast")]; + tensor var_6215_begin_0 = const()[name = tensor("op_6215_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_6215_end_0 = const()[name = tensor("op_6215_end_0"), val = tensor([2, 64, 1, 320])]; + tensor var_6215_end_mask_0 = const()[name = tensor("op_6215_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6215_cast = slice_by_index(begin = var_6215_begin_0, end = var_6215_end_0, end_mask = var_6215_end_mask_0, x = transpose_19)[name = tensor("op_6215_cast")]; + tensor var_6219_begin_0 = const()[name = tensor("op_6219_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_6219_end_0 = const()[name = tensor("op_6219_end_0"), val = tensor([2, 64, 1, 384])]; + tensor var_6219_end_mask_0 = const()[name = tensor("op_6219_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6219_cast = slice_by_index(begin = var_6219_begin_0, end = var_6219_end_0, end_mask = var_6219_end_mask_0, x = transpose_19)[name = tensor("op_6219_cast")]; + tensor var_6223_begin_0 = const()[name = tensor("op_6223_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_6223_end_0 = const()[name = tensor("op_6223_end_0"), val = tensor([2, 64, 1, 448])]; + tensor var_6223_end_mask_0 = const()[name = tensor("op_6223_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6223_cast = slice_by_index(begin = var_6223_begin_0, end = var_6223_end_0, end_mask = var_6223_end_mask_0, x = transpose_19)[name = tensor("op_6223_cast")]; + tensor var_6227_begin_0 = const()[name = tensor("op_6227_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_6227_end_0 = const()[name = tensor("op_6227_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_6227_end_mask_0 = const()[name = tensor("op_6227_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6227_cast = slice_by_index(begin = var_6227_begin_0, end = var_6227_end_0, end_mask = var_6227_end_mask_0, x = transpose_19)[name = tensor("op_6227_cast")]; + tensor var_6231_begin_0 = const()[name = tensor("op_6231_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_6231_end_0 = const()[name = tensor("op_6231_end_0"), val = tensor([2, 64, 1, 576])]; + tensor var_6231_end_mask_0 = const()[name = tensor("op_6231_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6231_cast = slice_by_index(begin = var_6231_begin_0, end = var_6231_end_0, end_mask = var_6231_end_mask_0, x = transpose_19)[name = tensor("op_6231_cast")]; + tensor var_6235_begin_0 = const()[name = tensor("op_6235_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_6235_end_0 = const()[name = tensor("op_6235_end_0"), val = tensor([2, 64, 1, 640])]; + tensor var_6235_end_mask_0 = const()[name = tensor("op_6235_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6235_cast = slice_by_index(begin = var_6235_begin_0, end = var_6235_end_0, end_mask = var_6235_end_mask_0, x = transpose_19)[name = tensor("op_6235_cast")]; + tensor var_6239_begin_0 = const()[name = tensor("op_6239_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_6239_end_0 = const()[name = tensor("op_6239_end_0"), val = tensor([2, 64, 1, 704])]; + tensor var_6239_end_mask_0 = const()[name = tensor("op_6239_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6239_cast = slice_by_index(begin = var_6239_begin_0, end = var_6239_end_0, end_mask = var_6239_end_mask_0, x = transpose_19)[name = tensor("op_6239_cast")]; + tensor var_6243_begin_0 = const()[name = tensor("op_6243_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_6243_end_0 = const()[name = tensor("op_6243_end_0"), val = tensor([2, 64, 1, 768])]; + tensor var_6243_end_mask_0 = const()[name = tensor("op_6243_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6243_cast = slice_by_index(begin = var_6243_begin_0, end = var_6243_end_0, end_mask = var_6243_end_mask_0, x = transpose_19)[name = tensor("op_6243_cast")]; + tensor var_6247_begin_0 = const()[name = tensor("op_6247_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_6247_end_0 = const()[name = tensor("op_6247_end_0"), val = tensor([2, 64, 1, 832])]; + tensor var_6247_end_mask_0 = const()[name = tensor("op_6247_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6247_cast = slice_by_index(begin = var_6247_begin_0, end = var_6247_end_0, end_mask = var_6247_end_mask_0, x = transpose_19)[name = tensor("op_6247_cast")]; + tensor var_6251_begin_0 = const()[name = tensor("op_6251_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_6251_end_0 = const()[name = tensor("op_6251_end_0"), val = tensor([2, 64, 1, 896])]; + tensor var_6251_end_mask_0 = const()[name = tensor("op_6251_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6251_cast = slice_by_index(begin = var_6251_begin_0, end = var_6251_end_0, end_mask = var_6251_end_mask_0, x = transpose_19)[name = tensor("op_6251_cast")]; + tensor var_6255_begin_0 = const()[name = tensor("op_6255_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_6255_end_0 = const()[name = tensor("op_6255_end_0"), val = tensor([2, 64, 1, 960])]; + tensor var_6255_end_mask_0 = const()[name = tensor("op_6255_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6255_cast = slice_by_index(begin = var_6255_begin_0, end = var_6255_end_0, end_mask = var_6255_end_mask_0, x = transpose_19)[name = tensor("op_6255_cast")]; + tensor var_6259_begin_0 = const()[name = tensor("op_6259_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_6259_end_0 = const()[name = tensor("op_6259_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_6259_end_mask_0 = const()[name = tensor("op_6259_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6259_cast = slice_by_index(begin = var_6259_begin_0, end = var_6259_end_0, end_mask = var_6259_end_mask_0, x = transpose_19)[name = tensor("op_6259_cast")]; + tensor var_6263_begin_0 = const()[name = tensor("op_6263_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_6263_end_0 = const()[name = tensor("op_6263_end_0"), val = tensor([2, 64, 1, 1088])]; + tensor var_6263_end_mask_0 = const()[name = tensor("op_6263_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6263_cast = slice_by_index(begin = var_6263_begin_0, end = var_6263_end_0, end_mask = var_6263_end_mask_0, x = transpose_19)[name = tensor("op_6263_cast")]; + tensor var_6267_begin_0 = const()[name = tensor("op_6267_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_6267_end_0 = const()[name = tensor("op_6267_end_0"), val = tensor([2, 64, 1, 1152])]; + tensor var_6267_end_mask_0 = const()[name = tensor("op_6267_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6267_cast = slice_by_index(begin = var_6267_begin_0, end = var_6267_end_0, end_mask = var_6267_end_mask_0, x = transpose_19)[name = tensor("op_6267_cast")]; + tensor var_6271_begin_0 = const()[name = tensor("op_6271_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_6271_end_0 = const()[name = tensor("op_6271_end_0"), val = tensor([2, 64, 1, 1216])]; + tensor var_6271_end_mask_0 = const()[name = tensor("op_6271_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6271_cast = slice_by_index(begin = var_6271_begin_0, end = var_6271_end_0, end_mask = var_6271_end_mask_0, x = transpose_19)[name = tensor("op_6271_cast")]; + tensor var_6275_begin_0 = const()[name = tensor("op_6275_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_6275_end_0 = const()[name = tensor("op_6275_end_0"), val = tensor([2, 64, 1, 1280])]; + tensor var_6275_end_mask_0 = const()[name = tensor("op_6275_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6275_cast = slice_by_index(begin = var_6275_begin_0, end = var_6275_end_0, end_mask = var_6275_end_mask_0, x = transpose_19)[name = tensor("op_6275_cast")]; + tensor var_6277_begin_0 = const()[name = tensor("op_6277_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6277_end_0 = const()[name = tensor("op_6277_end_0"), val = tensor([2, 64, 1, 64])]; + tensor var_6277_end_mask_0 = const()[name = tensor("op_6277_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6277_cast = slice_by_index(begin = var_6277_begin_0, end = var_6277_end_0, end_mask = var_6277_end_mask_0, x = v_25_cast)[name = tensor("op_6277_cast")]; + tensor var_6281_begin_0 = const()[name = tensor("op_6281_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_6281_end_0 = const()[name = tensor("op_6281_end_0"), val = tensor([2, 128, 1, 64])]; + tensor var_6281_end_mask_0 = const()[name = tensor("op_6281_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6281_cast = slice_by_index(begin = var_6281_begin_0, end = var_6281_end_0, end_mask = var_6281_end_mask_0, x = v_25_cast)[name = tensor("op_6281_cast")]; + tensor var_6285_begin_0 = const()[name = tensor("op_6285_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_6285_end_0 = const()[name = tensor("op_6285_end_0"), val = tensor([2, 192, 1, 64])]; + tensor var_6285_end_mask_0 = const()[name = tensor("op_6285_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6285_cast = slice_by_index(begin = var_6285_begin_0, end = var_6285_end_0, end_mask = var_6285_end_mask_0, x = v_25_cast)[name = tensor("op_6285_cast")]; + tensor var_6289_begin_0 = const()[name = tensor("op_6289_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_6289_end_0 = const()[name = tensor("op_6289_end_0"), val = tensor([2, 256, 1, 64])]; + tensor var_6289_end_mask_0 = const()[name = tensor("op_6289_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6289_cast = slice_by_index(begin = var_6289_begin_0, end = var_6289_end_0, end_mask = var_6289_end_mask_0, x = v_25_cast)[name = tensor("op_6289_cast")]; + tensor var_6293_begin_0 = const()[name = tensor("op_6293_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_6293_end_0 = const()[name = tensor("op_6293_end_0"), val = tensor([2, 320, 1, 64])]; + tensor var_6293_end_mask_0 = const()[name = tensor("op_6293_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6293_cast = slice_by_index(begin = var_6293_begin_0, end = var_6293_end_0, end_mask = var_6293_end_mask_0, x = v_25_cast)[name = tensor("op_6293_cast")]; + tensor var_6297_begin_0 = const()[name = tensor("op_6297_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_6297_end_0 = const()[name = tensor("op_6297_end_0"), val = tensor([2, 384, 1, 64])]; + tensor var_6297_end_mask_0 = const()[name = tensor("op_6297_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6297_cast = slice_by_index(begin = var_6297_begin_0, end = var_6297_end_0, end_mask = var_6297_end_mask_0, x = v_25_cast)[name = tensor("op_6297_cast")]; + tensor var_6301_begin_0 = const()[name = tensor("op_6301_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_6301_end_0 = const()[name = tensor("op_6301_end_0"), val = tensor([2, 448, 1, 64])]; + tensor var_6301_end_mask_0 = const()[name = tensor("op_6301_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6301_cast = slice_by_index(begin = var_6301_begin_0, end = var_6301_end_0, end_mask = var_6301_end_mask_0, x = v_25_cast)[name = tensor("op_6301_cast")]; + tensor var_6305_begin_0 = const()[name = tensor("op_6305_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_6305_end_0 = const()[name = tensor("op_6305_end_0"), val = tensor([2, 512, 1, 64])]; + tensor var_6305_end_mask_0 = const()[name = tensor("op_6305_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6305_cast = slice_by_index(begin = var_6305_begin_0, end = var_6305_end_0, end_mask = var_6305_end_mask_0, x = v_25_cast)[name = tensor("op_6305_cast")]; + tensor var_6309_begin_0 = const()[name = tensor("op_6309_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_6309_end_0 = const()[name = tensor("op_6309_end_0"), val = tensor([2, 576, 1, 64])]; + tensor var_6309_end_mask_0 = const()[name = tensor("op_6309_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6309_cast = slice_by_index(begin = var_6309_begin_0, end = var_6309_end_0, end_mask = var_6309_end_mask_0, x = v_25_cast)[name = tensor("op_6309_cast")]; + tensor var_6313_begin_0 = const()[name = tensor("op_6313_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_6313_end_0 = const()[name = tensor("op_6313_end_0"), val = tensor([2, 640, 1, 64])]; + tensor var_6313_end_mask_0 = const()[name = tensor("op_6313_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6313_cast = slice_by_index(begin = var_6313_begin_0, end = var_6313_end_0, end_mask = var_6313_end_mask_0, x = v_25_cast)[name = tensor("op_6313_cast")]; + tensor var_6317_begin_0 = const()[name = tensor("op_6317_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_6317_end_0 = const()[name = tensor("op_6317_end_0"), val = tensor([2, 704, 1, 64])]; + tensor var_6317_end_mask_0 = const()[name = tensor("op_6317_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6317_cast = slice_by_index(begin = var_6317_begin_0, end = var_6317_end_0, end_mask = var_6317_end_mask_0, x = v_25_cast)[name = tensor("op_6317_cast")]; + tensor var_6321_begin_0 = const()[name = tensor("op_6321_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_6321_end_0 = const()[name = tensor("op_6321_end_0"), val = tensor([2, 768, 1, 64])]; + tensor var_6321_end_mask_0 = const()[name = tensor("op_6321_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6321_cast = slice_by_index(begin = var_6321_begin_0, end = var_6321_end_0, end_mask = var_6321_end_mask_0, x = v_25_cast)[name = tensor("op_6321_cast")]; + tensor var_6325_begin_0 = const()[name = tensor("op_6325_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_6325_end_0 = const()[name = tensor("op_6325_end_0"), val = tensor([2, 832, 1, 64])]; + tensor var_6325_end_mask_0 = const()[name = tensor("op_6325_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6325_cast = slice_by_index(begin = var_6325_begin_0, end = var_6325_end_0, end_mask = var_6325_end_mask_0, x = v_25_cast)[name = tensor("op_6325_cast")]; + tensor var_6329_begin_0 = const()[name = tensor("op_6329_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_6329_end_0 = const()[name = tensor("op_6329_end_0"), val = tensor([2, 896, 1, 64])]; + tensor var_6329_end_mask_0 = const()[name = tensor("op_6329_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6329_cast = slice_by_index(begin = var_6329_begin_0, end = var_6329_end_0, end_mask = var_6329_end_mask_0, x = v_25_cast)[name = tensor("op_6329_cast")]; + tensor var_6333_begin_0 = const()[name = tensor("op_6333_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_6333_end_0 = const()[name = tensor("op_6333_end_0"), val = tensor([2, 960, 1, 64])]; + tensor var_6333_end_mask_0 = const()[name = tensor("op_6333_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6333_cast = slice_by_index(begin = var_6333_begin_0, end = var_6333_end_0, end_mask = var_6333_end_mask_0, x = v_25_cast)[name = tensor("op_6333_cast")]; + tensor var_6337_begin_0 = const()[name = tensor("op_6337_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_6337_end_0 = const()[name = tensor("op_6337_end_0"), val = tensor([2, 1024, 1, 64])]; + tensor var_6337_end_mask_0 = const()[name = tensor("op_6337_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6337_cast = slice_by_index(begin = var_6337_begin_0, end = var_6337_end_0, end_mask = var_6337_end_mask_0, x = v_25_cast)[name = tensor("op_6337_cast")]; + tensor var_6341_begin_0 = const()[name = tensor("op_6341_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_6341_end_0 = const()[name = tensor("op_6341_end_0"), val = tensor([2, 1088, 1, 64])]; + tensor var_6341_end_mask_0 = const()[name = tensor("op_6341_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6341_cast = slice_by_index(begin = var_6341_begin_0, end = var_6341_end_0, end_mask = var_6341_end_mask_0, x = v_25_cast)[name = tensor("op_6341_cast")]; + tensor var_6345_begin_0 = const()[name = tensor("op_6345_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_6345_end_0 = const()[name = tensor("op_6345_end_0"), val = tensor([2, 1152, 1, 64])]; + tensor var_6345_end_mask_0 = const()[name = tensor("op_6345_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6345_cast = slice_by_index(begin = var_6345_begin_0, end = var_6345_end_0, end_mask = var_6345_end_mask_0, x = v_25_cast)[name = tensor("op_6345_cast")]; + tensor var_6349_begin_0 = const()[name = tensor("op_6349_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_6349_end_0 = const()[name = tensor("op_6349_end_0"), val = tensor([2, 1216, 1, 64])]; + tensor var_6349_end_mask_0 = const()[name = tensor("op_6349_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6349_cast = slice_by_index(begin = var_6349_begin_0, end = var_6349_end_0, end_mask = var_6349_end_mask_0, x = v_25_cast)[name = tensor("op_6349_cast")]; + tensor var_6353_begin_0 = const()[name = tensor("op_6353_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_6353_end_0 = const()[name = tensor("op_6353_end_0"), val = tensor([2, 1280, 1, 64])]; + tensor var_6353_end_mask_0 = const()[name = tensor("op_6353_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6353_cast = slice_by_index(begin = var_6353_begin_0, end = var_6353_end_0, end_mask = var_6353_end_mask_0, x = v_25_cast)[name = tensor("op_6353_cast")]; + tensor var_6357_equation_0 = const()[name = tensor("op_6357_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6357_cast = einsum(equation = var_6357_equation_0, values = (var_6199_cast, var_6116_cast))[name = tensor("op_6357_cast")]; + tensor var_6358_to_fp16 = const()[name = tensor("op_6358_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_161_cast = mul(x = var_6357_cast, y = var_6358_to_fp16)[name = tensor("aw_161_cast")]; + tensor var_6361_equation_0 = const()[name = tensor("op_6361_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6361_cast = einsum(equation = var_6361_equation_0, values = (var_6203_cast, var_6120_cast))[name = tensor("op_6361_cast")]; + tensor var_6362_to_fp16 = const()[name = tensor("op_6362_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_163_cast = mul(x = var_6361_cast, y = var_6362_to_fp16)[name = tensor("aw_163_cast")]; + tensor var_6365_equation_0 = const()[name = tensor("op_6365_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6365_cast = einsum(equation = var_6365_equation_0, values = (var_6207_cast, var_6124_cast))[name = tensor("op_6365_cast")]; + tensor var_6366_to_fp16 = const()[name = tensor("op_6366_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_165_cast = mul(x = var_6365_cast, y = var_6366_to_fp16)[name = tensor("aw_165_cast")]; + tensor var_6369_equation_0 = const()[name = tensor("op_6369_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6369_cast = einsum(equation = var_6369_equation_0, values = (var_6211_cast, var_6128_cast))[name = tensor("op_6369_cast")]; + tensor var_6370_to_fp16 = const()[name = tensor("op_6370_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_167_cast = mul(x = var_6369_cast, y = var_6370_to_fp16)[name = tensor("aw_167_cast")]; + tensor var_6373_equation_0 = const()[name = tensor("op_6373_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6373_cast = einsum(equation = var_6373_equation_0, values = (var_6215_cast, var_6132_cast))[name = tensor("op_6373_cast")]; + tensor var_6374_to_fp16 = const()[name = tensor("op_6374_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_169_cast = mul(x = var_6373_cast, y = var_6374_to_fp16)[name = tensor("aw_169_cast")]; + tensor var_6377_equation_0 = const()[name = tensor("op_6377_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6377_cast = einsum(equation = var_6377_equation_0, values = (var_6219_cast, var_6136_cast))[name = tensor("op_6377_cast")]; + tensor var_6378_to_fp16 = const()[name = tensor("op_6378_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_171_cast = mul(x = var_6377_cast, y = var_6378_to_fp16)[name = tensor("aw_171_cast")]; + tensor var_6381_equation_0 = const()[name = tensor("op_6381_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6381_cast = einsum(equation = var_6381_equation_0, values = (var_6223_cast, var_6140_cast))[name = tensor("op_6381_cast")]; + tensor var_6382_to_fp16 = const()[name = tensor("op_6382_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_173_cast = mul(x = var_6381_cast, y = var_6382_to_fp16)[name = tensor("aw_173_cast")]; + tensor var_6385_equation_0 = const()[name = tensor("op_6385_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6385_cast = einsum(equation = var_6385_equation_0, values = (var_6227_cast, var_6144_cast))[name = tensor("op_6385_cast")]; + tensor var_6386_to_fp16 = const()[name = tensor("op_6386_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_175_cast = mul(x = var_6385_cast, y = var_6386_to_fp16)[name = tensor("aw_175_cast")]; + tensor var_6389_equation_0 = const()[name = tensor("op_6389_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6389_cast = einsum(equation = var_6389_equation_0, values = (var_6231_cast, var_6148_cast))[name = tensor("op_6389_cast")]; + tensor var_6390_to_fp16 = const()[name = tensor("op_6390_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_177_cast = mul(x = var_6389_cast, y = var_6390_to_fp16)[name = tensor("aw_177_cast")]; + tensor var_6393_equation_0 = const()[name = tensor("op_6393_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6393_cast = einsum(equation = var_6393_equation_0, values = (var_6235_cast, var_6152_cast))[name = tensor("op_6393_cast")]; + tensor var_6394_to_fp16 = const()[name = tensor("op_6394_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_179_cast = mul(x = var_6393_cast, y = var_6394_to_fp16)[name = tensor("aw_179_cast")]; + tensor var_6397_equation_0 = const()[name = tensor("op_6397_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6397_cast = einsum(equation = var_6397_equation_0, values = (var_6239_cast, var_6156_cast))[name = tensor("op_6397_cast")]; + tensor var_6398_to_fp16 = const()[name = tensor("op_6398_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_181_cast = mul(x = var_6397_cast, y = var_6398_to_fp16)[name = tensor("aw_181_cast")]; + tensor var_6401_equation_0 = const()[name = tensor("op_6401_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6401_cast = einsum(equation = var_6401_equation_0, values = (var_6243_cast, var_6160_cast))[name = tensor("op_6401_cast")]; + tensor var_6402_to_fp16 = const()[name = tensor("op_6402_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_183_cast = mul(x = var_6401_cast, y = var_6402_to_fp16)[name = tensor("aw_183_cast")]; + tensor var_6405_equation_0 = const()[name = tensor("op_6405_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6405_cast = einsum(equation = var_6405_equation_0, values = (var_6247_cast, var_6164_cast))[name = tensor("op_6405_cast")]; + tensor var_6406_to_fp16 = const()[name = tensor("op_6406_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_185_cast = mul(x = var_6405_cast, y = var_6406_to_fp16)[name = tensor("aw_185_cast")]; + tensor var_6409_equation_0 = const()[name = tensor("op_6409_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6409_cast = einsum(equation = var_6409_equation_0, values = (var_6251_cast, var_6168_cast))[name = tensor("op_6409_cast")]; + tensor var_6410_to_fp16 = const()[name = tensor("op_6410_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_187_cast = mul(x = var_6409_cast, y = var_6410_to_fp16)[name = tensor("aw_187_cast")]; + tensor var_6413_equation_0 = const()[name = tensor("op_6413_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6413_cast = einsum(equation = var_6413_equation_0, values = (var_6255_cast, var_6172_cast))[name = tensor("op_6413_cast")]; + tensor var_6414_to_fp16 = const()[name = tensor("op_6414_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_189_cast = mul(x = var_6413_cast, y = var_6414_to_fp16)[name = tensor("aw_189_cast")]; + tensor var_6417_equation_0 = const()[name = tensor("op_6417_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6417_cast = einsum(equation = var_6417_equation_0, values = (var_6259_cast, var_6176_cast))[name = tensor("op_6417_cast")]; + tensor var_6418_to_fp16 = const()[name = tensor("op_6418_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_191_cast = mul(x = var_6417_cast, y = var_6418_to_fp16)[name = tensor("aw_191_cast")]; + tensor var_6421_equation_0 = const()[name = tensor("op_6421_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6421_cast = einsum(equation = var_6421_equation_0, values = (var_6263_cast, var_6180_cast))[name = tensor("op_6421_cast")]; + tensor var_6422_to_fp16 = const()[name = tensor("op_6422_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_193_cast = mul(x = var_6421_cast, y = var_6422_to_fp16)[name = tensor("aw_193_cast")]; + tensor var_6425_equation_0 = const()[name = tensor("op_6425_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6425_cast = einsum(equation = var_6425_equation_0, values = (var_6267_cast, var_6184_cast))[name = tensor("op_6425_cast")]; + tensor var_6426_to_fp16 = const()[name = tensor("op_6426_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_195_cast = mul(x = var_6425_cast, y = var_6426_to_fp16)[name = tensor("aw_195_cast")]; + tensor var_6429_equation_0 = const()[name = tensor("op_6429_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6429_cast = einsum(equation = var_6429_equation_0, values = (var_6271_cast, var_6188_cast))[name = tensor("op_6429_cast")]; + tensor var_6430_to_fp16 = const()[name = tensor("op_6430_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_197_cast = mul(x = var_6429_cast, y = var_6430_to_fp16)[name = tensor("aw_197_cast")]; + tensor var_6433_equation_0 = const()[name = tensor("op_6433_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6433_cast = einsum(equation = var_6433_equation_0, values = (var_6275_cast, var_6192_cast))[name = tensor("op_6433_cast")]; + tensor var_6434_to_fp16 = const()[name = tensor("op_6434_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_199_cast = mul(x = var_6433_cast, y = var_6434_to_fp16)[name = tensor("aw_199_cast")]; + tensor var_6436_cast = softmax(axis = var_6000, x = aw_161_cast)[name = tensor("op_6436_cast")]; + tensor var_6437_cast = softmax(axis = var_6000, x = aw_163_cast)[name = tensor("op_6437_cast")]; + tensor var_6438_cast = softmax(axis = var_6000, x = aw_165_cast)[name = tensor("op_6438_cast")]; + tensor var_6439_cast = softmax(axis = var_6000, x = aw_167_cast)[name = tensor("op_6439_cast")]; + tensor var_6440_cast = softmax(axis = var_6000, x = aw_169_cast)[name = tensor("op_6440_cast")]; + tensor var_6441_cast = softmax(axis = var_6000, x = aw_171_cast)[name = tensor("op_6441_cast")]; + tensor var_6442_cast = softmax(axis = var_6000, x = aw_173_cast)[name = tensor("op_6442_cast")]; + tensor var_6443_cast = softmax(axis = var_6000, x = aw_175_cast)[name = tensor("op_6443_cast")]; + tensor var_6444_cast = softmax(axis = var_6000, x = aw_177_cast)[name = tensor("op_6444_cast")]; + tensor var_6445_cast = softmax(axis = var_6000, x = aw_179_cast)[name = tensor("op_6445_cast")]; + tensor var_6446_cast = softmax(axis = var_6000, x = aw_181_cast)[name = tensor("op_6446_cast")]; + tensor var_6447_cast = softmax(axis = var_6000, x = aw_183_cast)[name = tensor("op_6447_cast")]; + tensor var_6448_cast = softmax(axis = var_6000, x = aw_185_cast)[name = tensor("op_6448_cast")]; + tensor var_6449_cast = softmax(axis = var_6000, x = aw_187_cast)[name = tensor("op_6449_cast")]; + tensor var_6450_cast = softmax(axis = var_6000, x = aw_189_cast)[name = tensor("op_6450_cast")]; + tensor var_6451_cast = softmax(axis = var_6000, x = aw_191_cast)[name = tensor("op_6451_cast")]; + tensor var_6452_cast = softmax(axis = var_6000, x = aw_193_cast)[name = tensor("op_6452_cast")]; + tensor var_6453_cast = softmax(axis = var_6000, x = aw_195_cast)[name = tensor("op_6453_cast")]; + tensor var_6454_cast = softmax(axis = var_6000, x = aw_197_cast)[name = tensor("op_6454_cast")]; + tensor var_6455_cast = softmax(axis = var_6000, x = aw_199_cast)[name = tensor("op_6455_cast")]; + tensor var_6457_equation_0 = const()[name = tensor("op_6457_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6457_cast = einsum(equation = var_6457_equation_0, values = (var_6277_cast, var_6436_cast))[name = tensor("op_6457_cast")]; + tensor var_6459_equation_0 = const()[name = tensor("op_6459_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6459_cast = einsum(equation = var_6459_equation_0, values = (var_6281_cast, var_6437_cast))[name = tensor("op_6459_cast")]; + tensor var_6461_equation_0 = const()[name = tensor("op_6461_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6461_cast = einsum(equation = var_6461_equation_0, values = (var_6285_cast, var_6438_cast))[name = tensor("op_6461_cast")]; + tensor var_6463_equation_0 = const()[name = tensor("op_6463_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6463_cast = einsum(equation = var_6463_equation_0, values = (var_6289_cast, var_6439_cast))[name = tensor("op_6463_cast")]; + tensor var_6465_equation_0 = const()[name = tensor("op_6465_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6465_cast = einsum(equation = var_6465_equation_0, values = (var_6293_cast, var_6440_cast))[name = tensor("op_6465_cast")]; + tensor var_6467_equation_0 = const()[name = tensor("op_6467_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6467_cast = einsum(equation = var_6467_equation_0, values = (var_6297_cast, var_6441_cast))[name = tensor("op_6467_cast")]; + tensor var_6469_equation_0 = const()[name = tensor("op_6469_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6469_cast = einsum(equation = var_6469_equation_0, values = (var_6301_cast, var_6442_cast))[name = tensor("op_6469_cast")]; + tensor var_6471_equation_0 = const()[name = tensor("op_6471_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6471_cast = einsum(equation = var_6471_equation_0, values = (var_6305_cast, var_6443_cast))[name = tensor("op_6471_cast")]; + tensor var_6473_equation_0 = const()[name = tensor("op_6473_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6473_cast = einsum(equation = var_6473_equation_0, values = (var_6309_cast, var_6444_cast))[name = tensor("op_6473_cast")]; + tensor var_6475_equation_0 = const()[name = tensor("op_6475_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6475_cast = einsum(equation = var_6475_equation_0, values = (var_6313_cast, var_6445_cast))[name = tensor("op_6475_cast")]; + tensor var_6477_equation_0 = const()[name = tensor("op_6477_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6477_cast = einsum(equation = var_6477_equation_0, values = (var_6317_cast, var_6446_cast))[name = tensor("op_6477_cast")]; + tensor var_6479_equation_0 = const()[name = tensor("op_6479_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6479_cast = einsum(equation = var_6479_equation_0, values = (var_6321_cast, var_6447_cast))[name = tensor("op_6479_cast")]; + tensor var_6481_equation_0 = const()[name = tensor("op_6481_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6481_cast = einsum(equation = var_6481_equation_0, values = (var_6325_cast, var_6448_cast))[name = tensor("op_6481_cast")]; + tensor var_6483_equation_0 = const()[name = tensor("op_6483_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6483_cast = einsum(equation = var_6483_equation_0, values = (var_6329_cast, var_6449_cast))[name = tensor("op_6483_cast")]; + tensor var_6485_equation_0 = const()[name = tensor("op_6485_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6485_cast = einsum(equation = var_6485_equation_0, values = (var_6333_cast, var_6450_cast))[name = tensor("op_6485_cast")]; + tensor var_6487_equation_0 = const()[name = tensor("op_6487_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6487_cast = einsum(equation = var_6487_equation_0, values = (var_6337_cast, var_6451_cast))[name = tensor("op_6487_cast")]; + tensor var_6489_equation_0 = const()[name = tensor("op_6489_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6489_cast = einsum(equation = var_6489_equation_0, values = (var_6341_cast, var_6452_cast))[name = tensor("op_6489_cast")]; + tensor var_6491_equation_0 = const()[name = tensor("op_6491_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6491_cast = einsum(equation = var_6491_equation_0, values = (var_6345_cast, var_6453_cast))[name = tensor("op_6491_cast")]; + tensor var_6493_equation_0 = const()[name = tensor("op_6493_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6493_cast = einsum(equation = var_6493_equation_0, values = (var_6349_cast, var_6454_cast))[name = tensor("op_6493_cast")]; + tensor var_6495_equation_0 = const()[name = tensor("op_6495_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6495_cast = einsum(equation = var_6495_equation_0, values = (var_6353_cast, var_6455_cast))[name = tensor("op_6495_cast")]; + tensor input_215_interleave_0 = const()[name = tensor("input_215_interleave_0"), val = tensor(false)]; + tensor input_215_cast = concat(axis = var_6000, interleave = input_215_interleave_0, values = (var_6457_cast, var_6459_cast, var_6461_cast, var_6463_cast, var_6465_cast, var_6467_cast, var_6469_cast, var_6471_cast, var_6473_cast, var_6475_cast, var_6477_cast, var_6479_cast, var_6481_cast, var_6483_cast, var_6485_cast, var_6487_cast, var_6489_cast, var_6491_cast, var_6493_cast, var_6495_cast))[name = tensor("input_215_cast")]; + tensor var_6501 = const()[name = tensor("op_6501"), val = tensor([1, 1])]; + tensor var_6503 = const()[name = tensor("op_6503"), val = tensor([1, 1])]; + tensor var_6505_pad_type_0 = const()[name = tensor("op_6505_pad_type_0"), val = tensor("custom")]; + tensor var_6505_pad_0 = const()[name = tensor("op_6505_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor mid_block_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(218960000))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(220188864))), name = tensor("mid_block_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor mid_block_attentions_0_transformer_blocks_0_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_0_attn1_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(220189056)))]; + tensor var_6505_cast = conv(bias = mid_block_attentions_0_transformer_blocks_0_attn1_to_out_0_bias_to_fp16, dilations = var_6503, groups = var_6000, pad = var_6505_pad_0, pad_type = var_6505_pad_type_0, strides = var_6501, weight = mid_block_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16_palettized, x = input_215_cast)[name = tensor("op_6505_cast")]; + tensor inputs_39_cast = add(x = var_6505_cast, y = inputs_37_cast)[name = tensor("inputs_39_cast")]; + tensor var_6509 = const()[name = tensor("op_6509"), val = tensor([1])]; + tensor channels_mean_39_cast = reduce_mean(axes = var_6509, keep_dims = var_5995, x = inputs_39_cast)[name = tensor("channels_mean_39_cast")]; + tensor zero_mean_39_cast = sub(x = inputs_39_cast, y = channels_mean_39_cast)[name = tensor("zero_mean_39_cast")]; + tensor zero_mean_sq_39_cast = mul(x = zero_mean_39_cast, y = zero_mean_39_cast)[name = tensor("zero_mean_sq_39_cast")]; + tensor var_6513 = const()[name = tensor("op_6513"), val = tensor([1])]; + tensor var_6514_cast = reduce_mean(axes = var_6513, keep_dims = var_5995, x = zero_mean_sq_39_cast)[name = tensor("op_6514_cast")]; + tensor var_6515_to_fp16 = const()[name = tensor("op_6515_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_6516_cast = add(x = var_6514_cast, y = var_6515_to_fp16)[name = tensor("op_6516_cast")]; + tensor denom_39_epsilon_0_to_fp16 = const()[name = tensor("denom_39_epsilon_0_to_fp16"), val = tensor(0x1p-24)]; + tensor denom_39_cast = rsqrt(epsilon = denom_39_epsilon_0_to_fp16, x = var_6516_cast)[name = tensor("denom_39_cast")]; + tensor out_39_cast = mul(x = zero_mean_39_cast, y = denom_39_cast)[name = tensor("out_39_cast")]; + tensor var_6520_to_fp16 = const()[name = tensor("op_6520_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(220191680)))]; + tensor var_6521_cast = add(x = out_39_cast, y = var_6520_to_fp16)[name = tensor("op_6521_cast")]; + tensor var_6523_to_fp16 = const()[name = tensor("op_6523_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(220194304)))]; + tensor hidden_states_127_cast = mul(x = var_6521_cast, y = var_6523_to_fp16)[name = tensor("hidden_states_127_cast")]; + tensor var_6530 = const()[name = tensor("op_6530"), val = tensor([1, 1])]; + tensor var_6532 = const()[name = tensor("op_6532"), val = tensor([1, 1])]; + tensor q_27_pad_type_0 = const()[name = tensor("q_27_pad_type_0"), val = tensor("custom")]; + tensor q_27_pad_0 = const()[name = tensor("q_27_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor mid_block_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(220196928))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(221425792))), name = tensor("mid_block_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor q_27_cast = conv(dilations = var_6532, groups = var_6000, pad = q_27_pad_0, pad_type = q_27_pad_type_0, strides = var_6530, weight = mid_block_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16_palettized, x = hidden_states_127_cast)[name = tensor("q_27_cast")]; + tensor var_6536 = const()[name = tensor("op_6536"), val = tensor([1, 1])]; + tensor var_6538 = const()[name = tensor("op_6538"), val = tensor([1, 1])]; + tensor k_53_pad_type_0 = const()[name = tensor("k_53_pad_type_0"), val = tensor("custom")]; + tensor k_53_pad_0 = const()[name = tensor("k_53_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor mid_block_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(221425984))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(222409088))), name = tensor("mid_block_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized"), shape = tensor([1280, 1024, 1, 1])]; + tensor k_53_cast = conv(dilations = var_6538, groups = var_6000, pad = k_53_pad_0, pad_type = k_53_pad_type_0, strides = var_6536, weight = mid_block_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor("k_53_cast")]; + tensor var_6542 = const()[name = tensor("op_6542"), val = tensor([1, 1])]; + tensor var_6544 = const()[name = tensor("op_6544"), val = tensor([1, 1])]; + tensor v_27_pad_type_0 = const()[name = tensor("v_27_pad_type_0"), val = tensor("custom")]; + tensor v_27_pad_0 = const()[name = tensor("v_27_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor mid_block_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(222409280))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(223392384))), name = tensor("mid_block_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized"), shape = tensor([1280, 1024, 1, 1])]; + tensor v_27_cast = conv(dilations = var_6544, groups = var_6000, pad = v_27_pad_0, pad_type = v_27_pad_type_0, strides = var_6542, weight = mid_block_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor("v_27_cast")]; + tensor var_6548_begin_0 = const()[name = tensor("op_6548_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6548_end_0 = const()[name = tensor("op_6548_end_0"), val = tensor([2, 64, 1, 64])]; + tensor var_6548_end_mask_0 = const()[name = tensor("op_6548_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6548_cast = slice_by_index(begin = var_6548_begin_0, end = var_6548_end_0, end_mask = var_6548_end_mask_0, x = q_27_cast)[name = tensor("op_6548_cast")]; + tensor var_6552_begin_0 = const()[name = tensor("op_6552_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_6552_end_0 = const()[name = tensor("op_6552_end_0"), val = tensor([2, 128, 1, 64])]; + tensor var_6552_end_mask_0 = const()[name = tensor("op_6552_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6552_cast = slice_by_index(begin = var_6552_begin_0, end = var_6552_end_0, end_mask = var_6552_end_mask_0, x = q_27_cast)[name = tensor("op_6552_cast")]; + tensor var_6556_begin_0 = const()[name = tensor("op_6556_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_6556_end_0 = const()[name = tensor("op_6556_end_0"), val = tensor([2, 192, 1, 64])]; + tensor var_6556_end_mask_0 = const()[name = tensor("op_6556_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6556_cast = slice_by_index(begin = var_6556_begin_0, end = var_6556_end_0, end_mask = var_6556_end_mask_0, x = q_27_cast)[name = tensor("op_6556_cast")]; + tensor var_6560_begin_0 = const()[name = tensor("op_6560_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_6560_end_0 = const()[name = tensor("op_6560_end_0"), val = tensor([2, 256, 1, 64])]; + tensor var_6560_end_mask_0 = const()[name = tensor("op_6560_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6560_cast = slice_by_index(begin = var_6560_begin_0, end = var_6560_end_0, end_mask = var_6560_end_mask_0, x = q_27_cast)[name = tensor("op_6560_cast")]; + tensor var_6564_begin_0 = const()[name = tensor("op_6564_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_6564_end_0 = const()[name = tensor("op_6564_end_0"), val = tensor([2, 320, 1, 64])]; + tensor var_6564_end_mask_0 = const()[name = tensor("op_6564_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6564_cast = slice_by_index(begin = var_6564_begin_0, end = var_6564_end_0, end_mask = var_6564_end_mask_0, x = q_27_cast)[name = tensor("op_6564_cast")]; + tensor var_6568_begin_0 = const()[name = tensor("op_6568_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_6568_end_0 = const()[name = tensor("op_6568_end_0"), val = tensor([2, 384, 1, 64])]; + tensor var_6568_end_mask_0 = const()[name = tensor("op_6568_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6568_cast = slice_by_index(begin = var_6568_begin_0, end = var_6568_end_0, end_mask = var_6568_end_mask_0, x = q_27_cast)[name = tensor("op_6568_cast")]; + tensor var_6572_begin_0 = const()[name = tensor("op_6572_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_6572_end_0 = const()[name = tensor("op_6572_end_0"), val = tensor([2, 448, 1, 64])]; + tensor var_6572_end_mask_0 = const()[name = tensor("op_6572_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6572_cast = slice_by_index(begin = var_6572_begin_0, end = var_6572_end_0, end_mask = var_6572_end_mask_0, x = q_27_cast)[name = tensor("op_6572_cast")]; + tensor var_6576_begin_0 = const()[name = tensor("op_6576_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_6576_end_0 = const()[name = tensor("op_6576_end_0"), val = tensor([2, 512, 1, 64])]; + tensor var_6576_end_mask_0 = const()[name = tensor("op_6576_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6576_cast = slice_by_index(begin = var_6576_begin_0, end = var_6576_end_0, end_mask = var_6576_end_mask_0, x = q_27_cast)[name = tensor("op_6576_cast")]; + tensor var_6580_begin_0 = const()[name = tensor("op_6580_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_6580_end_0 = const()[name = tensor("op_6580_end_0"), val = tensor([2, 576, 1, 64])]; + tensor var_6580_end_mask_0 = const()[name = tensor("op_6580_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6580_cast = slice_by_index(begin = var_6580_begin_0, end = var_6580_end_0, end_mask = var_6580_end_mask_0, x = q_27_cast)[name = tensor("op_6580_cast")]; + tensor var_6584_begin_0 = const()[name = tensor("op_6584_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_6584_end_0 = const()[name = tensor("op_6584_end_0"), val = tensor([2, 640, 1, 64])]; + tensor var_6584_end_mask_0 = const()[name = tensor("op_6584_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6584_cast = slice_by_index(begin = var_6584_begin_0, end = var_6584_end_0, end_mask = var_6584_end_mask_0, x = q_27_cast)[name = tensor("op_6584_cast")]; + tensor var_6588_begin_0 = const()[name = tensor("op_6588_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_6588_end_0 = const()[name = tensor("op_6588_end_0"), val = tensor([2, 704, 1, 64])]; + tensor var_6588_end_mask_0 = const()[name = tensor("op_6588_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6588_cast = slice_by_index(begin = var_6588_begin_0, end = var_6588_end_0, end_mask = var_6588_end_mask_0, x = q_27_cast)[name = tensor("op_6588_cast")]; + tensor var_6592_begin_0 = const()[name = tensor("op_6592_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_6592_end_0 = const()[name = tensor("op_6592_end_0"), val = tensor([2, 768, 1, 64])]; + tensor var_6592_end_mask_0 = const()[name = tensor("op_6592_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6592_cast = slice_by_index(begin = var_6592_begin_0, end = var_6592_end_0, end_mask = var_6592_end_mask_0, x = q_27_cast)[name = tensor("op_6592_cast")]; + tensor var_6596_begin_0 = const()[name = tensor("op_6596_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_6596_end_0 = const()[name = tensor("op_6596_end_0"), val = tensor([2, 832, 1, 64])]; + tensor var_6596_end_mask_0 = const()[name = tensor("op_6596_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6596_cast = slice_by_index(begin = var_6596_begin_0, end = var_6596_end_0, end_mask = var_6596_end_mask_0, x = q_27_cast)[name = tensor("op_6596_cast")]; + tensor var_6600_begin_0 = const()[name = tensor("op_6600_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_6600_end_0 = const()[name = tensor("op_6600_end_0"), val = tensor([2, 896, 1, 64])]; + tensor var_6600_end_mask_0 = const()[name = tensor("op_6600_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6600_cast = slice_by_index(begin = var_6600_begin_0, end = var_6600_end_0, end_mask = var_6600_end_mask_0, x = q_27_cast)[name = tensor("op_6600_cast")]; + tensor var_6604_begin_0 = const()[name = tensor("op_6604_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_6604_end_0 = const()[name = tensor("op_6604_end_0"), val = tensor([2, 960, 1, 64])]; + tensor var_6604_end_mask_0 = const()[name = tensor("op_6604_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6604_cast = slice_by_index(begin = var_6604_begin_0, end = var_6604_end_0, end_mask = var_6604_end_mask_0, x = q_27_cast)[name = tensor("op_6604_cast")]; + tensor var_6608_begin_0 = const()[name = tensor("op_6608_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_6608_end_0 = const()[name = tensor("op_6608_end_0"), val = tensor([2, 1024, 1, 64])]; + tensor var_6608_end_mask_0 = const()[name = tensor("op_6608_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6608_cast = slice_by_index(begin = var_6608_begin_0, end = var_6608_end_0, end_mask = var_6608_end_mask_0, x = q_27_cast)[name = tensor("op_6608_cast")]; + tensor var_6612_begin_0 = const()[name = tensor("op_6612_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_6612_end_0 = const()[name = tensor("op_6612_end_0"), val = tensor([2, 1088, 1, 64])]; + tensor var_6612_end_mask_0 = const()[name = tensor("op_6612_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6612_cast = slice_by_index(begin = var_6612_begin_0, end = var_6612_end_0, end_mask = var_6612_end_mask_0, x = q_27_cast)[name = tensor("op_6612_cast")]; + tensor var_6616_begin_0 = const()[name = tensor("op_6616_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_6616_end_0 = const()[name = tensor("op_6616_end_0"), val = tensor([2, 1152, 1, 64])]; + tensor var_6616_end_mask_0 = const()[name = tensor("op_6616_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6616_cast = slice_by_index(begin = var_6616_begin_0, end = var_6616_end_0, end_mask = var_6616_end_mask_0, x = q_27_cast)[name = tensor("op_6616_cast")]; + tensor var_6620_begin_0 = const()[name = tensor("op_6620_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_6620_end_0 = const()[name = tensor("op_6620_end_0"), val = tensor([2, 1216, 1, 64])]; + tensor var_6620_end_mask_0 = const()[name = tensor("op_6620_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6620_cast = slice_by_index(begin = var_6620_begin_0, end = var_6620_end_0, end_mask = var_6620_end_mask_0, x = q_27_cast)[name = tensor("op_6620_cast")]; + tensor var_6624_begin_0 = const()[name = tensor("op_6624_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_6624_end_0 = const()[name = tensor("op_6624_end_0"), val = tensor([2, 1280, 1, 64])]; + tensor var_6624_end_mask_0 = const()[name = tensor("op_6624_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6624_cast = slice_by_index(begin = var_6624_begin_0, end = var_6624_end_0, end_mask = var_6624_end_mask_0, x = q_27_cast)[name = tensor("op_6624_cast")]; + tensor k_55_perm_0 = const()[name = tensor("k_55_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_6631_begin_0 = const()[name = tensor("op_6631_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6631_end_0 = const()[name = tensor("op_6631_end_0"), val = tensor([2, 77, 1, 64])]; + tensor var_6631_end_mask_0 = const()[name = tensor("op_6631_end_mask_0"), val = tensor([true, true, true, false])]; + tensor transpose_18 = transpose(perm = k_55_perm_0, x = k_53_cast)[name = tensor("transpose_18")]; + tensor var_6631_cast = slice_by_index(begin = var_6631_begin_0, end = var_6631_end_0, end_mask = var_6631_end_mask_0, x = transpose_18)[name = tensor("op_6631_cast")]; + tensor var_6635_begin_0 = const()[name = tensor("op_6635_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_6635_end_0 = const()[name = tensor("op_6635_end_0"), val = tensor([2, 77, 1, 128])]; + tensor var_6635_end_mask_0 = const()[name = tensor("op_6635_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6635_cast = slice_by_index(begin = var_6635_begin_0, end = var_6635_end_0, end_mask = var_6635_end_mask_0, x = transpose_18)[name = tensor("op_6635_cast")]; + tensor var_6639_begin_0 = const()[name = tensor("op_6639_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_6639_end_0 = const()[name = tensor("op_6639_end_0"), val = tensor([2, 77, 1, 192])]; + tensor var_6639_end_mask_0 = const()[name = tensor("op_6639_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6639_cast = slice_by_index(begin = var_6639_begin_0, end = var_6639_end_0, end_mask = var_6639_end_mask_0, x = transpose_18)[name = tensor("op_6639_cast")]; + tensor var_6643_begin_0 = const()[name = tensor("op_6643_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_6643_end_0 = const()[name = tensor("op_6643_end_0"), val = tensor([2, 77, 1, 256])]; + tensor var_6643_end_mask_0 = const()[name = tensor("op_6643_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6643_cast = slice_by_index(begin = var_6643_begin_0, end = var_6643_end_0, end_mask = var_6643_end_mask_0, x = transpose_18)[name = tensor("op_6643_cast")]; + tensor var_6647_begin_0 = const()[name = tensor("op_6647_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_6647_end_0 = const()[name = tensor("op_6647_end_0"), val = tensor([2, 77, 1, 320])]; + tensor var_6647_end_mask_0 = const()[name = tensor("op_6647_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6647_cast = slice_by_index(begin = var_6647_begin_0, end = var_6647_end_0, end_mask = var_6647_end_mask_0, x = transpose_18)[name = tensor("op_6647_cast")]; + tensor var_6651_begin_0 = const()[name = tensor("op_6651_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_6651_end_0 = const()[name = tensor("op_6651_end_0"), val = tensor([2, 77, 1, 384])]; + tensor var_6651_end_mask_0 = const()[name = tensor("op_6651_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6651_cast = slice_by_index(begin = var_6651_begin_0, end = var_6651_end_0, end_mask = var_6651_end_mask_0, x = transpose_18)[name = tensor("op_6651_cast")]; + tensor var_6655_begin_0 = const()[name = tensor("op_6655_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_6655_end_0 = const()[name = tensor("op_6655_end_0"), val = tensor([2, 77, 1, 448])]; + tensor var_6655_end_mask_0 = const()[name = tensor("op_6655_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6655_cast = slice_by_index(begin = var_6655_begin_0, end = var_6655_end_0, end_mask = var_6655_end_mask_0, x = transpose_18)[name = tensor("op_6655_cast")]; + tensor var_6659_begin_0 = const()[name = tensor("op_6659_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_6659_end_0 = const()[name = tensor("op_6659_end_0"), val = tensor([2, 77, 1, 512])]; + tensor var_6659_end_mask_0 = const()[name = tensor("op_6659_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6659_cast = slice_by_index(begin = var_6659_begin_0, end = var_6659_end_0, end_mask = var_6659_end_mask_0, x = transpose_18)[name = tensor("op_6659_cast")]; + tensor var_6663_begin_0 = const()[name = tensor("op_6663_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_6663_end_0 = const()[name = tensor("op_6663_end_0"), val = tensor([2, 77, 1, 576])]; + tensor var_6663_end_mask_0 = const()[name = tensor("op_6663_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6663_cast = slice_by_index(begin = var_6663_begin_0, end = var_6663_end_0, end_mask = var_6663_end_mask_0, x = transpose_18)[name = tensor("op_6663_cast")]; + tensor var_6667_begin_0 = const()[name = tensor("op_6667_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_6667_end_0 = const()[name = tensor("op_6667_end_0"), val = tensor([2, 77, 1, 640])]; + tensor var_6667_end_mask_0 = const()[name = tensor("op_6667_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6667_cast = slice_by_index(begin = var_6667_begin_0, end = var_6667_end_0, end_mask = var_6667_end_mask_0, x = transpose_18)[name = tensor("op_6667_cast")]; + tensor var_6671_begin_0 = const()[name = tensor("op_6671_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_6671_end_0 = const()[name = tensor("op_6671_end_0"), val = tensor([2, 77, 1, 704])]; + tensor var_6671_end_mask_0 = const()[name = tensor("op_6671_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6671_cast = slice_by_index(begin = var_6671_begin_0, end = var_6671_end_0, end_mask = var_6671_end_mask_0, x = transpose_18)[name = tensor("op_6671_cast")]; + tensor var_6675_begin_0 = const()[name = tensor("op_6675_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_6675_end_0 = const()[name = tensor("op_6675_end_0"), val = tensor([2, 77, 1, 768])]; + tensor var_6675_end_mask_0 = const()[name = tensor("op_6675_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6675_cast = slice_by_index(begin = var_6675_begin_0, end = var_6675_end_0, end_mask = var_6675_end_mask_0, x = transpose_18)[name = tensor("op_6675_cast")]; + tensor var_6679_begin_0 = const()[name = tensor("op_6679_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_6679_end_0 = const()[name = tensor("op_6679_end_0"), val = tensor([2, 77, 1, 832])]; + tensor var_6679_end_mask_0 = const()[name = tensor("op_6679_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6679_cast = slice_by_index(begin = var_6679_begin_0, end = var_6679_end_0, end_mask = var_6679_end_mask_0, x = transpose_18)[name = tensor("op_6679_cast")]; + tensor var_6683_begin_0 = const()[name = tensor("op_6683_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_6683_end_0 = const()[name = tensor("op_6683_end_0"), val = tensor([2, 77, 1, 896])]; + tensor var_6683_end_mask_0 = const()[name = tensor("op_6683_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6683_cast = slice_by_index(begin = var_6683_begin_0, end = var_6683_end_0, end_mask = var_6683_end_mask_0, x = transpose_18)[name = tensor("op_6683_cast")]; + tensor var_6687_begin_0 = const()[name = tensor("op_6687_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_6687_end_0 = const()[name = tensor("op_6687_end_0"), val = tensor([2, 77, 1, 960])]; + tensor var_6687_end_mask_0 = const()[name = tensor("op_6687_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6687_cast = slice_by_index(begin = var_6687_begin_0, end = var_6687_end_0, end_mask = var_6687_end_mask_0, x = transpose_18)[name = tensor("op_6687_cast")]; + tensor var_6691_begin_0 = const()[name = tensor("op_6691_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_6691_end_0 = const()[name = tensor("op_6691_end_0"), val = tensor([2, 77, 1, 1024])]; + tensor var_6691_end_mask_0 = const()[name = tensor("op_6691_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6691_cast = slice_by_index(begin = var_6691_begin_0, end = var_6691_end_0, end_mask = var_6691_end_mask_0, x = transpose_18)[name = tensor("op_6691_cast")]; + tensor var_6695_begin_0 = const()[name = tensor("op_6695_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_6695_end_0 = const()[name = tensor("op_6695_end_0"), val = tensor([2, 77, 1, 1088])]; + tensor var_6695_end_mask_0 = const()[name = tensor("op_6695_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6695_cast = slice_by_index(begin = var_6695_begin_0, end = var_6695_end_0, end_mask = var_6695_end_mask_0, x = transpose_18)[name = tensor("op_6695_cast")]; + tensor var_6699_begin_0 = const()[name = tensor("op_6699_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_6699_end_0 = const()[name = tensor("op_6699_end_0"), val = tensor([2, 77, 1, 1152])]; + tensor var_6699_end_mask_0 = const()[name = tensor("op_6699_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6699_cast = slice_by_index(begin = var_6699_begin_0, end = var_6699_end_0, end_mask = var_6699_end_mask_0, x = transpose_18)[name = tensor("op_6699_cast")]; + tensor var_6703_begin_0 = const()[name = tensor("op_6703_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_6703_end_0 = const()[name = tensor("op_6703_end_0"), val = tensor([2, 77, 1, 1216])]; + tensor var_6703_end_mask_0 = const()[name = tensor("op_6703_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6703_cast = slice_by_index(begin = var_6703_begin_0, end = var_6703_end_0, end_mask = var_6703_end_mask_0, x = transpose_18)[name = tensor("op_6703_cast")]; + tensor var_6707_begin_0 = const()[name = tensor("op_6707_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_6707_end_0 = const()[name = tensor("op_6707_end_0"), val = tensor([2, 77, 1, 1280])]; + tensor var_6707_end_mask_0 = const()[name = tensor("op_6707_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6707_cast = slice_by_index(begin = var_6707_begin_0, end = var_6707_end_0, end_mask = var_6707_end_mask_0, x = transpose_18)[name = tensor("op_6707_cast")]; + tensor var_6709_begin_0 = const()[name = tensor("op_6709_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6709_end_0 = const()[name = tensor("op_6709_end_0"), val = tensor([2, 64, 1, 77])]; + tensor var_6709_end_mask_0 = const()[name = tensor("op_6709_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6709_cast = slice_by_index(begin = var_6709_begin_0, end = var_6709_end_0, end_mask = var_6709_end_mask_0, x = v_27_cast)[name = tensor("op_6709_cast")]; + tensor var_6713_begin_0 = const()[name = tensor("op_6713_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_6713_end_0 = const()[name = tensor("op_6713_end_0"), val = tensor([2, 128, 1, 77])]; + tensor var_6713_end_mask_0 = const()[name = tensor("op_6713_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6713_cast = slice_by_index(begin = var_6713_begin_0, end = var_6713_end_0, end_mask = var_6713_end_mask_0, x = v_27_cast)[name = tensor("op_6713_cast")]; + tensor var_6717_begin_0 = const()[name = tensor("op_6717_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_6717_end_0 = const()[name = tensor("op_6717_end_0"), val = tensor([2, 192, 1, 77])]; + tensor var_6717_end_mask_0 = const()[name = tensor("op_6717_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6717_cast = slice_by_index(begin = var_6717_begin_0, end = var_6717_end_0, end_mask = var_6717_end_mask_0, x = v_27_cast)[name = tensor("op_6717_cast")]; + tensor var_6721_begin_0 = const()[name = tensor("op_6721_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_6721_end_0 = const()[name = tensor("op_6721_end_0"), val = tensor([2, 256, 1, 77])]; + tensor var_6721_end_mask_0 = const()[name = tensor("op_6721_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6721_cast = slice_by_index(begin = var_6721_begin_0, end = var_6721_end_0, end_mask = var_6721_end_mask_0, x = v_27_cast)[name = tensor("op_6721_cast")]; + tensor var_6725_begin_0 = const()[name = tensor("op_6725_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_6725_end_0 = const()[name = tensor("op_6725_end_0"), val = tensor([2, 320, 1, 77])]; + tensor var_6725_end_mask_0 = const()[name = tensor("op_6725_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6725_cast = slice_by_index(begin = var_6725_begin_0, end = var_6725_end_0, end_mask = var_6725_end_mask_0, x = v_27_cast)[name = tensor("op_6725_cast")]; + tensor var_6729_begin_0 = const()[name = tensor("op_6729_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_6729_end_0 = const()[name = tensor("op_6729_end_0"), val = tensor([2, 384, 1, 77])]; + tensor var_6729_end_mask_0 = const()[name = tensor("op_6729_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6729_cast = slice_by_index(begin = var_6729_begin_0, end = var_6729_end_0, end_mask = var_6729_end_mask_0, x = v_27_cast)[name = tensor("op_6729_cast")]; + tensor var_6733_begin_0 = const()[name = tensor("op_6733_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_6733_end_0 = const()[name = tensor("op_6733_end_0"), val = tensor([2, 448, 1, 77])]; + tensor var_6733_end_mask_0 = const()[name = tensor("op_6733_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6733_cast = slice_by_index(begin = var_6733_begin_0, end = var_6733_end_0, end_mask = var_6733_end_mask_0, x = v_27_cast)[name = tensor("op_6733_cast")]; + tensor var_6737_begin_0 = const()[name = tensor("op_6737_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_6737_end_0 = const()[name = tensor("op_6737_end_0"), val = tensor([2, 512, 1, 77])]; + tensor var_6737_end_mask_0 = const()[name = tensor("op_6737_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6737_cast = slice_by_index(begin = var_6737_begin_0, end = var_6737_end_0, end_mask = var_6737_end_mask_0, x = v_27_cast)[name = tensor("op_6737_cast")]; + tensor var_6741_begin_0 = const()[name = tensor("op_6741_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_6741_end_0 = const()[name = tensor("op_6741_end_0"), val = tensor([2, 576, 1, 77])]; + tensor var_6741_end_mask_0 = const()[name = tensor("op_6741_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6741_cast = slice_by_index(begin = var_6741_begin_0, end = var_6741_end_0, end_mask = var_6741_end_mask_0, x = v_27_cast)[name = tensor("op_6741_cast")]; + tensor var_6745_begin_0 = const()[name = tensor("op_6745_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_6745_end_0 = const()[name = tensor("op_6745_end_0"), val = tensor([2, 640, 1, 77])]; + tensor var_6745_end_mask_0 = const()[name = tensor("op_6745_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6745_cast = slice_by_index(begin = var_6745_begin_0, end = var_6745_end_0, end_mask = var_6745_end_mask_0, x = v_27_cast)[name = tensor("op_6745_cast")]; + tensor var_6749_begin_0 = const()[name = tensor("op_6749_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_6749_end_0 = const()[name = tensor("op_6749_end_0"), val = tensor([2, 704, 1, 77])]; + tensor var_6749_end_mask_0 = const()[name = tensor("op_6749_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6749_cast = slice_by_index(begin = var_6749_begin_0, end = var_6749_end_0, end_mask = var_6749_end_mask_0, x = v_27_cast)[name = tensor("op_6749_cast")]; + tensor var_6753_begin_0 = const()[name = tensor("op_6753_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_6753_end_0 = const()[name = tensor("op_6753_end_0"), val = tensor([2, 768, 1, 77])]; + tensor var_6753_end_mask_0 = const()[name = tensor("op_6753_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6753_cast = slice_by_index(begin = var_6753_begin_0, end = var_6753_end_0, end_mask = var_6753_end_mask_0, x = v_27_cast)[name = tensor("op_6753_cast")]; + tensor var_6757_begin_0 = const()[name = tensor("op_6757_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_6757_end_0 = const()[name = tensor("op_6757_end_0"), val = tensor([2, 832, 1, 77])]; + tensor var_6757_end_mask_0 = const()[name = tensor("op_6757_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6757_cast = slice_by_index(begin = var_6757_begin_0, end = var_6757_end_0, end_mask = var_6757_end_mask_0, x = v_27_cast)[name = tensor("op_6757_cast")]; + tensor var_6761_begin_0 = const()[name = tensor("op_6761_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_6761_end_0 = const()[name = tensor("op_6761_end_0"), val = tensor([2, 896, 1, 77])]; + tensor var_6761_end_mask_0 = const()[name = tensor("op_6761_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6761_cast = slice_by_index(begin = var_6761_begin_0, end = var_6761_end_0, end_mask = var_6761_end_mask_0, x = v_27_cast)[name = tensor("op_6761_cast")]; + tensor var_6765_begin_0 = const()[name = tensor("op_6765_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_6765_end_0 = const()[name = tensor("op_6765_end_0"), val = tensor([2, 960, 1, 77])]; + tensor var_6765_end_mask_0 = const()[name = tensor("op_6765_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6765_cast = slice_by_index(begin = var_6765_begin_0, end = var_6765_end_0, end_mask = var_6765_end_mask_0, x = v_27_cast)[name = tensor("op_6765_cast")]; + tensor var_6769_begin_0 = const()[name = tensor("op_6769_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_6769_end_0 = const()[name = tensor("op_6769_end_0"), val = tensor([2, 1024, 1, 77])]; + tensor var_6769_end_mask_0 = const()[name = tensor("op_6769_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6769_cast = slice_by_index(begin = var_6769_begin_0, end = var_6769_end_0, end_mask = var_6769_end_mask_0, x = v_27_cast)[name = tensor("op_6769_cast")]; + tensor var_6773_begin_0 = const()[name = tensor("op_6773_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_6773_end_0 = const()[name = tensor("op_6773_end_0"), val = tensor([2, 1088, 1, 77])]; + tensor var_6773_end_mask_0 = const()[name = tensor("op_6773_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6773_cast = slice_by_index(begin = var_6773_begin_0, end = var_6773_end_0, end_mask = var_6773_end_mask_0, x = v_27_cast)[name = tensor("op_6773_cast")]; + tensor var_6777_begin_0 = const()[name = tensor("op_6777_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_6777_end_0 = const()[name = tensor("op_6777_end_0"), val = tensor([2, 1152, 1, 77])]; + tensor var_6777_end_mask_0 = const()[name = tensor("op_6777_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6777_cast = slice_by_index(begin = var_6777_begin_0, end = var_6777_end_0, end_mask = var_6777_end_mask_0, x = v_27_cast)[name = tensor("op_6777_cast")]; + tensor var_6781_begin_0 = const()[name = tensor("op_6781_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_6781_end_0 = const()[name = tensor("op_6781_end_0"), val = tensor([2, 1216, 1, 77])]; + tensor var_6781_end_mask_0 = const()[name = tensor("op_6781_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6781_cast = slice_by_index(begin = var_6781_begin_0, end = var_6781_end_0, end_mask = var_6781_end_mask_0, x = v_27_cast)[name = tensor("op_6781_cast")]; + tensor var_6785_begin_0 = const()[name = tensor("op_6785_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_6785_end_0 = const()[name = tensor("op_6785_end_0"), val = tensor([2, 1280, 1, 77])]; + tensor var_6785_end_mask_0 = const()[name = tensor("op_6785_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6785_cast = slice_by_index(begin = var_6785_begin_0, end = var_6785_end_0, end_mask = var_6785_end_mask_0, x = v_27_cast)[name = tensor("op_6785_cast")]; + tensor var_6789_equation_0 = const()[name = tensor("op_6789_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6789_cast = einsum(equation = var_6789_equation_0, values = (var_6631_cast, var_6548_cast))[name = tensor("op_6789_cast")]; + tensor var_6790_to_fp16 = const()[name = tensor("op_6790_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_201_cast = mul(x = var_6789_cast, y = var_6790_to_fp16)[name = tensor("aw_201_cast")]; + tensor var_6793_equation_0 = const()[name = tensor("op_6793_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6793_cast = einsum(equation = var_6793_equation_0, values = (var_6635_cast, var_6552_cast))[name = tensor("op_6793_cast")]; + tensor var_6794_to_fp16 = const()[name = tensor("op_6794_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_203_cast = mul(x = var_6793_cast, y = var_6794_to_fp16)[name = tensor("aw_203_cast")]; + tensor var_6797_equation_0 = const()[name = tensor("op_6797_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6797_cast = einsum(equation = var_6797_equation_0, values = (var_6639_cast, var_6556_cast))[name = tensor("op_6797_cast")]; + tensor var_6798_to_fp16 = const()[name = tensor("op_6798_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_205_cast = mul(x = var_6797_cast, y = var_6798_to_fp16)[name = tensor("aw_205_cast")]; + tensor var_6801_equation_0 = const()[name = tensor("op_6801_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6801_cast = einsum(equation = var_6801_equation_0, values = (var_6643_cast, var_6560_cast))[name = tensor("op_6801_cast")]; + tensor var_6802_to_fp16 = const()[name = tensor("op_6802_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_207_cast = mul(x = var_6801_cast, y = var_6802_to_fp16)[name = tensor("aw_207_cast")]; + tensor var_6805_equation_0 = const()[name = tensor("op_6805_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6805_cast = einsum(equation = var_6805_equation_0, values = (var_6647_cast, var_6564_cast))[name = tensor("op_6805_cast")]; + tensor var_6806_to_fp16 = const()[name = tensor("op_6806_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_209_cast = mul(x = var_6805_cast, y = var_6806_to_fp16)[name = tensor("aw_209_cast")]; + tensor var_6809_equation_0 = const()[name = tensor("op_6809_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6809_cast = einsum(equation = var_6809_equation_0, values = (var_6651_cast, var_6568_cast))[name = tensor("op_6809_cast")]; + tensor var_6810_to_fp16 = const()[name = tensor("op_6810_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_211_cast = mul(x = var_6809_cast, y = var_6810_to_fp16)[name = tensor("aw_211_cast")]; + tensor var_6813_equation_0 = const()[name = tensor("op_6813_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6813_cast = einsum(equation = var_6813_equation_0, values = (var_6655_cast, var_6572_cast))[name = tensor("op_6813_cast")]; + tensor var_6814_to_fp16 = const()[name = tensor("op_6814_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_213_cast = mul(x = var_6813_cast, y = var_6814_to_fp16)[name = tensor("aw_213_cast")]; + tensor var_6817_equation_0 = const()[name = tensor("op_6817_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6817_cast = einsum(equation = var_6817_equation_0, values = (var_6659_cast, var_6576_cast))[name = tensor("op_6817_cast")]; + tensor var_6818_to_fp16 = const()[name = tensor("op_6818_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_215_cast = mul(x = var_6817_cast, y = var_6818_to_fp16)[name = tensor("aw_215_cast")]; + tensor var_6821_equation_0 = const()[name = tensor("op_6821_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6821_cast = einsum(equation = var_6821_equation_0, values = (var_6663_cast, var_6580_cast))[name = tensor("op_6821_cast")]; + tensor var_6822_to_fp16 = const()[name = tensor("op_6822_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_217_cast = mul(x = var_6821_cast, y = var_6822_to_fp16)[name = tensor("aw_217_cast")]; + tensor var_6825_equation_0 = const()[name = tensor("op_6825_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6825_cast = einsum(equation = var_6825_equation_0, values = (var_6667_cast, var_6584_cast))[name = tensor("op_6825_cast")]; + tensor var_6826_to_fp16 = const()[name = tensor("op_6826_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_219_cast = mul(x = var_6825_cast, y = var_6826_to_fp16)[name = tensor("aw_219_cast")]; + tensor var_6829_equation_0 = const()[name = tensor("op_6829_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6829_cast = einsum(equation = var_6829_equation_0, values = (var_6671_cast, var_6588_cast))[name = tensor("op_6829_cast")]; + tensor var_6830_to_fp16 = const()[name = tensor("op_6830_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_221_cast = mul(x = var_6829_cast, y = var_6830_to_fp16)[name = tensor("aw_221_cast")]; + tensor var_6833_equation_0 = const()[name = tensor("op_6833_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6833_cast = einsum(equation = var_6833_equation_0, values = (var_6675_cast, var_6592_cast))[name = tensor("op_6833_cast")]; + tensor var_6834_to_fp16 = const()[name = tensor("op_6834_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_223_cast = mul(x = var_6833_cast, y = var_6834_to_fp16)[name = tensor("aw_223_cast")]; + tensor var_6837_equation_0 = const()[name = tensor("op_6837_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6837_cast = einsum(equation = var_6837_equation_0, values = (var_6679_cast, var_6596_cast))[name = tensor("op_6837_cast")]; + tensor var_6838_to_fp16 = const()[name = tensor("op_6838_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_225_cast = mul(x = var_6837_cast, y = var_6838_to_fp16)[name = tensor("aw_225_cast")]; + tensor var_6841_equation_0 = const()[name = tensor("op_6841_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6841_cast = einsum(equation = var_6841_equation_0, values = (var_6683_cast, var_6600_cast))[name = tensor("op_6841_cast")]; + tensor var_6842_to_fp16 = const()[name = tensor("op_6842_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_227_cast = mul(x = var_6841_cast, y = var_6842_to_fp16)[name = tensor("aw_227_cast")]; + tensor var_6845_equation_0 = const()[name = tensor("op_6845_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6845_cast = einsum(equation = var_6845_equation_0, values = (var_6687_cast, var_6604_cast))[name = tensor("op_6845_cast")]; + tensor var_6846_to_fp16 = const()[name = tensor("op_6846_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_229_cast = mul(x = var_6845_cast, y = var_6846_to_fp16)[name = tensor("aw_229_cast")]; + tensor var_6849_equation_0 = const()[name = tensor("op_6849_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6849_cast = einsum(equation = var_6849_equation_0, values = (var_6691_cast, var_6608_cast))[name = tensor("op_6849_cast")]; + tensor var_6850_to_fp16 = const()[name = tensor("op_6850_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_231_cast = mul(x = var_6849_cast, y = var_6850_to_fp16)[name = tensor("aw_231_cast")]; + tensor var_6853_equation_0 = const()[name = tensor("op_6853_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6853_cast = einsum(equation = var_6853_equation_0, values = (var_6695_cast, var_6612_cast))[name = tensor("op_6853_cast")]; + tensor var_6854_to_fp16 = const()[name = tensor("op_6854_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_233_cast = mul(x = var_6853_cast, y = var_6854_to_fp16)[name = tensor("aw_233_cast")]; + tensor var_6857_equation_0 = const()[name = tensor("op_6857_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6857_cast = einsum(equation = var_6857_equation_0, values = (var_6699_cast, var_6616_cast))[name = tensor("op_6857_cast")]; + tensor var_6858_to_fp16 = const()[name = tensor("op_6858_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_235_cast = mul(x = var_6857_cast, y = var_6858_to_fp16)[name = tensor("aw_235_cast")]; + tensor var_6861_equation_0 = const()[name = tensor("op_6861_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6861_cast = einsum(equation = var_6861_equation_0, values = (var_6703_cast, var_6620_cast))[name = tensor("op_6861_cast")]; + tensor var_6862_to_fp16 = const()[name = tensor("op_6862_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_237_cast = mul(x = var_6861_cast, y = var_6862_to_fp16)[name = tensor("aw_237_cast")]; + tensor var_6865_equation_0 = const()[name = tensor("op_6865_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6865_cast = einsum(equation = var_6865_equation_0, values = (var_6707_cast, var_6624_cast))[name = tensor("op_6865_cast")]; + tensor var_6866_to_fp16 = const()[name = tensor("op_6866_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_239_cast = mul(x = var_6865_cast, y = var_6866_to_fp16)[name = tensor("aw_239_cast")]; + tensor var_6868_cast = softmax(axis = var_6000, x = aw_201_cast)[name = tensor("op_6868_cast")]; + tensor var_6869_cast = softmax(axis = var_6000, x = aw_203_cast)[name = tensor("op_6869_cast")]; + tensor var_6870_cast = softmax(axis = var_6000, x = aw_205_cast)[name = tensor("op_6870_cast")]; + tensor var_6871_cast = softmax(axis = var_6000, x = aw_207_cast)[name = tensor("op_6871_cast")]; + tensor var_6872_cast = softmax(axis = var_6000, x = aw_209_cast)[name = tensor("op_6872_cast")]; + tensor var_6873_cast = softmax(axis = var_6000, x = aw_211_cast)[name = tensor("op_6873_cast")]; + tensor var_6874_cast = softmax(axis = var_6000, x = aw_213_cast)[name = tensor("op_6874_cast")]; + tensor var_6875_cast = softmax(axis = var_6000, x = aw_215_cast)[name = tensor("op_6875_cast")]; + tensor var_6876_cast = softmax(axis = var_6000, x = aw_217_cast)[name = tensor("op_6876_cast")]; + tensor var_6877_cast = softmax(axis = var_6000, x = aw_219_cast)[name = tensor("op_6877_cast")]; + tensor var_6878_cast = softmax(axis = var_6000, x = aw_221_cast)[name = tensor("op_6878_cast")]; + tensor var_6879_cast = softmax(axis = var_6000, x = aw_223_cast)[name = tensor("op_6879_cast")]; + tensor var_6880_cast = softmax(axis = var_6000, x = aw_225_cast)[name = tensor("op_6880_cast")]; + tensor var_6881_cast = softmax(axis = var_6000, x = aw_227_cast)[name = tensor("op_6881_cast")]; + tensor var_6882_cast = softmax(axis = var_6000, x = aw_229_cast)[name = tensor("op_6882_cast")]; + tensor var_6883_cast = softmax(axis = var_6000, x = aw_231_cast)[name = tensor("op_6883_cast")]; + tensor var_6884_cast = softmax(axis = var_6000, x = aw_233_cast)[name = tensor("op_6884_cast")]; + tensor var_6885_cast = softmax(axis = var_6000, x = aw_235_cast)[name = tensor("op_6885_cast")]; + tensor var_6886_cast = softmax(axis = var_6000, x = aw_237_cast)[name = tensor("op_6886_cast")]; + tensor var_6887_cast = softmax(axis = var_6000, x = aw_239_cast)[name = tensor("op_6887_cast")]; + tensor var_6889_equation_0 = const()[name = tensor("op_6889_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6889_cast = einsum(equation = var_6889_equation_0, values = (var_6709_cast, var_6868_cast))[name = tensor("op_6889_cast")]; + tensor var_6891_equation_0 = const()[name = tensor("op_6891_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6891_cast = einsum(equation = var_6891_equation_0, values = (var_6713_cast, var_6869_cast))[name = tensor("op_6891_cast")]; + tensor var_6893_equation_0 = const()[name = tensor("op_6893_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6893_cast = einsum(equation = var_6893_equation_0, values = (var_6717_cast, var_6870_cast))[name = tensor("op_6893_cast")]; + tensor var_6895_equation_0 = const()[name = tensor("op_6895_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6895_cast = einsum(equation = var_6895_equation_0, values = (var_6721_cast, var_6871_cast))[name = tensor("op_6895_cast")]; + tensor var_6897_equation_0 = const()[name = tensor("op_6897_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6897_cast = einsum(equation = var_6897_equation_0, values = (var_6725_cast, var_6872_cast))[name = tensor("op_6897_cast")]; + tensor var_6899_equation_0 = const()[name = tensor("op_6899_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6899_cast = einsum(equation = var_6899_equation_0, values = (var_6729_cast, var_6873_cast))[name = tensor("op_6899_cast")]; + tensor var_6901_equation_0 = const()[name = tensor("op_6901_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6901_cast = einsum(equation = var_6901_equation_0, values = (var_6733_cast, var_6874_cast))[name = tensor("op_6901_cast")]; + tensor var_6903_equation_0 = const()[name = tensor("op_6903_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6903_cast = einsum(equation = var_6903_equation_0, values = (var_6737_cast, var_6875_cast))[name = tensor("op_6903_cast")]; + tensor var_6905_equation_0 = const()[name = tensor("op_6905_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6905_cast = einsum(equation = var_6905_equation_0, values = (var_6741_cast, var_6876_cast))[name = tensor("op_6905_cast")]; + tensor var_6907_equation_0 = const()[name = tensor("op_6907_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6907_cast = einsum(equation = var_6907_equation_0, values = (var_6745_cast, var_6877_cast))[name = tensor("op_6907_cast")]; + tensor var_6909_equation_0 = const()[name = tensor("op_6909_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6909_cast = einsum(equation = var_6909_equation_0, values = (var_6749_cast, var_6878_cast))[name = tensor("op_6909_cast")]; + tensor var_6911_equation_0 = const()[name = tensor("op_6911_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6911_cast = einsum(equation = var_6911_equation_0, values = (var_6753_cast, var_6879_cast))[name = tensor("op_6911_cast")]; + tensor var_6913_equation_0 = const()[name = tensor("op_6913_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6913_cast = einsum(equation = var_6913_equation_0, values = (var_6757_cast, var_6880_cast))[name = tensor("op_6913_cast")]; + tensor var_6915_equation_0 = const()[name = tensor("op_6915_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6915_cast = einsum(equation = var_6915_equation_0, values = (var_6761_cast, var_6881_cast))[name = tensor("op_6915_cast")]; + tensor var_6917_equation_0 = const()[name = tensor("op_6917_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6917_cast = einsum(equation = var_6917_equation_0, values = (var_6765_cast, var_6882_cast))[name = tensor("op_6917_cast")]; + tensor var_6919_equation_0 = const()[name = tensor("op_6919_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6919_cast = einsum(equation = var_6919_equation_0, values = (var_6769_cast, var_6883_cast))[name = tensor("op_6919_cast")]; + tensor var_6921_equation_0 = const()[name = tensor("op_6921_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6921_cast = einsum(equation = var_6921_equation_0, values = (var_6773_cast, var_6884_cast))[name = tensor("op_6921_cast")]; + tensor var_6923_equation_0 = const()[name = tensor("op_6923_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6923_cast = einsum(equation = var_6923_equation_0, values = (var_6777_cast, var_6885_cast))[name = tensor("op_6923_cast")]; + tensor var_6925_equation_0 = const()[name = tensor("op_6925_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6925_cast = einsum(equation = var_6925_equation_0, values = (var_6781_cast, var_6886_cast))[name = tensor("op_6925_cast")]; + tensor var_6927_equation_0 = const()[name = tensor("op_6927_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6927_cast = einsum(equation = var_6927_equation_0, values = (var_6785_cast, var_6887_cast))[name = tensor("op_6927_cast")]; + tensor input_217_interleave_0 = const()[name = tensor("input_217_interleave_0"), val = tensor(false)]; + tensor input_217_cast = concat(axis = var_6000, interleave = input_217_interleave_0, values = (var_6889_cast, var_6891_cast, var_6893_cast, var_6895_cast, var_6897_cast, var_6899_cast, var_6901_cast, var_6903_cast, var_6905_cast, var_6907_cast, var_6909_cast, var_6911_cast, var_6913_cast, var_6915_cast, var_6917_cast, var_6919_cast, var_6921_cast, var_6923_cast, var_6925_cast, var_6927_cast))[name = tensor("input_217_cast")]; + tensor var_6933 = const()[name = tensor("op_6933"), val = tensor([1, 1])]; + tensor var_6935 = const()[name = tensor("op_6935"), val = tensor([1, 1])]; + tensor var_6937_pad_type_0 = const()[name = tensor("op_6937_pad_type_0"), val = tensor("custom")]; + tensor var_6937_pad_0 = const()[name = tensor("op_6937_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor mid_block_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(223392576))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(224621440))), name = tensor("mid_block_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor mid_block_attentions_0_transformer_blocks_0_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_0_attn2_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(224621632)))]; + tensor var_6937_cast = conv(bias = mid_block_attentions_0_transformer_blocks_0_attn2_to_out_0_bias_to_fp16, dilations = var_6935, groups = var_6000, pad = var_6937_pad_0, pad_type = var_6937_pad_type_0, strides = var_6933, weight = mid_block_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16_palettized, x = input_217_cast)[name = tensor("op_6937_cast")]; + tensor inputs_41_cast = add(x = var_6937_cast, y = inputs_39_cast)[name = tensor("inputs_41_cast")]; + tensor var_6941 = const()[name = tensor("op_6941"), val = tensor([1])]; + tensor channels_mean_41_cast = reduce_mean(axes = var_6941, keep_dims = var_5995, x = inputs_41_cast)[name = tensor("channels_mean_41_cast")]; + tensor zero_mean_41_cast = sub(x = inputs_41_cast, y = channels_mean_41_cast)[name = tensor("zero_mean_41_cast")]; + tensor zero_mean_sq_41_cast = mul(x = zero_mean_41_cast, y = zero_mean_41_cast)[name = tensor("zero_mean_sq_41_cast")]; + tensor var_6945 = const()[name = tensor("op_6945"), val = tensor([1])]; + tensor var_6946_cast = reduce_mean(axes = var_6945, keep_dims = var_5995, x = zero_mean_sq_41_cast)[name = tensor("op_6946_cast")]; + tensor var_6947_to_fp16 = const()[name = tensor("op_6947_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_6948_cast = add(x = var_6946_cast, y = var_6947_to_fp16)[name = tensor("op_6948_cast")]; + tensor denom_41_epsilon_0_to_fp16 = const()[name = tensor("denom_41_epsilon_0_to_fp16"), val = tensor(0x1p-24)]; + tensor denom_41_cast = rsqrt(epsilon = denom_41_epsilon_0_to_fp16, x = var_6948_cast)[name = tensor("denom_41_cast")]; + tensor out_41_cast = mul(x = zero_mean_41_cast, y = denom_41_cast)[name = tensor("out_41_cast")]; + tensor var_6952_to_fp16 = const()[name = tensor("op_6952_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(224624256)))]; + tensor var_6953_cast = add(x = out_41_cast, y = var_6952_to_fp16)[name = tensor("op_6953_cast")]; + tensor var_6955_to_fp16 = const()[name = tensor("op_6955_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(224626880)))]; + tensor input_219_cast = mul(x = var_6953_cast, y = var_6955_to_fp16)[name = tensor("input_219_cast")]; + tensor var_6963 = const()[name = tensor("op_6963"), val = tensor([1, 1])]; + tensor var_6965 = const()[name = tensor("op_6965"), val = tensor([1, 1])]; + tensor var_6967_pad_type_0 = const()[name = tensor("op_6967_pad_type_0"), val = tensor("custom")]; + tensor var_6967_pad_0 = const()[name = tensor("op_6967_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor mid_block_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(224629504))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(234459968))), name = tensor("mid_block_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16_palettized"), shape = tensor([10240, 1280, 1, 1])]; + tensor mid_block_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(234460160))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(234467904))), name = tensor("mid_block_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16_palettized"), shape = tensor([10240])]; + tensor var_6967_cast = conv(bias = mid_block_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16_palettized, dilations = var_6965, groups = var_6000, pad = var_6967_pad_0, pad_type = var_6967_pad_type_0, strides = var_6963, weight = mid_block_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16_palettized, x = input_219_cast)[name = tensor("op_6967_cast")]; + tensor var_6968_split_sizes_0 = const()[name = tensor("op_6968_split_sizes_0"), val = tensor([5120, 5120])]; + tensor var_6968_axis_0 = const()[name = tensor("op_6968_axis_0"), val = tensor(1)]; + tensor var_6968_cast_0, tensor var_6968_cast_1 = split(axis = var_6968_axis_0, split_sizes = var_6968_split_sizes_0, x = var_6967_cast)[name = tensor("op_6968_cast")]; + tensor var_6970_mode_0 = const()[name = tensor("op_6970_mode_0"), val = tensor("EXACT")]; + tensor var_6970_cast = gelu(mode = var_6970_mode_0, x = var_6968_cast_1)[name = tensor("op_6970_cast")]; + tensor input_221_cast = mul(x = var_6968_cast_0, y = var_6970_cast)[name = tensor("input_221_cast")]; + tensor var_6974 = const()[name = tensor("op_6974"), val = tensor([1, 1])]; + tensor var_6976 = const()[name = tensor("op_6976"), val = tensor([1, 1])]; + tensor var_6978_pad_type_0 = const()[name = tensor("op_6978_pad_type_0"), val = tensor("custom")]; + tensor var_6978_pad_0 = const()[name = tensor("op_6978_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor mid_block_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(234468096))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(239383360))), name = tensor("mid_block_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16_palettized"), shape = tensor([1280, 5120, 1, 1])]; + tensor mid_block_attentions_0_transformer_blocks_0_ff_net_2_bias_to_fp16 = const()[name = tensor("mid_block_attentions_0_transformer_blocks_0_ff_net_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(239383552)))]; + tensor var_6978_cast = conv(bias = mid_block_attentions_0_transformer_blocks_0_ff_net_2_bias_to_fp16, dilations = var_6976, groups = var_6000, pad = var_6978_pad_0, pad_type = var_6978_pad_type_0, strides = var_6974, weight = mid_block_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16_palettized, x = input_221_cast)[name = tensor("op_6978_cast")]; + tensor hidden_states_131_cast = add(x = var_6978_cast, y = inputs_41_cast)[name = tensor("hidden_states_131_cast")]; + tensor var_6980 = const()[name = tensor("op_6980"), val = tensor([2, 1280, 8, 8])]; + tensor input_223_cast = reshape(shape = var_6980, x = hidden_states_131_cast)[name = tensor("input_223_cast")]; + tensor var_6984 = const()[name = tensor("op_6984"), val = tensor([1, 1])]; + tensor var_6986 = const()[name = tensor("op_6986"), val = tensor([1, 1])]; + tensor hidden_states_133_pad_type_0 = const()[name = tensor("hidden_states_133_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_133_pad_0 = const()[name = tensor("hidden_states_133_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor mid_block_attentions_0_proj_out_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(239386176))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(240615040))), name = tensor("mid_block_attentions_0_proj_out_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor mid_block_attentions_0_proj_out_bias_to_fp16 = const()[name = tensor("mid_block_attentions_0_proj_out_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(240615232)))]; + tensor hidden_states_133_cast = conv(bias = mid_block_attentions_0_proj_out_bias_to_fp16, dilations = var_6986, groups = var_6000, pad = hidden_states_133_pad_0, pad_type = hidden_states_133_pad_type_0, strides = var_6984, weight = mid_block_attentions_0_proj_out_weight_to_fp16_palettized, x = input_223_cast)[name = tensor("hidden_states_133_cast")]; + tensor input_225_cast = add(x = hidden_states_133_cast, y = hidden_states_121_cast)[name = tensor("input_225_cast")]; + tensor reshape_100_shape_0 = const()[name = tensor("reshape_100_shape_0"), val = tensor([2, 32, 40, 8, 8])]; + tensor reshape_100_cast = reshape(shape = reshape_100_shape_0, x = input_225_cast)[name = tensor("reshape_100_cast")]; + tensor reduce_mean_75_axes_0 = const()[name = tensor("reduce_mean_75_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_75_keep_dims_0 = const()[name = tensor("reduce_mean_75_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_75_cast = reduce_mean(axes = reduce_mean_75_axes_0, keep_dims = reduce_mean_75_keep_dims_0, x = reshape_100_cast)[name = tensor("reduce_mean_75_cast")]; + tensor sub_50_cast = sub(x = reshape_100_cast, y = reduce_mean_75_cast)[name = tensor("sub_50_cast")]; + tensor square_25_cast = square(x = sub_50_cast)[name = tensor("square_25_cast")]; + tensor reduce_mean_77_axes_0 = const()[name = tensor("reduce_mean_77_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_77_keep_dims_0 = const()[name = tensor("reduce_mean_77_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_77_cast = reduce_mean(axes = reduce_mean_77_axes_0, keep_dims = reduce_mean_77_keep_dims_0, x = square_25_cast)[name = tensor("reduce_mean_77_cast")]; + tensor add_50_y_0_to_fp16 = const()[name = tensor("add_50_y_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_50_cast = add(x = reduce_mean_77_cast, y = add_50_y_0_to_fp16)[name = tensor("add_50_cast")]; + tensor sqrt_25_cast = sqrt(x = add_50_cast)[name = tensor("sqrt_25_cast")]; + tensor real_div_25_cast = real_div(x = sub_50_cast, y = sqrt_25_cast)[name = tensor("real_div_25_cast")]; + tensor reshape_101_shape_0 = const()[name = tensor("reshape_101_shape_0"), val = tensor([2, 1280, 8, 8])]; + tensor reshape_101_cast = reshape(shape = reshape_101_shape_0, x = real_div_25_cast)[name = tensor("reshape_101_cast")]; + tensor add_51_gamma_0_to_fp16 = const()[name = tensor("add_51_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(240617856)))]; + tensor add_51_beta_0_to_fp16 = const()[name = tensor("add_51_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(240620480)))]; + tensor add_51_epsilon_0_to_fp16 = const()[name = tensor("add_51_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_51_cast = batch_norm(beta = add_51_beta_0_to_fp16, epsilon = add_51_epsilon_0_to_fp16, gamma = add_51_gamma_0_to_fp16, mean = add_27_mean_0_to_fp16, variance = add_27_variance_0_to_fp16, x = reshape_101_cast)[name = tensor("add_51_cast")]; + tensor input_229_cast = silu(x = add_51_cast)[name = tensor("input_229_cast")]; + tensor var_7001 = const()[name = tensor("op_7001"), val = tensor([1, 1])]; + tensor var_7003 = const()[name = tensor("op_7003"), val = tensor([1, 1])]; + tensor hidden_states_135_pad_type_0 = const()[name = tensor("hidden_states_135_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_135_pad_0 = const()[name = tensor("hidden_states_135_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor mid_block_resnets_1_conv1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(240623104))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(251682368))), name = tensor("mid_block_resnets_1_conv1_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 3, 3])]; + tensor mid_block_resnets_1_conv1_bias_to_fp16 = const()[name = tensor("mid_block_resnets_1_conv1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(251682560)))]; + tensor hidden_states_135_cast = conv(bias = mid_block_resnets_1_conv1_bias_to_fp16, dilations = var_7003, groups = var_6000, pad = hidden_states_135_pad_0, pad_type = hidden_states_135_pad_type_0, strides = var_7001, weight = mid_block_resnets_1_conv1_weight_to_fp16_palettized, x = input_229_cast)[name = tensor("hidden_states_135_cast")]; + tensor var_7009 = const()[name = tensor("op_7009"), val = tensor([1, 1])]; + tensor var_7011 = const()[name = tensor("op_7011"), val = tensor([1, 1])]; + tensor temb_19_pad_type_0 = const()[name = tensor("temb_19_pad_type_0"), val = tensor("custom")]; + tensor temb_19_pad_0 = const()[name = tensor("temb_19_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor mid_block_resnets_1_time_emb_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(251685184))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(252914048))), name = tensor("mid_block_resnets_1_time_emb_proj_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor mid_block_resnets_1_time_emb_proj_bias_to_fp16 = const()[name = tensor("mid_block_resnets_1_time_emb_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(252914240)))]; + tensor temb_19_cast = conv(bias = mid_block_resnets_1_time_emb_proj_bias_to_fp16, dilations = var_7011, groups = var_6000, pad = temb_19_pad_0, pad_type = temb_19_pad_type_0, strides = var_7009, weight = mid_block_resnets_1_time_emb_proj_weight_to_fp16_palettized, x = input_15_cast)[name = tensor("temb_19_cast")]; + tensor input_233_cast = add(x = hidden_states_135_cast, y = temb_19_cast)[name = tensor("input_233_cast")]; + tensor reshape_104_shape_0 = const()[name = tensor("reshape_104_shape_0"), val = tensor([2, 32, 40, 8, 8])]; + tensor reshape_104_cast = reshape(shape = reshape_104_shape_0, x = input_233_cast)[name = tensor("reshape_104_cast")]; + tensor reduce_mean_78_axes_0 = const()[name = tensor("reduce_mean_78_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_78_keep_dims_0 = const()[name = tensor("reduce_mean_78_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_78_cast = reduce_mean(axes = reduce_mean_78_axes_0, keep_dims = reduce_mean_78_keep_dims_0, x = reshape_104_cast)[name = tensor("reduce_mean_78_cast")]; + tensor sub_52_cast = sub(x = reshape_104_cast, y = reduce_mean_78_cast)[name = tensor("sub_52_cast")]; + tensor square_26_cast = square(x = sub_52_cast)[name = tensor("square_26_cast")]; + tensor reduce_mean_80_axes_0 = const()[name = tensor("reduce_mean_80_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_80_keep_dims_0 = const()[name = tensor("reduce_mean_80_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_80_cast = reduce_mean(axes = reduce_mean_80_axes_0, keep_dims = reduce_mean_80_keep_dims_0, x = square_26_cast)[name = tensor("reduce_mean_80_cast")]; + tensor add_52_y_0_to_fp16 = const()[name = tensor("add_52_y_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_52_cast = add(x = reduce_mean_80_cast, y = add_52_y_0_to_fp16)[name = tensor("add_52_cast")]; + tensor sqrt_26_cast = sqrt(x = add_52_cast)[name = tensor("sqrt_26_cast")]; + tensor real_div_26_cast = real_div(x = sub_52_cast, y = sqrt_26_cast)[name = tensor("real_div_26_cast")]; + tensor reshape_105_shape_0 = const()[name = tensor("reshape_105_shape_0"), val = tensor([2, 1280, 8, 8])]; + tensor reshape_105_cast = reshape(shape = reshape_105_shape_0, x = real_div_26_cast)[name = tensor("reshape_105_cast")]; + tensor add_53_gamma_0_to_fp16 = const()[name = tensor("add_53_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(252916864)))]; + tensor add_53_beta_0_to_fp16 = const()[name = tensor("add_53_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(252919488)))]; + tensor add_53_epsilon_0_to_fp16 = const()[name = tensor("add_53_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_53_cast = batch_norm(beta = add_53_beta_0_to_fp16, epsilon = add_53_epsilon_0_to_fp16, gamma = add_53_gamma_0_to_fp16, mean = add_27_mean_0_to_fp16, variance = add_27_variance_0_to_fp16, x = reshape_105_cast)[name = tensor("add_53_cast")]; + tensor input_237_cast = silu(x = add_53_cast)[name = tensor("input_237_cast")]; + tensor var_7021 = const()[name = tensor("op_7021"), val = tensor([1, 1])]; + tensor var_7023 = const()[name = tensor("op_7023"), val = tensor([1, 1])]; + tensor hidden_states_137_pad_type_0 = const()[name = tensor("hidden_states_137_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_137_pad_0 = const()[name = tensor("hidden_states_137_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor mid_block_resnets_1_conv2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(252922112))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(263981376))), name = tensor("mid_block_resnets_1_conv2_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 3, 3])]; + tensor mid_block_resnets_1_conv2_bias_to_fp16 = const()[name = tensor("mid_block_resnets_1_conv2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(263981568)))]; + tensor hidden_states_137_cast = conv(bias = mid_block_resnets_1_conv2_bias_to_fp16, dilations = var_7023, groups = var_6000, pad = hidden_states_137_pad_0, pad_type = hidden_states_137_pad_type_0, strides = var_7021, weight = mid_block_resnets_1_conv2_weight_to_fp16_palettized, x = input_237_cast)[name = tensor("hidden_states_137_cast")]; + tensor hidden_states_139_cast = add(x = input_225_cast, y = hidden_states_137_cast)[name = tensor("hidden_states_139_cast")]; + tensor var_7034 = const()[name = tensor("op_7034"), val = tensor(1)]; + tensor input_239_interleave_0 = const()[name = tensor("input_239_interleave_0"), val = tensor(false)]; + tensor input_239_cast = concat(axis = var_7034, interleave = input_239_interleave_0, values = (hidden_states_139_cast, input_199_cast))[name = tensor("input_239_cast")]; + tensor reshape_108_shape_0 = const()[name = tensor("reshape_108_shape_0"), val = tensor([2, 32, 80, 8, 8])]; + tensor reshape_108_cast = reshape(shape = reshape_108_shape_0, x = input_239_cast)[name = tensor("reshape_108_cast")]; + tensor reduce_mean_81_axes_0 = const()[name = tensor("reduce_mean_81_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_81_keep_dims_0 = const()[name = tensor("reduce_mean_81_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_81_cast = reduce_mean(axes = reduce_mean_81_axes_0, keep_dims = reduce_mean_81_keep_dims_0, x = reshape_108_cast)[name = tensor("reduce_mean_81_cast")]; + tensor sub_54_cast = sub(x = reshape_108_cast, y = reduce_mean_81_cast)[name = tensor("sub_54_cast")]; + tensor square_27_cast = square(x = sub_54_cast)[name = tensor("square_27_cast")]; + tensor reduce_mean_83_axes_0 = const()[name = tensor("reduce_mean_83_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_83_keep_dims_0 = const()[name = tensor("reduce_mean_83_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_83_cast = reduce_mean(axes = reduce_mean_83_axes_0, keep_dims = reduce_mean_83_keep_dims_0, x = square_27_cast)[name = tensor("reduce_mean_83_cast")]; + tensor add_54_y_0_to_fp16 = const()[name = tensor("add_54_y_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_54_cast = add(x = reduce_mean_83_cast, y = add_54_y_0_to_fp16)[name = tensor("add_54_cast")]; + tensor sqrt_27_cast = sqrt(x = add_54_cast)[name = tensor("sqrt_27_cast")]; + tensor real_div_27_cast = real_div(x = sub_54_cast, y = sqrt_27_cast)[name = tensor("real_div_27_cast")]; + tensor reshape_109_shape_0 = const()[name = tensor("reshape_109_shape_0"), val = tensor([2, 2560, 8, 8])]; + tensor reshape_109_cast = reshape(shape = reshape_109_shape_0, x = real_div_27_cast)[name = tensor("reshape_109_cast")]; + tensor add_55_mean_0_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(263984192))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(263986176))), name = tensor("add_55_mean_0_to_fp16_palettized"), shape = tensor([2560])]; + tensor add_55_variance_0_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(263986368))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(263988352))), name = tensor("add_55_variance_0_to_fp16_palettized"), shape = tensor([2560])]; + tensor add_55_gamma_0_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(263988544))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(263990528))), name = tensor("add_55_gamma_0_to_fp16_palettized"), shape = tensor([2560])]; + tensor add_55_beta_0_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(263990720))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(263992704))), name = tensor("add_55_beta_0_to_fp16_palettized"), shape = tensor([2560])]; + tensor add_55_epsilon_0_to_fp16 = const()[name = tensor("add_55_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_55_cast = batch_norm(beta = add_55_beta_0_to_fp16_palettized, epsilon = add_55_epsilon_0_to_fp16, gamma = add_55_gamma_0_to_fp16_palettized, mean = add_55_mean_0_to_fp16_palettized, variance = add_55_variance_0_to_fp16_palettized, x = reshape_109_cast)[name = tensor("add_55_cast")]; + tensor input_243_cast = silu(x = add_55_cast)[name = tensor("input_243_cast")]; + tensor var_7057 = const()[name = tensor("op_7057"), val = tensor([1, 1])]; + tensor var_7059 = const()[name = tensor("op_7059"), val = tensor([1, 1])]; + tensor hidden_states_141_pad_type_0 = const()[name = tensor("hidden_states_141_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_141_pad_0 = const()[name = tensor("hidden_states_141_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor up_blocks_0_resnets_0_conv1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(263992896))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(286111360))), name = tensor("up_blocks_0_resnets_0_conv1_weight_to_fp16_palettized"), shape = tensor([1280, 2560, 3, 3])]; + tensor up_blocks_0_resnets_0_conv1_bias_to_fp16 = const()[name = tensor("up_blocks_0_resnets_0_conv1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(286111552)))]; + tensor hidden_states_141_cast = conv(bias = up_blocks_0_resnets_0_conv1_bias_to_fp16, dilations = var_7059, groups = var_7034, pad = hidden_states_141_pad_0, pad_type = hidden_states_141_pad_type_0, strides = var_7057, weight = up_blocks_0_resnets_0_conv1_weight_to_fp16_palettized, x = input_243_cast)[name = tensor("hidden_states_141_cast")]; + tensor var_7065 = const()[name = tensor("op_7065"), val = tensor([1, 1])]; + tensor var_7067 = const()[name = tensor("op_7067"), val = tensor([1, 1])]; + tensor temb_21_pad_type_0 = const()[name = tensor("temb_21_pad_type_0"), val = tensor("custom")]; + tensor temb_21_pad_0 = const()[name = tensor("temb_21_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_resnets_0_time_emb_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(286114176))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(287343040))), name = tensor("up_blocks_0_resnets_0_time_emb_proj_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor up_blocks_0_resnets_0_time_emb_proj_bias_to_fp16 = const()[name = tensor("up_blocks_0_resnets_0_time_emb_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(287343232)))]; + tensor temb_21_cast = conv(bias = up_blocks_0_resnets_0_time_emb_proj_bias_to_fp16, dilations = var_7067, groups = var_7034, pad = temb_21_pad_0, pad_type = temb_21_pad_type_0, strides = var_7065, weight = up_blocks_0_resnets_0_time_emb_proj_weight_to_fp16_palettized, x = input_15_cast)[name = tensor("temb_21_cast")]; + tensor input_247_cast = add(x = hidden_states_141_cast, y = temb_21_cast)[name = tensor("input_247_cast")]; + tensor reshape_112_shape_0 = const()[name = tensor("reshape_112_shape_0"), val = tensor([2, 32, 40, 8, 8])]; + tensor reshape_112_cast = reshape(shape = reshape_112_shape_0, x = input_247_cast)[name = tensor("reshape_112_cast")]; + tensor reduce_mean_84_axes_0 = const()[name = tensor("reduce_mean_84_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_84_keep_dims_0 = const()[name = tensor("reduce_mean_84_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_84_cast = reduce_mean(axes = reduce_mean_84_axes_0, keep_dims = reduce_mean_84_keep_dims_0, x = reshape_112_cast)[name = tensor("reduce_mean_84_cast")]; + tensor sub_56_cast = sub(x = reshape_112_cast, y = reduce_mean_84_cast)[name = tensor("sub_56_cast")]; + tensor square_28_cast = square(x = sub_56_cast)[name = tensor("square_28_cast")]; + tensor reduce_mean_86_axes_0 = const()[name = tensor("reduce_mean_86_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_86_keep_dims_0 = const()[name = tensor("reduce_mean_86_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_86_cast = reduce_mean(axes = reduce_mean_86_axes_0, keep_dims = reduce_mean_86_keep_dims_0, x = square_28_cast)[name = tensor("reduce_mean_86_cast")]; + tensor add_56_y_0_to_fp16 = const()[name = tensor("add_56_y_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_56_cast = add(x = reduce_mean_86_cast, y = add_56_y_0_to_fp16)[name = tensor("add_56_cast")]; + tensor sqrt_28_cast = sqrt(x = add_56_cast)[name = tensor("sqrt_28_cast")]; + tensor real_div_28_cast = real_div(x = sub_56_cast, y = sqrt_28_cast)[name = tensor("real_div_28_cast")]; + tensor reshape_113_shape_0 = const()[name = tensor("reshape_113_shape_0"), val = tensor([2, 1280, 8, 8])]; + tensor reshape_113_cast = reshape(shape = reshape_113_shape_0, x = real_div_28_cast)[name = tensor("reshape_113_cast")]; + tensor add_57_gamma_0_to_fp16 = const()[name = tensor("add_57_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(287345856)))]; + tensor add_57_beta_0_to_fp16 = const()[name = tensor("add_57_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(287348480)))]; + tensor add_57_epsilon_0_to_fp16 = const()[name = tensor("add_57_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_57_cast = batch_norm(beta = add_57_beta_0_to_fp16, epsilon = add_57_epsilon_0_to_fp16, gamma = add_57_gamma_0_to_fp16, mean = add_27_mean_0_to_fp16, variance = add_27_variance_0_to_fp16, x = reshape_113_cast)[name = tensor("add_57_cast")]; + tensor input_251_cast = silu(x = add_57_cast)[name = tensor("input_251_cast")]; + tensor var_7077 = const()[name = tensor("op_7077"), val = tensor([1, 1])]; + tensor var_7079 = const()[name = tensor("op_7079"), val = tensor([1, 1])]; + tensor hidden_states_143_pad_type_0 = const()[name = tensor("hidden_states_143_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_143_pad_0 = const()[name = tensor("hidden_states_143_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor up_blocks_0_resnets_0_conv2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(287351104))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(298410368))), name = tensor("up_blocks_0_resnets_0_conv2_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 3, 3])]; + tensor up_blocks_0_resnets_0_conv2_bias_to_fp16 = const()[name = tensor("up_blocks_0_resnets_0_conv2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(298410560)))]; + tensor hidden_states_143_cast = conv(bias = up_blocks_0_resnets_0_conv2_bias_to_fp16, dilations = var_7079, groups = var_7034, pad = hidden_states_143_pad_0, pad_type = hidden_states_143_pad_type_0, strides = var_7077, weight = up_blocks_0_resnets_0_conv2_weight_to_fp16_palettized, x = input_251_cast)[name = tensor("hidden_states_143_cast")]; + tensor var_7084 = const()[name = tensor("op_7084"), val = tensor([1, 1])]; + tensor var_7086 = const()[name = tensor("op_7086"), val = tensor([1, 1])]; + tensor x_5_pad_type_0 = const()[name = tensor("x_5_pad_type_0"), val = tensor("custom")]; + tensor x_5_pad_0 = const()[name = tensor("x_5_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_resnets_0_conv_shortcut_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(298413184))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(300870848))), name = tensor("up_blocks_0_resnets_0_conv_shortcut_weight_to_fp16_palettized"), shape = tensor([1280, 2560, 1, 1])]; + tensor up_blocks_0_resnets_0_conv_shortcut_bias_to_fp16 = const()[name = tensor("up_blocks_0_resnets_0_conv_shortcut_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(300871040)))]; + tensor x_5_cast = conv(bias = up_blocks_0_resnets_0_conv_shortcut_bias_to_fp16, dilations = var_7086, groups = var_7034, pad = x_5_pad_0, pad_type = x_5_pad_type_0, strides = var_7084, weight = up_blocks_0_resnets_0_conv_shortcut_weight_to_fp16_palettized, x = input_239_cast)[name = tensor("x_5_cast")]; + tensor hidden_states_145_cast = add(x = x_5_cast, y = hidden_states_143_cast)[name = tensor("hidden_states_145_cast")]; + tensor input_253_interleave_0 = const()[name = tensor("input_253_interleave_0"), val = tensor(false)]; + tensor input_253_cast = concat(axis = var_7034, interleave = input_253_interleave_0, values = (hidden_states_145_cast, input_185_cast))[name = tensor("input_253_cast")]; + tensor reshape_116_shape_0 = const()[name = tensor("reshape_116_shape_0"), val = tensor([2, 32, 80, 8, 8])]; + tensor reshape_116_cast = reshape(shape = reshape_116_shape_0, x = input_253_cast)[name = tensor("reshape_116_cast")]; + tensor reduce_mean_87_axes_0 = const()[name = tensor("reduce_mean_87_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_87_keep_dims_0 = const()[name = tensor("reduce_mean_87_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_87_cast = reduce_mean(axes = reduce_mean_87_axes_0, keep_dims = reduce_mean_87_keep_dims_0, x = reshape_116_cast)[name = tensor("reduce_mean_87_cast")]; + tensor sub_58_cast = sub(x = reshape_116_cast, y = reduce_mean_87_cast)[name = tensor("sub_58_cast")]; + tensor square_29_cast = square(x = sub_58_cast)[name = tensor("square_29_cast")]; + tensor reduce_mean_89_axes_0 = const()[name = tensor("reduce_mean_89_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_89_keep_dims_0 = const()[name = tensor("reduce_mean_89_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_89_cast = reduce_mean(axes = reduce_mean_89_axes_0, keep_dims = reduce_mean_89_keep_dims_0, x = square_29_cast)[name = tensor("reduce_mean_89_cast")]; + tensor add_58_y_0_to_fp16 = const()[name = tensor("add_58_y_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_58_cast = add(x = reduce_mean_89_cast, y = add_58_y_0_to_fp16)[name = tensor("add_58_cast")]; + tensor sqrt_29_cast = sqrt(x = add_58_cast)[name = tensor("sqrt_29_cast")]; + tensor real_div_29_cast = real_div(x = sub_58_cast, y = sqrt_29_cast)[name = tensor("real_div_29_cast")]; + tensor reshape_117_shape_0 = const()[name = tensor("reshape_117_shape_0"), val = tensor([2, 2560, 8, 8])]; + tensor reshape_117_cast = reshape(shape = reshape_117_shape_0, x = real_div_29_cast)[name = tensor("reshape_117_cast")]; + tensor add_59_gamma_0_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(300873664))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(300875648))), name = tensor("add_59_gamma_0_to_fp16_palettized"), shape = tensor([2560])]; + tensor add_59_beta_0_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(300875840))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(300877824))), name = tensor("add_59_beta_0_to_fp16_palettized"), shape = tensor([2560])]; + tensor add_59_epsilon_0_to_fp16 = const()[name = tensor("add_59_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_59_cast = batch_norm(beta = add_59_beta_0_to_fp16_palettized, epsilon = add_59_epsilon_0_to_fp16, gamma = add_59_gamma_0_to_fp16_palettized, mean = add_55_mean_0_to_fp16_palettized, variance = add_55_variance_0_to_fp16_palettized, x = reshape_117_cast)[name = tensor("add_59_cast")]; + tensor input_257_cast = silu(x = add_59_cast)[name = tensor("input_257_cast")]; + tensor var_7104 = const()[name = tensor("op_7104"), val = tensor([1, 1])]; + tensor var_7106 = const()[name = tensor("op_7106"), val = tensor([1, 1])]; + tensor hidden_states_147_pad_type_0 = const()[name = tensor("hidden_states_147_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_147_pad_0 = const()[name = tensor("hidden_states_147_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor up_blocks_0_resnets_1_conv1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(300878016))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(322996480))), name = tensor("up_blocks_0_resnets_1_conv1_weight_to_fp16_palettized"), shape = tensor([1280, 2560, 3, 3])]; + tensor up_blocks_0_resnets_1_conv1_bias_to_fp16 = const()[name = tensor("up_blocks_0_resnets_1_conv1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(322996672)))]; + tensor hidden_states_147_cast = conv(bias = up_blocks_0_resnets_1_conv1_bias_to_fp16, dilations = var_7106, groups = var_7034, pad = hidden_states_147_pad_0, pad_type = hidden_states_147_pad_type_0, strides = var_7104, weight = up_blocks_0_resnets_1_conv1_weight_to_fp16_palettized, x = input_257_cast)[name = tensor("hidden_states_147_cast")]; + tensor var_7112 = const()[name = tensor("op_7112"), val = tensor([1, 1])]; + tensor var_7114 = const()[name = tensor("op_7114"), val = tensor([1, 1])]; + tensor temb_23_pad_type_0 = const()[name = tensor("temb_23_pad_type_0"), val = tensor("custom")]; + tensor temb_23_pad_0 = const()[name = tensor("temb_23_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_resnets_1_time_emb_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(322999296))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(324228160))), name = tensor("up_blocks_0_resnets_1_time_emb_proj_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor up_blocks_0_resnets_1_time_emb_proj_bias_to_fp16 = const()[name = tensor("up_blocks_0_resnets_1_time_emb_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(324228352)))]; + tensor temb_23_cast = conv(bias = up_blocks_0_resnets_1_time_emb_proj_bias_to_fp16, dilations = var_7114, groups = var_7034, pad = temb_23_pad_0, pad_type = temb_23_pad_type_0, strides = var_7112, weight = up_blocks_0_resnets_1_time_emb_proj_weight_to_fp16_palettized, x = input_15_cast)[name = tensor("temb_23_cast")]; + tensor input_261_cast = add(x = hidden_states_147_cast, y = temb_23_cast)[name = tensor("input_261_cast")]; + tensor reshape_120_shape_0 = const()[name = tensor("reshape_120_shape_0"), val = tensor([2, 32, 40, 8, 8])]; + tensor reshape_120_cast = reshape(shape = reshape_120_shape_0, x = input_261_cast)[name = tensor("reshape_120_cast")]; + tensor reduce_mean_90_axes_0 = const()[name = tensor("reduce_mean_90_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_90_keep_dims_0 = const()[name = tensor("reduce_mean_90_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_90_cast = reduce_mean(axes = reduce_mean_90_axes_0, keep_dims = reduce_mean_90_keep_dims_0, x = reshape_120_cast)[name = tensor("reduce_mean_90_cast")]; + tensor sub_60_cast = sub(x = reshape_120_cast, y = reduce_mean_90_cast)[name = tensor("sub_60_cast")]; + tensor square_30_cast = square(x = sub_60_cast)[name = tensor("square_30_cast")]; + tensor reduce_mean_92_axes_0 = const()[name = tensor("reduce_mean_92_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_92_keep_dims_0 = const()[name = tensor("reduce_mean_92_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_92_cast = reduce_mean(axes = reduce_mean_92_axes_0, keep_dims = reduce_mean_92_keep_dims_0, x = square_30_cast)[name = tensor("reduce_mean_92_cast")]; + tensor add_60_y_0_to_fp16 = const()[name = tensor("add_60_y_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_60_cast = add(x = reduce_mean_92_cast, y = add_60_y_0_to_fp16)[name = tensor("add_60_cast")]; + tensor sqrt_30_cast = sqrt(x = add_60_cast)[name = tensor("sqrt_30_cast")]; + tensor real_div_30_cast = real_div(x = sub_60_cast, y = sqrt_30_cast)[name = tensor("real_div_30_cast")]; + tensor reshape_121_shape_0 = const()[name = tensor("reshape_121_shape_0"), val = tensor([2, 1280, 8, 8])]; + tensor reshape_121_cast = reshape(shape = reshape_121_shape_0, x = real_div_30_cast)[name = tensor("reshape_121_cast")]; + tensor add_61_gamma_0_to_fp16 = const()[name = tensor("add_61_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(324230976)))]; + tensor add_61_beta_0_to_fp16 = const()[name = tensor("add_61_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(324233600)))]; + tensor add_61_epsilon_0_to_fp16 = const()[name = tensor("add_61_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_61_cast = batch_norm(beta = add_61_beta_0_to_fp16, epsilon = add_61_epsilon_0_to_fp16, gamma = add_61_gamma_0_to_fp16, mean = add_27_mean_0_to_fp16, variance = add_27_variance_0_to_fp16, x = reshape_121_cast)[name = tensor("add_61_cast")]; + tensor input_265_cast = silu(x = add_61_cast)[name = tensor("input_265_cast")]; + tensor var_7124 = const()[name = tensor("op_7124"), val = tensor([1, 1])]; + tensor var_7126 = const()[name = tensor("op_7126"), val = tensor([1, 1])]; + tensor hidden_states_149_pad_type_0 = const()[name = tensor("hidden_states_149_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_149_pad_0 = const()[name = tensor("hidden_states_149_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor up_blocks_0_resnets_1_conv2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(324236224))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(335295488))), name = tensor("up_blocks_0_resnets_1_conv2_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 3, 3])]; + tensor up_blocks_0_resnets_1_conv2_bias_to_fp16 = const()[name = tensor("up_blocks_0_resnets_1_conv2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(335295680)))]; + tensor hidden_states_149_cast = conv(bias = up_blocks_0_resnets_1_conv2_bias_to_fp16, dilations = var_7126, groups = var_7034, pad = hidden_states_149_pad_0, pad_type = hidden_states_149_pad_type_0, strides = var_7124, weight = up_blocks_0_resnets_1_conv2_weight_to_fp16_palettized, x = input_265_cast)[name = tensor("hidden_states_149_cast")]; + tensor var_7131 = const()[name = tensor("op_7131"), val = tensor([1, 1])]; + tensor var_7133 = const()[name = tensor("op_7133"), val = tensor([1, 1])]; + tensor x_7_pad_type_0 = const()[name = tensor("x_7_pad_type_0"), val = tensor("custom")]; + tensor x_7_pad_0 = const()[name = tensor("x_7_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_resnets_1_conv_shortcut_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(335298304))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(337755968))), name = tensor("up_blocks_0_resnets_1_conv_shortcut_weight_to_fp16_palettized"), shape = tensor([1280, 2560, 1, 1])]; + tensor up_blocks_0_resnets_1_conv_shortcut_bias_to_fp16 = const()[name = tensor("up_blocks_0_resnets_1_conv_shortcut_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(337756160)))]; + tensor x_7_cast = conv(bias = up_blocks_0_resnets_1_conv_shortcut_bias_to_fp16, dilations = var_7133, groups = var_7034, pad = x_7_pad_0, pad_type = x_7_pad_type_0, strides = var_7131, weight = up_blocks_0_resnets_1_conv_shortcut_weight_to_fp16_palettized, x = input_253_cast)[name = tensor("x_7_cast")]; + tensor hidden_states_151_cast = add(x = x_7_cast, y = hidden_states_149_cast)[name = tensor("hidden_states_151_cast")]; + tensor input_267_interleave_0 = const()[name = tensor("input_267_interleave_0"), val = tensor(false)]; + tensor input_267_cast = concat(axis = var_7034, interleave = input_267_interleave_0, values = (hidden_states_151_cast, input_171_cast))[name = tensor("input_267_cast")]; + tensor reshape_124_shape_0 = const()[name = tensor("reshape_124_shape_0"), val = tensor([2, 32, 80, 8, 8])]; + tensor reshape_124_cast = reshape(shape = reshape_124_shape_0, x = input_267_cast)[name = tensor("reshape_124_cast")]; + tensor reduce_mean_93_axes_0 = const()[name = tensor("reduce_mean_93_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_93_keep_dims_0 = const()[name = tensor("reduce_mean_93_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_93_cast = reduce_mean(axes = reduce_mean_93_axes_0, keep_dims = reduce_mean_93_keep_dims_0, x = reshape_124_cast)[name = tensor("reduce_mean_93_cast")]; + tensor sub_62_cast = sub(x = reshape_124_cast, y = reduce_mean_93_cast)[name = tensor("sub_62_cast")]; + tensor square_31_cast = square(x = sub_62_cast)[name = tensor("square_31_cast")]; + tensor reduce_mean_95_axes_0 = const()[name = tensor("reduce_mean_95_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_95_keep_dims_0 = const()[name = tensor("reduce_mean_95_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_95_cast = reduce_mean(axes = reduce_mean_95_axes_0, keep_dims = reduce_mean_95_keep_dims_0, x = square_31_cast)[name = tensor("reduce_mean_95_cast")]; + tensor add_62_y_0_to_fp16 = const()[name = tensor("add_62_y_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_62_cast = add(x = reduce_mean_95_cast, y = add_62_y_0_to_fp16)[name = tensor("add_62_cast")]; + tensor sqrt_31_cast = sqrt(x = add_62_cast)[name = tensor("sqrt_31_cast")]; + tensor real_div_31_cast = real_div(x = sub_62_cast, y = sqrt_31_cast)[name = tensor("real_div_31_cast")]; + tensor reshape_125_shape_0 = const()[name = tensor("reshape_125_shape_0"), val = tensor([2, 2560, 8, 8])]; + tensor reshape_125_cast = reshape(shape = reshape_125_shape_0, x = real_div_31_cast)[name = tensor("reshape_125_cast")]; + tensor add_63_gamma_0_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(337758784))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(337760768))), name = tensor("add_63_gamma_0_to_fp16_palettized"), shape = tensor([2560])]; + tensor add_63_beta_0_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(337760960))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(337762944))), name = tensor("add_63_beta_0_to_fp16_palettized"), shape = tensor([2560])]; + tensor add_63_epsilon_0_to_fp16 = const()[name = tensor("add_63_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_63_cast = batch_norm(beta = add_63_beta_0_to_fp16_palettized, epsilon = add_63_epsilon_0_to_fp16, gamma = add_63_gamma_0_to_fp16_palettized, mean = add_55_mean_0_to_fp16_palettized, variance = add_55_variance_0_to_fp16_palettized, x = reshape_125_cast)[name = tensor("add_63_cast")]; + tensor input_271_cast = silu(x = add_63_cast)[name = tensor("input_271_cast")]; + tensor var_7151 = const()[name = tensor("op_7151"), val = tensor([1, 1])]; + tensor var_7153 = const()[name = tensor("op_7153"), val = tensor([1, 1])]; + tensor hidden_states_153_pad_type_0 = const()[name = tensor("hidden_states_153_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_153_pad_0 = const()[name = tensor("hidden_states_153_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor up_blocks_0_resnets_2_conv1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(337763136))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(359881600))), name = tensor("up_blocks_0_resnets_2_conv1_weight_to_fp16_palettized"), shape = tensor([1280, 2560, 3, 3])]; + tensor up_blocks_0_resnets_2_conv1_bias_to_fp16 = const()[name = tensor("up_blocks_0_resnets_2_conv1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(359881792)))]; + tensor hidden_states_153_cast = conv(bias = up_blocks_0_resnets_2_conv1_bias_to_fp16, dilations = var_7153, groups = var_7034, pad = hidden_states_153_pad_0, pad_type = hidden_states_153_pad_type_0, strides = var_7151, weight = up_blocks_0_resnets_2_conv1_weight_to_fp16_palettized, x = input_271_cast)[name = tensor("hidden_states_153_cast")]; + tensor var_7159 = const()[name = tensor("op_7159"), val = tensor([1, 1])]; + tensor var_7161 = const()[name = tensor("op_7161"), val = tensor([1, 1])]; + tensor temb_25_pad_type_0 = const()[name = tensor("temb_25_pad_type_0"), val = tensor("custom")]; + tensor temb_25_pad_0 = const()[name = tensor("temb_25_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_resnets_2_time_emb_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(359884416))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(361113280))), name = tensor("up_blocks_0_resnets_2_time_emb_proj_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor up_blocks_0_resnets_2_time_emb_proj_bias_to_fp16 = const()[name = tensor("up_blocks_0_resnets_2_time_emb_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(361113472)))]; + tensor temb_25_cast = conv(bias = up_blocks_0_resnets_2_time_emb_proj_bias_to_fp16, dilations = var_7161, groups = var_7034, pad = temb_25_pad_0, pad_type = temb_25_pad_type_0, strides = var_7159, weight = up_blocks_0_resnets_2_time_emb_proj_weight_to_fp16_palettized, x = input_15_cast)[name = tensor("temb_25_cast")]; + tensor input_275_cast = add(x = hidden_states_153_cast, y = temb_25_cast)[name = tensor("input_275_cast")]; + tensor reshape_128_shape_0 = const()[name = tensor("reshape_128_shape_0"), val = tensor([2, 32, 40, 8, 8])]; + tensor reshape_128_cast = reshape(shape = reshape_128_shape_0, x = input_275_cast)[name = tensor("reshape_128_cast")]; + tensor reduce_mean_96_axes_0 = const()[name = tensor("reduce_mean_96_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_96_keep_dims_0 = const()[name = tensor("reduce_mean_96_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_96_cast = reduce_mean(axes = reduce_mean_96_axes_0, keep_dims = reduce_mean_96_keep_dims_0, x = reshape_128_cast)[name = tensor("reduce_mean_96_cast")]; + tensor sub_64_cast = sub(x = reshape_128_cast, y = reduce_mean_96_cast)[name = tensor("sub_64_cast")]; + tensor square_32_cast = square(x = sub_64_cast)[name = tensor("square_32_cast")]; + tensor reduce_mean_98_axes_0 = const()[name = tensor("reduce_mean_98_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_98_keep_dims_0 = const()[name = tensor("reduce_mean_98_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_98_cast = reduce_mean(axes = reduce_mean_98_axes_0, keep_dims = reduce_mean_98_keep_dims_0, x = square_32_cast)[name = tensor("reduce_mean_98_cast")]; + tensor add_64_y_0_to_fp16 = const()[name = tensor("add_64_y_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_64_cast = add(x = reduce_mean_98_cast, y = add_64_y_0_to_fp16)[name = tensor("add_64_cast")]; + tensor sqrt_32_cast = sqrt(x = add_64_cast)[name = tensor("sqrt_32_cast")]; + tensor real_div_32_cast = real_div(x = sub_64_cast, y = sqrt_32_cast)[name = tensor("real_div_32_cast")]; + tensor reshape_129_shape_0 = const()[name = tensor("reshape_129_shape_0"), val = tensor([2, 1280, 8, 8])]; + tensor reshape_129_cast = reshape(shape = reshape_129_shape_0, x = real_div_32_cast)[name = tensor("reshape_129_cast")]; + tensor add_65_gamma_0_to_fp16 = const()[name = tensor("add_65_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(361116096)))]; + tensor add_65_beta_0_to_fp16 = const()[name = tensor("add_65_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(361118720)))]; + tensor add_65_epsilon_0_to_fp16 = const()[name = tensor("add_65_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_65_cast = batch_norm(beta = add_65_beta_0_to_fp16, epsilon = add_65_epsilon_0_to_fp16, gamma = add_65_gamma_0_to_fp16, mean = add_27_mean_0_to_fp16, variance = add_27_variance_0_to_fp16, x = reshape_129_cast)[name = tensor("add_65_cast")]; + tensor input_279_cast = silu(x = add_65_cast)[name = tensor("input_279_cast")]; + tensor var_7171 = const()[name = tensor("op_7171"), val = tensor([1, 1])]; + tensor var_7173 = const()[name = tensor("op_7173"), val = tensor([1, 1])]; + tensor hidden_states_155_pad_type_0 = const()[name = tensor("hidden_states_155_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_155_pad_0 = const()[name = tensor("hidden_states_155_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor up_blocks_0_resnets_2_conv2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(361121344))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(372180608))), name = tensor("up_blocks_0_resnets_2_conv2_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 3, 3])]; + tensor up_blocks_0_resnets_2_conv2_bias_to_fp16 = const()[name = tensor("up_blocks_0_resnets_2_conv2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(372180800)))]; + tensor hidden_states_155_cast = conv(bias = up_blocks_0_resnets_2_conv2_bias_to_fp16, dilations = var_7173, groups = var_7034, pad = hidden_states_155_pad_0, pad_type = hidden_states_155_pad_type_0, strides = var_7171, weight = up_blocks_0_resnets_2_conv2_weight_to_fp16_palettized, x = input_279_cast)[name = tensor("hidden_states_155_cast")]; + tensor var_7178 = const()[name = tensor("op_7178"), val = tensor([1, 1])]; + tensor var_7180 = const()[name = tensor("op_7180"), val = tensor([1, 1])]; + tensor x_9_pad_type_0 = const()[name = tensor("x_9_pad_type_0"), val = tensor("custom")]; + tensor x_9_pad_0 = const()[name = tensor("x_9_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_0_resnets_2_conv_shortcut_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(372183424))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(374641088))), name = tensor("up_blocks_0_resnets_2_conv_shortcut_weight_to_fp16_palettized"), shape = tensor([1280, 2560, 1, 1])]; + tensor up_blocks_0_resnets_2_conv_shortcut_bias_to_fp16 = const()[name = tensor("up_blocks_0_resnets_2_conv_shortcut_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(374641280)))]; + tensor x_9_cast = conv(bias = up_blocks_0_resnets_2_conv_shortcut_bias_to_fp16, dilations = var_7180, groups = var_7034, pad = x_9_pad_0, pad_type = x_9_pad_type_0, strides = var_7178, weight = up_blocks_0_resnets_2_conv_shortcut_weight_to_fp16_palettized, x = input_267_cast)[name = tensor("x_9_cast")]; + tensor input_281_cast = add(x = x_9_cast, y = hidden_states_155_cast)[name = tensor("input_281_cast")]; + tensor input_283_scale_factor_height_0 = const()[name = tensor("input_283_scale_factor_height_0"), val = tensor(0x1p+1)]; + tensor input_283_scale_factor_width_0 = const()[name = tensor("input_283_scale_factor_width_0"), val = tensor(0x1p+1)]; + tensor input_283_cast = upsample_nearest_neighbor(scale_factor_height = input_283_scale_factor_height_0, scale_factor_width = input_283_scale_factor_width_0, x = input_281_cast)[name = tensor("input_283_cast")]; + tensor var_7189 = const()[name = tensor("op_7189"), val = tensor([1, 1])]; + tensor var_7191 = const()[name = tensor("op_7191"), val = tensor([1, 1])]; + tensor hidden_states_157_pad_type_0 = const()[name = tensor("hidden_states_157_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_157_pad_0 = const()[name = tensor("hidden_states_157_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor up_blocks_0_upsamplers_0_conv_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(374643904))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(385703168))), name = tensor("up_blocks_0_upsamplers_0_conv_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 3, 3])]; + tensor up_blocks_0_upsamplers_0_conv_bias_to_fp16 = const()[name = tensor("up_blocks_0_upsamplers_0_conv_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(385703360)))]; + tensor hidden_states_157_cast = conv(bias = up_blocks_0_upsamplers_0_conv_bias_to_fp16, dilations = var_7191, groups = var_7034, pad = hidden_states_157_pad_0, pad_type = hidden_states_157_pad_type_0, strides = var_7189, weight = up_blocks_0_upsamplers_0_conv_weight_to_fp16_palettized, x = input_283_cast)[name = tensor("hidden_states_157_cast")]; + tensor var_7223 = const()[name = tensor("op_7223"), val = tensor(true)]; + tensor var_7228 = const()[name = tensor("op_7228"), val = tensor(1)]; + tensor input_285_interleave_0 = const()[name = tensor("input_285_interleave_0"), val = tensor(false)]; + tensor input_285_cast = concat(axis = var_7228, interleave = input_285_interleave_0, values = (hidden_states_157_cast, input_169_cast))[name = tensor("input_285_cast")]; + tensor reshape_132_shape_0 = const()[name = tensor("reshape_132_shape_0"), val = tensor([2, 32, 80, 16, 16])]; + tensor reshape_132_cast = reshape(shape = reshape_132_shape_0, x = input_285_cast)[name = tensor("reshape_132_cast")]; + tensor reduce_mean_99_axes_0 = const()[name = tensor("reduce_mean_99_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_99_keep_dims_0 = const()[name = tensor("reduce_mean_99_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_99_cast = reduce_mean(axes = reduce_mean_99_axes_0, keep_dims = reduce_mean_99_keep_dims_0, x = reshape_132_cast)[name = tensor("reduce_mean_99_cast")]; + tensor sub_66_cast = sub(x = reshape_132_cast, y = reduce_mean_99_cast)[name = tensor("sub_66_cast")]; + tensor square_33_cast = square(x = sub_66_cast)[name = tensor("square_33_cast")]; + tensor reduce_mean_101_axes_0 = const()[name = tensor("reduce_mean_101_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_101_keep_dims_0 = const()[name = tensor("reduce_mean_101_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_101_cast = reduce_mean(axes = reduce_mean_101_axes_0, keep_dims = reduce_mean_101_keep_dims_0, x = square_33_cast)[name = tensor("reduce_mean_101_cast")]; + tensor add_66_y_0_to_fp16 = const()[name = tensor("add_66_y_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_66_cast = add(x = reduce_mean_101_cast, y = add_66_y_0_to_fp16)[name = tensor("add_66_cast")]; + tensor sqrt_33_cast = sqrt(x = add_66_cast)[name = tensor("sqrt_33_cast")]; + tensor real_div_33_cast = real_div(x = sub_66_cast, y = sqrt_33_cast)[name = tensor("real_div_33_cast")]; + tensor reshape_133_shape_0 = const()[name = tensor("reshape_133_shape_0"), val = tensor([2, 2560, 16, 16])]; + tensor reshape_133_cast = reshape(shape = reshape_133_shape_0, x = real_div_33_cast)[name = tensor("reshape_133_cast")]; + tensor add_67_gamma_0_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(385705984))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(385707968))), name = tensor("add_67_gamma_0_to_fp16_palettized"), shape = tensor([2560])]; + tensor add_67_beta_0_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(385708160))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(385710144))), name = tensor("add_67_beta_0_to_fp16_palettized"), shape = tensor([2560])]; + tensor add_67_epsilon_0_to_fp16 = const()[name = tensor("add_67_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_67_cast = batch_norm(beta = add_67_beta_0_to_fp16_palettized, epsilon = add_67_epsilon_0_to_fp16, gamma = add_67_gamma_0_to_fp16_palettized, mean = add_55_mean_0_to_fp16_palettized, variance = add_55_variance_0_to_fp16_palettized, x = reshape_133_cast)[name = tensor("add_67_cast")]; + tensor input_289_cast = silu(x = add_67_cast)[name = tensor("input_289_cast")]; + tensor var_7257 = const()[name = tensor("op_7257"), val = tensor([1, 1])]; + tensor var_7259 = const()[name = tensor("op_7259"), val = tensor([1, 1])]; + tensor hidden_states_159_pad_type_0 = const()[name = tensor("hidden_states_159_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_159_pad_0 = const()[name = tensor("hidden_states_159_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor up_blocks_1_resnets_0_conv1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(385710336))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(407828800))), name = tensor("up_blocks_1_resnets_0_conv1_weight_to_fp16_palettized"), shape = tensor([1280, 2560, 3, 3])]; + tensor up_blocks_1_resnets_0_conv1_bias_to_fp16 = const()[name = tensor("up_blocks_1_resnets_0_conv1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(407828992)))]; + tensor hidden_states_159_cast = conv(bias = up_blocks_1_resnets_0_conv1_bias_to_fp16, dilations = var_7259, groups = var_7228, pad = hidden_states_159_pad_0, pad_type = hidden_states_159_pad_type_0, strides = var_7257, weight = up_blocks_1_resnets_0_conv1_weight_to_fp16_palettized, x = input_289_cast)[name = tensor("hidden_states_159_cast")]; + tensor var_7265 = const()[name = tensor("op_7265"), val = tensor([1, 1])]; + tensor var_7267 = const()[name = tensor("op_7267"), val = tensor([1, 1])]; + tensor temb_27_pad_type_0 = const()[name = tensor("temb_27_pad_type_0"), val = tensor("custom")]; + tensor temb_27_pad_0 = const()[name = tensor("temb_27_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_1_resnets_0_time_emb_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(407831616))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(409060480))), name = tensor("up_blocks_1_resnets_0_time_emb_proj_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor up_blocks_1_resnets_0_time_emb_proj_bias_to_fp16 = const()[name = tensor("up_blocks_1_resnets_0_time_emb_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(409060672)))]; + tensor temb_27_cast = conv(bias = up_blocks_1_resnets_0_time_emb_proj_bias_to_fp16, dilations = var_7267, groups = var_7228, pad = temb_27_pad_0, pad_type = temb_27_pad_type_0, strides = var_7265, weight = up_blocks_1_resnets_0_time_emb_proj_weight_to_fp16_palettized, x = input_15_cast)[name = tensor("temb_27_cast")]; + tensor input_293_cast = add(x = hidden_states_159_cast, y = temb_27_cast)[name = tensor("input_293_cast")]; + tensor reshape_136_shape_0 = const()[name = tensor("reshape_136_shape_0"), val = tensor([2, 32, 40, 16, 16])]; + tensor reshape_136_cast = reshape(shape = reshape_136_shape_0, x = input_293_cast)[name = tensor("reshape_136_cast")]; + tensor reduce_mean_102_axes_0 = const()[name = tensor("reduce_mean_102_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_102_keep_dims_0 = const()[name = tensor("reduce_mean_102_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_102_cast = reduce_mean(axes = reduce_mean_102_axes_0, keep_dims = reduce_mean_102_keep_dims_0, x = reshape_136_cast)[name = tensor("reduce_mean_102_cast")]; + tensor sub_68_cast = sub(x = reshape_136_cast, y = reduce_mean_102_cast)[name = tensor("sub_68_cast")]; + tensor square_34_cast = square(x = sub_68_cast)[name = tensor("square_34_cast")]; + tensor reduce_mean_104_axes_0 = const()[name = tensor("reduce_mean_104_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_104_keep_dims_0 = const()[name = tensor("reduce_mean_104_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_104_cast = reduce_mean(axes = reduce_mean_104_axes_0, keep_dims = reduce_mean_104_keep_dims_0, x = square_34_cast)[name = tensor("reduce_mean_104_cast")]; + tensor add_68_y_0_to_fp16 = const()[name = tensor("add_68_y_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_68_cast = add(x = reduce_mean_104_cast, y = add_68_y_0_to_fp16)[name = tensor("add_68_cast")]; + tensor sqrt_34_cast = sqrt(x = add_68_cast)[name = tensor("sqrt_34_cast")]; + tensor real_div_34_cast = real_div(x = sub_68_cast, y = sqrt_34_cast)[name = tensor("real_div_34_cast")]; + tensor reshape_137_shape_0 = const()[name = tensor("reshape_137_shape_0"), val = tensor([2, 1280, 16, 16])]; + tensor reshape_137_cast = reshape(shape = reshape_137_shape_0, x = real_div_34_cast)[name = tensor("reshape_137_cast")]; + tensor add_69_gamma_0_to_fp16 = const()[name = tensor("add_69_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(409063296)))]; + tensor add_69_beta_0_to_fp16 = const()[name = tensor("add_69_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(409065920)))]; + tensor add_69_epsilon_0_to_fp16 = const()[name = tensor("add_69_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_69_cast = batch_norm(beta = add_69_beta_0_to_fp16, epsilon = add_69_epsilon_0_to_fp16, gamma = add_69_gamma_0_to_fp16, mean = add_27_mean_0_to_fp16, variance = add_27_variance_0_to_fp16, x = reshape_137_cast)[name = tensor("add_69_cast")]; + tensor input_297_cast = silu(x = add_69_cast)[name = tensor("input_297_cast")]; + tensor var_7277 = const()[name = tensor("op_7277"), val = tensor([1, 1])]; + tensor var_7279 = const()[name = tensor("op_7279"), val = tensor([1, 1])]; + tensor hidden_states_161_pad_type_0 = const()[name = tensor("hidden_states_161_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_161_pad_0 = const()[name = tensor("hidden_states_161_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor up_blocks_1_resnets_0_conv2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(409068544))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(420127808))), name = tensor("up_blocks_1_resnets_0_conv2_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 3, 3])]; + tensor up_blocks_1_resnets_0_conv2_bias_to_fp16 = const()[name = tensor("up_blocks_1_resnets_0_conv2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(420128000)))]; + tensor hidden_states_161_cast = conv(bias = up_blocks_1_resnets_0_conv2_bias_to_fp16, dilations = var_7279, groups = var_7228, pad = hidden_states_161_pad_0, pad_type = hidden_states_161_pad_type_0, strides = var_7277, weight = up_blocks_1_resnets_0_conv2_weight_to_fp16_palettized, x = input_297_cast)[name = tensor("hidden_states_161_cast")]; + tensor var_7284 = const()[name = tensor("op_7284"), val = tensor([1, 1])]; + tensor var_7286 = const()[name = tensor("op_7286"), val = tensor([1, 1])]; + tensor x_11_pad_type_0 = const()[name = tensor("x_11_pad_type_0"), val = tensor("custom")]; + tensor x_11_pad_0 = const()[name = tensor("x_11_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_1_resnets_0_conv_shortcut_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(420130624))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(422588288))), name = tensor("up_blocks_1_resnets_0_conv_shortcut_weight_to_fp16_palettized"), shape = tensor([1280, 2560, 1, 1])]; + tensor up_blocks_1_resnets_0_conv_shortcut_bias_to_fp16 = const()[name = tensor("up_blocks_1_resnets_0_conv_shortcut_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(422588480)))]; + tensor x_11_cast = conv(bias = up_blocks_1_resnets_0_conv_shortcut_bias_to_fp16, dilations = var_7286, groups = var_7228, pad = x_11_pad_0, pad_type = x_11_pad_type_0, strides = var_7284, weight = up_blocks_1_resnets_0_conv_shortcut_weight_to_fp16_palettized, x = input_285_cast)[name = tensor("x_11_cast")]; + tensor hidden_states_163_cast = add(x = x_11_cast, y = hidden_states_161_cast)[name = tensor("hidden_states_163_cast")]; + tensor reshape_140_shape_0 = const()[name = tensor("reshape_140_shape_0"), val = tensor([2, 32, 40, 16, 16])]; + tensor reshape_140_cast = reshape(shape = reshape_140_shape_0, x = hidden_states_163_cast)[name = tensor("reshape_140_cast")]; + tensor reduce_mean_105_axes_0 = const()[name = tensor("reduce_mean_105_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_105_keep_dims_0 = const()[name = tensor("reduce_mean_105_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_105_cast = reduce_mean(axes = reduce_mean_105_axes_0, keep_dims = reduce_mean_105_keep_dims_0, x = reshape_140_cast)[name = tensor("reduce_mean_105_cast")]; + tensor sub_70_cast = sub(x = reshape_140_cast, y = reduce_mean_105_cast)[name = tensor("sub_70_cast")]; + tensor square_35_cast = square(x = sub_70_cast)[name = tensor("square_35_cast")]; + tensor reduce_mean_107_axes_0 = const()[name = tensor("reduce_mean_107_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_107_keep_dims_0 = const()[name = tensor("reduce_mean_107_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_107_cast = reduce_mean(axes = reduce_mean_107_axes_0, keep_dims = reduce_mean_107_keep_dims_0, x = square_35_cast)[name = tensor("reduce_mean_107_cast")]; + tensor add_70_y_0_to_fp16 = const()[name = tensor("add_70_y_0_to_fp16"), val = tensor(0x1.1p-20)]; + tensor add_70_cast = add(x = reduce_mean_107_cast, y = add_70_y_0_to_fp16)[name = tensor("add_70_cast")]; + tensor sqrt_35_cast = sqrt(x = add_70_cast)[name = tensor("sqrt_35_cast")]; + tensor real_div_35_cast = real_div(x = sub_70_cast, y = sqrt_35_cast)[name = tensor("real_div_35_cast")]; + tensor reshape_141_shape_0 = const()[name = tensor("reshape_141_shape_0"), val = tensor([2, 1280, 16, 16])]; + tensor reshape_141_cast = reshape(shape = reshape_141_shape_0, x = real_div_35_cast)[name = tensor("reshape_141_cast")]; + tensor add_71_gamma_0_to_fp16 = const()[name = tensor("add_71_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(422591104)))]; + tensor add_71_beta_0_to_fp16 = const()[name = tensor("add_71_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(422593728)))]; + tensor add_71_epsilon_0_to_fp16 = const()[name = tensor("add_71_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_71_cast = batch_norm(beta = add_71_beta_0_to_fp16, epsilon = add_71_epsilon_0_to_fp16, gamma = add_71_gamma_0_to_fp16, mean = add_27_mean_0_to_fp16, variance = add_27_variance_0_to_fp16, x = reshape_141_cast)[name = tensor("add_71_cast")]; + tensor var_7306 = const()[name = tensor("op_7306"), val = tensor([1, 1])]; + tensor var_7308 = const()[name = tensor("op_7308"), val = tensor([1, 1])]; + tensor hidden_states_165_pad_type_0 = const()[name = tensor("hidden_states_165_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_165_pad_0 = const()[name = tensor("hidden_states_165_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_1_attentions_0_proj_in_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(422596352))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(423825216))), name = tensor("up_blocks_1_attentions_0_proj_in_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor up_blocks_1_attentions_0_proj_in_bias_to_fp16 = const()[name = tensor("up_blocks_1_attentions_0_proj_in_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(423825408)))]; + tensor hidden_states_165_cast = conv(bias = up_blocks_1_attentions_0_proj_in_bias_to_fp16, dilations = var_7308, groups = var_7228, pad = hidden_states_165_pad_0, pad_type = hidden_states_165_pad_type_0, strides = var_7306, weight = up_blocks_1_attentions_0_proj_in_weight_to_fp16_palettized, x = add_71_cast)[name = tensor("hidden_states_165_cast")]; + tensor var_7313 = const()[name = tensor("op_7313"), val = tensor([2, 1280, 1, 256])]; + tensor inputs_43_cast = reshape(shape = var_7313, x = hidden_states_165_cast)[name = tensor("inputs_43_cast")]; + tensor var_7323 = const()[name = tensor("op_7323"), val = tensor([1])]; + tensor channels_mean_43_cast = reduce_mean(axes = var_7323, keep_dims = var_7223, x = inputs_43_cast)[name = tensor("channels_mean_43_cast")]; + tensor zero_mean_43_cast = sub(x = inputs_43_cast, y = channels_mean_43_cast)[name = tensor("zero_mean_43_cast")]; + tensor zero_mean_sq_43_cast = mul(x = zero_mean_43_cast, y = zero_mean_43_cast)[name = tensor("zero_mean_sq_43_cast")]; + tensor var_7327 = const()[name = tensor("op_7327"), val = tensor([1])]; + tensor var_7328_cast = reduce_mean(axes = var_7327, keep_dims = var_7223, x = zero_mean_sq_43_cast)[name = tensor("op_7328_cast")]; + tensor var_7329_to_fp16 = const()[name = tensor("op_7329_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_7330_cast = add(x = var_7328_cast, y = var_7329_to_fp16)[name = tensor("op_7330_cast")]; + tensor denom_43_epsilon_0_to_fp16 = const()[name = tensor("denom_43_epsilon_0_to_fp16"), val = tensor(0x1p-24)]; + tensor denom_43_cast = rsqrt(epsilon = denom_43_epsilon_0_to_fp16, x = var_7330_cast)[name = tensor("denom_43_cast")]; + tensor out_43_cast = mul(x = zero_mean_43_cast, y = denom_43_cast)[name = tensor("out_43_cast")]; + tensor var_7334_to_fp16 = const()[name = tensor("op_7334_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(423828032)))]; + tensor var_7335_cast = add(x = out_43_cast, y = var_7334_to_fp16)[name = tensor("op_7335_cast")]; + tensor var_7337_to_fp16 = const()[name = tensor("op_7337_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(423830656)))]; + tensor hidden_states_167_cast = mul(x = var_7335_cast, y = var_7337_to_fp16)[name = tensor("hidden_states_167_cast")]; + tensor var_7344 = const()[name = tensor("op_7344"), val = tensor([1, 1])]; + tensor var_7346 = const()[name = tensor("op_7346"), val = tensor([1, 1])]; + tensor q_29_pad_type_0 = const()[name = tensor("q_29_pad_type_0"), val = tensor("custom")]; + tensor q_29_pad_0 = const()[name = tensor("q_29_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_1_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(423833280))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(425062144))), name = tensor("up_blocks_1_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor q_29_cast = conv(dilations = var_7346, groups = var_7228, pad = q_29_pad_0, pad_type = q_29_pad_type_0, strides = var_7344, weight = up_blocks_1_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16_palettized, x = hidden_states_167_cast)[name = tensor("q_29_cast")]; + tensor var_7350 = const()[name = tensor("op_7350"), val = tensor([1, 1])]; + tensor var_7352 = const()[name = tensor("op_7352"), val = tensor([1, 1])]; + tensor k_57_pad_type_0 = const()[name = tensor("k_57_pad_type_0"), val = tensor("custom")]; + tensor k_57_pad_0 = const()[name = tensor("k_57_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_1_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(425062336))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(426291200))), name = tensor("up_blocks_1_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor k_57_cast = conv(dilations = var_7352, groups = var_7228, pad = k_57_pad_0, pad_type = k_57_pad_type_0, strides = var_7350, weight = up_blocks_1_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16_palettized, x = hidden_states_167_cast)[name = tensor("k_57_cast")]; + tensor var_7356 = const()[name = tensor("op_7356"), val = tensor([1, 1])]; + tensor var_7358 = const()[name = tensor("op_7358"), val = tensor([1, 1])]; + tensor v_29_pad_type_0 = const()[name = tensor("v_29_pad_type_0"), val = tensor("custom")]; + tensor v_29_pad_0 = const()[name = tensor("v_29_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_1_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(426291392))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(427520256))), name = tensor("up_blocks_1_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor v_29_cast = conv(dilations = var_7358, groups = var_7228, pad = v_29_pad_0, pad_type = v_29_pad_type_0, strides = var_7356, weight = up_blocks_1_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16_palettized, x = hidden_states_167_cast)[name = tensor("v_29_cast")]; + tensor var_7362_begin_0 = const()[name = tensor("op_7362_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_7362_end_0 = const()[name = tensor("op_7362_end_0"), val = tensor([2, 64, 1, 256])]; + tensor var_7362_end_mask_0 = const()[name = tensor("op_7362_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7362_cast = slice_by_index(begin = var_7362_begin_0, end = var_7362_end_0, end_mask = var_7362_end_mask_0, x = q_29_cast)[name = tensor("op_7362_cast")]; + tensor var_7366_begin_0 = const()[name = tensor("op_7366_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_7366_end_0 = const()[name = tensor("op_7366_end_0"), val = tensor([2, 128, 1, 256])]; + tensor var_7366_end_mask_0 = const()[name = tensor("op_7366_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7366_cast = slice_by_index(begin = var_7366_begin_0, end = var_7366_end_0, end_mask = var_7366_end_mask_0, x = q_29_cast)[name = tensor("op_7366_cast")]; + tensor var_7370_begin_0 = const()[name = tensor("op_7370_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_7370_end_0 = const()[name = tensor("op_7370_end_0"), val = tensor([2, 192, 1, 256])]; + tensor var_7370_end_mask_0 = const()[name = tensor("op_7370_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7370_cast = slice_by_index(begin = var_7370_begin_0, end = var_7370_end_0, end_mask = var_7370_end_mask_0, x = q_29_cast)[name = tensor("op_7370_cast")]; + tensor var_7374_begin_0 = const()[name = tensor("op_7374_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_7374_end_0 = const()[name = tensor("op_7374_end_0"), val = tensor([2, 256, 1, 256])]; + tensor var_7374_end_mask_0 = const()[name = tensor("op_7374_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7374_cast = slice_by_index(begin = var_7374_begin_0, end = var_7374_end_0, end_mask = var_7374_end_mask_0, x = q_29_cast)[name = tensor("op_7374_cast")]; + tensor var_7378_begin_0 = const()[name = tensor("op_7378_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_7378_end_0 = const()[name = tensor("op_7378_end_0"), val = tensor([2, 320, 1, 256])]; + tensor var_7378_end_mask_0 = const()[name = tensor("op_7378_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7378_cast = slice_by_index(begin = var_7378_begin_0, end = var_7378_end_0, end_mask = var_7378_end_mask_0, x = q_29_cast)[name = tensor("op_7378_cast")]; + tensor var_7382_begin_0 = const()[name = tensor("op_7382_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_7382_end_0 = const()[name = tensor("op_7382_end_0"), val = tensor([2, 384, 1, 256])]; + tensor var_7382_end_mask_0 = const()[name = tensor("op_7382_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7382_cast = slice_by_index(begin = var_7382_begin_0, end = var_7382_end_0, end_mask = var_7382_end_mask_0, x = q_29_cast)[name = tensor("op_7382_cast")]; + tensor var_7386_begin_0 = const()[name = tensor("op_7386_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_7386_end_0 = const()[name = tensor("op_7386_end_0"), val = tensor([2, 448, 1, 256])]; + tensor var_7386_end_mask_0 = const()[name = tensor("op_7386_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7386_cast = slice_by_index(begin = var_7386_begin_0, end = var_7386_end_0, end_mask = var_7386_end_mask_0, x = q_29_cast)[name = tensor("op_7386_cast")]; + tensor var_7390_begin_0 = const()[name = tensor("op_7390_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_7390_end_0 = const()[name = tensor("op_7390_end_0"), val = tensor([2, 512, 1, 256])]; + tensor var_7390_end_mask_0 = const()[name = tensor("op_7390_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7390_cast = slice_by_index(begin = var_7390_begin_0, end = var_7390_end_0, end_mask = var_7390_end_mask_0, x = q_29_cast)[name = tensor("op_7390_cast")]; + tensor var_7394_begin_0 = const()[name = tensor("op_7394_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_7394_end_0 = const()[name = tensor("op_7394_end_0"), val = tensor([2, 576, 1, 256])]; + tensor var_7394_end_mask_0 = const()[name = tensor("op_7394_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7394_cast = slice_by_index(begin = var_7394_begin_0, end = var_7394_end_0, end_mask = var_7394_end_mask_0, x = q_29_cast)[name = tensor("op_7394_cast")]; + tensor var_7398_begin_0 = const()[name = tensor("op_7398_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_7398_end_0 = const()[name = tensor("op_7398_end_0"), val = tensor([2, 640, 1, 256])]; + tensor var_7398_end_mask_0 = const()[name = tensor("op_7398_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7398_cast = slice_by_index(begin = var_7398_begin_0, end = var_7398_end_0, end_mask = var_7398_end_mask_0, x = q_29_cast)[name = tensor("op_7398_cast")]; + tensor var_7402_begin_0 = const()[name = tensor("op_7402_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_7402_end_0 = const()[name = tensor("op_7402_end_0"), val = tensor([2, 704, 1, 256])]; + tensor var_7402_end_mask_0 = const()[name = tensor("op_7402_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7402_cast = slice_by_index(begin = var_7402_begin_0, end = var_7402_end_0, end_mask = var_7402_end_mask_0, x = q_29_cast)[name = tensor("op_7402_cast")]; + tensor var_7406_begin_0 = const()[name = tensor("op_7406_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_7406_end_0 = const()[name = tensor("op_7406_end_0"), val = tensor([2, 768, 1, 256])]; + tensor var_7406_end_mask_0 = const()[name = tensor("op_7406_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7406_cast = slice_by_index(begin = var_7406_begin_0, end = var_7406_end_0, end_mask = var_7406_end_mask_0, x = q_29_cast)[name = tensor("op_7406_cast")]; + tensor var_7410_begin_0 = const()[name = tensor("op_7410_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_7410_end_0 = const()[name = tensor("op_7410_end_0"), val = tensor([2, 832, 1, 256])]; + tensor var_7410_end_mask_0 = const()[name = tensor("op_7410_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7410_cast = slice_by_index(begin = var_7410_begin_0, end = var_7410_end_0, end_mask = var_7410_end_mask_0, x = q_29_cast)[name = tensor("op_7410_cast")]; + tensor var_7414_begin_0 = const()[name = tensor("op_7414_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_7414_end_0 = const()[name = tensor("op_7414_end_0"), val = tensor([2, 896, 1, 256])]; + tensor var_7414_end_mask_0 = const()[name = tensor("op_7414_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7414_cast = slice_by_index(begin = var_7414_begin_0, end = var_7414_end_0, end_mask = var_7414_end_mask_0, x = q_29_cast)[name = tensor("op_7414_cast")]; + tensor var_7418_begin_0 = const()[name = tensor("op_7418_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_7418_end_0 = const()[name = tensor("op_7418_end_0"), val = tensor([2, 960, 1, 256])]; + tensor var_7418_end_mask_0 = const()[name = tensor("op_7418_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7418_cast = slice_by_index(begin = var_7418_begin_0, end = var_7418_end_0, end_mask = var_7418_end_mask_0, x = q_29_cast)[name = tensor("op_7418_cast")]; + tensor var_7422_begin_0 = const()[name = tensor("op_7422_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_7422_end_0 = const()[name = tensor("op_7422_end_0"), val = tensor([2, 1024, 1, 256])]; + tensor var_7422_end_mask_0 = const()[name = tensor("op_7422_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7422_cast = slice_by_index(begin = var_7422_begin_0, end = var_7422_end_0, end_mask = var_7422_end_mask_0, x = q_29_cast)[name = tensor("op_7422_cast")]; + tensor var_7426_begin_0 = const()[name = tensor("op_7426_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_7426_end_0 = const()[name = tensor("op_7426_end_0"), val = tensor([2, 1088, 1, 256])]; + tensor var_7426_end_mask_0 = const()[name = tensor("op_7426_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7426_cast = slice_by_index(begin = var_7426_begin_0, end = var_7426_end_0, end_mask = var_7426_end_mask_0, x = q_29_cast)[name = tensor("op_7426_cast")]; + tensor var_7430_begin_0 = const()[name = tensor("op_7430_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_7430_end_0 = const()[name = tensor("op_7430_end_0"), val = tensor([2, 1152, 1, 256])]; + tensor var_7430_end_mask_0 = const()[name = tensor("op_7430_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7430_cast = slice_by_index(begin = var_7430_begin_0, end = var_7430_end_0, end_mask = var_7430_end_mask_0, x = q_29_cast)[name = tensor("op_7430_cast")]; + tensor var_7434_begin_0 = const()[name = tensor("op_7434_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_7434_end_0 = const()[name = tensor("op_7434_end_0"), val = tensor([2, 1216, 1, 256])]; + tensor var_7434_end_mask_0 = const()[name = tensor("op_7434_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7434_cast = slice_by_index(begin = var_7434_begin_0, end = var_7434_end_0, end_mask = var_7434_end_mask_0, x = q_29_cast)[name = tensor("op_7434_cast")]; + tensor var_7438_begin_0 = const()[name = tensor("op_7438_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_7438_end_0 = const()[name = tensor("op_7438_end_0"), val = tensor([2, 1280, 1, 256])]; + tensor var_7438_end_mask_0 = const()[name = tensor("op_7438_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7438_cast = slice_by_index(begin = var_7438_begin_0, end = var_7438_end_0, end_mask = var_7438_end_mask_0, x = q_29_cast)[name = tensor("op_7438_cast")]; + tensor k_59_perm_0 = const()[name = tensor("k_59_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_7445_begin_0 = const()[name = tensor("op_7445_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_7445_end_0 = const()[name = tensor("op_7445_end_0"), val = tensor([2, 256, 1, 64])]; + tensor var_7445_end_mask_0 = const()[name = tensor("op_7445_end_mask_0"), val = tensor([true, true, true, false])]; + tensor transpose_17 = transpose(perm = k_59_perm_0, x = k_57_cast)[name = tensor("transpose_17")]; + tensor var_7445_cast = slice_by_index(begin = var_7445_begin_0, end = var_7445_end_0, end_mask = var_7445_end_mask_0, x = transpose_17)[name = tensor("op_7445_cast")]; + tensor var_7449_begin_0 = const()[name = tensor("op_7449_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_7449_end_0 = const()[name = tensor("op_7449_end_0"), val = tensor([2, 256, 1, 128])]; + tensor var_7449_end_mask_0 = const()[name = tensor("op_7449_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7449_cast = slice_by_index(begin = var_7449_begin_0, end = var_7449_end_0, end_mask = var_7449_end_mask_0, x = transpose_17)[name = tensor("op_7449_cast")]; + tensor var_7453_begin_0 = const()[name = tensor("op_7453_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_7453_end_0 = const()[name = tensor("op_7453_end_0"), val = tensor([2, 256, 1, 192])]; + tensor var_7453_end_mask_0 = const()[name = tensor("op_7453_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7453_cast = slice_by_index(begin = var_7453_begin_0, end = var_7453_end_0, end_mask = var_7453_end_mask_0, x = transpose_17)[name = tensor("op_7453_cast")]; + tensor var_7457_begin_0 = const()[name = tensor("op_7457_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_7457_end_0 = const()[name = tensor("op_7457_end_0"), val = tensor([2, 256, 1, 256])]; + tensor var_7457_end_mask_0 = const()[name = tensor("op_7457_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7457_cast = slice_by_index(begin = var_7457_begin_0, end = var_7457_end_0, end_mask = var_7457_end_mask_0, x = transpose_17)[name = tensor("op_7457_cast")]; + tensor var_7461_begin_0 = const()[name = tensor("op_7461_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_7461_end_0 = const()[name = tensor("op_7461_end_0"), val = tensor([2, 256, 1, 320])]; + tensor var_7461_end_mask_0 = const()[name = tensor("op_7461_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7461_cast = slice_by_index(begin = var_7461_begin_0, end = var_7461_end_0, end_mask = var_7461_end_mask_0, x = transpose_17)[name = tensor("op_7461_cast")]; + tensor var_7465_begin_0 = const()[name = tensor("op_7465_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_7465_end_0 = const()[name = tensor("op_7465_end_0"), val = tensor([2, 256, 1, 384])]; + tensor var_7465_end_mask_0 = const()[name = tensor("op_7465_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7465_cast = slice_by_index(begin = var_7465_begin_0, end = var_7465_end_0, end_mask = var_7465_end_mask_0, x = transpose_17)[name = tensor("op_7465_cast")]; + tensor var_7469_begin_0 = const()[name = tensor("op_7469_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_7469_end_0 = const()[name = tensor("op_7469_end_0"), val = tensor([2, 256, 1, 448])]; + tensor var_7469_end_mask_0 = const()[name = tensor("op_7469_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7469_cast = slice_by_index(begin = var_7469_begin_0, end = var_7469_end_0, end_mask = var_7469_end_mask_0, x = transpose_17)[name = tensor("op_7469_cast")]; + tensor var_7473_begin_0 = const()[name = tensor("op_7473_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_7473_end_0 = const()[name = tensor("op_7473_end_0"), val = tensor([2, 256, 1, 512])]; + tensor var_7473_end_mask_0 = const()[name = tensor("op_7473_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7473_cast = slice_by_index(begin = var_7473_begin_0, end = var_7473_end_0, end_mask = var_7473_end_mask_0, x = transpose_17)[name = tensor("op_7473_cast")]; + tensor var_7477_begin_0 = const()[name = tensor("op_7477_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_7477_end_0 = const()[name = tensor("op_7477_end_0"), val = tensor([2, 256, 1, 576])]; + tensor var_7477_end_mask_0 = const()[name = tensor("op_7477_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7477_cast = slice_by_index(begin = var_7477_begin_0, end = var_7477_end_0, end_mask = var_7477_end_mask_0, x = transpose_17)[name = tensor("op_7477_cast")]; + tensor var_7481_begin_0 = const()[name = tensor("op_7481_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_7481_end_0 = const()[name = tensor("op_7481_end_0"), val = tensor([2, 256, 1, 640])]; + tensor var_7481_end_mask_0 = const()[name = tensor("op_7481_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7481_cast = slice_by_index(begin = var_7481_begin_0, end = var_7481_end_0, end_mask = var_7481_end_mask_0, x = transpose_17)[name = tensor("op_7481_cast")]; + tensor var_7485_begin_0 = const()[name = tensor("op_7485_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_7485_end_0 = const()[name = tensor("op_7485_end_0"), val = tensor([2, 256, 1, 704])]; + tensor var_7485_end_mask_0 = const()[name = tensor("op_7485_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7485_cast = slice_by_index(begin = var_7485_begin_0, end = var_7485_end_0, end_mask = var_7485_end_mask_0, x = transpose_17)[name = tensor("op_7485_cast")]; + tensor var_7489_begin_0 = const()[name = tensor("op_7489_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_7489_end_0 = const()[name = tensor("op_7489_end_0"), val = tensor([2, 256, 1, 768])]; + tensor var_7489_end_mask_0 = const()[name = tensor("op_7489_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7489_cast = slice_by_index(begin = var_7489_begin_0, end = var_7489_end_0, end_mask = var_7489_end_mask_0, x = transpose_17)[name = tensor("op_7489_cast")]; + tensor var_7493_begin_0 = const()[name = tensor("op_7493_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_7493_end_0 = const()[name = tensor("op_7493_end_0"), val = tensor([2, 256, 1, 832])]; + tensor var_7493_end_mask_0 = const()[name = tensor("op_7493_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7493_cast = slice_by_index(begin = var_7493_begin_0, end = var_7493_end_0, end_mask = var_7493_end_mask_0, x = transpose_17)[name = tensor("op_7493_cast")]; + tensor var_7497_begin_0 = const()[name = tensor("op_7497_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_7497_end_0 = const()[name = tensor("op_7497_end_0"), val = tensor([2, 256, 1, 896])]; + tensor var_7497_end_mask_0 = const()[name = tensor("op_7497_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7497_cast = slice_by_index(begin = var_7497_begin_0, end = var_7497_end_0, end_mask = var_7497_end_mask_0, x = transpose_17)[name = tensor("op_7497_cast")]; + tensor var_7501_begin_0 = const()[name = tensor("op_7501_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_7501_end_0 = const()[name = tensor("op_7501_end_0"), val = tensor([2, 256, 1, 960])]; + tensor var_7501_end_mask_0 = const()[name = tensor("op_7501_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7501_cast = slice_by_index(begin = var_7501_begin_0, end = var_7501_end_0, end_mask = var_7501_end_mask_0, x = transpose_17)[name = tensor("op_7501_cast")]; + tensor var_7505_begin_0 = const()[name = tensor("op_7505_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_7505_end_0 = const()[name = tensor("op_7505_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_7505_end_mask_0 = const()[name = tensor("op_7505_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7505_cast = slice_by_index(begin = var_7505_begin_0, end = var_7505_end_0, end_mask = var_7505_end_mask_0, x = transpose_17)[name = tensor("op_7505_cast")]; + tensor var_7509_begin_0 = const()[name = tensor("op_7509_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_7509_end_0 = const()[name = tensor("op_7509_end_0"), val = tensor([2, 256, 1, 1088])]; + tensor var_7509_end_mask_0 = const()[name = tensor("op_7509_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7509_cast = slice_by_index(begin = var_7509_begin_0, end = var_7509_end_0, end_mask = var_7509_end_mask_0, x = transpose_17)[name = tensor("op_7509_cast")]; + tensor var_7513_begin_0 = const()[name = tensor("op_7513_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_7513_end_0 = const()[name = tensor("op_7513_end_0"), val = tensor([2, 256, 1, 1152])]; + tensor var_7513_end_mask_0 = const()[name = tensor("op_7513_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7513_cast = slice_by_index(begin = var_7513_begin_0, end = var_7513_end_0, end_mask = var_7513_end_mask_0, x = transpose_17)[name = tensor("op_7513_cast")]; + tensor var_7517_begin_0 = const()[name = tensor("op_7517_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_7517_end_0 = const()[name = tensor("op_7517_end_0"), val = tensor([2, 256, 1, 1216])]; + tensor var_7517_end_mask_0 = const()[name = tensor("op_7517_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7517_cast = slice_by_index(begin = var_7517_begin_0, end = var_7517_end_0, end_mask = var_7517_end_mask_0, x = transpose_17)[name = tensor("op_7517_cast")]; + tensor var_7521_begin_0 = const()[name = tensor("op_7521_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_7521_end_0 = const()[name = tensor("op_7521_end_0"), val = tensor([2, 256, 1, 1280])]; + tensor var_7521_end_mask_0 = const()[name = tensor("op_7521_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7521_cast = slice_by_index(begin = var_7521_begin_0, end = var_7521_end_0, end_mask = var_7521_end_mask_0, x = transpose_17)[name = tensor("op_7521_cast")]; + tensor var_7523_begin_0 = const()[name = tensor("op_7523_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_7523_end_0 = const()[name = tensor("op_7523_end_0"), val = tensor([2, 64, 1, 256])]; + tensor var_7523_end_mask_0 = const()[name = tensor("op_7523_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7523_cast = slice_by_index(begin = var_7523_begin_0, end = var_7523_end_0, end_mask = var_7523_end_mask_0, x = v_29_cast)[name = tensor("op_7523_cast")]; + tensor var_7527_begin_0 = const()[name = tensor("op_7527_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_7527_end_0 = const()[name = tensor("op_7527_end_0"), val = tensor([2, 128, 1, 256])]; + tensor var_7527_end_mask_0 = const()[name = tensor("op_7527_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7527_cast = slice_by_index(begin = var_7527_begin_0, end = var_7527_end_0, end_mask = var_7527_end_mask_0, x = v_29_cast)[name = tensor("op_7527_cast")]; + tensor var_7531_begin_0 = const()[name = tensor("op_7531_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_7531_end_0 = const()[name = tensor("op_7531_end_0"), val = tensor([2, 192, 1, 256])]; + tensor var_7531_end_mask_0 = const()[name = tensor("op_7531_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7531_cast = slice_by_index(begin = var_7531_begin_0, end = var_7531_end_0, end_mask = var_7531_end_mask_0, x = v_29_cast)[name = tensor("op_7531_cast")]; + tensor var_7535_begin_0 = const()[name = tensor("op_7535_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_7535_end_0 = const()[name = tensor("op_7535_end_0"), val = tensor([2, 256, 1, 256])]; + tensor var_7535_end_mask_0 = const()[name = tensor("op_7535_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7535_cast = slice_by_index(begin = var_7535_begin_0, end = var_7535_end_0, end_mask = var_7535_end_mask_0, x = v_29_cast)[name = tensor("op_7535_cast")]; + tensor var_7539_begin_0 = const()[name = tensor("op_7539_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_7539_end_0 = const()[name = tensor("op_7539_end_0"), val = tensor([2, 320, 1, 256])]; + tensor var_7539_end_mask_0 = const()[name = tensor("op_7539_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7539_cast = slice_by_index(begin = var_7539_begin_0, end = var_7539_end_0, end_mask = var_7539_end_mask_0, x = v_29_cast)[name = tensor("op_7539_cast")]; + tensor var_7543_begin_0 = const()[name = tensor("op_7543_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_7543_end_0 = const()[name = tensor("op_7543_end_0"), val = tensor([2, 384, 1, 256])]; + tensor var_7543_end_mask_0 = const()[name = tensor("op_7543_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7543_cast = slice_by_index(begin = var_7543_begin_0, end = var_7543_end_0, end_mask = var_7543_end_mask_0, x = v_29_cast)[name = tensor("op_7543_cast")]; + tensor var_7547_begin_0 = const()[name = tensor("op_7547_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_7547_end_0 = const()[name = tensor("op_7547_end_0"), val = tensor([2, 448, 1, 256])]; + tensor var_7547_end_mask_0 = const()[name = tensor("op_7547_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7547_cast = slice_by_index(begin = var_7547_begin_0, end = var_7547_end_0, end_mask = var_7547_end_mask_0, x = v_29_cast)[name = tensor("op_7547_cast")]; + tensor var_7551_begin_0 = const()[name = tensor("op_7551_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_7551_end_0 = const()[name = tensor("op_7551_end_0"), val = tensor([2, 512, 1, 256])]; + tensor var_7551_end_mask_0 = const()[name = tensor("op_7551_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7551_cast = slice_by_index(begin = var_7551_begin_0, end = var_7551_end_0, end_mask = var_7551_end_mask_0, x = v_29_cast)[name = tensor("op_7551_cast")]; + tensor var_7555_begin_0 = const()[name = tensor("op_7555_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_7555_end_0 = const()[name = tensor("op_7555_end_0"), val = tensor([2, 576, 1, 256])]; + tensor var_7555_end_mask_0 = const()[name = tensor("op_7555_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7555_cast = slice_by_index(begin = var_7555_begin_0, end = var_7555_end_0, end_mask = var_7555_end_mask_0, x = v_29_cast)[name = tensor("op_7555_cast")]; + tensor var_7559_begin_0 = const()[name = tensor("op_7559_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_7559_end_0 = const()[name = tensor("op_7559_end_0"), val = tensor([2, 640, 1, 256])]; + tensor var_7559_end_mask_0 = const()[name = tensor("op_7559_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7559_cast = slice_by_index(begin = var_7559_begin_0, end = var_7559_end_0, end_mask = var_7559_end_mask_0, x = v_29_cast)[name = tensor("op_7559_cast")]; + tensor var_7563_begin_0 = const()[name = tensor("op_7563_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_7563_end_0 = const()[name = tensor("op_7563_end_0"), val = tensor([2, 704, 1, 256])]; + tensor var_7563_end_mask_0 = const()[name = tensor("op_7563_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7563_cast = slice_by_index(begin = var_7563_begin_0, end = var_7563_end_0, end_mask = var_7563_end_mask_0, x = v_29_cast)[name = tensor("op_7563_cast")]; + tensor var_7567_begin_0 = const()[name = tensor("op_7567_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_7567_end_0 = const()[name = tensor("op_7567_end_0"), val = tensor([2, 768, 1, 256])]; + tensor var_7567_end_mask_0 = const()[name = tensor("op_7567_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7567_cast = slice_by_index(begin = var_7567_begin_0, end = var_7567_end_0, end_mask = var_7567_end_mask_0, x = v_29_cast)[name = tensor("op_7567_cast")]; + tensor var_7571_begin_0 = const()[name = tensor("op_7571_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_7571_end_0 = const()[name = tensor("op_7571_end_0"), val = tensor([2, 832, 1, 256])]; + tensor var_7571_end_mask_0 = const()[name = tensor("op_7571_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7571_cast = slice_by_index(begin = var_7571_begin_0, end = var_7571_end_0, end_mask = var_7571_end_mask_0, x = v_29_cast)[name = tensor("op_7571_cast")]; + tensor var_7575_begin_0 = const()[name = tensor("op_7575_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_7575_end_0 = const()[name = tensor("op_7575_end_0"), val = tensor([2, 896, 1, 256])]; + tensor var_7575_end_mask_0 = const()[name = tensor("op_7575_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7575_cast = slice_by_index(begin = var_7575_begin_0, end = var_7575_end_0, end_mask = var_7575_end_mask_0, x = v_29_cast)[name = tensor("op_7575_cast")]; + tensor var_7579_begin_0 = const()[name = tensor("op_7579_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_7579_end_0 = const()[name = tensor("op_7579_end_0"), val = tensor([2, 960, 1, 256])]; + tensor var_7579_end_mask_0 = const()[name = tensor("op_7579_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7579_cast = slice_by_index(begin = var_7579_begin_0, end = var_7579_end_0, end_mask = var_7579_end_mask_0, x = v_29_cast)[name = tensor("op_7579_cast")]; + tensor var_7583_begin_0 = const()[name = tensor("op_7583_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_7583_end_0 = const()[name = tensor("op_7583_end_0"), val = tensor([2, 1024, 1, 256])]; + tensor var_7583_end_mask_0 = const()[name = tensor("op_7583_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7583_cast = slice_by_index(begin = var_7583_begin_0, end = var_7583_end_0, end_mask = var_7583_end_mask_0, x = v_29_cast)[name = tensor("op_7583_cast")]; + tensor var_7587_begin_0 = const()[name = tensor("op_7587_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_7587_end_0 = const()[name = tensor("op_7587_end_0"), val = tensor([2, 1088, 1, 256])]; + tensor var_7587_end_mask_0 = const()[name = tensor("op_7587_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7587_cast = slice_by_index(begin = var_7587_begin_0, end = var_7587_end_0, end_mask = var_7587_end_mask_0, x = v_29_cast)[name = tensor("op_7587_cast")]; + tensor var_7591_begin_0 = const()[name = tensor("op_7591_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_7591_end_0 = const()[name = tensor("op_7591_end_0"), val = tensor([2, 1152, 1, 256])]; + tensor var_7591_end_mask_0 = const()[name = tensor("op_7591_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7591_cast = slice_by_index(begin = var_7591_begin_0, end = var_7591_end_0, end_mask = var_7591_end_mask_0, x = v_29_cast)[name = tensor("op_7591_cast")]; + tensor var_7595_begin_0 = const()[name = tensor("op_7595_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_7595_end_0 = const()[name = tensor("op_7595_end_0"), val = tensor([2, 1216, 1, 256])]; + tensor var_7595_end_mask_0 = const()[name = tensor("op_7595_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7595_cast = slice_by_index(begin = var_7595_begin_0, end = var_7595_end_0, end_mask = var_7595_end_mask_0, x = v_29_cast)[name = tensor("op_7595_cast")]; + tensor var_7599_begin_0 = const()[name = tensor("op_7599_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_7599_end_0 = const()[name = tensor("op_7599_end_0"), val = tensor([2, 1280, 1, 256])]; + tensor var_7599_end_mask_0 = const()[name = tensor("op_7599_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7599_cast = slice_by_index(begin = var_7599_begin_0, end = var_7599_end_0, end_mask = var_7599_end_mask_0, x = v_29_cast)[name = tensor("op_7599_cast")]; + tensor var_7603_equation_0 = const()[name = tensor("op_7603_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7603_cast = einsum(equation = var_7603_equation_0, values = (var_7445_cast, var_7362_cast))[name = tensor("op_7603_cast")]; + tensor var_7604_to_fp16 = const()[name = tensor("op_7604_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_241_cast = mul(x = var_7603_cast, y = var_7604_to_fp16)[name = tensor("aw_241_cast")]; + tensor var_7607_equation_0 = const()[name = tensor("op_7607_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7607_cast = einsum(equation = var_7607_equation_0, values = (var_7449_cast, var_7366_cast))[name = tensor("op_7607_cast")]; + tensor var_7608_to_fp16 = const()[name = tensor("op_7608_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_243_cast = mul(x = var_7607_cast, y = var_7608_to_fp16)[name = tensor("aw_243_cast")]; + tensor var_7611_equation_0 = const()[name = tensor("op_7611_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7611_cast = einsum(equation = var_7611_equation_0, values = (var_7453_cast, var_7370_cast))[name = tensor("op_7611_cast")]; + tensor var_7612_to_fp16 = const()[name = tensor("op_7612_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_245_cast = mul(x = var_7611_cast, y = var_7612_to_fp16)[name = tensor("aw_245_cast")]; + tensor var_7615_equation_0 = const()[name = tensor("op_7615_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7615_cast = einsum(equation = var_7615_equation_0, values = (var_7457_cast, var_7374_cast))[name = tensor("op_7615_cast")]; + tensor var_7616_to_fp16 = const()[name = tensor("op_7616_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_247_cast = mul(x = var_7615_cast, y = var_7616_to_fp16)[name = tensor("aw_247_cast")]; + tensor var_7619_equation_0 = const()[name = tensor("op_7619_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7619_cast = einsum(equation = var_7619_equation_0, values = (var_7461_cast, var_7378_cast))[name = tensor("op_7619_cast")]; + tensor var_7620_to_fp16 = const()[name = tensor("op_7620_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_249_cast = mul(x = var_7619_cast, y = var_7620_to_fp16)[name = tensor("aw_249_cast")]; + tensor var_7623_equation_0 = const()[name = tensor("op_7623_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7623_cast = einsum(equation = var_7623_equation_0, values = (var_7465_cast, var_7382_cast))[name = tensor("op_7623_cast")]; + tensor var_7624_to_fp16 = const()[name = tensor("op_7624_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_251_cast = mul(x = var_7623_cast, y = var_7624_to_fp16)[name = tensor("aw_251_cast")]; + tensor var_7627_equation_0 = const()[name = tensor("op_7627_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7627_cast = einsum(equation = var_7627_equation_0, values = (var_7469_cast, var_7386_cast))[name = tensor("op_7627_cast")]; + tensor var_7628_to_fp16 = const()[name = tensor("op_7628_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_253_cast = mul(x = var_7627_cast, y = var_7628_to_fp16)[name = tensor("aw_253_cast")]; + tensor var_7631_equation_0 = const()[name = tensor("op_7631_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7631_cast = einsum(equation = var_7631_equation_0, values = (var_7473_cast, var_7390_cast))[name = tensor("op_7631_cast")]; + tensor var_7632_to_fp16 = const()[name = tensor("op_7632_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_255_cast = mul(x = var_7631_cast, y = var_7632_to_fp16)[name = tensor("aw_255_cast")]; + tensor var_7635_equation_0 = const()[name = tensor("op_7635_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7635_cast = einsum(equation = var_7635_equation_0, values = (var_7477_cast, var_7394_cast))[name = tensor("op_7635_cast")]; + tensor var_7636_to_fp16 = const()[name = tensor("op_7636_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_257_cast = mul(x = var_7635_cast, y = var_7636_to_fp16)[name = tensor("aw_257_cast")]; + tensor var_7639_equation_0 = const()[name = tensor("op_7639_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7639_cast = einsum(equation = var_7639_equation_0, values = (var_7481_cast, var_7398_cast))[name = tensor("op_7639_cast")]; + tensor var_7640_to_fp16 = const()[name = tensor("op_7640_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_259_cast = mul(x = var_7639_cast, y = var_7640_to_fp16)[name = tensor("aw_259_cast")]; + tensor var_7643_equation_0 = const()[name = tensor("op_7643_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7643_cast = einsum(equation = var_7643_equation_0, values = (var_7485_cast, var_7402_cast))[name = tensor("op_7643_cast")]; + tensor var_7644_to_fp16 = const()[name = tensor("op_7644_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_261_cast = mul(x = var_7643_cast, y = var_7644_to_fp16)[name = tensor("aw_261_cast")]; + tensor var_7647_equation_0 = const()[name = tensor("op_7647_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7647_cast = einsum(equation = var_7647_equation_0, values = (var_7489_cast, var_7406_cast))[name = tensor("op_7647_cast")]; + tensor var_7648_to_fp16 = const()[name = tensor("op_7648_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_263_cast = mul(x = var_7647_cast, y = var_7648_to_fp16)[name = tensor("aw_263_cast")]; + tensor var_7651_equation_0 = const()[name = tensor("op_7651_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7651_cast = einsum(equation = var_7651_equation_0, values = (var_7493_cast, var_7410_cast))[name = tensor("op_7651_cast")]; + tensor var_7652_to_fp16 = const()[name = tensor("op_7652_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_265_cast = mul(x = var_7651_cast, y = var_7652_to_fp16)[name = tensor("aw_265_cast")]; + tensor var_7655_equation_0 = const()[name = tensor("op_7655_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7655_cast = einsum(equation = var_7655_equation_0, values = (var_7497_cast, var_7414_cast))[name = tensor("op_7655_cast")]; + tensor var_7656_to_fp16 = const()[name = tensor("op_7656_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_267_cast = mul(x = var_7655_cast, y = var_7656_to_fp16)[name = tensor("aw_267_cast")]; + tensor var_7659_equation_0 = const()[name = tensor("op_7659_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7659_cast = einsum(equation = var_7659_equation_0, values = (var_7501_cast, var_7418_cast))[name = tensor("op_7659_cast")]; + tensor var_7660_to_fp16 = const()[name = tensor("op_7660_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_269_cast = mul(x = var_7659_cast, y = var_7660_to_fp16)[name = tensor("aw_269_cast")]; + tensor var_7663_equation_0 = const()[name = tensor("op_7663_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7663_cast = einsum(equation = var_7663_equation_0, values = (var_7505_cast, var_7422_cast))[name = tensor("op_7663_cast")]; + tensor var_7664_to_fp16 = const()[name = tensor("op_7664_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_271_cast = mul(x = var_7663_cast, y = var_7664_to_fp16)[name = tensor("aw_271_cast")]; + tensor var_7667_equation_0 = const()[name = tensor("op_7667_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7667_cast = einsum(equation = var_7667_equation_0, values = (var_7509_cast, var_7426_cast))[name = tensor("op_7667_cast")]; + tensor var_7668_to_fp16 = const()[name = tensor("op_7668_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_273_cast = mul(x = var_7667_cast, y = var_7668_to_fp16)[name = tensor("aw_273_cast")]; + tensor var_7671_equation_0 = const()[name = tensor("op_7671_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7671_cast = einsum(equation = var_7671_equation_0, values = (var_7513_cast, var_7430_cast))[name = tensor("op_7671_cast")]; + tensor var_7672_to_fp16 = const()[name = tensor("op_7672_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_275_cast = mul(x = var_7671_cast, y = var_7672_to_fp16)[name = tensor("aw_275_cast")]; + tensor var_7675_equation_0 = const()[name = tensor("op_7675_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7675_cast = einsum(equation = var_7675_equation_0, values = (var_7517_cast, var_7434_cast))[name = tensor("op_7675_cast")]; + tensor var_7676_to_fp16 = const()[name = tensor("op_7676_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_277_cast = mul(x = var_7675_cast, y = var_7676_to_fp16)[name = tensor("aw_277_cast")]; + tensor var_7679_equation_0 = const()[name = tensor("op_7679_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7679_cast = einsum(equation = var_7679_equation_0, values = (var_7521_cast, var_7438_cast))[name = tensor("op_7679_cast")]; + tensor var_7680_to_fp16 = const()[name = tensor("op_7680_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_279_cast = mul(x = var_7679_cast, y = var_7680_to_fp16)[name = tensor("aw_279_cast")]; + tensor var_7682_cast = softmax(axis = var_7228, x = aw_241_cast)[name = tensor("op_7682_cast")]; + tensor var_7683_cast = softmax(axis = var_7228, x = aw_243_cast)[name = tensor("op_7683_cast")]; + tensor var_7684_cast = softmax(axis = var_7228, x = aw_245_cast)[name = tensor("op_7684_cast")]; + tensor var_7685_cast = softmax(axis = var_7228, x = aw_247_cast)[name = tensor("op_7685_cast")]; + tensor var_7686_cast = softmax(axis = var_7228, x = aw_249_cast)[name = tensor("op_7686_cast")]; + tensor var_7687_cast = softmax(axis = var_7228, x = aw_251_cast)[name = tensor("op_7687_cast")]; + tensor var_7688_cast = softmax(axis = var_7228, x = aw_253_cast)[name = tensor("op_7688_cast")]; + tensor var_7689_cast = softmax(axis = var_7228, x = aw_255_cast)[name = tensor("op_7689_cast")]; + tensor var_7690_cast = softmax(axis = var_7228, x = aw_257_cast)[name = tensor("op_7690_cast")]; + tensor var_7691_cast = softmax(axis = var_7228, x = aw_259_cast)[name = tensor("op_7691_cast")]; + tensor var_7692_cast = softmax(axis = var_7228, x = aw_261_cast)[name = tensor("op_7692_cast")]; + tensor var_7693_cast = softmax(axis = var_7228, x = aw_263_cast)[name = tensor("op_7693_cast")]; + tensor var_7694_cast = softmax(axis = var_7228, x = aw_265_cast)[name = tensor("op_7694_cast")]; + tensor var_7695_cast = softmax(axis = var_7228, x = aw_267_cast)[name = tensor("op_7695_cast")]; + tensor var_7696_cast = softmax(axis = var_7228, x = aw_269_cast)[name = tensor("op_7696_cast")]; + tensor var_7697_cast = softmax(axis = var_7228, x = aw_271_cast)[name = tensor("op_7697_cast")]; + tensor var_7698_cast = softmax(axis = var_7228, x = aw_273_cast)[name = tensor("op_7698_cast")]; + tensor var_7699_cast = softmax(axis = var_7228, x = aw_275_cast)[name = tensor("op_7699_cast")]; + tensor var_7700_cast = softmax(axis = var_7228, x = aw_277_cast)[name = tensor("op_7700_cast")]; + tensor var_7701_cast = softmax(axis = var_7228, x = aw_279_cast)[name = tensor("op_7701_cast")]; + tensor var_7703_equation_0 = const()[name = tensor("op_7703_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_7703_cast = einsum(equation = var_7703_equation_0, values = (var_7523_cast, var_7682_cast))[name = tensor("op_7703_cast")]; + tensor var_7705_equation_0 = const()[name = tensor("op_7705_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_7705_cast = einsum(equation = var_7705_equation_0, values = (var_7527_cast, var_7683_cast))[name = tensor("op_7705_cast")]; + tensor var_7707_equation_0 = const()[name = tensor("op_7707_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_7707_cast = einsum(equation = var_7707_equation_0, values = (var_7531_cast, var_7684_cast))[name = tensor("op_7707_cast")]; + tensor var_7709_equation_0 = const()[name = tensor("op_7709_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_7709_cast = einsum(equation = var_7709_equation_0, values = (var_7535_cast, var_7685_cast))[name = tensor("op_7709_cast")]; + tensor var_7711_equation_0 = const()[name = tensor("op_7711_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_7711_cast = einsum(equation = var_7711_equation_0, values = (var_7539_cast, var_7686_cast))[name = tensor("op_7711_cast")]; + tensor var_7713_equation_0 = const()[name = tensor("op_7713_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_7713_cast = einsum(equation = var_7713_equation_0, values = (var_7543_cast, var_7687_cast))[name = tensor("op_7713_cast")]; + tensor var_7715_equation_0 = const()[name = tensor("op_7715_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_7715_cast = einsum(equation = var_7715_equation_0, values = (var_7547_cast, var_7688_cast))[name = tensor("op_7715_cast")]; + tensor var_7717_equation_0 = const()[name = tensor("op_7717_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_7717_cast = einsum(equation = var_7717_equation_0, values = (var_7551_cast, var_7689_cast))[name = tensor("op_7717_cast")]; + tensor var_7719_equation_0 = const()[name = tensor("op_7719_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_7719_cast = einsum(equation = var_7719_equation_0, values = (var_7555_cast, var_7690_cast))[name = tensor("op_7719_cast")]; + tensor var_7721_equation_0 = const()[name = tensor("op_7721_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_7721_cast = einsum(equation = var_7721_equation_0, values = (var_7559_cast, var_7691_cast))[name = tensor("op_7721_cast")]; + tensor var_7723_equation_0 = const()[name = tensor("op_7723_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_7723_cast = einsum(equation = var_7723_equation_0, values = (var_7563_cast, var_7692_cast))[name = tensor("op_7723_cast")]; + tensor var_7725_equation_0 = const()[name = tensor("op_7725_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_7725_cast = einsum(equation = var_7725_equation_0, values = (var_7567_cast, var_7693_cast))[name = tensor("op_7725_cast")]; + tensor var_7727_equation_0 = const()[name = tensor("op_7727_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_7727_cast = einsum(equation = var_7727_equation_0, values = (var_7571_cast, var_7694_cast))[name = tensor("op_7727_cast")]; + tensor var_7729_equation_0 = const()[name = tensor("op_7729_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_7729_cast = einsum(equation = var_7729_equation_0, values = (var_7575_cast, var_7695_cast))[name = tensor("op_7729_cast")]; + tensor var_7731_equation_0 = const()[name = tensor("op_7731_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_7731_cast = einsum(equation = var_7731_equation_0, values = (var_7579_cast, var_7696_cast))[name = tensor("op_7731_cast")]; + tensor var_7733_equation_0 = const()[name = tensor("op_7733_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_7733_cast = einsum(equation = var_7733_equation_0, values = (var_7583_cast, var_7697_cast))[name = tensor("op_7733_cast")]; + tensor var_7735_equation_0 = const()[name = tensor("op_7735_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_7735_cast = einsum(equation = var_7735_equation_0, values = (var_7587_cast, var_7698_cast))[name = tensor("op_7735_cast")]; + tensor var_7737_equation_0 = const()[name = tensor("op_7737_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_7737_cast = einsum(equation = var_7737_equation_0, values = (var_7591_cast, var_7699_cast))[name = tensor("op_7737_cast")]; + tensor var_7739_equation_0 = const()[name = tensor("op_7739_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_7739_cast = einsum(equation = var_7739_equation_0, values = (var_7595_cast, var_7700_cast))[name = tensor("op_7739_cast")]; + tensor var_7741_equation_0 = const()[name = tensor("op_7741_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_7741_cast = einsum(equation = var_7741_equation_0, values = (var_7599_cast, var_7701_cast))[name = tensor("op_7741_cast")]; + tensor input_301_interleave_0 = const()[name = tensor("input_301_interleave_0"), val = tensor(false)]; + tensor input_301_cast = concat(axis = var_7228, interleave = input_301_interleave_0, values = (var_7703_cast, var_7705_cast, var_7707_cast, var_7709_cast, var_7711_cast, var_7713_cast, var_7715_cast, var_7717_cast, var_7719_cast, var_7721_cast, var_7723_cast, var_7725_cast, var_7727_cast, var_7729_cast, var_7731_cast, var_7733_cast, var_7735_cast, var_7737_cast, var_7739_cast, var_7741_cast))[name = tensor("input_301_cast")]; + tensor var_7747 = const()[name = tensor("op_7747"), val = tensor([1, 1])]; + tensor var_7749 = const()[name = tensor("op_7749"), val = tensor([1, 1])]; + tensor var_7751_pad_type_0 = const()[name = tensor("op_7751_pad_type_0"), val = tensor("custom")]; + tensor var_7751_pad_0 = const()[name = tensor("op_7751_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_1_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(427520448))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(428749312))), name = tensor("up_blocks_1_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor up_blocks_1_attentions_0_transformer_blocks_0_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("up_blocks_1_attentions_0_transformer_blocks_0_attn1_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(428749504)))]; + tensor var_7751_cast = conv(bias = up_blocks_1_attentions_0_transformer_blocks_0_attn1_to_out_0_bias_to_fp16, dilations = var_7749, groups = var_7228, pad = var_7751_pad_0, pad_type = var_7751_pad_type_0, strides = var_7747, weight = up_blocks_1_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16_palettized, x = input_301_cast)[name = tensor("op_7751_cast")]; + tensor inputs_45_cast = add(x = var_7751_cast, y = inputs_43_cast)[name = tensor("inputs_45_cast")]; + tensor var_7755 = const()[name = tensor("op_7755"), val = tensor([1])]; + tensor channels_mean_45_cast = reduce_mean(axes = var_7755, keep_dims = var_7223, x = inputs_45_cast)[name = tensor("channels_mean_45_cast")]; + tensor zero_mean_45_cast = sub(x = inputs_45_cast, y = channels_mean_45_cast)[name = tensor("zero_mean_45_cast")]; + tensor zero_mean_sq_45_cast = mul(x = zero_mean_45_cast, y = zero_mean_45_cast)[name = tensor("zero_mean_sq_45_cast")]; + tensor var_7759 = const()[name = tensor("op_7759"), val = tensor([1])]; + tensor var_7760_cast = reduce_mean(axes = var_7759, keep_dims = var_7223, x = zero_mean_sq_45_cast)[name = tensor("op_7760_cast")]; + tensor var_7761_to_fp16 = const()[name = tensor("op_7761_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_7762_cast = add(x = var_7760_cast, y = var_7761_to_fp16)[name = tensor("op_7762_cast")]; + tensor denom_45_epsilon_0_to_fp16 = const()[name = tensor("denom_45_epsilon_0_to_fp16"), val = tensor(0x1p-24)]; + tensor denom_45_cast = rsqrt(epsilon = denom_45_epsilon_0_to_fp16, x = var_7762_cast)[name = tensor("denom_45_cast")]; + tensor out_45_cast = mul(x = zero_mean_45_cast, y = denom_45_cast)[name = tensor("out_45_cast")]; + tensor var_7766_to_fp16 = const()[name = tensor("op_7766_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(428752128)))]; + tensor var_7767_cast = add(x = out_45_cast, y = var_7766_to_fp16)[name = tensor("op_7767_cast")]; + tensor var_7769_to_fp16 = const()[name = tensor("op_7769_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(428754752)))]; + tensor hidden_states_169_cast = mul(x = var_7767_cast, y = var_7769_to_fp16)[name = tensor("hidden_states_169_cast")]; + tensor var_7776 = const()[name = tensor("op_7776"), val = tensor([1, 1])]; + tensor var_7778 = const()[name = tensor("op_7778"), val = tensor([1, 1])]; + tensor q_31_pad_type_0 = const()[name = tensor("q_31_pad_type_0"), val = tensor("custom")]; + tensor q_31_pad_0 = const()[name = tensor("q_31_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_1_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(428757376))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(429986240))), name = tensor("up_blocks_1_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor q_31_cast = conv(dilations = var_7778, groups = var_7228, pad = q_31_pad_0, pad_type = q_31_pad_type_0, strides = var_7776, weight = up_blocks_1_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16_palettized, x = hidden_states_169_cast)[name = tensor("q_31_cast")]; + tensor var_7782 = const()[name = tensor("op_7782"), val = tensor([1, 1])]; + tensor var_7784 = const()[name = tensor("op_7784"), val = tensor([1, 1])]; + tensor k_61_pad_type_0 = const()[name = tensor("k_61_pad_type_0"), val = tensor("custom")]; + tensor k_61_pad_0 = const()[name = tensor("k_61_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_1_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(429986432))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(430969536))), name = tensor("up_blocks_1_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized"), shape = tensor([1280, 1024, 1, 1])]; + tensor k_61_cast = conv(dilations = var_7784, groups = var_7228, pad = k_61_pad_0, pad_type = k_61_pad_type_0, strides = var_7782, weight = up_blocks_1_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor("k_61_cast")]; + tensor var_7788 = const()[name = tensor("op_7788"), val = tensor([1, 1])]; + tensor var_7790 = const()[name = tensor("op_7790"), val = tensor([1, 1])]; + tensor v_31_pad_type_0 = const()[name = tensor("v_31_pad_type_0"), val = tensor("custom")]; + tensor v_31_pad_0 = const()[name = tensor("v_31_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_1_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(430969728))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(431952832))), name = tensor("up_blocks_1_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized"), shape = tensor([1280, 1024, 1, 1])]; + tensor v_31_cast = conv(dilations = var_7790, groups = var_7228, pad = v_31_pad_0, pad_type = v_31_pad_type_0, strides = var_7788, weight = up_blocks_1_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor("v_31_cast")]; + tensor var_7794_begin_0 = const()[name = tensor("op_7794_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_7794_end_0 = const()[name = tensor("op_7794_end_0"), val = tensor([2, 64, 1, 256])]; + tensor var_7794_end_mask_0 = const()[name = tensor("op_7794_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7794_cast = slice_by_index(begin = var_7794_begin_0, end = var_7794_end_0, end_mask = var_7794_end_mask_0, x = q_31_cast)[name = tensor("op_7794_cast")]; + tensor var_7798_begin_0 = const()[name = tensor("op_7798_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_7798_end_0 = const()[name = tensor("op_7798_end_0"), val = tensor([2, 128, 1, 256])]; + tensor var_7798_end_mask_0 = const()[name = tensor("op_7798_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7798_cast = slice_by_index(begin = var_7798_begin_0, end = var_7798_end_0, end_mask = var_7798_end_mask_0, x = q_31_cast)[name = tensor("op_7798_cast")]; + tensor var_7802_begin_0 = const()[name = tensor("op_7802_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_7802_end_0 = const()[name = tensor("op_7802_end_0"), val = tensor([2, 192, 1, 256])]; + tensor var_7802_end_mask_0 = const()[name = tensor("op_7802_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7802_cast = slice_by_index(begin = var_7802_begin_0, end = var_7802_end_0, end_mask = var_7802_end_mask_0, x = q_31_cast)[name = tensor("op_7802_cast")]; + tensor var_7806_begin_0 = const()[name = tensor("op_7806_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_7806_end_0 = const()[name = tensor("op_7806_end_0"), val = tensor([2, 256, 1, 256])]; + tensor var_7806_end_mask_0 = const()[name = tensor("op_7806_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7806_cast = slice_by_index(begin = var_7806_begin_0, end = var_7806_end_0, end_mask = var_7806_end_mask_0, x = q_31_cast)[name = tensor("op_7806_cast")]; + tensor var_7810_begin_0 = const()[name = tensor("op_7810_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_7810_end_0 = const()[name = tensor("op_7810_end_0"), val = tensor([2, 320, 1, 256])]; + tensor var_7810_end_mask_0 = const()[name = tensor("op_7810_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7810_cast = slice_by_index(begin = var_7810_begin_0, end = var_7810_end_0, end_mask = var_7810_end_mask_0, x = q_31_cast)[name = tensor("op_7810_cast")]; + tensor var_7814_begin_0 = const()[name = tensor("op_7814_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_7814_end_0 = const()[name = tensor("op_7814_end_0"), val = tensor([2, 384, 1, 256])]; + tensor var_7814_end_mask_0 = const()[name = tensor("op_7814_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7814_cast = slice_by_index(begin = var_7814_begin_0, end = var_7814_end_0, end_mask = var_7814_end_mask_0, x = q_31_cast)[name = tensor("op_7814_cast")]; + tensor var_7818_begin_0 = const()[name = tensor("op_7818_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_7818_end_0 = const()[name = tensor("op_7818_end_0"), val = tensor([2, 448, 1, 256])]; + tensor var_7818_end_mask_0 = const()[name = tensor("op_7818_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7818_cast = slice_by_index(begin = var_7818_begin_0, end = var_7818_end_0, end_mask = var_7818_end_mask_0, x = q_31_cast)[name = tensor("op_7818_cast")]; + tensor var_7822_begin_0 = const()[name = tensor("op_7822_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_7822_end_0 = const()[name = tensor("op_7822_end_0"), val = tensor([2, 512, 1, 256])]; + tensor var_7822_end_mask_0 = const()[name = tensor("op_7822_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7822_cast = slice_by_index(begin = var_7822_begin_0, end = var_7822_end_0, end_mask = var_7822_end_mask_0, x = q_31_cast)[name = tensor("op_7822_cast")]; + tensor var_7826_begin_0 = const()[name = tensor("op_7826_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_7826_end_0 = const()[name = tensor("op_7826_end_0"), val = tensor([2, 576, 1, 256])]; + tensor var_7826_end_mask_0 = const()[name = tensor("op_7826_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7826_cast = slice_by_index(begin = var_7826_begin_0, end = var_7826_end_0, end_mask = var_7826_end_mask_0, x = q_31_cast)[name = tensor("op_7826_cast")]; + tensor var_7830_begin_0 = const()[name = tensor("op_7830_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_7830_end_0 = const()[name = tensor("op_7830_end_0"), val = tensor([2, 640, 1, 256])]; + tensor var_7830_end_mask_0 = const()[name = tensor("op_7830_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7830_cast = slice_by_index(begin = var_7830_begin_0, end = var_7830_end_0, end_mask = var_7830_end_mask_0, x = q_31_cast)[name = tensor("op_7830_cast")]; + tensor var_7834_begin_0 = const()[name = tensor("op_7834_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_7834_end_0 = const()[name = tensor("op_7834_end_0"), val = tensor([2, 704, 1, 256])]; + tensor var_7834_end_mask_0 = const()[name = tensor("op_7834_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7834_cast = slice_by_index(begin = var_7834_begin_0, end = var_7834_end_0, end_mask = var_7834_end_mask_0, x = q_31_cast)[name = tensor("op_7834_cast")]; + tensor var_7838_begin_0 = const()[name = tensor("op_7838_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_7838_end_0 = const()[name = tensor("op_7838_end_0"), val = tensor([2, 768, 1, 256])]; + tensor var_7838_end_mask_0 = const()[name = tensor("op_7838_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7838_cast = slice_by_index(begin = var_7838_begin_0, end = var_7838_end_0, end_mask = var_7838_end_mask_0, x = q_31_cast)[name = tensor("op_7838_cast")]; + tensor var_7842_begin_0 = const()[name = tensor("op_7842_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_7842_end_0 = const()[name = tensor("op_7842_end_0"), val = tensor([2, 832, 1, 256])]; + tensor var_7842_end_mask_0 = const()[name = tensor("op_7842_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7842_cast = slice_by_index(begin = var_7842_begin_0, end = var_7842_end_0, end_mask = var_7842_end_mask_0, x = q_31_cast)[name = tensor("op_7842_cast")]; + tensor var_7846_begin_0 = const()[name = tensor("op_7846_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_7846_end_0 = const()[name = tensor("op_7846_end_0"), val = tensor([2, 896, 1, 256])]; + tensor var_7846_end_mask_0 = const()[name = tensor("op_7846_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7846_cast = slice_by_index(begin = var_7846_begin_0, end = var_7846_end_0, end_mask = var_7846_end_mask_0, x = q_31_cast)[name = tensor("op_7846_cast")]; + tensor var_7850_begin_0 = const()[name = tensor("op_7850_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_7850_end_0 = const()[name = tensor("op_7850_end_0"), val = tensor([2, 960, 1, 256])]; + tensor var_7850_end_mask_0 = const()[name = tensor("op_7850_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7850_cast = slice_by_index(begin = var_7850_begin_0, end = var_7850_end_0, end_mask = var_7850_end_mask_0, x = q_31_cast)[name = tensor("op_7850_cast")]; + tensor var_7854_begin_0 = const()[name = tensor("op_7854_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_7854_end_0 = const()[name = tensor("op_7854_end_0"), val = tensor([2, 1024, 1, 256])]; + tensor var_7854_end_mask_0 = const()[name = tensor("op_7854_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7854_cast = slice_by_index(begin = var_7854_begin_0, end = var_7854_end_0, end_mask = var_7854_end_mask_0, x = q_31_cast)[name = tensor("op_7854_cast")]; + tensor var_7858_begin_0 = const()[name = tensor("op_7858_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_7858_end_0 = const()[name = tensor("op_7858_end_0"), val = tensor([2, 1088, 1, 256])]; + tensor var_7858_end_mask_0 = const()[name = tensor("op_7858_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7858_cast = slice_by_index(begin = var_7858_begin_0, end = var_7858_end_0, end_mask = var_7858_end_mask_0, x = q_31_cast)[name = tensor("op_7858_cast")]; + tensor var_7862_begin_0 = const()[name = tensor("op_7862_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_7862_end_0 = const()[name = tensor("op_7862_end_0"), val = tensor([2, 1152, 1, 256])]; + tensor var_7862_end_mask_0 = const()[name = tensor("op_7862_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7862_cast = slice_by_index(begin = var_7862_begin_0, end = var_7862_end_0, end_mask = var_7862_end_mask_0, x = q_31_cast)[name = tensor("op_7862_cast")]; + tensor var_7866_begin_0 = const()[name = tensor("op_7866_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_7866_end_0 = const()[name = tensor("op_7866_end_0"), val = tensor([2, 1216, 1, 256])]; + tensor var_7866_end_mask_0 = const()[name = tensor("op_7866_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7866_cast = slice_by_index(begin = var_7866_begin_0, end = var_7866_end_0, end_mask = var_7866_end_mask_0, x = q_31_cast)[name = tensor("op_7866_cast")]; + tensor var_7870_begin_0 = const()[name = tensor("op_7870_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_7870_end_0 = const()[name = tensor("op_7870_end_0"), val = tensor([2, 1280, 1, 256])]; + tensor var_7870_end_mask_0 = const()[name = tensor("op_7870_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7870_cast = slice_by_index(begin = var_7870_begin_0, end = var_7870_end_0, end_mask = var_7870_end_mask_0, x = q_31_cast)[name = tensor("op_7870_cast")]; + tensor k_63_perm_0 = const()[name = tensor("k_63_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_7877_begin_0 = const()[name = tensor("op_7877_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_7877_end_0 = const()[name = tensor("op_7877_end_0"), val = tensor([2, 77, 1, 64])]; + tensor var_7877_end_mask_0 = const()[name = tensor("op_7877_end_mask_0"), val = tensor([true, true, true, false])]; + tensor transpose_16 = transpose(perm = k_63_perm_0, x = k_61_cast)[name = tensor("transpose_16")]; + tensor var_7877_cast = slice_by_index(begin = var_7877_begin_0, end = var_7877_end_0, end_mask = var_7877_end_mask_0, x = transpose_16)[name = tensor("op_7877_cast")]; + tensor var_7881_begin_0 = const()[name = tensor("op_7881_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_7881_end_0 = const()[name = tensor("op_7881_end_0"), val = tensor([2, 77, 1, 128])]; + tensor var_7881_end_mask_0 = const()[name = tensor("op_7881_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7881_cast = slice_by_index(begin = var_7881_begin_0, end = var_7881_end_0, end_mask = var_7881_end_mask_0, x = transpose_16)[name = tensor("op_7881_cast")]; + tensor var_7885_begin_0 = const()[name = tensor("op_7885_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_7885_end_0 = const()[name = tensor("op_7885_end_0"), val = tensor([2, 77, 1, 192])]; + tensor var_7885_end_mask_0 = const()[name = tensor("op_7885_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7885_cast = slice_by_index(begin = var_7885_begin_0, end = var_7885_end_0, end_mask = var_7885_end_mask_0, x = transpose_16)[name = tensor("op_7885_cast")]; + tensor var_7889_begin_0 = const()[name = tensor("op_7889_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_7889_end_0 = const()[name = tensor("op_7889_end_0"), val = tensor([2, 77, 1, 256])]; + tensor var_7889_end_mask_0 = const()[name = tensor("op_7889_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7889_cast = slice_by_index(begin = var_7889_begin_0, end = var_7889_end_0, end_mask = var_7889_end_mask_0, x = transpose_16)[name = tensor("op_7889_cast")]; + tensor var_7893_begin_0 = const()[name = tensor("op_7893_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_7893_end_0 = const()[name = tensor("op_7893_end_0"), val = tensor([2, 77, 1, 320])]; + tensor var_7893_end_mask_0 = const()[name = tensor("op_7893_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7893_cast = slice_by_index(begin = var_7893_begin_0, end = var_7893_end_0, end_mask = var_7893_end_mask_0, x = transpose_16)[name = tensor("op_7893_cast")]; + tensor var_7897_begin_0 = const()[name = tensor("op_7897_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_7897_end_0 = const()[name = tensor("op_7897_end_0"), val = tensor([2, 77, 1, 384])]; + tensor var_7897_end_mask_0 = const()[name = tensor("op_7897_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7897_cast = slice_by_index(begin = var_7897_begin_0, end = var_7897_end_0, end_mask = var_7897_end_mask_0, x = transpose_16)[name = tensor("op_7897_cast")]; + tensor var_7901_begin_0 = const()[name = tensor("op_7901_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_7901_end_0 = const()[name = tensor("op_7901_end_0"), val = tensor([2, 77, 1, 448])]; + tensor var_7901_end_mask_0 = const()[name = tensor("op_7901_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7901_cast = slice_by_index(begin = var_7901_begin_0, end = var_7901_end_0, end_mask = var_7901_end_mask_0, x = transpose_16)[name = tensor("op_7901_cast")]; + tensor var_7905_begin_0 = const()[name = tensor("op_7905_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_7905_end_0 = const()[name = tensor("op_7905_end_0"), val = tensor([2, 77, 1, 512])]; + tensor var_7905_end_mask_0 = const()[name = tensor("op_7905_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7905_cast = slice_by_index(begin = var_7905_begin_0, end = var_7905_end_0, end_mask = var_7905_end_mask_0, x = transpose_16)[name = tensor("op_7905_cast")]; + tensor var_7909_begin_0 = const()[name = tensor("op_7909_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_7909_end_0 = const()[name = tensor("op_7909_end_0"), val = tensor([2, 77, 1, 576])]; + tensor var_7909_end_mask_0 = const()[name = tensor("op_7909_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7909_cast = slice_by_index(begin = var_7909_begin_0, end = var_7909_end_0, end_mask = var_7909_end_mask_0, x = transpose_16)[name = tensor("op_7909_cast")]; + tensor var_7913_begin_0 = const()[name = tensor("op_7913_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_7913_end_0 = const()[name = tensor("op_7913_end_0"), val = tensor([2, 77, 1, 640])]; + tensor var_7913_end_mask_0 = const()[name = tensor("op_7913_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7913_cast = slice_by_index(begin = var_7913_begin_0, end = var_7913_end_0, end_mask = var_7913_end_mask_0, x = transpose_16)[name = tensor("op_7913_cast")]; + tensor var_7917_begin_0 = const()[name = tensor("op_7917_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_7917_end_0 = const()[name = tensor("op_7917_end_0"), val = tensor([2, 77, 1, 704])]; + tensor var_7917_end_mask_0 = const()[name = tensor("op_7917_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7917_cast = slice_by_index(begin = var_7917_begin_0, end = var_7917_end_0, end_mask = var_7917_end_mask_0, x = transpose_16)[name = tensor("op_7917_cast")]; + tensor var_7921_begin_0 = const()[name = tensor("op_7921_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_7921_end_0 = const()[name = tensor("op_7921_end_0"), val = tensor([2, 77, 1, 768])]; + tensor var_7921_end_mask_0 = const()[name = tensor("op_7921_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7921_cast = slice_by_index(begin = var_7921_begin_0, end = var_7921_end_0, end_mask = var_7921_end_mask_0, x = transpose_16)[name = tensor("op_7921_cast")]; + tensor var_7925_begin_0 = const()[name = tensor("op_7925_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_7925_end_0 = const()[name = tensor("op_7925_end_0"), val = tensor([2, 77, 1, 832])]; + tensor var_7925_end_mask_0 = const()[name = tensor("op_7925_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7925_cast = slice_by_index(begin = var_7925_begin_0, end = var_7925_end_0, end_mask = var_7925_end_mask_0, x = transpose_16)[name = tensor("op_7925_cast")]; + tensor var_7929_begin_0 = const()[name = tensor("op_7929_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_7929_end_0 = const()[name = tensor("op_7929_end_0"), val = tensor([2, 77, 1, 896])]; + tensor var_7929_end_mask_0 = const()[name = tensor("op_7929_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7929_cast = slice_by_index(begin = var_7929_begin_0, end = var_7929_end_0, end_mask = var_7929_end_mask_0, x = transpose_16)[name = tensor("op_7929_cast")]; + tensor var_7933_begin_0 = const()[name = tensor("op_7933_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_7933_end_0 = const()[name = tensor("op_7933_end_0"), val = tensor([2, 77, 1, 960])]; + tensor var_7933_end_mask_0 = const()[name = tensor("op_7933_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7933_cast = slice_by_index(begin = var_7933_begin_0, end = var_7933_end_0, end_mask = var_7933_end_mask_0, x = transpose_16)[name = tensor("op_7933_cast")]; + tensor var_7937_begin_0 = const()[name = tensor("op_7937_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_7937_end_0 = const()[name = tensor("op_7937_end_0"), val = tensor([2, 77, 1, 1024])]; + tensor var_7937_end_mask_0 = const()[name = tensor("op_7937_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7937_cast = slice_by_index(begin = var_7937_begin_0, end = var_7937_end_0, end_mask = var_7937_end_mask_0, x = transpose_16)[name = tensor("op_7937_cast")]; + tensor var_7941_begin_0 = const()[name = tensor("op_7941_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_7941_end_0 = const()[name = tensor("op_7941_end_0"), val = tensor([2, 77, 1, 1088])]; + tensor var_7941_end_mask_0 = const()[name = tensor("op_7941_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7941_cast = slice_by_index(begin = var_7941_begin_0, end = var_7941_end_0, end_mask = var_7941_end_mask_0, x = transpose_16)[name = tensor("op_7941_cast")]; + tensor var_7945_begin_0 = const()[name = tensor("op_7945_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_7945_end_0 = const()[name = tensor("op_7945_end_0"), val = tensor([2, 77, 1, 1152])]; + tensor var_7945_end_mask_0 = const()[name = tensor("op_7945_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7945_cast = slice_by_index(begin = var_7945_begin_0, end = var_7945_end_0, end_mask = var_7945_end_mask_0, x = transpose_16)[name = tensor("op_7945_cast")]; + tensor var_7949_begin_0 = const()[name = tensor("op_7949_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_7949_end_0 = const()[name = tensor("op_7949_end_0"), val = tensor([2, 77, 1, 1216])]; + tensor var_7949_end_mask_0 = const()[name = tensor("op_7949_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7949_cast = slice_by_index(begin = var_7949_begin_0, end = var_7949_end_0, end_mask = var_7949_end_mask_0, x = transpose_16)[name = tensor("op_7949_cast")]; + tensor var_7953_begin_0 = const()[name = tensor("op_7953_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_7953_end_0 = const()[name = tensor("op_7953_end_0"), val = tensor([2, 77, 1, 1280])]; + tensor var_7953_end_mask_0 = const()[name = tensor("op_7953_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7953_cast = slice_by_index(begin = var_7953_begin_0, end = var_7953_end_0, end_mask = var_7953_end_mask_0, x = transpose_16)[name = tensor("op_7953_cast")]; + tensor var_7955_begin_0 = const()[name = tensor("op_7955_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_7955_end_0 = const()[name = tensor("op_7955_end_0"), val = tensor([2, 64, 1, 77])]; + tensor var_7955_end_mask_0 = const()[name = tensor("op_7955_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7955_cast = slice_by_index(begin = var_7955_begin_0, end = var_7955_end_0, end_mask = var_7955_end_mask_0, x = v_31_cast)[name = tensor("op_7955_cast")]; + tensor var_7959_begin_0 = const()[name = tensor("op_7959_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_7959_end_0 = const()[name = tensor("op_7959_end_0"), val = tensor([2, 128, 1, 77])]; + tensor var_7959_end_mask_0 = const()[name = tensor("op_7959_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7959_cast = slice_by_index(begin = var_7959_begin_0, end = var_7959_end_0, end_mask = var_7959_end_mask_0, x = v_31_cast)[name = tensor("op_7959_cast")]; + tensor var_7963_begin_0 = const()[name = tensor("op_7963_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_7963_end_0 = const()[name = tensor("op_7963_end_0"), val = tensor([2, 192, 1, 77])]; + tensor var_7963_end_mask_0 = const()[name = tensor("op_7963_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7963_cast = slice_by_index(begin = var_7963_begin_0, end = var_7963_end_0, end_mask = var_7963_end_mask_0, x = v_31_cast)[name = tensor("op_7963_cast")]; + tensor var_7967_begin_0 = const()[name = tensor("op_7967_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_7967_end_0 = const()[name = tensor("op_7967_end_0"), val = tensor([2, 256, 1, 77])]; + tensor var_7967_end_mask_0 = const()[name = tensor("op_7967_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7967_cast = slice_by_index(begin = var_7967_begin_0, end = var_7967_end_0, end_mask = var_7967_end_mask_0, x = v_31_cast)[name = tensor("op_7967_cast")]; + tensor var_7971_begin_0 = const()[name = tensor("op_7971_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_7971_end_0 = const()[name = tensor("op_7971_end_0"), val = tensor([2, 320, 1, 77])]; + tensor var_7971_end_mask_0 = const()[name = tensor("op_7971_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7971_cast = slice_by_index(begin = var_7971_begin_0, end = var_7971_end_0, end_mask = var_7971_end_mask_0, x = v_31_cast)[name = tensor("op_7971_cast")]; + tensor var_7975_begin_0 = const()[name = tensor("op_7975_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_7975_end_0 = const()[name = tensor("op_7975_end_0"), val = tensor([2, 384, 1, 77])]; + tensor var_7975_end_mask_0 = const()[name = tensor("op_7975_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7975_cast = slice_by_index(begin = var_7975_begin_0, end = var_7975_end_0, end_mask = var_7975_end_mask_0, x = v_31_cast)[name = tensor("op_7975_cast")]; + tensor var_7979_begin_0 = const()[name = tensor("op_7979_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_7979_end_0 = const()[name = tensor("op_7979_end_0"), val = tensor([2, 448, 1, 77])]; + tensor var_7979_end_mask_0 = const()[name = tensor("op_7979_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7979_cast = slice_by_index(begin = var_7979_begin_0, end = var_7979_end_0, end_mask = var_7979_end_mask_0, x = v_31_cast)[name = tensor("op_7979_cast")]; + tensor var_7983_begin_0 = const()[name = tensor("op_7983_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_7983_end_0 = const()[name = tensor("op_7983_end_0"), val = tensor([2, 512, 1, 77])]; + tensor var_7983_end_mask_0 = const()[name = tensor("op_7983_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7983_cast = slice_by_index(begin = var_7983_begin_0, end = var_7983_end_0, end_mask = var_7983_end_mask_0, x = v_31_cast)[name = tensor("op_7983_cast")]; + tensor var_7987_begin_0 = const()[name = tensor("op_7987_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_7987_end_0 = const()[name = tensor("op_7987_end_0"), val = tensor([2, 576, 1, 77])]; + tensor var_7987_end_mask_0 = const()[name = tensor("op_7987_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7987_cast = slice_by_index(begin = var_7987_begin_0, end = var_7987_end_0, end_mask = var_7987_end_mask_0, x = v_31_cast)[name = tensor("op_7987_cast")]; + tensor var_7991_begin_0 = const()[name = tensor("op_7991_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_7991_end_0 = const()[name = tensor("op_7991_end_0"), val = tensor([2, 640, 1, 77])]; + tensor var_7991_end_mask_0 = const()[name = tensor("op_7991_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7991_cast = slice_by_index(begin = var_7991_begin_0, end = var_7991_end_0, end_mask = var_7991_end_mask_0, x = v_31_cast)[name = tensor("op_7991_cast")]; + tensor var_7995_begin_0 = const()[name = tensor("op_7995_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_7995_end_0 = const()[name = tensor("op_7995_end_0"), val = tensor([2, 704, 1, 77])]; + tensor var_7995_end_mask_0 = const()[name = tensor("op_7995_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7995_cast = slice_by_index(begin = var_7995_begin_0, end = var_7995_end_0, end_mask = var_7995_end_mask_0, x = v_31_cast)[name = tensor("op_7995_cast")]; + tensor var_7999_begin_0 = const()[name = tensor("op_7999_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_7999_end_0 = const()[name = tensor("op_7999_end_0"), val = tensor([2, 768, 1, 77])]; + tensor var_7999_end_mask_0 = const()[name = tensor("op_7999_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7999_cast = slice_by_index(begin = var_7999_begin_0, end = var_7999_end_0, end_mask = var_7999_end_mask_0, x = v_31_cast)[name = tensor("op_7999_cast")]; + tensor var_8003_begin_0 = const()[name = tensor("op_8003_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_8003_end_0 = const()[name = tensor("op_8003_end_0"), val = tensor([2, 832, 1, 77])]; + tensor var_8003_end_mask_0 = const()[name = tensor("op_8003_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8003_cast = slice_by_index(begin = var_8003_begin_0, end = var_8003_end_0, end_mask = var_8003_end_mask_0, x = v_31_cast)[name = tensor("op_8003_cast")]; + tensor var_8007_begin_0 = const()[name = tensor("op_8007_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_8007_end_0 = const()[name = tensor("op_8007_end_0"), val = tensor([2, 896, 1, 77])]; + tensor var_8007_end_mask_0 = const()[name = tensor("op_8007_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8007_cast = slice_by_index(begin = var_8007_begin_0, end = var_8007_end_0, end_mask = var_8007_end_mask_0, x = v_31_cast)[name = tensor("op_8007_cast")]; + tensor var_8011_begin_0 = const()[name = tensor("op_8011_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_8011_end_0 = const()[name = tensor("op_8011_end_0"), val = tensor([2, 960, 1, 77])]; + tensor var_8011_end_mask_0 = const()[name = tensor("op_8011_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8011_cast = slice_by_index(begin = var_8011_begin_0, end = var_8011_end_0, end_mask = var_8011_end_mask_0, x = v_31_cast)[name = tensor("op_8011_cast")]; + tensor var_8015_begin_0 = const()[name = tensor("op_8015_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_8015_end_0 = const()[name = tensor("op_8015_end_0"), val = tensor([2, 1024, 1, 77])]; + tensor var_8015_end_mask_0 = const()[name = tensor("op_8015_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8015_cast = slice_by_index(begin = var_8015_begin_0, end = var_8015_end_0, end_mask = var_8015_end_mask_0, x = v_31_cast)[name = tensor("op_8015_cast")]; + tensor var_8019_begin_0 = const()[name = tensor("op_8019_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_8019_end_0 = const()[name = tensor("op_8019_end_0"), val = tensor([2, 1088, 1, 77])]; + tensor var_8019_end_mask_0 = const()[name = tensor("op_8019_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8019_cast = slice_by_index(begin = var_8019_begin_0, end = var_8019_end_0, end_mask = var_8019_end_mask_0, x = v_31_cast)[name = tensor("op_8019_cast")]; + tensor var_8023_begin_0 = const()[name = tensor("op_8023_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_8023_end_0 = const()[name = tensor("op_8023_end_0"), val = tensor([2, 1152, 1, 77])]; + tensor var_8023_end_mask_0 = const()[name = tensor("op_8023_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8023_cast = slice_by_index(begin = var_8023_begin_0, end = var_8023_end_0, end_mask = var_8023_end_mask_0, x = v_31_cast)[name = tensor("op_8023_cast")]; + tensor var_8027_begin_0 = const()[name = tensor("op_8027_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_8027_end_0 = const()[name = tensor("op_8027_end_0"), val = tensor([2, 1216, 1, 77])]; + tensor var_8027_end_mask_0 = const()[name = tensor("op_8027_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8027_cast = slice_by_index(begin = var_8027_begin_0, end = var_8027_end_0, end_mask = var_8027_end_mask_0, x = v_31_cast)[name = tensor("op_8027_cast")]; + tensor var_8031_begin_0 = const()[name = tensor("op_8031_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_8031_end_0 = const()[name = tensor("op_8031_end_0"), val = tensor([2, 1280, 1, 77])]; + tensor var_8031_end_mask_0 = const()[name = tensor("op_8031_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8031_cast = slice_by_index(begin = var_8031_begin_0, end = var_8031_end_0, end_mask = var_8031_end_mask_0, x = v_31_cast)[name = tensor("op_8031_cast")]; + tensor var_8035_equation_0 = const()[name = tensor("op_8035_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8035_cast = einsum(equation = var_8035_equation_0, values = (var_7877_cast, var_7794_cast))[name = tensor("op_8035_cast")]; + tensor var_8036_to_fp16 = const()[name = tensor("op_8036_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_281_cast = mul(x = var_8035_cast, y = var_8036_to_fp16)[name = tensor("aw_281_cast")]; + tensor var_8039_equation_0 = const()[name = tensor("op_8039_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8039_cast = einsum(equation = var_8039_equation_0, values = (var_7881_cast, var_7798_cast))[name = tensor("op_8039_cast")]; + tensor var_8040_to_fp16 = const()[name = tensor("op_8040_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_283_cast = mul(x = var_8039_cast, y = var_8040_to_fp16)[name = tensor("aw_283_cast")]; + tensor var_8043_equation_0 = const()[name = tensor("op_8043_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8043_cast = einsum(equation = var_8043_equation_0, values = (var_7885_cast, var_7802_cast))[name = tensor("op_8043_cast")]; + tensor var_8044_to_fp16 = const()[name = tensor("op_8044_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_285_cast = mul(x = var_8043_cast, y = var_8044_to_fp16)[name = tensor("aw_285_cast")]; + tensor var_8047_equation_0 = const()[name = tensor("op_8047_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8047_cast = einsum(equation = var_8047_equation_0, values = (var_7889_cast, var_7806_cast))[name = tensor("op_8047_cast")]; + tensor var_8048_to_fp16 = const()[name = tensor("op_8048_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_287_cast = mul(x = var_8047_cast, y = var_8048_to_fp16)[name = tensor("aw_287_cast")]; + tensor var_8051_equation_0 = const()[name = tensor("op_8051_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8051_cast = einsum(equation = var_8051_equation_0, values = (var_7893_cast, var_7810_cast))[name = tensor("op_8051_cast")]; + tensor var_8052_to_fp16 = const()[name = tensor("op_8052_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_289_cast = mul(x = var_8051_cast, y = var_8052_to_fp16)[name = tensor("aw_289_cast")]; + tensor var_8055_equation_0 = const()[name = tensor("op_8055_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8055_cast = einsum(equation = var_8055_equation_0, values = (var_7897_cast, var_7814_cast))[name = tensor("op_8055_cast")]; + tensor var_8056_to_fp16 = const()[name = tensor("op_8056_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_291_cast = mul(x = var_8055_cast, y = var_8056_to_fp16)[name = tensor("aw_291_cast")]; + tensor var_8059_equation_0 = const()[name = tensor("op_8059_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8059_cast = einsum(equation = var_8059_equation_0, values = (var_7901_cast, var_7818_cast))[name = tensor("op_8059_cast")]; + tensor var_8060_to_fp16 = const()[name = tensor("op_8060_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_293_cast = mul(x = var_8059_cast, y = var_8060_to_fp16)[name = tensor("aw_293_cast")]; + tensor var_8063_equation_0 = const()[name = tensor("op_8063_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8063_cast = einsum(equation = var_8063_equation_0, values = (var_7905_cast, var_7822_cast))[name = tensor("op_8063_cast")]; + tensor var_8064_to_fp16 = const()[name = tensor("op_8064_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_295_cast = mul(x = var_8063_cast, y = var_8064_to_fp16)[name = tensor("aw_295_cast")]; + tensor var_8067_equation_0 = const()[name = tensor("op_8067_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8067_cast = einsum(equation = var_8067_equation_0, values = (var_7909_cast, var_7826_cast))[name = tensor("op_8067_cast")]; + tensor var_8068_to_fp16 = const()[name = tensor("op_8068_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_297_cast = mul(x = var_8067_cast, y = var_8068_to_fp16)[name = tensor("aw_297_cast")]; + tensor var_8071_equation_0 = const()[name = tensor("op_8071_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8071_cast = einsum(equation = var_8071_equation_0, values = (var_7913_cast, var_7830_cast))[name = tensor("op_8071_cast")]; + tensor var_8072_to_fp16 = const()[name = tensor("op_8072_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_299_cast = mul(x = var_8071_cast, y = var_8072_to_fp16)[name = tensor("aw_299_cast")]; + tensor var_8075_equation_0 = const()[name = tensor("op_8075_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8075_cast = einsum(equation = var_8075_equation_0, values = (var_7917_cast, var_7834_cast))[name = tensor("op_8075_cast")]; + tensor var_8076_to_fp16 = const()[name = tensor("op_8076_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_301_cast = mul(x = var_8075_cast, y = var_8076_to_fp16)[name = tensor("aw_301_cast")]; + tensor var_8079_equation_0 = const()[name = tensor("op_8079_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8079_cast = einsum(equation = var_8079_equation_0, values = (var_7921_cast, var_7838_cast))[name = tensor("op_8079_cast")]; + tensor var_8080_to_fp16 = const()[name = tensor("op_8080_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_303_cast = mul(x = var_8079_cast, y = var_8080_to_fp16)[name = tensor("aw_303_cast")]; + tensor var_8083_equation_0 = const()[name = tensor("op_8083_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8083_cast = einsum(equation = var_8083_equation_0, values = (var_7925_cast, var_7842_cast))[name = tensor("op_8083_cast")]; + tensor var_8084_to_fp16 = const()[name = tensor("op_8084_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_305_cast = mul(x = var_8083_cast, y = var_8084_to_fp16)[name = tensor("aw_305_cast")]; + tensor var_8087_equation_0 = const()[name = tensor("op_8087_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8087_cast = einsum(equation = var_8087_equation_0, values = (var_7929_cast, var_7846_cast))[name = tensor("op_8087_cast")]; + tensor var_8088_to_fp16 = const()[name = tensor("op_8088_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_307_cast = mul(x = var_8087_cast, y = var_8088_to_fp16)[name = tensor("aw_307_cast")]; + tensor var_8091_equation_0 = const()[name = tensor("op_8091_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8091_cast = einsum(equation = var_8091_equation_0, values = (var_7933_cast, var_7850_cast))[name = tensor("op_8091_cast")]; + tensor var_8092_to_fp16 = const()[name = tensor("op_8092_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_309_cast = mul(x = var_8091_cast, y = var_8092_to_fp16)[name = tensor("aw_309_cast")]; + tensor var_8095_equation_0 = const()[name = tensor("op_8095_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8095_cast = einsum(equation = var_8095_equation_0, values = (var_7937_cast, var_7854_cast))[name = tensor("op_8095_cast")]; + tensor var_8096_to_fp16 = const()[name = tensor("op_8096_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_311_cast = mul(x = var_8095_cast, y = var_8096_to_fp16)[name = tensor("aw_311_cast")]; + tensor var_8099_equation_0 = const()[name = tensor("op_8099_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8099_cast = einsum(equation = var_8099_equation_0, values = (var_7941_cast, var_7858_cast))[name = tensor("op_8099_cast")]; + tensor var_8100_to_fp16 = const()[name = tensor("op_8100_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_313_cast = mul(x = var_8099_cast, y = var_8100_to_fp16)[name = tensor("aw_313_cast")]; + tensor var_8103_equation_0 = const()[name = tensor("op_8103_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8103_cast = einsum(equation = var_8103_equation_0, values = (var_7945_cast, var_7862_cast))[name = tensor("op_8103_cast")]; + tensor var_8104_to_fp16 = const()[name = tensor("op_8104_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_315_cast = mul(x = var_8103_cast, y = var_8104_to_fp16)[name = tensor("aw_315_cast")]; + tensor var_8107_equation_0 = const()[name = tensor("op_8107_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8107_cast = einsum(equation = var_8107_equation_0, values = (var_7949_cast, var_7866_cast))[name = tensor("op_8107_cast")]; + tensor var_8108_to_fp16 = const()[name = tensor("op_8108_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_317_cast = mul(x = var_8107_cast, y = var_8108_to_fp16)[name = tensor("aw_317_cast")]; + tensor var_8111_equation_0 = const()[name = tensor("op_8111_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8111_cast = einsum(equation = var_8111_equation_0, values = (var_7953_cast, var_7870_cast))[name = tensor("op_8111_cast")]; + tensor var_8112_to_fp16 = const()[name = tensor("op_8112_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_319_cast = mul(x = var_8111_cast, y = var_8112_to_fp16)[name = tensor("aw_319_cast")]; + tensor var_8114_cast = softmax(axis = var_7228, x = aw_281_cast)[name = tensor("op_8114_cast")]; + tensor var_8115_cast = softmax(axis = var_7228, x = aw_283_cast)[name = tensor("op_8115_cast")]; + tensor var_8116_cast = softmax(axis = var_7228, x = aw_285_cast)[name = tensor("op_8116_cast")]; + tensor var_8117_cast = softmax(axis = var_7228, x = aw_287_cast)[name = tensor("op_8117_cast")]; + tensor var_8118_cast = softmax(axis = var_7228, x = aw_289_cast)[name = tensor("op_8118_cast")]; + tensor var_8119_cast = softmax(axis = var_7228, x = aw_291_cast)[name = tensor("op_8119_cast")]; + tensor var_8120_cast = softmax(axis = var_7228, x = aw_293_cast)[name = tensor("op_8120_cast")]; + tensor var_8121_cast = softmax(axis = var_7228, x = aw_295_cast)[name = tensor("op_8121_cast")]; + tensor var_8122_cast = softmax(axis = var_7228, x = aw_297_cast)[name = tensor("op_8122_cast")]; + tensor var_8123_cast = softmax(axis = var_7228, x = aw_299_cast)[name = tensor("op_8123_cast")]; + tensor var_8124_cast = softmax(axis = var_7228, x = aw_301_cast)[name = tensor("op_8124_cast")]; + tensor var_8125_cast = softmax(axis = var_7228, x = aw_303_cast)[name = tensor("op_8125_cast")]; + tensor var_8126_cast = softmax(axis = var_7228, x = aw_305_cast)[name = tensor("op_8126_cast")]; + tensor var_8127_cast = softmax(axis = var_7228, x = aw_307_cast)[name = tensor("op_8127_cast")]; + tensor var_8128_cast = softmax(axis = var_7228, x = aw_309_cast)[name = tensor("op_8128_cast")]; + tensor var_8129_cast = softmax(axis = var_7228, x = aw_311_cast)[name = tensor("op_8129_cast")]; + tensor var_8130_cast = softmax(axis = var_7228, x = aw_313_cast)[name = tensor("op_8130_cast")]; + tensor var_8131_cast = softmax(axis = var_7228, x = aw_315_cast)[name = tensor("op_8131_cast")]; + tensor var_8132_cast = softmax(axis = var_7228, x = aw_317_cast)[name = tensor("op_8132_cast")]; + tensor var_8133_cast = softmax(axis = var_7228, x = aw_319_cast)[name = tensor("op_8133_cast")]; + tensor var_8135_equation_0 = const()[name = tensor("op_8135_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8135_cast = einsum(equation = var_8135_equation_0, values = (var_7955_cast, var_8114_cast))[name = tensor("op_8135_cast")]; + tensor var_8137_equation_0 = const()[name = tensor("op_8137_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8137_cast = einsum(equation = var_8137_equation_0, values = (var_7959_cast, var_8115_cast))[name = tensor("op_8137_cast")]; + tensor var_8139_equation_0 = const()[name = tensor("op_8139_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8139_cast = einsum(equation = var_8139_equation_0, values = (var_7963_cast, var_8116_cast))[name = tensor("op_8139_cast")]; + tensor var_8141_equation_0 = const()[name = tensor("op_8141_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8141_cast = einsum(equation = var_8141_equation_0, values = (var_7967_cast, var_8117_cast))[name = tensor("op_8141_cast")]; + tensor var_8143_equation_0 = const()[name = tensor("op_8143_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8143_cast = einsum(equation = var_8143_equation_0, values = (var_7971_cast, var_8118_cast))[name = tensor("op_8143_cast")]; + tensor var_8145_equation_0 = const()[name = tensor("op_8145_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8145_cast = einsum(equation = var_8145_equation_0, values = (var_7975_cast, var_8119_cast))[name = tensor("op_8145_cast")]; + tensor var_8147_equation_0 = const()[name = tensor("op_8147_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8147_cast = einsum(equation = var_8147_equation_0, values = (var_7979_cast, var_8120_cast))[name = tensor("op_8147_cast")]; + tensor var_8149_equation_0 = const()[name = tensor("op_8149_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8149_cast = einsum(equation = var_8149_equation_0, values = (var_7983_cast, var_8121_cast))[name = tensor("op_8149_cast")]; + tensor var_8151_equation_0 = const()[name = tensor("op_8151_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8151_cast = einsum(equation = var_8151_equation_0, values = (var_7987_cast, var_8122_cast))[name = tensor("op_8151_cast")]; + tensor var_8153_equation_0 = const()[name = tensor("op_8153_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8153_cast = einsum(equation = var_8153_equation_0, values = (var_7991_cast, var_8123_cast))[name = tensor("op_8153_cast")]; + tensor var_8155_equation_0 = const()[name = tensor("op_8155_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8155_cast = einsum(equation = var_8155_equation_0, values = (var_7995_cast, var_8124_cast))[name = tensor("op_8155_cast")]; + tensor var_8157_equation_0 = const()[name = tensor("op_8157_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8157_cast = einsum(equation = var_8157_equation_0, values = (var_7999_cast, var_8125_cast))[name = tensor("op_8157_cast")]; + tensor var_8159_equation_0 = const()[name = tensor("op_8159_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8159_cast = einsum(equation = var_8159_equation_0, values = (var_8003_cast, var_8126_cast))[name = tensor("op_8159_cast")]; + tensor var_8161_equation_0 = const()[name = tensor("op_8161_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8161_cast = einsum(equation = var_8161_equation_0, values = (var_8007_cast, var_8127_cast))[name = tensor("op_8161_cast")]; + tensor var_8163_equation_0 = const()[name = tensor("op_8163_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8163_cast = einsum(equation = var_8163_equation_0, values = (var_8011_cast, var_8128_cast))[name = tensor("op_8163_cast")]; + tensor var_8165_equation_0 = const()[name = tensor("op_8165_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8165_cast = einsum(equation = var_8165_equation_0, values = (var_8015_cast, var_8129_cast))[name = tensor("op_8165_cast")]; + tensor var_8167_equation_0 = const()[name = tensor("op_8167_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8167_cast = einsum(equation = var_8167_equation_0, values = (var_8019_cast, var_8130_cast))[name = tensor("op_8167_cast")]; + tensor var_8169_equation_0 = const()[name = tensor("op_8169_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8169_cast = einsum(equation = var_8169_equation_0, values = (var_8023_cast, var_8131_cast))[name = tensor("op_8169_cast")]; + tensor var_8171_equation_0 = const()[name = tensor("op_8171_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8171_cast = einsum(equation = var_8171_equation_0, values = (var_8027_cast, var_8132_cast))[name = tensor("op_8171_cast")]; + tensor var_8173_equation_0 = const()[name = tensor("op_8173_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8173_cast = einsum(equation = var_8173_equation_0, values = (var_8031_cast, var_8133_cast))[name = tensor("op_8173_cast")]; + tensor input_303_interleave_0 = const()[name = tensor("input_303_interleave_0"), val = tensor(false)]; + tensor input_303_cast = concat(axis = var_7228, interleave = input_303_interleave_0, values = (var_8135_cast, var_8137_cast, var_8139_cast, var_8141_cast, var_8143_cast, var_8145_cast, var_8147_cast, var_8149_cast, var_8151_cast, var_8153_cast, var_8155_cast, var_8157_cast, var_8159_cast, var_8161_cast, var_8163_cast, var_8165_cast, var_8167_cast, var_8169_cast, var_8171_cast, var_8173_cast))[name = tensor("input_303_cast")]; + tensor var_8179 = const()[name = tensor("op_8179"), val = tensor([1, 1])]; + tensor var_8181 = const()[name = tensor("op_8181"), val = tensor([1, 1])]; + tensor var_8183_pad_type_0 = const()[name = tensor("op_8183_pad_type_0"), val = tensor("custom")]; + tensor var_8183_pad_0 = const()[name = tensor("op_8183_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_1_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(431953024))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(433181888))), name = tensor("up_blocks_1_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor up_blocks_1_attentions_0_transformer_blocks_0_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("up_blocks_1_attentions_0_transformer_blocks_0_attn2_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(433182080)))]; + tensor var_8183_cast = conv(bias = up_blocks_1_attentions_0_transformer_blocks_0_attn2_to_out_0_bias_to_fp16, dilations = var_8181, groups = var_7228, pad = var_8183_pad_0, pad_type = var_8183_pad_type_0, strides = var_8179, weight = up_blocks_1_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16_palettized, x = input_303_cast)[name = tensor("op_8183_cast")]; + tensor inputs_47_cast = add(x = var_8183_cast, y = inputs_45_cast)[name = tensor("inputs_47_cast")]; + tensor var_8187 = const()[name = tensor("op_8187"), val = tensor([1])]; + tensor channels_mean_47_cast = reduce_mean(axes = var_8187, keep_dims = var_7223, x = inputs_47_cast)[name = tensor("channels_mean_47_cast")]; + tensor zero_mean_47_cast = sub(x = inputs_47_cast, y = channels_mean_47_cast)[name = tensor("zero_mean_47_cast")]; + tensor zero_mean_sq_47_cast = mul(x = zero_mean_47_cast, y = zero_mean_47_cast)[name = tensor("zero_mean_sq_47_cast")]; + tensor var_8191 = const()[name = tensor("op_8191"), val = tensor([1])]; + tensor var_8192_cast = reduce_mean(axes = var_8191, keep_dims = var_7223, x = zero_mean_sq_47_cast)[name = tensor("op_8192_cast")]; + tensor var_8193_to_fp16 = const()[name = tensor("op_8193_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_8194_cast = add(x = var_8192_cast, y = var_8193_to_fp16)[name = tensor("op_8194_cast")]; + tensor denom_47_epsilon_0_to_fp16 = const()[name = tensor("denom_47_epsilon_0_to_fp16"), val = tensor(0x1p-24)]; + tensor denom_47_cast = rsqrt(epsilon = denom_47_epsilon_0_to_fp16, x = var_8194_cast)[name = tensor("denom_47_cast")]; + tensor out_47_cast = mul(x = zero_mean_47_cast, y = denom_47_cast)[name = tensor("out_47_cast")]; + tensor var_8198_to_fp16 = const()[name = tensor("op_8198_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(433184704)))]; + tensor var_8199_cast = add(x = out_47_cast, y = var_8198_to_fp16)[name = tensor("op_8199_cast")]; + tensor var_8201_to_fp16 = const()[name = tensor("op_8201_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(433187328)))]; + tensor input_305_cast = mul(x = var_8199_cast, y = var_8201_to_fp16)[name = tensor("input_305_cast")]; + tensor var_8209 = const()[name = tensor("op_8209"), val = tensor([1, 1])]; + tensor var_8211 = const()[name = tensor("op_8211"), val = tensor([1, 1])]; + tensor var_8213_pad_type_0 = const()[name = tensor("op_8213_pad_type_0"), val = tensor("custom")]; + tensor var_8213_pad_0 = const()[name = tensor("op_8213_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_1_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(433189952))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(443020416))), name = tensor("up_blocks_1_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16_palettized"), shape = tensor([10240, 1280, 1, 1])]; + tensor up_blocks_1_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(443020608))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(443028352))), name = tensor("up_blocks_1_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16_palettized"), shape = tensor([10240])]; + tensor var_8213_cast = conv(bias = up_blocks_1_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16_palettized, dilations = var_8211, groups = var_7228, pad = var_8213_pad_0, pad_type = var_8213_pad_type_0, strides = var_8209, weight = up_blocks_1_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16_palettized, x = input_305_cast)[name = tensor("op_8213_cast")]; + tensor var_8214_split_sizes_0 = const()[name = tensor("op_8214_split_sizes_0"), val = tensor([5120, 5120])]; + tensor var_8214_axis_0 = const()[name = tensor("op_8214_axis_0"), val = tensor(1)]; + tensor var_8214_cast_0, tensor var_8214_cast_1 = split(axis = var_8214_axis_0, split_sizes = var_8214_split_sizes_0, x = var_8213_cast)[name = tensor("op_8214_cast")]; + tensor var_8216_mode_0 = const()[name = tensor("op_8216_mode_0"), val = tensor("EXACT")]; + tensor var_8216_cast = gelu(mode = var_8216_mode_0, x = var_8214_cast_1)[name = tensor("op_8216_cast")]; + tensor input_307_cast = mul(x = var_8214_cast_0, y = var_8216_cast)[name = tensor("input_307_cast")]; + tensor var_8220 = const()[name = tensor("op_8220"), val = tensor([1, 1])]; + tensor var_8222 = const()[name = tensor("op_8222"), val = tensor([1, 1])]; + tensor var_8224_pad_type_0 = const()[name = tensor("op_8224_pad_type_0"), val = tensor("custom")]; + tensor var_8224_pad_0 = const()[name = tensor("op_8224_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_1_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(443028544))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(447943808))), name = tensor("up_blocks_1_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16_palettized"), shape = tensor([1280, 5120, 1, 1])]; + tensor up_blocks_1_attentions_0_transformer_blocks_0_ff_net_2_bias_to_fp16 = const()[name = tensor("up_blocks_1_attentions_0_transformer_blocks_0_ff_net_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(447944000)))]; + tensor var_8224_cast = conv(bias = up_blocks_1_attentions_0_transformer_blocks_0_ff_net_2_bias_to_fp16, dilations = var_8222, groups = var_7228, pad = var_8224_pad_0, pad_type = var_8224_pad_type_0, strides = var_8220, weight = up_blocks_1_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16_palettized, x = input_307_cast)[name = tensor("op_8224_cast")]; + tensor hidden_states_173_cast = add(x = var_8224_cast, y = inputs_47_cast)[name = tensor("hidden_states_173_cast")]; + tensor var_8226 = const()[name = tensor("op_8226"), val = tensor([2, 1280, 16, 16])]; + tensor input_309_cast = reshape(shape = var_8226, x = hidden_states_173_cast)[name = tensor("input_309_cast")]; + tensor var_8230 = const()[name = tensor("op_8230"), val = tensor([1, 1])]; + tensor var_8232 = const()[name = tensor("op_8232"), val = tensor([1, 1])]; + tensor hidden_states_175_pad_type_0 = const()[name = tensor("hidden_states_175_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_175_pad_0 = const()[name = tensor("hidden_states_175_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_1_attentions_0_proj_out_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(447946624))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(449175488))), name = tensor("up_blocks_1_attentions_0_proj_out_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor up_blocks_1_attentions_0_proj_out_bias_to_fp16 = const()[name = tensor("up_blocks_1_attentions_0_proj_out_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(449175680)))]; + tensor hidden_states_175_cast = conv(bias = up_blocks_1_attentions_0_proj_out_bias_to_fp16, dilations = var_8232, groups = var_7228, pad = hidden_states_175_pad_0, pad_type = hidden_states_175_pad_type_0, strides = var_8230, weight = up_blocks_1_attentions_0_proj_out_weight_to_fp16_palettized, x = input_309_cast)[name = tensor("hidden_states_175_cast")]; + tensor hidden_states_177_cast = add(x = hidden_states_175_cast, y = hidden_states_163_cast)[name = tensor("hidden_states_177_cast")]; + tensor input_311_interleave_0 = const()[name = tensor("input_311_interleave_0"), val = tensor(false)]; + tensor input_311_cast = concat(axis = var_7228, interleave = input_311_interleave_0, values = (hidden_states_177_cast, input_143_cast))[name = tensor("input_311_cast")]; + tensor reshape_144_shape_0 = const()[name = tensor("reshape_144_shape_0"), val = tensor([2, 32, 80, 16, 16])]; + tensor reshape_144_cast = reshape(shape = reshape_144_shape_0, x = input_311_cast)[name = tensor("reshape_144_cast")]; + tensor reduce_mean_108_axes_0 = const()[name = tensor("reduce_mean_108_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_108_keep_dims_0 = const()[name = tensor("reduce_mean_108_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_108_cast = reduce_mean(axes = reduce_mean_108_axes_0, keep_dims = reduce_mean_108_keep_dims_0, x = reshape_144_cast)[name = tensor("reduce_mean_108_cast")]; + tensor sub_72_cast = sub(x = reshape_144_cast, y = reduce_mean_108_cast)[name = tensor("sub_72_cast")]; + tensor square_36_cast = square(x = sub_72_cast)[name = tensor("square_36_cast")]; + tensor reduce_mean_110_axes_0 = const()[name = tensor("reduce_mean_110_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_110_keep_dims_0 = const()[name = tensor("reduce_mean_110_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_110_cast = reduce_mean(axes = reduce_mean_110_axes_0, keep_dims = reduce_mean_110_keep_dims_0, x = square_36_cast)[name = tensor("reduce_mean_110_cast")]; + tensor add_72_y_0_to_fp16 = const()[name = tensor("add_72_y_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_72_cast = add(x = reduce_mean_110_cast, y = add_72_y_0_to_fp16)[name = tensor("add_72_cast")]; + tensor sqrt_36_cast = sqrt(x = add_72_cast)[name = tensor("sqrt_36_cast")]; + tensor real_div_36_cast = real_div(x = sub_72_cast, y = sqrt_36_cast)[name = tensor("real_div_36_cast")]; + tensor reshape_145_shape_0 = const()[name = tensor("reshape_145_shape_0"), val = tensor([2, 2560, 16, 16])]; + tensor reshape_145_cast = reshape(shape = reshape_145_shape_0, x = real_div_36_cast)[name = tensor("reshape_145_cast")]; + tensor add_73_gamma_0_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(449178304))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(449180288))), name = tensor("add_73_gamma_0_to_fp16_palettized"), shape = tensor([2560])]; + tensor add_73_beta_0_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(449180480))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(449182464))), name = tensor("add_73_beta_0_to_fp16_palettized"), shape = tensor([2560])]; + tensor add_73_epsilon_0_to_fp16 = const()[name = tensor("add_73_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_73_cast = batch_norm(beta = add_73_beta_0_to_fp16_palettized, epsilon = add_73_epsilon_0_to_fp16, gamma = add_73_gamma_0_to_fp16_palettized, mean = add_55_mean_0_to_fp16_palettized, variance = add_55_variance_0_to_fp16_palettized, x = reshape_145_cast)[name = tensor("add_73_cast")]; + tensor input_315_cast = silu(x = add_73_cast)[name = tensor("input_315_cast")]; + tensor var_8250 = const()[name = tensor("op_8250"), val = tensor([1, 1])]; + tensor var_8252 = const()[name = tensor("op_8252"), val = tensor([1, 1])]; + tensor hidden_states_179_pad_type_0 = const()[name = tensor("hidden_states_179_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_179_pad_0 = const()[name = tensor("hidden_states_179_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor up_blocks_1_resnets_1_conv1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(449182656))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(471301120))), name = tensor("up_blocks_1_resnets_1_conv1_weight_to_fp16_palettized"), shape = tensor([1280, 2560, 3, 3])]; + tensor up_blocks_1_resnets_1_conv1_bias_to_fp16 = const()[name = tensor("up_blocks_1_resnets_1_conv1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(471301312)))]; + tensor hidden_states_179_cast = conv(bias = up_blocks_1_resnets_1_conv1_bias_to_fp16, dilations = var_8252, groups = var_7228, pad = hidden_states_179_pad_0, pad_type = hidden_states_179_pad_type_0, strides = var_8250, weight = up_blocks_1_resnets_1_conv1_weight_to_fp16_palettized, x = input_315_cast)[name = tensor("hidden_states_179_cast")]; + tensor var_8258 = const()[name = tensor("op_8258"), val = tensor([1, 1])]; + tensor var_8260 = const()[name = tensor("op_8260"), val = tensor([1, 1])]; + tensor temb_29_pad_type_0 = const()[name = tensor("temb_29_pad_type_0"), val = tensor("custom")]; + tensor temb_29_pad_0 = const()[name = tensor("temb_29_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_1_resnets_1_time_emb_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(471303936))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(472532800))), name = tensor("up_blocks_1_resnets_1_time_emb_proj_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor up_blocks_1_resnets_1_time_emb_proj_bias_to_fp16 = const()[name = tensor("up_blocks_1_resnets_1_time_emb_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(472532992)))]; + tensor temb_29_cast = conv(bias = up_blocks_1_resnets_1_time_emb_proj_bias_to_fp16, dilations = var_8260, groups = var_7228, pad = temb_29_pad_0, pad_type = temb_29_pad_type_0, strides = var_8258, weight = up_blocks_1_resnets_1_time_emb_proj_weight_to_fp16_palettized, x = input_15_cast)[name = tensor("temb_29_cast")]; + tensor input_319_cast = add(x = hidden_states_179_cast, y = temb_29_cast)[name = tensor("input_319_cast")]; + tensor reshape_148_shape_0 = const()[name = tensor("reshape_148_shape_0"), val = tensor([2, 32, 40, 16, 16])]; + tensor reshape_148_cast = reshape(shape = reshape_148_shape_0, x = input_319_cast)[name = tensor("reshape_148_cast")]; + tensor reduce_mean_111_axes_0 = const()[name = tensor("reduce_mean_111_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_111_keep_dims_0 = const()[name = tensor("reduce_mean_111_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_111_cast = reduce_mean(axes = reduce_mean_111_axes_0, keep_dims = reduce_mean_111_keep_dims_0, x = reshape_148_cast)[name = tensor("reduce_mean_111_cast")]; + tensor sub_74_cast = sub(x = reshape_148_cast, y = reduce_mean_111_cast)[name = tensor("sub_74_cast")]; + tensor square_37_cast = square(x = sub_74_cast)[name = tensor("square_37_cast")]; + tensor reduce_mean_113_axes_0 = const()[name = tensor("reduce_mean_113_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_113_keep_dims_0 = const()[name = tensor("reduce_mean_113_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_113_cast = reduce_mean(axes = reduce_mean_113_axes_0, keep_dims = reduce_mean_113_keep_dims_0, x = square_37_cast)[name = tensor("reduce_mean_113_cast")]; + tensor add_74_y_0_to_fp16 = const()[name = tensor("add_74_y_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_74_cast = add(x = reduce_mean_113_cast, y = add_74_y_0_to_fp16)[name = tensor("add_74_cast")]; + tensor sqrt_37_cast = sqrt(x = add_74_cast)[name = tensor("sqrt_37_cast")]; + tensor real_div_37_cast = real_div(x = sub_74_cast, y = sqrt_37_cast)[name = tensor("real_div_37_cast")]; + tensor reshape_149_shape_0 = const()[name = tensor("reshape_149_shape_0"), val = tensor([2, 1280, 16, 16])]; + tensor reshape_149_cast = reshape(shape = reshape_149_shape_0, x = real_div_37_cast)[name = tensor("reshape_149_cast")]; + tensor add_75_gamma_0_to_fp16 = const()[name = tensor("add_75_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(472535616)))]; + tensor add_75_beta_0_to_fp16 = const()[name = tensor("add_75_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(472538240)))]; + tensor add_75_epsilon_0_to_fp16 = const()[name = tensor("add_75_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_75_cast = batch_norm(beta = add_75_beta_0_to_fp16, epsilon = add_75_epsilon_0_to_fp16, gamma = add_75_gamma_0_to_fp16, mean = add_27_mean_0_to_fp16, variance = add_27_variance_0_to_fp16, x = reshape_149_cast)[name = tensor("add_75_cast")]; + tensor input_323_cast = silu(x = add_75_cast)[name = tensor("input_323_cast")]; + tensor var_8270 = const()[name = tensor("op_8270"), val = tensor([1, 1])]; + tensor var_8272 = const()[name = tensor("op_8272"), val = tensor([1, 1])]; + tensor hidden_states_181_pad_type_0 = const()[name = tensor("hidden_states_181_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_181_pad_0 = const()[name = tensor("hidden_states_181_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor up_blocks_1_resnets_1_conv2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(472540864))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(483600128))), name = tensor("up_blocks_1_resnets_1_conv2_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 3, 3])]; + tensor up_blocks_1_resnets_1_conv2_bias_to_fp16 = const()[name = tensor("up_blocks_1_resnets_1_conv2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(483600320)))]; + tensor hidden_states_181_cast = conv(bias = up_blocks_1_resnets_1_conv2_bias_to_fp16, dilations = var_8272, groups = var_7228, pad = hidden_states_181_pad_0, pad_type = hidden_states_181_pad_type_0, strides = var_8270, weight = up_blocks_1_resnets_1_conv2_weight_to_fp16_palettized, x = input_323_cast)[name = tensor("hidden_states_181_cast")]; + tensor var_8277 = const()[name = tensor("op_8277"), val = tensor([1, 1])]; + tensor var_8279 = const()[name = tensor("op_8279"), val = tensor([1, 1])]; + tensor x_13_pad_type_0 = const()[name = tensor("x_13_pad_type_0"), val = tensor("custom")]; + tensor x_13_pad_0 = const()[name = tensor("x_13_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_1_resnets_1_conv_shortcut_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(483602944))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(486060608))), name = tensor("up_blocks_1_resnets_1_conv_shortcut_weight_to_fp16_palettized"), shape = tensor([1280, 2560, 1, 1])]; + tensor up_blocks_1_resnets_1_conv_shortcut_bias_to_fp16 = const()[name = tensor("up_blocks_1_resnets_1_conv_shortcut_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(486060800)))]; + tensor x_13_cast = conv(bias = up_blocks_1_resnets_1_conv_shortcut_bias_to_fp16, dilations = var_8279, groups = var_7228, pad = x_13_pad_0, pad_type = x_13_pad_type_0, strides = var_8277, weight = up_blocks_1_resnets_1_conv_shortcut_weight_to_fp16_palettized, x = input_311_cast)[name = tensor("x_13_cast")]; + tensor hidden_states_183_cast = add(x = x_13_cast, y = hidden_states_181_cast)[name = tensor("hidden_states_183_cast")]; + tensor reshape_152_shape_0 = const()[name = tensor("reshape_152_shape_0"), val = tensor([2, 32, 40, 16, 16])]; + tensor reshape_152_cast = reshape(shape = reshape_152_shape_0, x = hidden_states_183_cast)[name = tensor("reshape_152_cast")]; + tensor reduce_mean_114_axes_0 = const()[name = tensor("reduce_mean_114_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_114_keep_dims_0 = const()[name = tensor("reduce_mean_114_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_114_cast = reduce_mean(axes = reduce_mean_114_axes_0, keep_dims = reduce_mean_114_keep_dims_0, x = reshape_152_cast)[name = tensor("reduce_mean_114_cast")]; + tensor sub_76_cast = sub(x = reshape_152_cast, y = reduce_mean_114_cast)[name = tensor("sub_76_cast")]; + tensor square_38_cast = square(x = sub_76_cast)[name = tensor("square_38_cast")]; + tensor reduce_mean_116_axes_0 = const()[name = tensor("reduce_mean_116_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_116_keep_dims_0 = const()[name = tensor("reduce_mean_116_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_116_cast = reduce_mean(axes = reduce_mean_116_axes_0, keep_dims = reduce_mean_116_keep_dims_0, x = square_38_cast)[name = tensor("reduce_mean_116_cast")]; + tensor add_76_y_0_to_fp16 = const()[name = tensor("add_76_y_0_to_fp16"), val = tensor(0x1.1p-20)]; + tensor add_76_cast = add(x = reduce_mean_116_cast, y = add_76_y_0_to_fp16)[name = tensor("add_76_cast")]; + tensor sqrt_38_cast = sqrt(x = add_76_cast)[name = tensor("sqrt_38_cast")]; + tensor real_div_38_cast = real_div(x = sub_76_cast, y = sqrt_38_cast)[name = tensor("real_div_38_cast")]; + tensor reshape_153_shape_0 = const()[name = tensor("reshape_153_shape_0"), val = tensor([2, 1280, 16, 16])]; + tensor reshape_153_cast = reshape(shape = reshape_153_shape_0, x = real_div_38_cast)[name = tensor("reshape_153_cast")]; + tensor add_77_gamma_0_to_fp16 = const()[name = tensor("add_77_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(486063424)))]; + tensor add_77_beta_0_to_fp16 = const()[name = tensor("add_77_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(486066048)))]; + tensor add_77_epsilon_0_to_fp16 = const()[name = tensor("add_77_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_77_cast = batch_norm(beta = add_77_beta_0_to_fp16, epsilon = add_77_epsilon_0_to_fp16, gamma = add_77_gamma_0_to_fp16, mean = add_27_mean_0_to_fp16, variance = add_27_variance_0_to_fp16, x = reshape_153_cast)[name = tensor("add_77_cast")]; + tensor var_8299 = const()[name = tensor("op_8299"), val = tensor([1, 1])]; + tensor var_8301 = const()[name = tensor("op_8301"), val = tensor([1, 1])]; + tensor hidden_states_185_pad_type_0 = const()[name = tensor("hidden_states_185_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_185_pad_0 = const()[name = tensor("hidden_states_185_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_1_attentions_1_proj_in_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(486068672))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(487297536))), name = tensor("up_blocks_1_attentions_1_proj_in_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor up_blocks_1_attentions_1_proj_in_bias_to_fp16 = const()[name = tensor("up_blocks_1_attentions_1_proj_in_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(487297728)))]; + tensor hidden_states_185_cast = conv(bias = up_blocks_1_attentions_1_proj_in_bias_to_fp16, dilations = var_8301, groups = var_7228, pad = hidden_states_185_pad_0, pad_type = hidden_states_185_pad_type_0, strides = var_8299, weight = up_blocks_1_attentions_1_proj_in_weight_to_fp16_palettized, x = add_77_cast)[name = tensor("hidden_states_185_cast")]; + tensor var_8306 = const()[name = tensor("op_8306"), val = tensor([2, 1280, 1, 256])]; + tensor inputs_49_cast = reshape(shape = var_8306, x = hidden_states_185_cast)[name = tensor("inputs_49_cast")]; + tensor var_8316 = const()[name = tensor("op_8316"), val = tensor([1])]; + tensor channels_mean_49_cast = reduce_mean(axes = var_8316, keep_dims = var_7223, x = inputs_49_cast)[name = tensor("channels_mean_49_cast")]; + tensor zero_mean_49_cast = sub(x = inputs_49_cast, y = channels_mean_49_cast)[name = tensor("zero_mean_49_cast")]; + tensor zero_mean_sq_49_cast = mul(x = zero_mean_49_cast, y = zero_mean_49_cast)[name = tensor("zero_mean_sq_49_cast")]; + tensor var_8320 = const()[name = tensor("op_8320"), val = tensor([1])]; + tensor var_8321_cast = reduce_mean(axes = var_8320, keep_dims = var_7223, x = zero_mean_sq_49_cast)[name = tensor("op_8321_cast")]; + tensor var_8322_to_fp16 = const()[name = tensor("op_8322_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_8323_cast = add(x = var_8321_cast, y = var_8322_to_fp16)[name = tensor("op_8323_cast")]; + tensor denom_49_epsilon_0_to_fp16 = const()[name = tensor("denom_49_epsilon_0_to_fp16"), val = tensor(0x1p-24)]; + tensor denom_49_cast = rsqrt(epsilon = denom_49_epsilon_0_to_fp16, x = var_8323_cast)[name = tensor("denom_49_cast")]; + tensor out_49_cast = mul(x = zero_mean_49_cast, y = denom_49_cast)[name = tensor("out_49_cast")]; + tensor var_8327_to_fp16 = const()[name = tensor("op_8327_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(487300352)))]; + tensor var_8328_cast = add(x = out_49_cast, y = var_8327_to_fp16)[name = tensor("op_8328_cast")]; + tensor var_8330_to_fp16 = const()[name = tensor("op_8330_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(487302976)))]; + tensor hidden_states_187_cast = mul(x = var_8328_cast, y = var_8330_to_fp16)[name = tensor("hidden_states_187_cast")]; + tensor var_8337 = const()[name = tensor("op_8337"), val = tensor([1, 1])]; + tensor var_8339 = const()[name = tensor("op_8339"), val = tensor([1, 1])]; + tensor q_33_pad_type_0 = const()[name = tensor("q_33_pad_type_0"), val = tensor("custom")]; + tensor q_33_pad_0 = const()[name = tensor("q_33_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_1_attentions_1_transformer_blocks_0_attn1_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(487305600))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(488534464))), name = tensor("up_blocks_1_attentions_1_transformer_blocks_0_attn1_to_q_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor q_33_cast = conv(dilations = var_8339, groups = var_7228, pad = q_33_pad_0, pad_type = q_33_pad_type_0, strides = var_8337, weight = up_blocks_1_attentions_1_transformer_blocks_0_attn1_to_q_weight_to_fp16_palettized, x = hidden_states_187_cast)[name = tensor("q_33_cast")]; + tensor var_8343 = const()[name = tensor("op_8343"), val = tensor([1, 1])]; + tensor var_8345 = const()[name = tensor("op_8345"), val = tensor([1, 1])]; + tensor k_65_pad_type_0 = const()[name = tensor("k_65_pad_type_0"), val = tensor("custom")]; + tensor k_65_pad_0 = const()[name = tensor("k_65_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_1_attentions_1_transformer_blocks_0_attn1_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(488534656))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(489763520))), name = tensor("up_blocks_1_attentions_1_transformer_blocks_0_attn1_to_k_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor k_65_cast = conv(dilations = var_8345, groups = var_7228, pad = k_65_pad_0, pad_type = k_65_pad_type_0, strides = var_8343, weight = up_blocks_1_attentions_1_transformer_blocks_0_attn1_to_k_weight_to_fp16_palettized, x = hidden_states_187_cast)[name = tensor("k_65_cast")]; + tensor var_8349 = const()[name = tensor("op_8349"), val = tensor([1, 1])]; + tensor var_8351 = const()[name = tensor("op_8351"), val = tensor([1, 1])]; + tensor v_33_pad_type_0 = const()[name = tensor("v_33_pad_type_0"), val = tensor("custom")]; + tensor v_33_pad_0 = const()[name = tensor("v_33_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_1_attentions_1_transformer_blocks_0_attn1_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(489763712))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(490992576))), name = tensor("up_blocks_1_attentions_1_transformer_blocks_0_attn1_to_v_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor v_33_cast = conv(dilations = var_8351, groups = var_7228, pad = v_33_pad_0, pad_type = v_33_pad_type_0, strides = var_8349, weight = up_blocks_1_attentions_1_transformer_blocks_0_attn1_to_v_weight_to_fp16_palettized, x = hidden_states_187_cast)[name = tensor("v_33_cast")]; + tensor var_8355_begin_0 = const()[name = tensor("op_8355_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8355_end_0 = const()[name = tensor("op_8355_end_0"), val = tensor([2, 64, 1, 256])]; + tensor var_8355_end_mask_0 = const()[name = tensor("op_8355_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8355_cast = slice_by_index(begin = var_8355_begin_0, end = var_8355_end_0, end_mask = var_8355_end_mask_0, x = q_33_cast)[name = tensor("op_8355_cast")]; + tensor var_8359_begin_0 = const()[name = tensor("op_8359_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_8359_end_0 = const()[name = tensor("op_8359_end_0"), val = tensor([2, 128, 1, 256])]; + tensor var_8359_end_mask_0 = const()[name = tensor("op_8359_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8359_cast = slice_by_index(begin = var_8359_begin_0, end = var_8359_end_0, end_mask = var_8359_end_mask_0, x = q_33_cast)[name = tensor("op_8359_cast")]; + tensor var_8363_begin_0 = const()[name = tensor("op_8363_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_8363_end_0 = const()[name = tensor("op_8363_end_0"), val = tensor([2, 192, 1, 256])]; + tensor var_8363_end_mask_0 = const()[name = tensor("op_8363_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8363_cast = slice_by_index(begin = var_8363_begin_0, end = var_8363_end_0, end_mask = var_8363_end_mask_0, x = q_33_cast)[name = tensor("op_8363_cast")]; + tensor var_8367_begin_0 = const()[name = tensor("op_8367_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_8367_end_0 = const()[name = tensor("op_8367_end_0"), val = tensor([2, 256, 1, 256])]; + tensor var_8367_end_mask_0 = const()[name = tensor("op_8367_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8367_cast = slice_by_index(begin = var_8367_begin_0, end = var_8367_end_0, end_mask = var_8367_end_mask_0, x = q_33_cast)[name = tensor("op_8367_cast")]; + tensor var_8371_begin_0 = const()[name = tensor("op_8371_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_8371_end_0 = const()[name = tensor("op_8371_end_0"), val = tensor([2, 320, 1, 256])]; + tensor var_8371_end_mask_0 = const()[name = tensor("op_8371_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8371_cast = slice_by_index(begin = var_8371_begin_0, end = var_8371_end_0, end_mask = var_8371_end_mask_0, x = q_33_cast)[name = tensor("op_8371_cast")]; + tensor var_8375_begin_0 = const()[name = tensor("op_8375_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_8375_end_0 = const()[name = tensor("op_8375_end_0"), val = tensor([2, 384, 1, 256])]; + tensor var_8375_end_mask_0 = const()[name = tensor("op_8375_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8375_cast = slice_by_index(begin = var_8375_begin_0, end = var_8375_end_0, end_mask = var_8375_end_mask_0, x = q_33_cast)[name = tensor("op_8375_cast")]; + tensor var_8379_begin_0 = const()[name = tensor("op_8379_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_8379_end_0 = const()[name = tensor("op_8379_end_0"), val = tensor([2, 448, 1, 256])]; + tensor var_8379_end_mask_0 = const()[name = tensor("op_8379_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8379_cast = slice_by_index(begin = var_8379_begin_0, end = var_8379_end_0, end_mask = var_8379_end_mask_0, x = q_33_cast)[name = tensor("op_8379_cast")]; + tensor var_8383_begin_0 = const()[name = tensor("op_8383_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_8383_end_0 = const()[name = tensor("op_8383_end_0"), val = tensor([2, 512, 1, 256])]; + tensor var_8383_end_mask_0 = const()[name = tensor("op_8383_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8383_cast = slice_by_index(begin = var_8383_begin_0, end = var_8383_end_0, end_mask = var_8383_end_mask_0, x = q_33_cast)[name = tensor("op_8383_cast")]; + tensor var_8387_begin_0 = const()[name = tensor("op_8387_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_8387_end_0 = const()[name = tensor("op_8387_end_0"), val = tensor([2, 576, 1, 256])]; + tensor var_8387_end_mask_0 = const()[name = tensor("op_8387_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8387_cast = slice_by_index(begin = var_8387_begin_0, end = var_8387_end_0, end_mask = var_8387_end_mask_0, x = q_33_cast)[name = tensor("op_8387_cast")]; + tensor var_8391_begin_0 = const()[name = tensor("op_8391_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_8391_end_0 = const()[name = tensor("op_8391_end_0"), val = tensor([2, 640, 1, 256])]; + tensor var_8391_end_mask_0 = const()[name = tensor("op_8391_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8391_cast = slice_by_index(begin = var_8391_begin_0, end = var_8391_end_0, end_mask = var_8391_end_mask_0, x = q_33_cast)[name = tensor("op_8391_cast")]; + tensor var_8395_begin_0 = const()[name = tensor("op_8395_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_8395_end_0 = const()[name = tensor("op_8395_end_0"), val = tensor([2, 704, 1, 256])]; + tensor var_8395_end_mask_0 = const()[name = tensor("op_8395_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8395_cast = slice_by_index(begin = var_8395_begin_0, end = var_8395_end_0, end_mask = var_8395_end_mask_0, x = q_33_cast)[name = tensor("op_8395_cast")]; + tensor var_8399_begin_0 = const()[name = tensor("op_8399_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_8399_end_0 = const()[name = tensor("op_8399_end_0"), val = tensor([2, 768, 1, 256])]; + tensor var_8399_end_mask_0 = const()[name = tensor("op_8399_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8399_cast = slice_by_index(begin = var_8399_begin_0, end = var_8399_end_0, end_mask = var_8399_end_mask_0, x = q_33_cast)[name = tensor("op_8399_cast")]; + tensor var_8403_begin_0 = const()[name = tensor("op_8403_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_8403_end_0 = const()[name = tensor("op_8403_end_0"), val = tensor([2, 832, 1, 256])]; + tensor var_8403_end_mask_0 = const()[name = tensor("op_8403_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8403_cast = slice_by_index(begin = var_8403_begin_0, end = var_8403_end_0, end_mask = var_8403_end_mask_0, x = q_33_cast)[name = tensor("op_8403_cast")]; + tensor var_8407_begin_0 = const()[name = tensor("op_8407_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_8407_end_0 = const()[name = tensor("op_8407_end_0"), val = tensor([2, 896, 1, 256])]; + tensor var_8407_end_mask_0 = const()[name = tensor("op_8407_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8407_cast = slice_by_index(begin = var_8407_begin_0, end = var_8407_end_0, end_mask = var_8407_end_mask_0, x = q_33_cast)[name = tensor("op_8407_cast")]; + tensor var_8411_begin_0 = const()[name = tensor("op_8411_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_8411_end_0 = const()[name = tensor("op_8411_end_0"), val = tensor([2, 960, 1, 256])]; + tensor var_8411_end_mask_0 = const()[name = tensor("op_8411_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8411_cast = slice_by_index(begin = var_8411_begin_0, end = var_8411_end_0, end_mask = var_8411_end_mask_0, x = q_33_cast)[name = tensor("op_8411_cast")]; + tensor var_8415_begin_0 = const()[name = tensor("op_8415_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_8415_end_0 = const()[name = tensor("op_8415_end_0"), val = tensor([2, 1024, 1, 256])]; + tensor var_8415_end_mask_0 = const()[name = tensor("op_8415_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8415_cast = slice_by_index(begin = var_8415_begin_0, end = var_8415_end_0, end_mask = var_8415_end_mask_0, x = q_33_cast)[name = tensor("op_8415_cast")]; + tensor var_8419_begin_0 = const()[name = tensor("op_8419_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_8419_end_0 = const()[name = tensor("op_8419_end_0"), val = tensor([2, 1088, 1, 256])]; + tensor var_8419_end_mask_0 = const()[name = tensor("op_8419_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8419_cast = slice_by_index(begin = var_8419_begin_0, end = var_8419_end_0, end_mask = var_8419_end_mask_0, x = q_33_cast)[name = tensor("op_8419_cast")]; + tensor var_8423_begin_0 = const()[name = tensor("op_8423_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_8423_end_0 = const()[name = tensor("op_8423_end_0"), val = tensor([2, 1152, 1, 256])]; + tensor var_8423_end_mask_0 = const()[name = tensor("op_8423_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8423_cast = slice_by_index(begin = var_8423_begin_0, end = var_8423_end_0, end_mask = var_8423_end_mask_0, x = q_33_cast)[name = tensor("op_8423_cast")]; + tensor var_8427_begin_0 = const()[name = tensor("op_8427_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_8427_end_0 = const()[name = tensor("op_8427_end_0"), val = tensor([2, 1216, 1, 256])]; + tensor var_8427_end_mask_0 = const()[name = tensor("op_8427_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8427_cast = slice_by_index(begin = var_8427_begin_0, end = var_8427_end_0, end_mask = var_8427_end_mask_0, x = q_33_cast)[name = tensor("op_8427_cast")]; + tensor var_8431_begin_0 = const()[name = tensor("op_8431_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_8431_end_0 = const()[name = tensor("op_8431_end_0"), val = tensor([2, 1280, 1, 256])]; + tensor var_8431_end_mask_0 = const()[name = tensor("op_8431_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8431_cast = slice_by_index(begin = var_8431_begin_0, end = var_8431_end_0, end_mask = var_8431_end_mask_0, x = q_33_cast)[name = tensor("op_8431_cast")]; + tensor k_67_perm_0 = const()[name = tensor("k_67_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_8438_begin_0 = const()[name = tensor("op_8438_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8438_end_0 = const()[name = tensor("op_8438_end_0"), val = tensor([2, 256, 1, 64])]; + tensor var_8438_end_mask_0 = const()[name = tensor("op_8438_end_mask_0"), val = tensor([true, true, true, false])]; + tensor transpose_15 = transpose(perm = k_67_perm_0, x = k_65_cast)[name = tensor("transpose_15")]; + tensor var_8438_cast = slice_by_index(begin = var_8438_begin_0, end = var_8438_end_0, end_mask = var_8438_end_mask_0, x = transpose_15)[name = tensor("op_8438_cast")]; + tensor var_8442_begin_0 = const()[name = tensor("op_8442_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_8442_end_0 = const()[name = tensor("op_8442_end_0"), val = tensor([2, 256, 1, 128])]; + tensor var_8442_end_mask_0 = const()[name = tensor("op_8442_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8442_cast = slice_by_index(begin = var_8442_begin_0, end = var_8442_end_0, end_mask = var_8442_end_mask_0, x = transpose_15)[name = tensor("op_8442_cast")]; + tensor var_8446_begin_0 = const()[name = tensor("op_8446_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_8446_end_0 = const()[name = tensor("op_8446_end_0"), val = tensor([2, 256, 1, 192])]; + tensor var_8446_end_mask_0 = const()[name = tensor("op_8446_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8446_cast = slice_by_index(begin = var_8446_begin_0, end = var_8446_end_0, end_mask = var_8446_end_mask_0, x = transpose_15)[name = tensor("op_8446_cast")]; + tensor var_8450_begin_0 = const()[name = tensor("op_8450_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_8450_end_0 = const()[name = tensor("op_8450_end_0"), val = tensor([2, 256, 1, 256])]; + tensor var_8450_end_mask_0 = const()[name = tensor("op_8450_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8450_cast = slice_by_index(begin = var_8450_begin_0, end = var_8450_end_0, end_mask = var_8450_end_mask_0, x = transpose_15)[name = tensor("op_8450_cast")]; + tensor var_8454_begin_0 = const()[name = tensor("op_8454_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_8454_end_0 = const()[name = tensor("op_8454_end_0"), val = tensor([2, 256, 1, 320])]; + tensor var_8454_end_mask_0 = const()[name = tensor("op_8454_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8454_cast = slice_by_index(begin = var_8454_begin_0, end = var_8454_end_0, end_mask = var_8454_end_mask_0, x = transpose_15)[name = tensor("op_8454_cast")]; + tensor var_8458_begin_0 = const()[name = tensor("op_8458_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_8458_end_0 = const()[name = tensor("op_8458_end_0"), val = tensor([2, 256, 1, 384])]; + tensor var_8458_end_mask_0 = const()[name = tensor("op_8458_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8458_cast = slice_by_index(begin = var_8458_begin_0, end = var_8458_end_0, end_mask = var_8458_end_mask_0, x = transpose_15)[name = tensor("op_8458_cast")]; + tensor var_8462_begin_0 = const()[name = tensor("op_8462_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_8462_end_0 = const()[name = tensor("op_8462_end_0"), val = tensor([2, 256, 1, 448])]; + tensor var_8462_end_mask_0 = const()[name = tensor("op_8462_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8462_cast = slice_by_index(begin = var_8462_begin_0, end = var_8462_end_0, end_mask = var_8462_end_mask_0, x = transpose_15)[name = tensor("op_8462_cast")]; + tensor var_8466_begin_0 = const()[name = tensor("op_8466_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_8466_end_0 = const()[name = tensor("op_8466_end_0"), val = tensor([2, 256, 1, 512])]; + tensor var_8466_end_mask_0 = const()[name = tensor("op_8466_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8466_cast = slice_by_index(begin = var_8466_begin_0, end = var_8466_end_0, end_mask = var_8466_end_mask_0, x = transpose_15)[name = tensor("op_8466_cast")]; + tensor var_8470_begin_0 = const()[name = tensor("op_8470_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_8470_end_0 = const()[name = tensor("op_8470_end_0"), val = tensor([2, 256, 1, 576])]; + tensor var_8470_end_mask_0 = const()[name = tensor("op_8470_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8470_cast = slice_by_index(begin = var_8470_begin_0, end = var_8470_end_0, end_mask = var_8470_end_mask_0, x = transpose_15)[name = tensor("op_8470_cast")]; + tensor var_8474_begin_0 = const()[name = tensor("op_8474_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_8474_end_0 = const()[name = tensor("op_8474_end_0"), val = tensor([2, 256, 1, 640])]; + tensor var_8474_end_mask_0 = const()[name = tensor("op_8474_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8474_cast = slice_by_index(begin = var_8474_begin_0, end = var_8474_end_0, end_mask = var_8474_end_mask_0, x = transpose_15)[name = tensor("op_8474_cast")]; + tensor var_8478_begin_0 = const()[name = tensor("op_8478_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_8478_end_0 = const()[name = tensor("op_8478_end_0"), val = tensor([2, 256, 1, 704])]; + tensor var_8478_end_mask_0 = const()[name = tensor("op_8478_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8478_cast = slice_by_index(begin = var_8478_begin_0, end = var_8478_end_0, end_mask = var_8478_end_mask_0, x = transpose_15)[name = tensor("op_8478_cast")]; + tensor var_8482_begin_0 = const()[name = tensor("op_8482_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_8482_end_0 = const()[name = tensor("op_8482_end_0"), val = tensor([2, 256, 1, 768])]; + tensor var_8482_end_mask_0 = const()[name = tensor("op_8482_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8482_cast = slice_by_index(begin = var_8482_begin_0, end = var_8482_end_0, end_mask = var_8482_end_mask_0, x = transpose_15)[name = tensor("op_8482_cast")]; + tensor var_8486_begin_0 = const()[name = tensor("op_8486_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_8486_end_0 = const()[name = tensor("op_8486_end_0"), val = tensor([2, 256, 1, 832])]; + tensor var_8486_end_mask_0 = const()[name = tensor("op_8486_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8486_cast = slice_by_index(begin = var_8486_begin_0, end = var_8486_end_0, end_mask = var_8486_end_mask_0, x = transpose_15)[name = tensor("op_8486_cast")]; + tensor var_8490_begin_0 = const()[name = tensor("op_8490_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_8490_end_0 = const()[name = tensor("op_8490_end_0"), val = tensor([2, 256, 1, 896])]; + tensor var_8490_end_mask_0 = const()[name = tensor("op_8490_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8490_cast = slice_by_index(begin = var_8490_begin_0, end = var_8490_end_0, end_mask = var_8490_end_mask_0, x = transpose_15)[name = tensor("op_8490_cast")]; + tensor var_8494_begin_0 = const()[name = tensor("op_8494_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_8494_end_0 = const()[name = tensor("op_8494_end_0"), val = tensor([2, 256, 1, 960])]; + tensor var_8494_end_mask_0 = const()[name = tensor("op_8494_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8494_cast = slice_by_index(begin = var_8494_begin_0, end = var_8494_end_0, end_mask = var_8494_end_mask_0, x = transpose_15)[name = tensor("op_8494_cast")]; + tensor var_8498_begin_0 = const()[name = tensor("op_8498_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_8498_end_0 = const()[name = tensor("op_8498_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_8498_end_mask_0 = const()[name = tensor("op_8498_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8498_cast = slice_by_index(begin = var_8498_begin_0, end = var_8498_end_0, end_mask = var_8498_end_mask_0, x = transpose_15)[name = tensor("op_8498_cast")]; + tensor var_8502_begin_0 = const()[name = tensor("op_8502_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_8502_end_0 = const()[name = tensor("op_8502_end_0"), val = tensor([2, 256, 1, 1088])]; + tensor var_8502_end_mask_0 = const()[name = tensor("op_8502_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8502_cast = slice_by_index(begin = var_8502_begin_0, end = var_8502_end_0, end_mask = var_8502_end_mask_0, x = transpose_15)[name = tensor("op_8502_cast")]; + tensor var_8506_begin_0 = const()[name = tensor("op_8506_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_8506_end_0 = const()[name = tensor("op_8506_end_0"), val = tensor([2, 256, 1, 1152])]; + tensor var_8506_end_mask_0 = const()[name = tensor("op_8506_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8506_cast = slice_by_index(begin = var_8506_begin_0, end = var_8506_end_0, end_mask = var_8506_end_mask_0, x = transpose_15)[name = tensor("op_8506_cast")]; + tensor var_8510_begin_0 = const()[name = tensor("op_8510_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_8510_end_0 = const()[name = tensor("op_8510_end_0"), val = tensor([2, 256, 1, 1216])]; + tensor var_8510_end_mask_0 = const()[name = tensor("op_8510_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8510_cast = slice_by_index(begin = var_8510_begin_0, end = var_8510_end_0, end_mask = var_8510_end_mask_0, x = transpose_15)[name = tensor("op_8510_cast")]; + tensor var_8514_begin_0 = const()[name = tensor("op_8514_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_8514_end_0 = const()[name = tensor("op_8514_end_0"), val = tensor([2, 256, 1, 1280])]; + tensor var_8514_end_mask_0 = const()[name = tensor("op_8514_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8514_cast = slice_by_index(begin = var_8514_begin_0, end = var_8514_end_0, end_mask = var_8514_end_mask_0, x = transpose_15)[name = tensor("op_8514_cast")]; + tensor var_8516_begin_0 = const()[name = tensor("op_8516_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8516_end_0 = const()[name = tensor("op_8516_end_0"), val = tensor([2, 64, 1, 256])]; + tensor var_8516_end_mask_0 = const()[name = tensor("op_8516_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8516_cast = slice_by_index(begin = var_8516_begin_0, end = var_8516_end_0, end_mask = var_8516_end_mask_0, x = v_33_cast)[name = tensor("op_8516_cast")]; + tensor var_8520_begin_0 = const()[name = tensor("op_8520_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_8520_end_0 = const()[name = tensor("op_8520_end_0"), val = tensor([2, 128, 1, 256])]; + tensor var_8520_end_mask_0 = const()[name = tensor("op_8520_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8520_cast = slice_by_index(begin = var_8520_begin_0, end = var_8520_end_0, end_mask = var_8520_end_mask_0, x = v_33_cast)[name = tensor("op_8520_cast")]; + tensor var_8524_begin_0 = const()[name = tensor("op_8524_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_8524_end_0 = const()[name = tensor("op_8524_end_0"), val = tensor([2, 192, 1, 256])]; + tensor var_8524_end_mask_0 = const()[name = tensor("op_8524_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8524_cast = slice_by_index(begin = var_8524_begin_0, end = var_8524_end_0, end_mask = var_8524_end_mask_0, x = v_33_cast)[name = tensor("op_8524_cast")]; + tensor var_8528_begin_0 = const()[name = tensor("op_8528_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_8528_end_0 = const()[name = tensor("op_8528_end_0"), val = tensor([2, 256, 1, 256])]; + tensor var_8528_end_mask_0 = const()[name = tensor("op_8528_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8528_cast = slice_by_index(begin = var_8528_begin_0, end = var_8528_end_0, end_mask = var_8528_end_mask_0, x = v_33_cast)[name = tensor("op_8528_cast")]; + tensor var_8532_begin_0 = const()[name = tensor("op_8532_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_8532_end_0 = const()[name = tensor("op_8532_end_0"), val = tensor([2, 320, 1, 256])]; + tensor var_8532_end_mask_0 = const()[name = tensor("op_8532_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8532_cast = slice_by_index(begin = var_8532_begin_0, end = var_8532_end_0, end_mask = var_8532_end_mask_0, x = v_33_cast)[name = tensor("op_8532_cast")]; + tensor var_8536_begin_0 = const()[name = tensor("op_8536_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_8536_end_0 = const()[name = tensor("op_8536_end_0"), val = tensor([2, 384, 1, 256])]; + tensor var_8536_end_mask_0 = const()[name = tensor("op_8536_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8536_cast = slice_by_index(begin = var_8536_begin_0, end = var_8536_end_0, end_mask = var_8536_end_mask_0, x = v_33_cast)[name = tensor("op_8536_cast")]; + tensor var_8540_begin_0 = const()[name = tensor("op_8540_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_8540_end_0 = const()[name = tensor("op_8540_end_0"), val = tensor([2, 448, 1, 256])]; + tensor var_8540_end_mask_0 = const()[name = tensor("op_8540_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8540_cast = slice_by_index(begin = var_8540_begin_0, end = var_8540_end_0, end_mask = var_8540_end_mask_0, x = v_33_cast)[name = tensor("op_8540_cast")]; + tensor var_8544_begin_0 = const()[name = tensor("op_8544_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_8544_end_0 = const()[name = tensor("op_8544_end_0"), val = tensor([2, 512, 1, 256])]; + tensor var_8544_end_mask_0 = const()[name = tensor("op_8544_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8544_cast = slice_by_index(begin = var_8544_begin_0, end = var_8544_end_0, end_mask = var_8544_end_mask_0, x = v_33_cast)[name = tensor("op_8544_cast")]; + tensor var_8548_begin_0 = const()[name = tensor("op_8548_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_8548_end_0 = const()[name = tensor("op_8548_end_0"), val = tensor([2, 576, 1, 256])]; + tensor var_8548_end_mask_0 = const()[name = tensor("op_8548_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8548_cast = slice_by_index(begin = var_8548_begin_0, end = var_8548_end_0, end_mask = var_8548_end_mask_0, x = v_33_cast)[name = tensor("op_8548_cast")]; + tensor var_8552_begin_0 = const()[name = tensor("op_8552_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_8552_end_0 = const()[name = tensor("op_8552_end_0"), val = tensor([2, 640, 1, 256])]; + tensor var_8552_end_mask_0 = const()[name = tensor("op_8552_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8552_cast = slice_by_index(begin = var_8552_begin_0, end = var_8552_end_0, end_mask = var_8552_end_mask_0, x = v_33_cast)[name = tensor("op_8552_cast")]; + tensor var_8556_begin_0 = const()[name = tensor("op_8556_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_8556_end_0 = const()[name = tensor("op_8556_end_0"), val = tensor([2, 704, 1, 256])]; + tensor var_8556_end_mask_0 = const()[name = tensor("op_8556_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8556_cast = slice_by_index(begin = var_8556_begin_0, end = var_8556_end_0, end_mask = var_8556_end_mask_0, x = v_33_cast)[name = tensor("op_8556_cast")]; + tensor var_8560_begin_0 = const()[name = tensor("op_8560_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_8560_end_0 = const()[name = tensor("op_8560_end_0"), val = tensor([2, 768, 1, 256])]; + tensor var_8560_end_mask_0 = const()[name = tensor("op_8560_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8560_cast = slice_by_index(begin = var_8560_begin_0, end = var_8560_end_0, end_mask = var_8560_end_mask_0, x = v_33_cast)[name = tensor("op_8560_cast")]; + tensor var_8564_begin_0 = const()[name = tensor("op_8564_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_8564_end_0 = const()[name = tensor("op_8564_end_0"), val = tensor([2, 832, 1, 256])]; + tensor var_8564_end_mask_0 = const()[name = tensor("op_8564_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8564_cast = slice_by_index(begin = var_8564_begin_0, end = var_8564_end_0, end_mask = var_8564_end_mask_0, x = v_33_cast)[name = tensor("op_8564_cast")]; + tensor var_8568_begin_0 = const()[name = tensor("op_8568_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_8568_end_0 = const()[name = tensor("op_8568_end_0"), val = tensor([2, 896, 1, 256])]; + tensor var_8568_end_mask_0 = const()[name = tensor("op_8568_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8568_cast = slice_by_index(begin = var_8568_begin_0, end = var_8568_end_0, end_mask = var_8568_end_mask_0, x = v_33_cast)[name = tensor("op_8568_cast")]; + tensor var_8572_begin_0 = const()[name = tensor("op_8572_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_8572_end_0 = const()[name = tensor("op_8572_end_0"), val = tensor([2, 960, 1, 256])]; + tensor var_8572_end_mask_0 = const()[name = tensor("op_8572_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8572_cast = slice_by_index(begin = var_8572_begin_0, end = var_8572_end_0, end_mask = var_8572_end_mask_0, x = v_33_cast)[name = tensor("op_8572_cast")]; + tensor var_8576_begin_0 = const()[name = tensor("op_8576_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_8576_end_0 = const()[name = tensor("op_8576_end_0"), val = tensor([2, 1024, 1, 256])]; + tensor var_8576_end_mask_0 = const()[name = tensor("op_8576_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8576_cast = slice_by_index(begin = var_8576_begin_0, end = var_8576_end_0, end_mask = var_8576_end_mask_0, x = v_33_cast)[name = tensor("op_8576_cast")]; + tensor var_8580_begin_0 = const()[name = tensor("op_8580_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_8580_end_0 = const()[name = tensor("op_8580_end_0"), val = tensor([2, 1088, 1, 256])]; + tensor var_8580_end_mask_0 = const()[name = tensor("op_8580_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8580_cast = slice_by_index(begin = var_8580_begin_0, end = var_8580_end_0, end_mask = var_8580_end_mask_0, x = v_33_cast)[name = tensor("op_8580_cast")]; + tensor var_8584_begin_0 = const()[name = tensor("op_8584_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_8584_end_0 = const()[name = tensor("op_8584_end_0"), val = tensor([2, 1152, 1, 256])]; + tensor var_8584_end_mask_0 = const()[name = tensor("op_8584_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8584_cast = slice_by_index(begin = var_8584_begin_0, end = var_8584_end_0, end_mask = var_8584_end_mask_0, x = v_33_cast)[name = tensor("op_8584_cast")]; + tensor var_8588_begin_0 = const()[name = tensor("op_8588_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_8588_end_0 = const()[name = tensor("op_8588_end_0"), val = tensor([2, 1216, 1, 256])]; + tensor var_8588_end_mask_0 = const()[name = tensor("op_8588_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8588_cast = slice_by_index(begin = var_8588_begin_0, end = var_8588_end_0, end_mask = var_8588_end_mask_0, x = v_33_cast)[name = tensor("op_8588_cast")]; + tensor var_8592_begin_0 = const()[name = tensor("op_8592_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_8592_end_0 = const()[name = tensor("op_8592_end_0"), val = tensor([2, 1280, 1, 256])]; + tensor var_8592_end_mask_0 = const()[name = tensor("op_8592_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8592_cast = slice_by_index(begin = var_8592_begin_0, end = var_8592_end_0, end_mask = var_8592_end_mask_0, x = v_33_cast)[name = tensor("op_8592_cast")]; + tensor var_8596_equation_0 = const()[name = tensor("op_8596_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8596_cast = einsum(equation = var_8596_equation_0, values = (var_8438_cast, var_8355_cast))[name = tensor("op_8596_cast")]; + tensor var_8597_to_fp16 = const()[name = tensor("op_8597_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_321_cast = mul(x = var_8596_cast, y = var_8597_to_fp16)[name = tensor("aw_321_cast")]; + tensor var_8600_equation_0 = const()[name = tensor("op_8600_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8600_cast = einsum(equation = var_8600_equation_0, values = (var_8442_cast, var_8359_cast))[name = tensor("op_8600_cast")]; + tensor var_8601_to_fp16 = const()[name = tensor("op_8601_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_323_cast = mul(x = var_8600_cast, y = var_8601_to_fp16)[name = tensor("aw_323_cast")]; + tensor var_8604_equation_0 = const()[name = tensor("op_8604_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8604_cast = einsum(equation = var_8604_equation_0, values = (var_8446_cast, var_8363_cast))[name = tensor("op_8604_cast")]; + tensor var_8605_to_fp16 = const()[name = tensor("op_8605_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_325_cast = mul(x = var_8604_cast, y = var_8605_to_fp16)[name = tensor("aw_325_cast")]; + tensor var_8608_equation_0 = const()[name = tensor("op_8608_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8608_cast = einsum(equation = var_8608_equation_0, values = (var_8450_cast, var_8367_cast))[name = tensor("op_8608_cast")]; + tensor var_8609_to_fp16 = const()[name = tensor("op_8609_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_327_cast = mul(x = var_8608_cast, y = var_8609_to_fp16)[name = tensor("aw_327_cast")]; + tensor var_8612_equation_0 = const()[name = tensor("op_8612_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8612_cast = einsum(equation = var_8612_equation_0, values = (var_8454_cast, var_8371_cast))[name = tensor("op_8612_cast")]; + tensor var_8613_to_fp16 = const()[name = tensor("op_8613_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_329_cast = mul(x = var_8612_cast, y = var_8613_to_fp16)[name = tensor("aw_329_cast")]; + tensor var_8616_equation_0 = const()[name = tensor("op_8616_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8616_cast = einsum(equation = var_8616_equation_0, values = (var_8458_cast, var_8375_cast))[name = tensor("op_8616_cast")]; + tensor var_8617_to_fp16 = const()[name = tensor("op_8617_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_331_cast = mul(x = var_8616_cast, y = var_8617_to_fp16)[name = tensor("aw_331_cast")]; + tensor var_8620_equation_0 = const()[name = tensor("op_8620_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8620_cast = einsum(equation = var_8620_equation_0, values = (var_8462_cast, var_8379_cast))[name = tensor("op_8620_cast")]; + tensor var_8621_to_fp16 = const()[name = tensor("op_8621_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_333_cast = mul(x = var_8620_cast, y = var_8621_to_fp16)[name = tensor("aw_333_cast")]; + tensor var_8624_equation_0 = const()[name = tensor("op_8624_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8624_cast = einsum(equation = var_8624_equation_0, values = (var_8466_cast, var_8383_cast))[name = tensor("op_8624_cast")]; + tensor var_8625_to_fp16 = const()[name = tensor("op_8625_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_335_cast = mul(x = var_8624_cast, y = var_8625_to_fp16)[name = tensor("aw_335_cast")]; + tensor var_8628_equation_0 = const()[name = tensor("op_8628_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8628_cast = einsum(equation = var_8628_equation_0, values = (var_8470_cast, var_8387_cast))[name = tensor("op_8628_cast")]; + tensor var_8629_to_fp16 = const()[name = tensor("op_8629_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_337_cast = mul(x = var_8628_cast, y = var_8629_to_fp16)[name = tensor("aw_337_cast")]; + tensor var_8632_equation_0 = const()[name = tensor("op_8632_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8632_cast = einsum(equation = var_8632_equation_0, values = (var_8474_cast, var_8391_cast))[name = tensor("op_8632_cast")]; + tensor var_8633_to_fp16 = const()[name = tensor("op_8633_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_339_cast = mul(x = var_8632_cast, y = var_8633_to_fp16)[name = tensor("aw_339_cast")]; + tensor var_8636_equation_0 = const()[name = tensor("op_8636_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8636_cast = einsum(equation = var_8636_equation_0, values = (var_8478_cast, var_8395_cast))[name = tensor("op_8636_cast")]; + tensor var_8637_to_fp16 = const()[name = tensor("op_8637_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_341_cast = mul(x = var_8636_cast, y = var_8637_to_fp16)[name = tensor("aw_341_cast")]; + tensor var_8640_equation_0 = const()[name = tensor("op_8640_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8640_cast = einsum(equation = var_8640_equation_0, values = (var_8482_cast, var_8399_cast))[name = tensor("op_8640_cast")]; + tensor var_8641_to_fp16 = const()[name = tensor("op_8641_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_343_cast = mul(x = var_8640_cast, y = var_8641_to_fp16)[name = tensor("aw_343_cast")]; + tensor var_8644_equation_0 = const()[name = tensor("op_8644_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8644_cast = einsum(equation = var_8644_equation_0, values = (var_8486_cast, var_8403_cast))[name = tensor("op_8644_cast")]; + tensor var_8645_to_fp16 = const()[name = tensor("op_8645_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_345_cast = mul(x = var_8644_cast, y = var_8645_to_fp16)[name = tensor("aw_345_cast")]; + tensor var_8648_equation_0 = const()[name = tensor("op_8648_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8648_cast = einsum(equation = var_8648_equation_0, values = (var_8490_cast, var_8407_cast))[name = tensor("op_8648_cast")]; + tensor var_8649_to_fp16 = const()[name = tensor("op_8649_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_347_cast = mul(x = var_8648_cast, y = var_8649_to_fp16)[name = tensor("aw_347_cast")]; + tensor var_8652_equation_0 = const()[name = tensor("op_8652_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8652_cast = einsum(equation = var_8652_equation_0, values = (var_8494_cast, var_8411_cast))[name = tensor("op_8652_cast")]; + tensor var_8653_to_fp16 = const()[name = tensor("op_8653_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_349_cast = mul(x = var_8652_cast, y = var_8653_to_fp16)[name = tensor("aw_349_cast")]; + tensor var_8656_equation_0 = const()[name = tensor("op_8656_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8656_cast = einsum(equation = var_8656_equation_0, values = (var_8498_cast, var_8415_cast))[name = tensor("op_8656_cast")]; + tensor var_8657_to_fp16 = const()[name = tensor("op_8657_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_351_cast = mul(x = var_8656_cast, y = var_8657_to_fp16)[name = tensor("aw_351_cast")]; + tensor var_8660_equation_0 = const()[name = tensor("op_8660_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8660_cast = einsum(equation = var_8660_equation_0, values = (var_8502_cast, var_8419_cast))[name = tensor("op_8660_cast")]; + tensor var_8661_to_fp16 = const()[name = tensor("op_8661_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_353_cast = mul(x = var_8660_cast, y = var_8661_to_fp16)[name = tensor("aw_353_cast")]; + tensor var_8664_equation_0 = const()[name = tensor("op_8664_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8664_cast = einsum(equation = var_8664_equation_0, values = (var_8506_cast, var_8423_cast))[name = tensor("op_8664_cast")]; + tensor var_8665_to_fp16 = const()[name = tensor("op_8665_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_355_cast = mul(x = var_8664_cast, y = var_8665_to_fp16)[name = tensor("aw_355_cast")]; + tensor var_8668_equation_0 = const()[name = tensor("op_8668_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8668_cast = einsum(equation = var_8668_equation_0, values = (var_8510_cast, var_8427_cast))[name = tensor("op_8668_cast")]; + tensor var_8669_to_fp16 = const()[name = tensor("op_8669_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_357_cast = mul(x = var_8668_cast, y = var_8669_to_fp16)[name = tensor("aw_357_cast")]; + tensor var_8672_equation_0 = const()[name = tensor("op_8672_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_8672_cast = einsum(equation = var_8672_equation_0, values = (var_8514_cast, var_8431_cast))[name = tensor("op_8672_cast")]; + tensor var_8673_to_fp16 = const()[name = tensor("op_8673_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_359_cast = mul(x = var_8672_cast, y = var_8673_to_fp16)[name = tensor("aw_359_cast")]; + tensor var_8675_cast = softmax(axis = var_7228, x = aw_321_cast)[name = tensor("op_8675_cast")]; + tensor var_8676_cast = softmax(axis = var_7228, x = aw_323_cast)[name = tensor("op_8676_cast")]; + tensor var_8677_cast = softmax(axis = var_7228, x = aw_325_cast)[name = tensor("op_8677_cast")]; + tensor var_8678_cast = softmax(axis = var_7228, x = aw_327_cast)[name = tensor("op_8678_cast")]; + tensor var_8679_cast = softmax(axis = var_7228, x = aw_329_cast)[name = tensor("op_8679_cast")]; + tensor var_8680_cast = softmax(axis = var_7228, x = aw_331_cast)[name = tensor("op_8680_cast")]; + tensor var_8681_cast = softmax(axis = var_7228, x = aw_333_cast)[name = tensor("op_8681_cast")]; + tensor var_8682_cast = softmax(axis = var_7228, x = aw_335_cast)[name = tensor("op_8682_cast")]; + tensor var_8683_cast = softmax(axis = var_7228, x = aw_337_cast)[name = tensor("op_8683_cast")]; + tensor var_8684_cast = softmax(axis = var_7228, x = aw_339_cast)[name = tensor("op_8684_cast")]; + tensor var_8685_cast = softmax(axis = var_7228, x = aw_341_cast)[name = tensor("op_8685_cast")]; + tensor var_8686_cast = softmax(axis = var_7228, x = aw_343_cast)[name = tensor("op_8686_cast")]; + tensor var_8687_cast = softmax(axis = var_7228, x = aw_345_cast)[name = tensor("op_8687_cast")]; + tensor var_8688_cast = softmax(axis = var_7228, x = aw_347_cast)[name = tensor("op_8688_cast")]; + tensor var_8689_cast = softmax(axis = var_7228, x = aw_349_cast)[name = tensor("op_8689_cast")]; + tensor var_8690_cast = softmax(axis = var_7228, x = aw_351_cast)[name = tensor("op_8690_cast")]; + tensor var_8691_cast = softmax(axis = var_7228, x = aw_353_cast)[name = tensor("op_8691_cast")]; + tensor var_8692_cast = softmax(axis = var_7228, x = aw_355_cast)[name = tensor("op_8692_cast")]; + tensor var_8693_cast = softmax(axis = var_7228, x = aw_357_cast)[name = tensor("op_8693_cast")]; + tensor var_8694_cast = softmax(axis = var_7228, x = aw_359_cast)[name = tensor("op_8694_cast")]; + tensor var_8696_equation_0 = const()[name = tensor("op_8696_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8696_cast = einsum(equation = var_8696_equation_0, values = (var_8516_cast, var_8675_cast))[name = tensor("op_8696_cast")]; + tensor var_8698_equation_0 = const()[name = tensor("op_8698_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8698_cast = einsum(equation = var_8698_equation_0, values = (var_8520_cast, var_8676_cast))[name = tensor("op_8698_cast")]; + tensor var_8700_equation_0 = const()[name = tensor("op_8700_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8700_cast = einsum(equation = var_8700_equation_0, values = (var_8524_cast, var_8677_cast))[name = tensor("op_8700_cast")]; + tensor var_8702_equation_0 = const()[name = tensor("op_8702_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8702_cast = einsum(equation = var_8702_equation_0, values = (var_8528_cast, var_8678_cast))[name = tensor("op_8702_cast")]; + tensor var_8704_equation_0 = const()[name = tensor("op_8704_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8704_cast = einsum(equation = var_8704_equation_0, values = (var_8532_cast, var_8679_cast))[name = tensor("op_8704_cast")]; + tensor var_8706_equation_0 = const()[name = tensor("op_8706_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8706_cast = einsum(equation = var_8706_equation_0, values = (var_8536_cast, var_8680_cast))[name = tensor("op_8706_cast")]; + tensor var_8708_equation_0 = const()[name = tensor("op_8708_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8708_cast = einsum(equation = var_8708_equation_0, values = (var_8540_cast, var_8681_cast))[name = tensor("op_8708_cast")]; + tensor var_8710_equation_0 = const()[name = tensor("op_8710_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8710_cast = einsum(equation = var_8710_equation_0, values = (var_8544_cast, var_8682_cast))[name = tensor("op_8710_cast")]; + tensor var_8712_equation_0 = const()[name = tensor("op_8712_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8712_cast = einsum(equation = var_8712_equation_0, values = (var_8548_cast, var_8683_cast))[name = tensor("op_8712_cast")]; + tensor var_8714_equation_0 = const()[name = tensor("op_8714_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8714_cast = einsum(equation = var_8714_equation_0, values = (var_8552_cast, var_8684_cast))[name = tensor("op_8714_cast")]; + tensor var_8716_equation_0 = const()[name = tensor("op_8716_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8716_cast = einsum(equation = var_8716_equation_0, values = (var_8556_cast, var_8685_cast))[name = tensor("op_8716_cast")]; + tensor var_8718_equation_0 = const()[name = tensor("op_8718_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8718_cast = einsum(equation = var_8718_equation_0, values = (var_8560_cast, var_8686_cast))[name = tensor("op_8718_cast")]; + tensor var_8720_equation_0 = const()[name = tensor("op_8720_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8720_cast = einsum(equation = var_8720_equation_0, values = (var_8564_cast, var_8687_cast))[name = tensor("op_8720_cast")]; + tensor var_8722_equation_0 = const()[name = tensor("op_8722_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8722_cast = einsum(equation = var_8722_equation_0, values = (var_8568_cast, var_8688_cast))[name = tensor("op_8722_cast")]; + tensor var_8724_equation_0 = const()[name = tensor("op_8724_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8724_cast = einsum(equation = var_8724_equation_0, values = (var_8572_cast, var_8689_cast))[name = tensor("op_8724_cast")]; + tensor var_8726_equation_0 = const()[name = tensor("op_8726_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8726_cast = einsum(equation = var_8726_equation_0, values = (var_8576_cast, var_8690_cast))[name = tensor("op_8726_cast")]; + tensor var_8728_equation_0 = const()[name = tensor("op_8728_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8728_cast = einsum(equation = var_8728_equation_0, values = (var_8580_cast, var_8691_cast))[name = tensor("op_8728_cast")]; + tensor var_8730_equation_0 = const()[name = tensor("op_8730_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8730_cast = einsum(equation = var_8730_equation_0, values = (var_8584_cast, var_8692_cast))[name = tensor("op_8730_cast")]; + tensor var_8732_equation_0 = const()[name = tensor("op_8732_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8732_cast = einsum(equation = var_8732_equation_0, values = (var_8588_cast, var_8693_cast))[name = tensor("op_8732_cast")]; + tensor var_8734_equation_0 = const()[name = tensor("op_8734_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8734_cast = einsum(equation = var_8734_equation_0, values = (var_8592_cast, var_8694_cast))[name = tensor("op_8734_cast")]; + tensor input_327_interleave_0 = const()[name = tensor("input_327_interleave_0"), val = tensor(false)]; + tensor input_327_cast = concat(axis = var_7228, interleave = input_327_interleave_0, values = (var_8696_cast, var_8698_cast, var_8700_cast, var_8702_cast, var_8704_cast, var_8706_cast, var_8708_cast, var_8710_cast, var_8712_cast, var_8714_cast, var_8716_cast, var_8718_cast, var_8720_cast, var_8722_cast, var_8724_cast, var_8726_cast, var_8728_cast, var_8730_cast, var_8732_cast, var_8734_cast))[name = tensor("input_327_cast")]; + tensor var_8740 = const()[name = tensor("op_8740"), val = tensor([1, 1])]; + tensor var_8742 = const()[name = tensor("op_8742"), val = tensor([1, 1])]; + tensor var_8744_pad_type_0 = const()[name = tensor("op_8744_pad_type_0"), val = tensor("custom")]; + tensor var_8744_pad_0 = const()[name = tensor("op_8744_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_1_attentions_1_transformer_blocks_0_attn1_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(490992768))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(492221632))), name = tensor("up_blocks_1_attentions_1_transformer_blocks_0_attn1_to_out_0_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor up_blocks_1_attentions_1_transformer_blocks_0_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("up_blocks_1_attentions_1_transformer_blocks_0_attn1_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(492221824)))]; + tensor var_8744_cast = conv(bias = up_blocks_1_attentions_1_transformer_blocks_0_attn1_to_out_0_bias_to_fp16, dilations = var_8742, groups = var_7228, pad = var_8744_pad_0, pad_type = var_8744_pad_type_0, strides = var_8740, weight = up_blocks_1_attentions_1_transformer_blocks_0_attn1_to_out_0_weight_to_fp16_palettized, x = input_327_cast)[name = tensor("op_8744_cast")]; + tensor inputs_51_cast = add(x = var_8744_cast, y = inputs_49_cast)[name = tensor("inputs_51_cast")]; + tensor var_8748 = const()[name = tensor("op_8748"), val = tensor([1])]; + tensor channels_mean_51_cast = reduce_mean(axes = var_8748, keep_dims = var_7223, x = inputs_51_cast)[name = tensor("channels_mean_51_cast")]; + tensor zero_mean_51_cast = sub(x = inputs_51_cast, y = channels_mean_51_cast)[name = tensor("zero_mean_51_cast")]; + tensor zero_mean_sq_51_cast = mul(x = zero_mean_51_cast, y = zero_mean_51_cast)[name = tensor("zero_mean_sq_51_cast")]; + tensor var_8752 = const()[name = tensor("op_8752"), val = tensor([1])]; + tensor var_8753_cast = reduce_mean(axes = var_8752, keep_dims = var_7223, x = zero_mean_sq_51_cast)[name = tensor("op_8753_cast")]; + tensor var_8754_to_fp16 = const()[name = tensor("op_8754_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_8755_cast = add(x = var_8753_cast, y = var_8754_to_fp16)[name = tensor("op_8755_cast")]; + tensor denom_51_epsilon_0_to_fp16 = const()[name = tensor("denom_51_epsilon_0_to_fp16"), val = tensor(0x1p-24)]; + tensor denom_51_cast = rsqrt(epsilon = denom_51_epsilon_0_to_fp16, x = var_8755_cast)[name = tensor("denom_51_cast")]; + tensor out_51_cast = mul(x = zero_mean_51_cast, y = denom_51_cast)[name = tensor("out_51_cast")]; + tensor var_8759_to_fp16 = const()[name = tensor("op_8759_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(492224448)))]; + tensor var_8760_cast = add(x = out_51_cast, y = var_8759_to_fp16)[name = tensor("op_8760_cast")]; + tensor var_8762_to_fp16 = const()[name = tensor("op_8762_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(492227072)))]; + tensor hidden_states_189_cast = mul(x = var_8760_cast, y = var_8762_to_fp16)[name = tensor("hidden_states_189_cast")]; + tensor var_8769 = const()[name = tensor("op_8769"), val = tensor([1, 1])]; + tensor var_8771 = const()[name = tensor("op_8771"), val = tensor([1, 1])]; + tensor q_35_pad_type_0 = const()[name = tensor("q_35_pad_type_0"), val = tensor("custom")]; + tensor q_35_pad_0 = const()[name = tensor("q_35_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_1_attentions_1_transformer_blocks_0_attn2_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(492229696))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(493458560))), name = tensor("up_blocks_1_attentions_1_transformer_blocks_0_attn2_to_q_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor q_35_cast = conv(dilations = var_8771, groups = var_7228, pad = q_35_pad_0, pad_type = q_35_pad_type_0, strides = var_8769, weight = up_blocks_1_attentions_1_transformer_blocks_0_attn2_to_q_weight_to_fp16_palettized, x = hidden_states_189_cast)[name = tensor("q_35_cast")]; + tensor var_8775 = const()[name = tensor("op_8775"), val = tensor([1, 1])]; + tensor var_8777 = const()[name = tensor("op_8777"), val = tensor([1, 1])]; + tensor k_69_pad_type_0 = const()[name = tensor("k_69_pad_type_0"), val = tensor("custom")]; + tensor k_69_pad_0 = const()[name = tensor("k_69_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_1_attentions_1_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(493458752))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(494441856))), name = tensor("up_blocks_1_attentions_1_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized"), shape = tensor([1280, 1024, 1, 1])]; + tensor k_69_cast = conv(dilations = var_8777, groups = var_7228, pad = k_69_pad_0, pad_type = k_69_pad_type_0, strides = var_8775, weight = up_blocks_1_attentions_1_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor("k_69_cast")]; + tensor var_8781 = const()[name = tensor("op_8781"), val = tensor([1, 1])]; + tensor var_8783 = const()[name = tensor("op_8783"), val = tensor([1, 1])]; + tensor v_35_pad_type_0 = const()[name = tensor("v_35_pad_type_0"), val = tensor("custom")]; + tensor v_35_pad_0 = const()[name = tensor("v_35_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_1_attentions_1_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(494442048))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(495425152))), name = tensor("up_blocks_1_attentions_1_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized"), shape = tensor([1280, 1024, 1, 1])]; + tensor v_35_cast = conv(dilations = var_8783, groups = var_7228, pad = v_35_pad_0, pad_type = v_35_pad_type_0, strides = var_8781, weight = up_blocks_1_attentions_1_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor("v_35_cast")]; + tensor var_8787_begin_0 = const()[name = tensor("op_8787_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8787_end_0 = const()[name = tensor("op_8787_end_0"), val = tensor([2, 64, 1, 256])]; + tensor var_8787_end_mask_0 = const()[name = tensor("op_8787_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8787_cast = slice_by_index(begin = var_8787_begin_0, end = var_8787_end_0, end_mask = var_8787_end_mask_0, x = q_35_cast)[name = tensor("op_8787_cast")]; + tensor var_8791_begin_0 = const()[name = tensor("op_8791_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_8791_end_0 = const()[name = tensor("op_8791_end_0"), val = tensor([2, 128, 1, 256])]; + tensor var_8791_end_mask_0 = const()[name = tensor("op_8791_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8791_cast = slice_by_index(begin = var_8791_begin_0, end = var_8791_end_0, end_mask = var_8791_end_mask_0, x = q_35_cast)[name = tensor("op_8791_cast")]; + tensor var_8795_begin_0 = const()[name = tensor("op_8795_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_8795_end_0 = const()[name = tensor("op_8795_end_0"), val = tensor([2, 192, 1, 256])]; + tensor var_8795_end_mask_0 = const()[name = tensor("op_8795_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8795_cast = slice_by_index(begin = var_8795_begin_0, end = var_8795_end_0, end_mask = var_8795_end_mask_0, x = q_35_cast)[name = tensor("op_8795_cast")]; + tensor var_8799_begin_0 = const()[name = tensor("op_8799_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_8799_end_0 = const()[name = tensor("op_8799_end_0"), val = tensor([2, 256, 1, 256])]; + tensor var_8799_end_mask_0 = const()[name = tensor("op_8799_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8799_cast = slice_by_index(begin = var_8799_begin_0, end = var_8799_end_0, end_mask = var_8799_end_mask_0, x = q_35_cast)[name = tensor("op_8799_cast")]; + tensor var_8803_begin_0 = const()[name = tensor("op_8803_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_8803_end_0 = const()[name = tensor("op_8803_end_0"), val = tensor([2, 320, 1, 256])]; + tensor var_8803_end_mask_0 = const()[name = tensor("op_8803_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8803_cast = slice_by_index(begin = var_8803_begin_0, end = var_8803_end_0, end_mask = var_8803_end_mask_0, x = q_35_cast)[name = tensor("op_8803_cast")]; + tensor var_8807_begin_0 = const()[name = tensor("op_8807_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_8807_end_0 = const()[name = tensor("op_8807_end_0"), val = tensor([2, 384, 1, 256])]; + tensor var_8807_end_mask_0 = const()[name = tensor("op_8807_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8807_cast = slice_by_index(begin = var_8807_begin_0, end = var_8807_end_0, end_mask = var_8807_end_mask_0, x = q_35_cast)[name = tensor("op_8807_cast")]; + tensor var_8811_begin_0 = const()[name = tensor("op_8811_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_8811_end_0 = const()[name = tensor("op_8811_end_0"), val = tensor([2, 448, 1, 256])]; + tensor var_8811_end_mask_0 = const()[name = tensor("op_8811_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8811_cast = slice_by_index(begin = var_8811_begin_0, end = var_8811_end_0, end_mask = var_8811_end_mask_0, x = q_35_cast)[name = tensor("op_8811_cast")]; + tensor var_8815_begin_0 = const()[name = tensor("op_8815_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_8815_end_0 = const()[name = tensor("op_8815_end_0"), val = tensor([2, 512, 1, 256])]; + tensor var_8815_end_mask_0 = const()[name = tensor("op_8815_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8815_cast = slice_by_index(begin = var_8815_begin_0, end = var_8815_end_0, end_mask = var_8815_end_mask_0, x = q_35_cast)[name = tensor("op_8815_cast")]; + tensor var_8819_begin_0 = const()[name = tensor("op_8819_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_8819_end_0 = const()[name = tensor("op_8819_end_0"), val = tensor([2, 576, 1, 256])]; + tensor var_8819_end_mask_0 = const()[name = tensor("op_8819_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8819_cast = slice_by_index(begin = var_8819_begin_0, end = var_8819_end_0, end_mask = var_8819_end_mask_0, x = q_35_cast)[name = tensor("op_8819_cast")]; + tensor var_8823_begin_0 = const()[name = tensor("op_8823_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_8823_end_0 = const()[name = tensor("op_8823_end_0"), val = tensor([2, 640, 1, 256])]; + tensor var_8823_end_mask_0 = const()[name = tensor("op_8823_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8823_cast = slice_by_index(begin = var_8823_begin_0, end = var_8823_end_0, end_mask = var_8823_end_mask_0, x = q_35_cast)[name = tensor("op_8823_cast")]; + tensor var_8827_begin_0 = const()[name = tensor("op_8827_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_8827_end_0 = const()[name = tensor("op_8827_end_0"), val = tensor([2, 704, 1, 256])]; + tensor var_8827_end_mask_0 = const()[name = tensor("op_8827_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8827_cast = slice_by_index(begin = var_8827_begin_0, end = var_8827_end_0, end_mask = var_8827_end_mask_0, x = q_35_cast)[name = tensor("op_8827_cast")]; + tensor var_8831_begin_0 = const()[name = tensor("op_8831_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_8831_end_0 = const()[name = tensor("op_8831_end_0"), val = tensor([2, 768, 1, 256])]; + tensor var_8831_end_mask_0 = const()[name = tensor("op_8831_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8831_cast = slice_by_index(begin = var_8831_begin_0, end = var_8831_end_0, end_mask = var_8831_end_mask_0, x = q_35_cast)[name = tensor("op_8831_cast")]; + tensor var_8835_begin_0 = const()[name = tensor("op_8835_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_8835_end_0 = const()[name = tensor("op_8835_end_0"), val = tensor([2, 832, 1, 256])]; + tensor var_8835_end_mask_0 = const()[name = tensor("op_8835_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8835_cast = slice_by_index(begin = var_8835_begin_0, end = var_8835_end_0, end_mask = var_8835_end_mask_0, x = q_35_cast)[name = tensor("op_8835_cast")]; + tensor var_8839_begin_0 = const()[name = tensor("op_8839_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_8839_end_0 = const()[name = tensor("op_8839_end_0"), val = tensor([2, 896, 1, 256])]; + tensor var_8839_end_mask_0 = const()[name = tensor("op_8839_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8839_cast = slice_by_index(begin = var_8839_begin_0, end = var_8839_end_0, end_mask = var_8839_end_mask_0, x = q_35_cast)[name = tensor("op_8839_cast")]; + tensor var_8843_begin_0 = const()[name = tensor("op_8843_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_8843_end_0 = const()[name = tensor("op_8843_end_0"), val = tensor([2, 960, 1, 256])]; + tensor var_8843_end_mask_0 = const()[name = tensor("op_8843_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8843_cast = slice_by_index(begin = var_8843_begin_0, end = var_8843_end_0, end_mask = var_8843_end_mask_0, x = q_35_cast)[name = tensor("op_8843_cast")]; + tensor var_8847_begin_0 = const()[name = tensor("op_8847_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_8847_end_0 = const()[name = tensor("op_8847_end_0"), val = tensor([2, 1024, 1, 256])]; + tensor var_8847_end_mask_0 = const()[name = tensor("op_8847_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8847_cast = slice_by_index(begin = var_8847_begin_0, end = var_8847_end_0, end_mask = var_8847_end_mask_0, x = q_35_cast)[name = tensor("op_8847_cast")]; + tensor var_8851_begin_0 = const()[name = tensor("op_8851_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_8851_end_0 = const()[name = tensor("op_8851_end_0"), val = tensor([2, 1088, 1, 256])]; + tensor var_8851_end_mask_0 = const()[name = tensor("op_8851_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8851_cast = slice_by_index(begin = var_8851_begin_0, end = var_8851_end_0, end_mask = var_8851_end_mask_0, x = q_35_cast)[name = tensor("op_8851_cast")]; + tensor var_8855_begin_0 = const()[name = tensor("op_8855_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_8855_end_0 = const()[name = tensor("op_8855_end_0"), val = tensor([2, 1152, 1, 256])]; + tensor var_8855_end_mask_0 = const()[name = tensor("op_8855_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8855_cast = slice_by_index(begin = var_8855_begin_0, end = var_8855_end_0, end_mask = var_8855_end_mask_0, x = q_35_cast)[name = tensor("op_8855_cast")]; + tensor var_8859_begin_0 = const()[name = tensor("op_8859_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_8859_end_0 = const()[name = tensor("op_8859_end_0"), val = tensor([2, 1216, 1, 256])]; + tensor var_8859_end_mask_0 = const()[name = tensor("op_8859_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8859_cast = slice_by_index(begin = var_8859_begin_0, end = var_8859_end_0, end_mask = var_8859_end_mask_0, x = q_35_cast)[name = tensor("op_8859_cast")]; + tensor var_8863_begin_0 = const()[name = tensor("op_8863_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_8863_end_0 = const()[name = tensor("op_8863_end_0"), val = tensor([2, 1280, 1, 256])]; + tensor var_8863_end_mask_0 = const()[name = tensor("op_8863_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8863_cast = slice_by_index(begin = var_8863_begin_0, end = var_8863_end_0, end_mask = var_8863_end_mask_0, x = q_35_cast)[name = tensor("op_8863_cast")]; + tensor k_71_perm_0 = const()[name = tensor("k_71_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_8870_begin_0 = const()[name = tensor("op_8870_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8870_end_0 = const()[name = tensor("op_8870_end_0"), val = tensor([2, 77, 1, 64])]; + tensor var_8870_end_mask_0 = const()[name = tensor("op_8870_end_mask_0"), val = tensor([true, true, true, false])]; + tensor transpose_14 = transpose(perm = k_71_perm_0, x = k_69_cast)[name = tensor("transpose_14")]; + tensor var_8870_cast = slice_by_index(begin = var_8870_begin_0, end = var_8870_end_0, end_mask = var_8870_end_mask_0, x = transpose_14)[name = tensor("op_8870_cast")]; + tensor var_8874_begin_0 = const()[name = tensor("op_8874_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_8874_end_0 = const()[name = tensor("op_8874_end_0"), val = tensor([2, 77, 1, 128])]; + tensor var_8874_end_mask_0 = const()[name = tensor("op_8874_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8874_cast = slice_by_index(begin = var_8874_begin_0, end = var_8874_end_0, end_mask = var_8874_end_mask_0, x = transpose_14)[name = tensor("op_8874_cast")]; + tensor var_8878_begin_0 = const()[name = tensor("op_8878_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_8878_end_0 = const()[name = tensor("op_8878_end_0"), val = tensor([2, 77, 1, 192])]; + tensor var_8878_end_mask_0 = const()[name = tensor("op_8878_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8878_cast = slice_by_index(begin = var_8878_begin_0, end = var_8878_end_0, end_mask = var_8878_end_mask_0, x = transpose_14)[name = tensor("op_8878_cast")]; + tensor var_8882_begin_0 = const()[name = tensor("op_8882_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_8882_end_0 = const()[name = tensor("op_8882_end_0"), val = tensor([2, 77, 1, 256])]; + tensor var_8882_end_mask_0 = const()[name = tensor("op_8882_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8882_cast = slice_by_index(begin = var_8882_begin_0, end = var_8882_end_0, end_mask = var_8882_end_mask_0, x = transpose_14)[name = tensor("op_8882_cast")]; + tensor var_8886_begin_0 = const()[name = tensor("op_8886_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_8886_end_0 = const()[name = tensor("op_8886_end_0"), val = tensor([2, 77, 1, 320])]; + tensor var_8886_end_mask_0 = const()[name = tensor("op_8886_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8886_cast = slice_by_index(begin = var_8886_begin_0, end = var_8886_end_0, end_mask = var_8886_end_mask_0, x = transpose_14)[name = tensor("op_8886_cast")]; + tensor var_8890_begin_0 = const()[name = tensor("op_8890_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_8890_end_0 = const()[name = tensor("op_8890_end_0"), val = tensor([2, 77, 1, 384])]; + tensor var_8890_end_mask_0 = const()[name = tensor("op_8890_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8890_cast = slice_by_index(begin = var_8890_begin_0, end = var_8890_end_0, end_mask = var_8890_end_mask_0, x = transpose_14)[name = tensor("op_8890_cast")]; + tensor var_8894_begin_0 = const()[name = tensor("op_8894_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_8894_end_0 = const()[name = tensor("op_8894_end_0"), val = tensor([2, 77, 1, 448])]; + tensor var_8894_end_mask_0 = const()[name = tensor("op_8894_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8894_cast = slice_by_index(begin = var_8894_begin_0, end = var_8894_end_0, end_mask = var_8894_end_mask_0, x = transpose_14)[name = tensor("op_8894_cast")]; + tensor var_8898_begin_0 = const()[name = tensor("op_8898_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_8898_end_0 = const()[name = tensor("op_8898_end_0"), val = tensor([2, 77, 1, 512])]; + tensor var_8898_end_mask_0 = const()[name = tensor("op_8898_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8898_cast = slice_by_index(begin = var_8898_begin_0, end = var_8898_end_0, end_mask = var_8898_end_mask_0, x = transpose_14)[name = tensor("op_8898_cast")]; + tensor var_8902_begin_0 = const()[name = tensor("op_8902_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_8902_end_0 = const()[name = tensor("op_8902_end_0"), val = tensor([2, 77, 1, 576])]; + tensor var_8902_end_mask_0 = const()[name = tensor("op_8902_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8902_cast = slice_by_index(begin = var_8902_begin_0, end = var_8902_end_0, end_mask = var_8902_end_mask_0, x = transpose_14)[name = tensor("op_8902_cast")]; + tensor var_8906_begin_0 = const()[name = tensor("op_8906_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_8906_end_0 = const()[name = tensor("op_8906_end_0"), val = tensor([2, 77, 1, 640])]; + tensor var_8906_end_mask_0 = const()[name = tensor("op_8906_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8906_cast = slice_by_index(begin = var_8906_begin_0, end = var_8906_end_0, end_mask = var_8906_end_mask_0, x = transpose_14)[name = tensor("op_8906_cast")]; + tensor var_8910_begin_0 = const()[name = tensor("op_8910_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_8910_end_0 = const()[name = tensor("op_8910_end_0"), val = tensor([2, 77, 1, 704])]; + tensor var_8910_end_mask_0 = const()[name = tensor("op_8910_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8910_cast = slice_by_index(begin = var_8910_begin_0, end = var_8910_end_0, end_mask = var_8910_end_mask_0, x = transpose_14)[name = tensor("op_8910_cast")]; + tensor var_8914_begin_0 = const()[name = tensor("op_8914_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_8914_end_0 = const()[name = tensor("op_8914_end_0"), val = tensor([2, 77, 1, 768])]; + tensor var_8914_end_mask_0 = const()[name = tensor("op_8914_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8914_cast = slice_by_index(begin = var_8914_begin_0, end = var_8914_end_0, end_mask = var_8914_end_mask_0, x = transpose_14)[name = tensor("op_8914_cast")]; + tensor var_8918_begin_0 = const()[name = tensor("op_8918_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_8918_end_0 = const()[name = tensor("op_8918_end_0"), val = tensor([2, 77, 1, 832])]; + tensor var_8918_end_mask_0 = const()[name = tensor("op_8918_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8918_cast = slice_by_index(begin = var_8918_begin_0, end = var_8918_end_0, end_mask = var_8918_end_mask_0, x = transpose_14)[name = tensor("op_8918_cast")]; + tensor var_8922_begin_0 = const()[name = tensor("op_8922_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_8922_end_0 = const()[name = tensor("op_8922_end_0"), val = tensor([2, 77, 1, 896])]; + tensor var_8922_end_mask_0 = const()[name = tensor("op_8922_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8922_cast = slice_by_index(begin = var_8922_begin_0, end = var_8922_end_0, end_mask = var_8922_end_mask_0, x = transpose_14)[name = tensor("op_8922_cast")]; + tensor var_8926_begin_0 = const()[name = tensor("op_8926_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_8926_end_0 = const()[name = tensor("op_8926_end_0"), val = tensor([2, 77, 1, 960])]; + tensor var_8926_end_mask_0 = const()[name = tensor("op_8926_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8926_cast = slice_by_index(begin = var_8926_begin_0, end = var_8926_end_0, end_mask = var_8926_end_mask_0, x = transpose_14)[name = tensor("op_8926_cast")]; + tensor var_8930_begin_0 = const()[name = tensor("op_8930_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_8930_end_0 = const()[name = tensor("op_8930_end_0"), val = tensor([2, 77, 1, 1024])]; + tensor var_8930_end_mask_0 = const()[name = tensor("op_8930_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8930_cast = slice_by_index(begin = var_8930_begin_0, end = var_8930_end_0, end_mask = var_8930_end_mask_0, x = transpose_14)[name = tensor("op_8930_cast")]; + tensor var_8934_begin_0 = const()[name = tensor("op_8934_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_8934_end_0 = const()[name = tensor("op_8934_end_0"), val = tensor([2, 77, 1, 1088])]; + tensor var_8934_end_mask_0 = const()[name = tensor("op_8934_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8934_cast = slice_by_index(begin = var_8934_begin_0, end = var_8934_end_0, end_mask = var_8934_end_mask_0, x = transpose_14)[name = tensor("op_8934_cast")]; + tensor var_8938_begin_0 = const()[name = tensor("op_8938_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_8938_end_0 = const()[name = tensor("op_8938_end_0"), val = tensor([2, 77, 1, 1152])]; + tensor var_8938_end_mask_0 = const()[name = tensor("op_8938_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8938_cast = slice_by_index(begin = var_8938_begin_0, end = var_8938_end_0, end_mask = var_8938_end_mask_0, x = transpose_14)[name = tensor("op_8938_cast")]; + tensor var_8942_begin_0 = const()[name = tensor("op_8942_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_8942_end_0 = const()[name = tensor("op_8942_end_0"), val = tensor([2, 77, 1, 1216])]; + tensor var_8942_end_mask_0 = const()[name = tensor("op_8942_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8942_cast = slice_by_index(begin = var_8942_begin_0, end = var_8942_end_0, end_mask = var_8942_end_mask_0, x = transpose_14)[name = tensor("op_8942_cast")]; + tensor var_8946_begin_0 = const()[name = tensor("op_8946_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_8946_end_0 = const()[name = tensor("op_8946_end_0"), val = tensor([2, 77, 1, 1280])]; + tensor var_8946_end_mask_0 = const()[name = tensor("op_8946_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8946_cast = slice_by_index(begin = var_8946_begin_0, end = var_8946_end_0, end_mask = var_8946_end_mask_0, x = transpose_14)[name = tensor("op_8946_cast")]; + tensor var_8948_begin_0 = const()[name = tensor("op_8948_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8948_end_0 = const()[name = tensor("op_8948_end_0"), val = tensor([2, 64, 1, 77])]; + tensor var_8948_end_mask_0 = const()[name = tensor("op_8948_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8948_cast = slice_by_index(begin = var_8948_begin_0, end = var_8948_end_0, end_mask = var_8948_end_mask_0, x = v_35_cast)[name = tensor("op_8948_cast")]; + tensor var_8952_begin_0 = const()[name = tensor("op_8952_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_8952_end_0 = const()[name = tensor("op_8952_end_0"), val = tensor([2, 128, 1, 77])]; + tensor var_8952_end_mask_0 = const()[name = tensor("op_8952_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8952_cast = slice_by_index(begin = var_8952_begin_0, end = var_8952_end_0, end_mask = var_8952_end_mask_0, x = v_35_cast)[name = tensor("op_8952_cast")]; + tensor var_8956_begin_0 = const()[name = tensor("op_8956_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_8956_end_0 = const()[name = tensor("op_8956_end_0"), val = tensor([2, 192, 1, 77])]; + tensor var_8956_end_mask_0 = const()[name = tensor("op_8956_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8956_cast = slice_by_index(begin = var_8956_begin_0, end = var_8956_end_0, end_mask = var_8956_end_mask_0, x = v_35_cast)[name = tensor("op_8956_cast")]; + tensor var_8960_begin_0 = const()[name = tensor("op_8960_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_8960_end_0 = const()[name = tensor("op_8960_end_0"), val = tensor([2, 256, 1, 77])]; + tensor var_8960_end_mask_0 = const()[name = tensor("op_8960_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8960_cast = slice_by_index(begin = var_8960_begin_0, end = var_8960_end_0, end_mask = var_8960_end_mask_0, x = v_35_cast)[name = tensor("op_8960_cast")]; + tensor var_8964_begin_0 = const()[name = tensor("op_8964_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_8964_end_0 = const()[name = tensor("op_8964_end_0"), val = tensor([2, 320, 1, 77])]; + tensor var_8964_end_mask_0 = const()[name = tensor("op_8964_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8964_cast = slice_by_index(begin = var_8964_begin_0, end = var_8964_end_0, end_mask = var_8964_end_mask_0, x = v_35_cast)[name = tensor("op_8964_cast")]; + tensor var_8968_begin_0 = const()[name = tensor("op_8968_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_8968_end_0 = const()[name = tensor("op_8968_end_0"), val = tensor([2, 384, 1, 77])]; + tensor var_8968_end_mask_0 = const()[name = tensor("op_8968_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8968_cast = slice_by_index(begin = var_8968_begin_0, end = var_8968_end_0, end_mask = var_8968_end_mask_0, x = v_35_cast)[name = tensor("op_8968_cast")]; + tensor var_8972_begin_0 = const()[name = tensor("op_8972_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_8972_end_0 = const()[name = tensor("op_8972_end_0"), val = tensor([2, 448, 1, 77])]; + tensor var_8972_end_mask_0 = const()[name = tensor("op_8972_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8972_cast = slice_by_index(begin = var_8972_begin_0, end = var_8972_end_0, end_mask = var_8972_end_mask_0, x = v_35_cast)[name = tensor("op_8972_cast")]; + tensor var_8976_begin_0 = const()[name = tensor("op_8976_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_8976_end_0 = const()[name = tensor("op_8976_end_0"), val = tensor([2, 512, 1, 77])]; + tensor var_8976_end_mask_0 = const()[name = tensor("op_8976_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8976_cast = slice_by_index(begin = var_8976_begin_0, end = var_8976_end_0, end_mask = var_8976_end_mask_0, x = v_35_cast)[name = tensor("op_8976_cast")]; + tensor var_8980_begin_0 = const()[name = tensor("op_8980_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_8980_end_0 = const()[name = tensor("op_8980_end_0"), val = tensor([2, 576, 1, 77])]; + tensor var_8980_end_mask_0 = const()[name = tensor("op_8980_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8980_cast = slice_by_index(begin = var_8980_begin_0, end = var_8980_end_0, end_mask = var_8980_end_mask_0, x = v_35_cast)[name = tensor("op_8980_cast")]; + tensor var_8984_begin_0 = const()[name = tensor("op_8984_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_8984_end_0 = const()[name = tensor("op_8984_end_0"), val = tensor([2, 640, 1, 77])]; + tensor var_8984_end_mask_0 = const()[name = tensor("op_8984_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8984_cast = slice_by_index(begin = var_8984_begin_0, end = var_8984_end_0, end_mask = var_8984_end_mask_0, x = v_35_cast)[name = tensor("op_8984_cast")]; + tensor var_8988_begin_0 = const()[name = tensor("op_8988_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_8988_end_0 = const()[name = tensor("op_8988_end_0"), val = tensor([2, 704, 1, 77])]; + tensor var_8988_end_mask_0 = const()[name = tensor("op_8988_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8988_cast = slice_by_index(begin = var_8988_begin_0, end = var_8988_end_0, end_mask = var_8988_end_mask_0, x = v_35_cast)[name = tensor("op_8988_cast")]; + tensor var_8992_begin_0 = const()[name = tensor("op_8992_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_8992_end_0 = const()[name = tensor("op_8992_end_0"), val = tensor([2, 768, 1, 77])]; + tensor var_8992_end_mask_0 = const()[name = tensor("op_8992_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8992_cast = slice_by_index(begin = var_8992_begin_0, end = var_8992_end_0, end_mask = var_8992_end_mask_0, x = v_35_cast)[name = tensor("op_8992_cast")]; + tensor var_8996_begin_0 = const()[name = tensor("op_8996_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_8996_end_0 = const()[name = tensor("op_8996_end_0"), val = tensor([2, 832, 1, 77])]; + tensor var_8996_end_mask_0 = const()[name = tensor("op_8996_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8996_cast = slice_by_index(begin = var_8996_begin_0, end = var_8996_end_0, end_mask = var_8996_end_mask_0, x = v_35_cast)[name = tensor("op_8996_cast")]; + tensor var_9000_begin_0 = const()[name = tensor("op_9000_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_9000_end_0 = const()[name = tensor("op_9000_end_0"), val = tensor([2, 896, 1, 77])]; + tensor var_9000_end_mask_0 = const()[name = tensor("op_9000_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9000_cast = slice_by_index(begin = var_9000_begin_0, end = var_9000_end_0, end_mask = var_9000_end_mask_0, x = v_35_cast)[name = tensor("op_9000_cast")]; + tensor var_9004_begin_0 = const()[name = tensor("op_9004_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_9004_end_0 = const()[name = tensor("op_9004_end_0"), val = tensor([2, 960, 1, 77])]; + tensor var_9004_end_mask_0 = const()[name = tensor("op_9004_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9004_cast = slice_by_index(begin = var_9004_begin_0, end = var_9004_end_0, end_mask = var_9004_end_mask_0, x = v_35_cast)[name = tensor("op_9004_cast")]; + tensor var_9008_begin_0 = const()[name = tensor("op_9008_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_9008_end_0 = const()[name = tensor("op_9008_end_0"), val = tensor([2, 1024, 1, 77])]; + tensor var_9008_end_mask_0 = const()[name = tensor("op_9008_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9008_cast = slice_by_index(begin = var_9008_begin_0, end = var_9008_end_0, end_mask = var_9008_end_mask_0, x = v_35_cast)[name = tensor("op_9008_cast")]; + tensor var_9012_begin_0 = const()[name = tensor("op_9012_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_9012_end_0 = const()[name = tensor("op_9012_end_0"), val = tensor([2, 1088, 1, 77])]; + tensor var_9012_end_mask_0 = const()[name = tensor("op_9012_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9012_cast = slice_by_index(begin = var_9012_begin_0, end = var_9012_end_0, end_mask = var_9012_end_mask_0, x = v_35_cast)[name = tensor("op_9012_cast")]; + tensor var_9016_begin_0 = const()[name = tensor("op_9016_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_9016_end_0 = const()[name = tensor("op_9016_end_0"), val = tensor([2, 1152, 1, 77])]; + tensor var_9016_end_mask_0 = const()[name = tensor("op_9016_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9016_cast = slice_by_index(begin = var_9016_begin_0, end = var_9016_end_0, end_mask = var_9016_end_mask_0, x = v_35_cast)[name = tensor("op_9016_cast")]; + tensor var_9020_begin_0 = const()[name = tensor("op_9020_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_9020_end_0 = const()[name = tensor("op_9020_end_0"), val = tensor([2, 1216, 1, 77])]; + tensor var_9020_end_mask_0 = const()[name = tensor("op_9020_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9020_cast = slice_by_index(begin = var_9020_begin_0, end = var_9020_end_0, end_mask = var_9020_end_mask_0, x = v_35_cast)[name = tensor("op_9020_cast")]; + tensor var_9024_begin_0 = const()[name = tensor("op_9024_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_9024_end_0 = const()[name = tensor("op_9024_end_0"), val = tensor([2, 1280, 1, 77])]; + tensor var_9024_end_mask_0 = const()[name = tensor("op_9024_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9024_cast = slice_by_index(begin = var_9024_begin_0, end = var_9024_end_0, end_mask = var_9024_end_mask_0, x = v_35_cast)[name = tensor("op_9024_cast")]; + tensor var_9028_equation_0 = const()[name = tensor("op_9028_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9028_cast = einsum(equation = var_9028_equation_0, values = (var_8870_cast, var_8787_cast))[name = tensor("op_9028_cast")]; + tensor var_9029_to_fp16 = const()[name = tensor("op_9029_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_361_cast = mul(x = var_9028_cast, y = var_9029_to_fp16)[name = tensor("aw_361_cast")]; + tensor var_9032_equation_0 = const()[name = tensor("op_9032_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9032_cast = einsum(equation = var_9032_equation_0, values = (var_8874_cast, var_8791_cast))[name = tensor("op_9032_cast")]; + tensor var_9033_to_fp16 = const()[name = tensor("op_9033_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_363_cast = mul(x = var_9032_cast, y = var_9033_to_fp16)[name = tensor("aw_363_cast")]; + tensor var_9036_equation_0 = const()[name = tensor("op_9036_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9036_cast = einsum(equation = var_9036_equation_0, values = (var_8878_cast, var_8795_cast))[name = tensor("op_9036_cast")]; + tensor var_9037_to_fp16 = const()[name = tensor("op_9037_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_365_cast = mul(x = var_9036_cast, y = var_9037_to_fp16)[name = tensor("aw_365_cast")]; + tensor var_9040_equation_0 = const()[name = tensor("op_9040_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9040_cast = einsum(equation = var_9040_equation_0, values = (var_8882_cast, var_8799_cast))[name = tensor("op_9040_cast")]; + tensor var_9041_to_fp16 = const()[name = tensor("op_9041_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_367_cast = mul(x = var_9040_cast, y = var_9041_to_fp16)[name = tensor("aw_367_cast")]; + tensor var_9044_equation_0 = const()[name = tensor("op_9044_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9044_cast = einsum(equation = var_9044_equation_0, values = (var_8886_cast, var_8803_cast))[name = tensor("op_9044_cast")]; + tensor var_9045_to_fp16 = const()[name = tensor("op_9045_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_369_cast = mul(x = var_9044_cast, y = var_9045_to_fp16)[name = tensor("aw_369_cast")]; + tensor var_9048_equation_0 = const()[name = tensor("op_9048_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9048_cast = einsum(equation = var_9048_equation_0, values = (var_8890_cast, var_8807_cast))[name = tensor("op_9048_cast")]; + tensor var_9049_to_fp16 = const()[name = tensor("op_9049_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_371_cast = mul(x = var_9048_cast, y = var_9049_to_fp16)[name = tensor("aw_371_cast")]; + tensor var_9052_equation_0 = const()[name = tensor("op_9052_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9052_cast = einsum(equation = var_9052_equation_0, values = (var_8894_cast, var_8811_cast))[name = tensor("op_9052_cast")]; + tensor var_9053_to_fp16 = const()[name = tensor("op_9053_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_373_cast = mul(x = var_9052_cast, y = var_9053_to_fp16)[name = tensor("aw_373_cast")]; + tensor var_9056_equation_0 = const()[name = tensor("op_9056_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9056_cast = einsum(equation = var_9056_equation_0, values = (var_8898_cast, var_8815_cast))[name = tensor("op_9056_cast")]; + tensor var_9057_to_fp16 = const()[name = tensor("op_9057_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_375_cast = mul(x = var_9056_cast, y = var_9057_to_fp16)[name = tensor("aw_375_cast")]; + tensor var_9060_equation_0 = const()[name = tensor("op_9060_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9060_cast = einsum(equation = var_9060_equation_0, values = (var_8902_cast, var_8819_cast))[name = tensor("op_9060_cast")]; + tensor var_9061_to_fp16 = const()[name = tensor("op_9061_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_377_cast = mul(x = var_9060_cast, y = var_9061_to_fp16)[name = tensor("aw_377_cast")]; + tensor var_9064_equation_0 = const()[name = tensor("op_9064_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9064_cast = einsum(equation = var_9064_equation_0, values = (var_8906_cast, var_8823_cast))[name = tensor("op_9064_cast")]; + tensor var_9065_to_fp16 = const()[name = tensor("op_9065_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_379_cast = mul(x = var_9064_cast, y = var_9065_to_fp16)[name = tensor("aw_379_cast")]; + tensor var_9068_equation_0 = const()[name = tensor("op_9068_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9068_cast = einsum(equation = var_9068_equation_0, values = (var_8910_cast, var_8827_cast))[name = tensor("op_9068_cast")]; + tensor var_9069_to_fp16 = const()[name = tensor("op_9069_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_381_cast = mul(x = var_9068_cast, y = var_9069_to_fp16)[name = tensor("aw_381_cast")]; + tensor var_9072_equation_0 = const()[name = tensor("op_9072_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9072_cast = einsum(equation = var_9072_equation_0, values = (var_8914_cast, var_8831_cast))[name = tensor("op_9072_cast")]; + tensor var_9073_to_fp16 = const()[name = tensor("op_9073_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_383_cast = mul(x = var_9072_cast, y = var_9073_to_fp16)[name = tensor("aw_383_cast")]; + tensor var_9076_equation_0 = const()[name = tensor("op_9076_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9076_cast = einsum(equation = var_9076_equation_0, values = (var_8918_cast, var_8835_cast))[name = tensor("op_9076_cast")]; + tensor var_9077_to_fp16 = const()[name = tensor("op_9077_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_385_cast = mul(x = var_9076_cast, y = var_9077_to_fp16)[name = tensor("aw_385_cast")]; + tensor var_9080_equation_0 = const()[name = tensor("op_9080_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9080_cast = einsum(equation = var_9080_equation_0, values = (var_8922_cast, var_8839_cast))[name = tensor("op_9080_cast")]; + tensor var_9081_to_fp16 = const()[name = tensor("op_9081_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_387_cast = mul(x = var_9080_cast, y = var_9081_to_fp16)[name = tensor("aw_387_cast")]; + tensor var_9084_equation_0 = const()[name = tensor("op_9084_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9084_cast = einsum(equation = var_9084_equation_0, values = (var_8926_cast, var_8843_cast))[name = tensor("op_9084_cast")]; + tensor var_9085_to_fp16 = const()[name = tensor("op_9085_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_389_cast = mul(x = var_9084_cast, y = var_9085_to_fp16)[name = tensor("aw_389_cast")]; + tensor var_9088_equation_0 = const()[name = tensor("op_9088_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9088_cast = einsum(equation = var_9088_equation_0, values = (var_8930_cast, var_8847_cast))[name = tensor("op_9088_cast")]; + tensor var_9089_to_fp16 = const()[name = tensor("op_9089_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_391_cast = mul(x = var_9088_cast, y = var_9089_to_fp16)[name = tensor("aw_391_cast")]; + tensor var_9092_equation_0 = const()[name = tensor("op_9092_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9092_cast = einsum(equation = var_9092_equation_0, values = (var_8934_cast, var_8851_cast))[name = tensor("op_9092_cast")]; + tensor var_9093_to_fp16 = const()[name = tensor("op_9093_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_393_cast = mul(x = var_9092_cast, y = var_9093_to_fp16)[name = tensor("aw_393_cast")]; + tensor var_9096_equation_0 = const()[name = tensor("op_9096_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9096_cast = einsum(equation = var_9096_equation_0, values = (var_8938_cast, var_8855_cast))[name = tensor("op_9096_cast")]; + tensor var_9097_to_fp16 = const()[name = tensor("op_9097_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_395_cast = mul(x = var_9096_cast, y = var_9097_to_fp16)[name = tensor("aw_395_cast")]; + tensor var_9100_equation_0 = const()[name = tensor("op_9100_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9100_cast = einsum(equation = var_9100_equation_0, values = (var_8942_cast, var_8859_cast))[name = tensor("op_9100_cast")]; + tensor var_9101_to_fp16 = const()[name = tensor("op_9101_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_397_cast = mul(x = var_9100_cast, y = var_9101_to_fp16)[name = tensor("aw_397_cast")]; + tensor var_9104_equation_0 = const()[name = tensor("op_9104_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9104_cast = einsum(equation = var_9104_equation_0, values = (var_8946_cast, var_8863_cast))[name = tensor("op_9104_cast")]; + tensor var_9105_to_fp16 = const()[name = tensor("op_9105_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_399_cast = mul(x = var_9104_cast, y = var_9105_to_fp16)[name = tensor("aw_399_cast")]; + tensor var_9107_cast = softmax(axis = var_7228, x = aw_361_cast)[name = tensor("op_9107_cast")]; + tensor var_9108_cast = softmax(axis = var_7228, x = aw_363_cast)[name = tensor("op_9108_cast")]; + tensor var_9109_cast = softmax(axis = var_7228, x = aw_365_cast)[name = tensor("op_9109_cast")]; + tensor var_9110_cast = softmax(axis = var_7228, x = aw_367_cast)[name = tensor("op_9110_cast")]; + tensor var_9111_cast = softmax(axis = var_7228, x = aw_369_cast)[name = tensor("op_9111_cast")]; + tensor var_9112_cast = softmax(axis = var_7228, x = aw_371_cast)[name = tensor("op_9112_cast")]; + tensor var_9113_cast = softmax(axis = var_7228, x = aw_373_cast)[name = tensor("op_9113_cast")]; + tensor var_9114_cast = softmax(axis = var_7228, x = aw_375_cast)[name = tensor("op_9114_cast")]; + tensor var_9115_cast = softmax(axis = var_7228, x = aw_377_cast)[name = tensor("op_9115_cast")]; + tensor var_9116_cast = softmax(axis = var_7228, x = aw_379_cast)[name = tensor("op_9116_cast")]; + tensor var_9117_cast = softmax(axis = var_7228, x = aw_381_cast)[name = tensor("op_9117_cast")]; + tensor var_9118_cast = softmax(axis = var_7228, x = aw_383_cast)[name = tensor("op_9118_cast")]; + tensor var_9119_cast = softmax(axis = var_7228, x = aw_385_cast)[name = tensor("op_9119_cast")]; + tensor var_9120_cast = softmax(axis = var_7228, x = aw_387_cast)[name = tensor("op_9120_cast")]; + tensor var_9121_cast = softmax(axis = var_7228, x = aw_389_cast)[name = tensor("op_9121_cast")]; + tensor var_9122_cast = softmax(axis = var_7228, x = aw_391_cast)[name = tensor("op_9122_cast")]; + tensor var_9123_cast = softmax(axis = var_7228, x = aw_393_cast)[name = tensor("op_9123_cast")]; + tensor var_9124_cast = softmax(axis = var_7228, x = aw_395_cast)[name = tensor("op_9124_cast")]; + tensor var_9125_cast = softmax(axis = var_7228, x = aw_397_cast)[name = tensor("op_9125_cast")]; + tensor var_9126_cast = softmax(axis = var_7228, x = aw_399_cast)[name = tensor("op_9126_cast")]; + tensor var_9128_equation_0 = const()[name = tensor("op_9128_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9128_cast = einsum(equation = var_9128_equation_0, values = (var_8948_cast, var_9107_cast))[name = tensor("op_9128_cast")]; + tensor var_9130_equation_0 = const()[name = tensor("op_9130_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9130_cast = einsum(equation = var_9130_equation_0, values = (var_8952_cast, var_9108_cast))[name = tensor("op_9130_cast")]; + tensor var_9132_equation_0 = const()[name = tensor("op_9132_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9132_cast = einsum(equation = var_9132_equation_0, values = (var_8956_cast, var_9109_cast))[name = tensor("op_9132_cast")]; + tensor var_9134_equation_0 = const()[name = tensor("op_9134_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9134_cast = einsum(equation = var_9134_equation_0, values = (var_8960_cast, var_9110_cast))[name = tensor("op_9134_cast")]; + tensor var_9136_equation_0 = const()[name = tensor("op_9136_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9136_cast = einsum(equation = var_9136_equation_0, values = (var_8964_cast, var_9111_cast))[name = tensor("op_9136_cast")]; + tensor var_9138_equation_0 = const()[name = tensor("op_9138_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9138_cast = einsum(equation = var_9138_equation_0, values = (var_8968_cast, var_9112_cast))[name = tensor("op_9138_cast")]; + tensor var_9140_equation_0 = const()[name = tensor("op_9140_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9140_cast = einsum(equation = var_9140_equation_0, values = (var_8972_cast, var_9113_cast))[name = tensor("op_9140_cast")]; + tensor var_9142_equation_0 = const()[name = tensor("op_9142_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9142_cast = einsum(equation = var_9142_equation_0, values = (var_8976_cast, var_9114_cast))[name = tensor("op_9142_cast")]; + tensor var_9144_equation_0 = const()[name = tensor("op_9144_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9144_cast = einsum(equation = var_9144_equation_0, values = (var_8980_cast, var_9115_cast))[name = tensor("op_9144_cast")]; + tensor var_9146_equation_0 = const()[name = tensor("op_9146_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9146_cast = einsum(equation = var_9146_equation_0, values = (var_8984_cast, var_9116_cast))[name = tensor("op_9146_cast")]; + tensor var_9148_equation_0 = const()[name = tensor("op_9148_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9148_cast = einsum(equation = var_9148_equation_0, values = (var_8988_cast, var_9117_cast))[name = tensor("op_9148_cast")]; + tensor var_9150_equation_0 = const()[name = tensor("op_9150_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9150_cast = einsum(equation = var_9150_equation_0, values = (var_8992_cast, var_9118_cast))[name = tensor("op_9150_cast")]; + tensor var_9152_equation_0 = const()[name = tensor("op_9152_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9152_cast = einsum(equation = var_9152_equation_0, values = (var_8996_cast, var_9119_cast))[name = tensor("op_9152_cast")]; + tensor var_9154_equation_0 = const()[name = tensor("op_9154_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9154_cast = einsum(equation = var_9154_equation_0, values = (var_9000_cast, var_9120_cast))[name = tensor("op_9154_cast")]; + tensor var_9156_equation_0 = const()[name = tensor("op_9156_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9156_cast = einsum(equation = var_9156_equation_0, values = (var_9004_cast, var_9121_cast))[name = tensor("op_9156_cast")]; + tensor var_9158_equation_0 = const()[name = tensor("op_9158_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9158_cast = einsum(equation = var_9158_equation_0, values = (var_9008_cast, var_9122_cast))[name = tensor("op_9158_cast")]; + tensor var_9160_equation_0 = const()[name = tensor("op_9160_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9160_cast = einsum(equation = var_9160_equation_0, values = (var_9012_cast, var_9123_cast))[name = tensor("op_9160_cast")]; + tensor var_9162_equation_0 = const()[name = tensor("op_9162_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9162_cast = einsum(equation = var_9162_equation_0, values = (var_9016_cast, var_9124_cast))[name = tensor("op_9162_cast")]; + tensor var_9164_equation_0 = const()[name = tensor("op_9164_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9164_cast = einsum(equation = var_9164_equation_0, values = (var_9020_cast, var_9125_cast))[name = tensor("op_9164_cast")]; + tensor var_9166_equation_0 = const()[name = tensor("op_9166_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9166_cast = einsum(equation = var_9166_equation_0, values = (var_9024_cast, var_9126_cast))[name = tensor("op_9166_cast")]; + tensor input_329_interleave_0 = const()[name = tensor("input_329_interleave_0"), val = tensor(false)]; + tensor input_329_cast = concat(axis = var_7228, interleave = input_329_interleave_0, values = (var_9128_cast, var_9130_cast, var_9132_cast, var_9134_cast, var_9136_cast, var_9138_cast, var_9140_cast, var_9142_cast, var_9144_cast, var_9146_cast, var_9148_cast, var_9150_cast, var_9152_cast, var_9154_cast, var_9156_cast, var_9158_cast, var_9160_cast, var_9162_cast, var_9164_cast, var_9166_cast))[name = tensor("input_329_cast")]; + tensor var_9172 = const()[name = tensor("op_9172"), val = tensor([1, 1])]; + tensor var_9174 = const()[name = tensor("op_9174"), val = tensor([1, 1])]; + tensor var_9176_pad_type_0 = const()[name = tensor("op_9176_pad_type_0"), val = tensor("custom")]; + tensor var_9176_pad_0 = const()[name = tensor("op_9176_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_1_attentions_1_transformer_blocks_0_attn2_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(495425344))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(496654208))), name = tensor("up_blocks_1_attentions_1_transformer_blocks_0_attn2_to_out_0_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor up_blocks_1_attentions_1_transformer_blocks_0_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("up_blocks_1_attentions_1_transformer_blocks_0_attn2_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(496654400)))]; + tensor var_9176_cast = conv(bias = up_blocks_1_attentions_1_transformer_blocks_0_attn2_to_out_0_bias_to_fp16, dilations = var_9174, groups = var_7228, pad = var_9176_pad_0, pad_type = var_9176_pad_type_0, strides = var_9172, weight = up_blocks_1_attentions_1_transformer_blocks_0_attn2_to_out_0_weight_to_fp16_palettized, x = input_329_cast)[name = tensor("op_9176_cast")]; + tensor inputs_53_cast = add(x = var_9176_cast, y = inputs_51_cast)[name = tensor("inputs_53_cast")]; + tensor var_9180 = const()[name = tensor("op_9180"), val = tensor([1])]; + tensor channels_mean_53_cast = reduce_mean(axes = var_9180, keep_dims = var_7223, x = inputs_53_cast)[name = tensor("channels_mean_53_cast")]; + tensor zero_mean_53_cast = sub(x = inputs_53_cast, y = channels_mean_53_cast)[name = tensor("zero_mean_53_cast")]; + tensor zero_mean_sq_53_cast = mul(x = zero_mean_53_cast, y = zero_mean_53_cast)[name = tensor("zero_mean_sq_53_cast")]; + tensor var_9184 = const()[name = tensor("op_9184"), val = tensor([1])]; + tensor var_9185_cast = reduce_mean(axes = var_9184, keep_dims = var_7223, x = zero_mean_sq_53_cast)[name = tensor("op_9185_cast")]; + tensor var_9186_to_fp16 = const()[name = tensor("op_9186_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_9187_cast = add(x = var_9185_cast, y = var_9186_to_fp16)[name = tensor("op_9187_cast")]; + tensor denom_53_epsilon_0_to_fp16 = const()[name = tensor("denom_53_epsilon_0_to_fp16"), val = tensor(0x1p-24)]; + tensor denom_53_cast = rsqrt(epsilon = denom_53_epsilon_0_to_fp16, x = var_9187_cast)[name = tensor("denom_53_cast")]; + tensor out_53_cast = mul(x = zero_mean_53_cast, y = denom_53_cast)[name = tensor("out_53_cast")]; + tensor var_9191_to_fp16 = const()[name = tensor("op_9191_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(496657024)))]; + tensor var_9192_cast = add(x = out_53_cast, y = var_9191_to_fp16)[name = tensor("op_9192_cast")]; + tensor var_9194_to_fp16 = const()[name = tensor("op_9194_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(496659648)))]; + tensor input_331_cast = mul(x = var_9192_cast, y = var_9194_to_fp16)[name = tensor("input_331_cast")]; + tensor var_9202 = const()[name = tensor("op_9202"), val = tensor([1, 1])]; + tensor var_9204 = const()[name = tensor("op_9204"), val = tensor([1, 1])]; + tensor var_9206_pad_type_0 = const()[name = tensor("op_9206_pad_type_0"), val = tensor("custom")]; + tensor var_9206_pad_0 = const()[name = tensor("op_9206_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_1_attentions_1_transformer_blocks_0_ff_net_0_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(496662272))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(506492736))), name = tensor("up_blocks_1_attentions_1_transformer_blocks_0_ff_net_0_proj_weight_to_fp16_palettized"), shape = tensor([10240, 1280, 1, 1])]; + tensor up_blocks_1_attentions_1_transformer_blocks_0_ff_net_0_proj_bias_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(506492928))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(506500672))), name = tensor("up_blocks_1_attentions_1_transformer_blocks_0_ff_net_0_proj_bias_to_fp16_palettized"), shape = tensor([10240])]; + tensor var_9206_cast = conv(bias = up_blocks_1_attentions_1_transformer_blocks_0_ff_net_0_proj_bias_to_fp16_palettized, dilations = var_9204, groups = var_7228, pad = var_9206_pad_0, pad_type = var_9206_pad_type_0, strides = var_9202, weight = up_blocks_1_attentions_1_transformer_blocks_0_ff_net_0_proj_weight_to_fp16_palettized, x = input_331_cast)[name = tensor("op_9206_cast")]; + tensor var_9207_split_sizes_0 = const()[name = tensor("op_9207_split_sizes_0"), val = tensor([5120, 5120])]; + tensor var_9207_axis_0 = const()[name = tensor("op_9207_axis_0"), val = tensor(1)]; + tensor var_9207_cast_0, tensor var_9207_cast_1 = split(axis = var_9207_axis_0, split_sizes = var_9207_split_sizes_0, x = var_9206_cast)[name = tensor("op_9207_cast")]; + tensor var_9209_mode_0 = const()[name = tensor("op_9209_mode_0"), val = tensor("EXACT")]; + tensor var_9209_cast = gelu(mode = var_9209_mode_0, x = var_9207_cast_1)[name = tensor("op_9209_cast")]; + tensor input_333_cast = mul(x = var_9207_cast_0, y = var_9209_cast)[name = tensor("input_333_cast")]; + tensor var_9213 = const()[name = tensor("op_9213"), val = tensor([1, 1])]; + tensor var_9215 = const()[name = tensor("op_9215"), val = tensor([1, 1])]; + tensor var_9217_pad_type_0 = const()[name = tensor("op_9217_pad_type_0"), val = tensor("custom")]; + tensor var_9217_pad_0 = const()[name = tensor("op_9217_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_1_attentions_1_transformer_blocks_0_ff_net_2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(506500864))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(511416128))), name = tensor("up_blocks_1_attentions_1_transformer_blocks_0_ff_net_2_weight_to_fp16_palettized"), shape = tensor([1280, 5120, 1, 1])]; + tensor up_blocks_1_attentions_1_transformer_blocks_0_ff_net_2_bias_to_fp16 = const()[name = tensor("up_blocks_1_attentions_1_transformer_blocks_0_ff_net_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(511416320)))]; + tensor var_9217_cast = conv(bias = up_blocks_1_attentions_1_transformer_blocks_0_ff_net_2_bias_to_fp16, dilations = var_9215, groups = var_7228, pad = var_9217_pad_0, pad_type = var_9217_pad_type_0, strides = var_9213, weight = up_blocks_1_attentions_1_transformer_blocks_0_ff_net_2_weight_to_fp16_palettized, x = input_333_cast)[name = tensor("op_9217_cast")]; + tensor hidden_states_193_cast = add(x = var_9217_cast, y = inputs_53_cast)[name = tensor("hidden_states_193_cast")]; + tensor var_9219 = const()[name = tensor("op_9219"), val = tensor([2, 1280, 16, 16])]; + tensor input_335_cast = reshape(shape = var_9219, x = hidden_states_193_cast)[name = tensor("input_335_cast")]; + tensor var_9223 = const()[name = tensor("op_9223"), val = tensor([1, 1])]; + tensor var_9225 = const()[name = tensor("op_9225"), val = tensor([1, 1])]; + tensor hidden_states_195_pad_type_0 = const()[name = tensor("hidden_states_195_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_195_pad_0 = const()[name = tensor("hidden_states_195_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_1_attentions_1_proj_out_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(511418944))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(512647808))), name = tensor("up_blocks_1_attentions_1_proj_out_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor up_blocks_1_attentions_1_proj_out_bias_to_fp16 = const()[name = tensor("up_blocks_1_attentions_1_proj_out_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(512648000)))]; + tensor hidden_states_195_cast = conv(bias = up_blocks_1_attentions_1_proj_out_bias_to_fp16, dilations = var_9225, groups = var_7228, pad = hidden_states_195_pad_0, pad_type = hidden_states_195_pad_type_0, strides = var_9223, weight = up_blocks_1_attentions_1_proj_out_weight_to_fp16_palettized, x = input_335_cast)[name = tensor("hidden_states_195_cast")]; + tensor hidden_states_197_cast = add(x = hidden_states_195_cast, y = hidden_states_183_cast)[name = tensor("hidden_states_197_cast")]; + tensor input_337_interleave_0 = const()[name = tensor("input_337_interleave_0"), val = tensor(false)]; + tensor input_337_cast = concat(axis = var_7228, interleave = input_337_interleave_0, values = (hidden_states_197_cast, input_117_cast))[name = tensor("input_337_cast")]; + tensor reshape_156_shape_0 = const()[name = tensor("reshape_156_shape_0"), val = tensor([2, 32, 60, 16, 16])]; + tensor reshape_156_cast = reshape(shape = reshape_156_shape_0, x = input_337_cast)[name = tensor("reshape_156_cast")]; + tensor reduce_mean_117_axes_0 = const()[name = tensor("reduce_mean_117_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_117_keep_dims_0 = const()[name = tensor("reduce_mean_117_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_117_cast = reduce_mean(axes = reduce_mean_117_axes_0, keep_dims = reduce_mean_117_keep_dims_0, x = reshape_156_cast)[name = tensor("reduce_mean_117_cast")]; + tensor sub_78_cast = sub(x = reshape_156_cast, y = reduce_mean_117_cast)[name = tensor("sub_78_cast")]; + tensor square_39_cast = square(x = sub_78_cast)[name = tensor("square_39_cast")]; + tensor reduce_mean_119_axes_0 = const()[name = tensor("reduce_mean_119_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_119_keep_dims_0 = const()[name = tensor("reduce_mean_119_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_119_cast = reduce_mean(axes = reduce_mean_119_axes_0, keep_dims = reduce_mean_119_keep_dims_0, x = square_39_cast)[name = tensor("reduce_mean_119_cast")]; + tensor add_78_y_0_to_fp16 = const()[name = tensor("add_78_y_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_78_cast = add(x = reduce_mean_119_cast, y = add_78_y_0_to_fp16)[name = tensor("add_78_cast")]; + tensor sqrt_39_cast = sqrt(x = add_78_cast)[name = tensor("sqrt_39_cast")]; + tensor real_div_39_cast = real_div(x = sub_78_cast, y = sqrt_39_cast)[name = tensor("real_div_39_cast")]; + tensor reshape_157_shape_0 = const()[name = tensor("reshape_157_shape_0"), val = tensor([2, 1920, 16, 16])]; + tensor reshape_157_cast = reshape(shape = reshape_157_shape_0, x = real_div_39_cast)[name = tensor("reshape_157_cast")]; + tensor add_79_mean_0_to_fp16 = const()[name = tensor("add_79_mean_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(512650624)))]; + tensor add_79_variance_0_to_fp16 = const()[name = tensor("add_79_variance_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(512654528)))]; + tensor add_79_gamma_0_to_fp16 = const()[name = tensor("add_79_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(512658432)))]; + tensor add_79_beta_0_to_fp16 = const()[name = tensor("add_79_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(512662336)))]; + tensor add_79_epsilon_0_to_fp16 = const()[name = tensor("add_79_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_79_cast = batch_norm(beta = add_79_beta_0_to_fp16, epsilon = add_79_epsilon_0_to_fp16, gamma = add_79_gamma_0_to_fp16, mean = add_79_mean_0_to_fp16, variance = add_79_variance_0_to_fp16, x = reshape_157_cast)[name = tensor("add_79_cast")]; + tensor input_341_cast = silu(x = add_79_cast)[name = tensor("input_341_cast")]; + tensor var_9243 = const()[name = tensor("op_9243"), val = tensor([1, 1])]; + tensor var_9245 = const()[name = tensor("op_9245"), val = tensor([1, 1])]; + tensor hidden_states_199_pad_type_0 = const()[name = tensor("hidden_states_199_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_199_pad_0 = const()[name = tensor("hidden_states_199_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor up_blocks_1_resnets_2_conv1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(512666240))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(529255104))), name = tensor("up_blocks_1_resnets_2_conv1_weight_to_fp16_palettized"), shape = tensor([1280, 1920, 3, 3])]; + tensor up_blocks_1_resnets_2_conv1_bias_to_fp16 = const()[name = tensor("up_blocks_1_resnets_2_conv1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(529255296)))]; + tensor hidden_states_199_cast = conv(bias = up_blocks_1_resnets_2_conv1_bias_to_fp16, dilations = var_9245, groups = var_7228, pad = hidden_states_199_pad_0, pad_type = hidden_states_199_pad_type_0, strides = var_9243, weight = up_blocks_1_resnets_2_conv1_weight_to_fp16_palettized, x = input_341_cast)[name = tensor("hidden_states_199_cast")]; + tensor var_9251 = const()[name = tensor("op_9251"), val = tensor([1, 1])]; + tensor var_9253 = const()[name = tensor("op_9253"), val = tensor([1, 1])]; + tensor temb_31_pad_type_0 = const()[name = tensor("temb_31_pad_type_0"), val = tensor("custom")]; + tensor temb_31_pad_0 = const()[name = tensor("temb_31_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_1_resnets_2_time_emb_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(529257920))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(530486784))), name = tensor("up_blocks_1_resnets_2_time_emb_proj_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor up_blocks_1_resnets_2_time_emb_proj_bias_to_fp16 = const()[name = tensor("up_blocks_1_resnets_2_time_emb_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(530486976)))]; + tensor temb_31_cast = conv(bias = up_blocks_1_resnets_2_time_emb_proj_bias_to_fp16, dilations = var_9253, groups = var_7228, pad = temb_31_pad_0, pad_type = temb_31_pad_type_0, strides = var_9251, weight = up_blocks_1_resnets_2_time_emb_proj_weight_to_fp16_palettized, x = input_15_cast)[name = tensor("temb_31_cast")]; + tensor input_345_cast = add(x = hidden_states_199_cast, y = temb_31_cast)[name = tensor("input_345_cast")]; + tensor reshape_160_shape_0 = const()[name = tensor("reshape_160_shape_0"), val = tensor([2, 32, 40, 16, 16])]; + tensor reshape_160_cast = reshape(shape = reshape_160_shape_0, x = input_345_cast)[name = tensor("reshape_160_cast")]; + tensor reduce_mean_120_axes_0 = const()[name = tensor("reduce_mean_120_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_120_keep_dims_0 = const()[name = tensor("reduce_mean_120_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_120_cast = reduce_mean(axes = reduce_mean_120_axes_0, keep_dims = reduce_mean_120_keep_dims_0, x = reshape_160_cast)[name = tensor("reduce_mean_120_cast")]; + tensor sub_80_cast = sub(x = reshape_160_cast, y = reduce_mean_120_cast)[name = tensor("sub_80_cast")]; + tensor square_40_cast = square(x = sub_80_cast)[name = tensor("square_40_cast")]; + tensor reduce_mean_122_axes_0 = const()[name = tensor("reduce_mean_122_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_122_keep_dims_0 = const()[name = tensor("reduce_mean_122_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_122_cast = reduce_mean(axes = reduce_mean_122_axes_0, keep_dims = reduce_mean_122_keep_dims_0, x = square_40_cast)[name = tensor("reduce_mean_122_cast")]; + tensor add_80_y_0_to_fp16 = const()[name = tensor("add_80_y_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_80_cast = add(x = reduce_mean_122_cast, y = add_80_y_0_to_fp16)[name = tensor("add_80_cast")]; + tensor sqrt_40_cast = sqrt(x = add_80_cast)[name = tensor("sqrt_40_cast")]; + tensor real_div_40_cast = real_div(x = sub_80_cast, y = sqrt_40_cast)[name = tensor("real_div_40_cast")]; + tensor reshape_161_shape_0 = const()[name = tensor("reshape_161_shape_0"), val = tensor([2, 1280, 16, 16])]; + tensor reshape_161_cast = reshape(shape = reshape_161_shape_0, x = real_div_40_cast)[name = tensor("reshape_161_cast")]; + tensor add_81_gamma_0_to_fp16 = const()[name = tensor("add_81_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(530489600)))]; + tensor add_81_beta_0_to_fp16 = const()[name = tensor("add_81_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(530492224)))]; + tensor add_81_epsilon_0_to_fp16 = const()[name = tensor("add_81_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_81_cast = batch_norm(beta = add_81_beta_0_to_fp16, epsilon = add_81_epsilon_0_to_fp16, gamma = add_81_gamma_0_to_fp16, mean = add_27_mean_0_to_fp16, variance = add_27_variance_0_to_fp16, x = reshape_161_cast)[name = tensor("add_81_cast")]; + tensor input_349_cast = silu(x = add_81_cast)[name = tensor("input_349_cast")]; + tensor var_9263 = const()[name = tensor("op_9263"), val = tensor([1, 1])]; + tensor var_9265 = const()[name = tensor("op_9265"), val = tensor([1, 1])]; + tensor hidden_states_201_pad_type_0 = const()[name = tensor("hidden_states_201_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_201_pad_0 = const()[name = tensor("hidden_states_201_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor up_blocks_1_resnets_2_conv2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(530494848))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(541554112))), name = tensor("up_blocks_1_resnets_2_conv2_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 3, 3])]; + tensor up_blocks_1_resnets_2_conv2_bias_to_fp16 = const()[name = tensor("up_blocks_1_resnets_2_conv2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(541554304)))]; + tensor hidden_states_201_cast = conv(bias = up_blocks_1_resnets_2_conv2_bias_to_fp16, dilations = var_9265, groups = var_7228, pad = hidden_states_201_pad_0, pad_type = hidden_states_201_pad_type_0, strides = var_9263, weight = up_blocks_1_resnets_2_conv2_weight_to_fp16_palettized, x = input_349_cast)[name = tensor("hidden_states_201_cast")]; + tensor var_9270 = const()[name = tensor("op_9270"), val = tensor([1, 1])]; + tensor var_9272 = const()[name = tensor("op_9272"), val = tensor([1, 1])]; + tensor x_15_pad_type_0 = const()[name = tensor("x_15_pad_type_0"), val = tensor("custom")]; + tensor x_15_pad_0 = const()[name = tensor("x_15_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_1_resnets_2_conv_shortcut_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(541556928))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(543400192))), name = tensor("up_blocks_1_resnets_2_conv_shortcut_weight_to_fp16_palettized"), shape = tensor([1280, 1920, 1, 1])]; + tensor up_blocks_1_resnets_2_conv_shortcut_bias_to_fp16 = const()[name = tensor("up_blocks_1_resnets_2_conv_shortcut_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(543400384)))]; + tensor x_15_cast = conv(bias = up_blocks_1_resnets_2_conv_shortcut_bias_to_fp16, dilations = var_9272, groups = var_7228, pad = x_15_pad_0, pad_type = x_15_pad_type_0, strides = var_9270, weight = up_blocks_1_resnets_2_conv_shortcut_weight_to_fp16_palettized, x = input_337_cast)[name = tensor("x_15_cast")]; + tensor hidden_states_203_cast = add(x = x_15_cast, y = hidden_states_201_cast)[name = tensor("hidden_states_203_cast")]; + tensor reshape_164_shape_0 = const()[name = tensor("reshape_164_shape_0"), val = tensor([2, 32, 40, 16, 16])]; + tensor reshape_164_cast = reshape(shape = reshape_164_shape_0, x = hidden_states_203_cast)[name = tensor("reshape_164_cast")]; + tensor reduce_mean_123_axes_0 = const()[name = tensor("reduce_mean_123_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_123_keep_dims_0 = const()[name = tensor("reduce_mean_123_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_123_cast = reduce_mean(axes = reduce_mean_123_axes_0, keep_dims = reduce_mean_123_keep_dims_0, x = reshape_164_cast)[name = tensor("reduce_mean_123_cast")]; + tensor sub_82_cast = sub(x = reshape_164_cast, y = reduce_mean_123_cast)[name = tensor("sub_82_cast")]; + tensor square_41_cast = square(x = sub_82_cast)[name = tensor("square_41_cast")]; + tensor reduce_mean_125_axes_0 = const()[name = tensor("reduce_mean_125_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_125_keep_dims_0 = const()[name = tensor("reduce_mean_125_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_125_cast = reduce_mean(axes = reduce_mean_125_axes_0, keep_dims = reduce_mean_125_keep_dims_0, x = square_41_cast)[name = tensor("reduce_mean_125_cast")]; + tensor add_82_y_0_to_fp16 = const()[name = tensor("add_82_y_0_to_fp16"), val = tensor(0x1.1p-20)]; + tensor add_82_cast = add(x = reduce_mean_125_cast, y = add_82_y_0_to_fp16)[name = tensor("add_82_cast")]; + tensor sqrt_41_cast = sqrt(x = add_82_cast)[name = tensor("sqrt_41_cast")]; + tensor real_div_41_cast = real_div(x = sub_82_cast, y = sqrt_41_cast)[name = tensor("real_div_41_cast")]; + tensor reshape_165_shape_0 = const()[name = tensor("reshape_165_shape_0"), val = tensor([2, 1280, 16, 16])]; + tensor reshape_165_cast = reshape(shape = reshape_165_shape_0, x = real_div_41_cast)[name = tensor("reshape_165_cast")]; + tensor add_83_gamma_0_to_fp16 = const()[name = tensor("add_83_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(543403008)))]; + tensor add_83_beta_0_to_fp16 = const()[name = tensor("add_83_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(543405632)))]; + tensor add_83_epsilon_0_to_fp16 = const()[name = tensor("add_83_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_83_cast = batch_norm(beta = add_83_beta_0_to_fp16, epsilon = add_83_epsilon_0_to_fp16, gamma = add_83_gamma_0_to_fp16, mean = add_27_mean_0_to_fp16, variance = add_27_variance_0_to_fp16, x = reshape_165_cast)[name = tensor("add_83_cast")]; + tensor var_9292 = const()[name = tensor("op_9292"), val = tensor([1, 1])]; + tensor var_9294 = const()[name = tensor("op_9294"), val = tensor([1, 1])]; + tensor hidden_states_205_pad_type_0 = const()[name = tensor("hidden_states_205_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_205_pad_0 = const()[name = tensor("hidden_states_205_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_1_attentions_2_proj_in_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(543408256))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(544637120))), name = tensor("up_blocks_1_attentions_2_proj_in_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor up_blocks_1_attentions_2_proj_in_bias_to_fp16 = const()[name = tensor("up_blocks_1_attentions_2_proj_in_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(544637312)))]; + tensor hidden_states_205_cast = conv(bias = up_blocks_1_attentions_2_proj_in_bias_to_fp16, dilations = var_9294, groups = var_7228, pad = hidden_states_205_pad_0, pad_type = hidden_states_205_pad_type_0, strides = var_9292, weight = up_blocks_1_attentions_2_proj_in_weight_to_fp16_palettized, x = add_83_cast)[name = tensor("hidden_states_205_cast")]; + tensor var_9299 = const()[name = tensor("op_9299"), val = tensor([2, 1280, 1, 256])]; + tensor inputs_55_cast = reshape(shape = var_9299, x = hidden_states_205_cast)[name = tensor("inputs_55_cast")]; + tensor var_9309 = const()[name = tensor("op_9309"), val = tensor([1])]; + tensor channels_mean_55_cast = reduce_mean(axes = var_9309, keep_dims = var_7223, x = inputs_55_cast)[name = tensor("channels_mean_55_cast")]; + tensor zero_mean_55_cast = sub(x = inputs_55_cast, y = channels_mean_55_cast)[name = tensor("zero_mean_55_cast")]; + tensor zero_mean_sq_55_cast = mul(x = zero_mean_55_cast, y = zero_mean_55_cast)[name = tensor("zero_mean_sq_55_cast")]; + tensor var_9313 = const()[name = tensor("op_9313"), val = tensor([1])]; + tensor var_9314_cast = reduce_mean(axes = var_9313, keep_dims = var_7223, x = zero_mean_sq_55_cast)[name = tensor("op_9314_cast")]; + tensor var_9315_to_fp16 = const()[name = tensor("op_9315_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_9316_cast = add(x = var_9314_cast, y = var_9315_to_fp16)[name = tensor("op_9316_cast")]; + tensor denom_55_epsilon_0_to_fp16 = const()[name = tensor("denom_55_epsilon_0_to_fp16"), val = tensor(0x1p-24)]; + tensor denom_55_cast = rsqrt(epsilon = denom_55_epsilon_0_to_fp16, x = var_9316_cast)[name = tensor("denom_55_cast")]; + tensor out_55_cast = mul(x = zero_mean_55_cast, y = denom_55_cast)[name = tensor("out_55_cast")]; + tensor var_9320_to_fp16 = const()[name = tensor("op_9320_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(544639936)))]; + tensor var_9321_cast = add(x = out_55_cast, y = var_9320_to_fp16)[name = tensor("op_9321_cast")]; + tensor var_9323_to_fp16 = const()[name = tensor("op_9323_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(544642560)))]; + tensor hidden_states_207_cast = mul(x = var_9321_cast, y = var_9323_to_fp16)[name = tensor("hidden_states_207_cast")]; + tensor var_9330 = const()[name = tensor("op_9330"), val = tensor([1, 1])]; + tensor var_9332 = const()[name = tensor("op_9332"), val = tensor([1, 1])]; + tensor q_37_pad_type_0 = const()[name = tensor("q_37_pad_type_0"), val = tensor("custom")]; + tensor q_37_pad_0 = const()[name = tensor("q_37_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_1_attentions_2_transformer_blocks_0_attn1_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(544645184))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(545874048))), name = tensor("up_blocks_1_attentions_2_transformer_blocks_0_attn1_to_q_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor q_37_cast = conv(dilations = var_9332, groups = var_7228, pad = q_37_pad_0, pad_type = q_37_pad_type_0, strides = var_9330, weight = up_blocks_1_attentions_2_transformer_blocks_0_attn1_to_q_weight_to_fp16_palettized, x = hidden_states_207_cast)[name = tensor("q_37_cast")]; + tensor var_9336 = const()[name = tensor("op_9336"), val = tensor([1, 1])]; + tensor var_9338 = const()[name = tensor("op_9338"), val = tensor([1, 1])]; + tensor k_73_pad_type_0 = const()[name = tensor("k_73_pad_type_0"), val = tensor("custom")]; + tensor k_73_pad_0 = const()[name = tensor("k_73_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_1_attentions_2_transformer_blocks_0_attn1_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(545874240))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(547103104))), name = tensor("up_blocks_1_attentions_2_transformer_blocks_0_attn1_to_k_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor k_73_cast = conv(dilations = var_9338, groups = var_7228, pad = k_73_pad_0, pad_type = k_73_pad_type_0, strides = var_9336, weight = up_blocks_1_attentions_2_transformer_blocks_0_attn1_to_k_weight_to_fp16_palettized, x = hidden_states_207_cast)[name = tensor("k_73_cast")]; + tensor var_9342 = const()[name = tensor("op_9342"), val = tensor([1, 1])]; + tensor var_9344 = const()[name = tensor("op_9344"), val = tensor([1, 1])]; + tensor v_37_pad_type_0 = const()[name = tensor("v_37_pad_type_0"), val = tensor("custom")]; + tensor v_37_pad_0 = const()[name = tensor("v_37_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_1_attentions_2_transformer_blocks_0_attn1_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(547103296))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(548332160))), name = tensor("up_blocks_1_attentions_2_transformer_blocks_0_attn1_to_v_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor v_37_cast = conv(dilations = var_9344, groups = var_7228, pad = v_37_pad_0, pad_type = v_37_pad_type_0, strides = var_9342, weight = up_blocks_1_attentions_2_transformer_blocks_0_attn1_to_v_weight_to_fp16_palettized, x = hidden_states_207_cast)[name = tensor("v_37_cast")]; + tensor var_9348_begin_0 = const()[name = tensor("op_9348_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_9348_end_0 = const()[name = tensor("op_9348_end_0"), val = tensor([2, 64, 1, 256])]; + tensor var_9348_end_mask_0 = const()[name = tensor("op_9348_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9348_cast = slice_by_index(begin = var_9348_begin_0, end = var_9348_end_0, end_mask = var_9348_end_mask_0, x = q_37_cast)[name = tensor("op_9348_cast")]; + tensor var_9352_begin_0 = const()[name = tensor("op_9352_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_9352_end_0 = const()[name = tensor("op_9352_end_0"), val = tensor([2, 128, 1, 256])]; + tensor var_9352_end_mask_0 = const()[name = tensor("op_9352_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9352_cast = slice_by_index(begin = var_9352_begin_0, end = var_9352_end_0, end_mask = var_9352_end_mask_0, x = q_37_cast)[name = tensor("op_9352_cast")]; + tensor var_9356_begin_0 = const()[name = tensor("op_9356_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_9356_end_0 = const()[name = tensor("op_9356_end_0"), val = tensor([2, 192, 1, 256])]; + tensor var_9356_end_mask_0 = const()[name = tensor("op_9356_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9356_cast = slice_by_index(begin = var_9356_begin_0, end = var_9356_end_0, end_mask = var_9356_end_mask_0, x = q_37_cast)[name = tensor("op_9356_cast")]; + tensor var_9360_begin_0 = const()[name = tensor("op_9360_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_9360_end_0 = const()[name = tensor("op_9360_end_0"), val = tensor([2, 256, 1, 256])]; + tensor var_9360_end_mask_0 = const()[name = tensor("op_9360_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9360_cast = slice_by_index(begin = var_9360_begin_0, end = var_9360_end_0, end_mask = var_9360_end_mask_0, x = q_37_cast)[name = tensor("op_9360_cast")]; + tensor var_9364_begin_0 = const()[name = tensor("op_9364_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_9364_end_0 = const()[name = tensor("op_9364_end_0"), val = tensor([2, 320, 1, 256])]; + tensor var_9364_end_mask_0 = const()[name = tensor("op_9364_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9364_cast = slice_by_index(begin = var_9364_begin_0, end = var_9364_end_0, end_mask = var_9364_end_mask_0, x = q_37_cast)[name = tensor("op_9364_cast")]; + tensor var_9368_begin_0 = const()[name = tensor("op_9368_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_9368_end_0 = const()[name = tensor("op_9368_end_0"), val = tensor([2, 384, 1, 256])]; + tensor var_9368_end_mask_0 = const()[name = tensor("op_9368_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9368_cast = slice_by_index(begin = var_9368_begin_0, end = var_9368_end_0, end_mask = var_9368_end_mask_0, x = q_37_cast)[name = tensor("op_9368_cast")]; + tensor var_9372_begin_0 = const()[name = tensor("op_9372_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_9372_end_0 = const()[name = tensor("op_9372_end_0"), val = tensor([2, 448, 1, 256])]; + tensor var_9372_end_mask_0 = const()[name = tensor("op_9372_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9372_cast = slice_by_index(begin = var_9372_begin_0, end = var_9372_end_0, end_mask = var_9372_end_mask_0, x = q_37_cast)[name = tensor("op_9372_cast")]; + tensor var_9376_begin_0 = const()[name = tensor("op_9376_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_9376_end_0 = const()[name = tensor("op_9376_end_0"), val = tensor([2, 512, 1, 256])]; + tensor var_9376_end_mask_0 = const()[name = tensor("op_9376_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9376_cast = slice_by_index(begin = var_9376_begin_0, end = var_9376_end_0, end_mask = var_9376_end_mask_0, x = q_37_cast)[name = tensor("op_9376_cast")]; + tensor var_9380_begin_0 = const()[name = tensor("op_9380_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_9380_end_0 = const()[name = tensor("op_9380_end_0"), val = tensor([2, 576, 1, 256])]; + tensor var_9380_end_mask_0 = const()[name = tensor("op_9380_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9380_cast = slice_by_index(begin = var_9380_begin_0, end = var_9380_end_0, end_mask = var_9380_end_mask_0, x = q_37_cast)[name = tensor("op_9380_cast")]; + tensor var_9384_begin_0 = const()[name = tensor("op_9384_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_9384_end_0 = const()[name = tensor("op_9384_end_0"), val = tensor([2, 640, 1, 256])]; + tensor var_9384_end_mask_0 = const()[name = tensor("op_9384_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9384_cast = slice_by_index(begin = var_9384_begin_0, end = var_9384_end_0, end_mask = var_9384_end_mask_0, x = q_37_cast)[name = tensor("op_9384_cast")]; + tensor var_9388_begin_0 = const()[name = tensor("op_9388_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_9388_end_0 = const()[name = tensor("op_9388_end_0"), val = tensor([2, 704, 1, 256])]; + tensor var_9388_end_mask_0 = const()[name = tensor("op_9388_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9388_cast = slice_by_index(begin = var_9388_begin_0, end = var_9388_end_0, end_mask = var_9388_end_mask_0, x = q_37_cast)[name = tensor("op_9388_cast")]; + tensor var_9392_begin_0 = const()[name = tensor("op_9392_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_9392_end_0 = const()[name = tensor("op_9392_end_0"), val = tensor([2, 768, 1, 256])]; + tensor var_9392_end_mask_0 = const()[name = tensor("op_9392_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9392_cast = slice_by_index(begin = var_9392_begin_0, end = var_9392_end_0, end_mask = var_9392_end_mask_0, x = q_37_cast)[name = tensor("op_9392_cast")]; + tensor var_9396_begin_0 = const()[name = tensor("op_9396_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_9396_end_0 = const()[name = tensor("op_9396_end_0"), val = tensor([2, 832, 1, 256])]; + tensor var_9396_end_mask_0 = const()[name = tensor("op_9396_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9396_cast = slice_by_index(begin = var_9396_begin_0, end = var_9396_end_0, end_mask = var_9396_end_mask_0, x = q_37_cast)[name = tensor("op_9396_cast")]; + tensor var_9400_begin_0 = const()[name = tensor("op_9400_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_9400_end_0 = const()[name = tensor("op_9400_end_0"), val = tensor([2, 896, 1, 256])]; + tensor var_9400_end_mask_0 = const()[name = tensor("op_9400_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9400_cast = slice_by_index(begin = var_9400_begin_0, end = var_9400_end_0, end_mask = var_9400_end_mask_0, x = q_37_cast)[name = tensor("op_9400_cast")]; + tensor var_9404_begin_0 = const()[name = tensor("op_9404_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_9404_end_0 = const()[name = tensor("op_9404_end_0"), val = tensor([2, 960, 1, 256])]; + tensor var_9404_end_mask_0 = const()[name = tensor("op_9404_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9404_cast = slice_by_index(begin = var_9404_begin_0, end = var_9404_end_0, end_mask = var_9404_end_mask_0, x = q_37_cast)[name = tensor("op_9404_cast")]; + tensor var_9408_begin_0 = const()[name = tensor("op_9408_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_9408_end_0 = const()[name = tensor("op_9408_end_0"), val = tensor([2, 1024, 1, 256])]; + tensor var_9408_end_mask_0 = const()[name = tensor("op_9408_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9408_cast = slice_by_index(begin = var_9408_begin_0, end = var_9408_end_0, end_mask = var_9408_end_mask_0, x = q_37_cast)[name = tensor("op_9408_cast")]; + tensor var_9412_begin_0 = const()[name = tensor("op_9412_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_9412_end_0 = const()[name = tensor("op_9412_end_0"), val = tensor([2, 1088, 1, 256])]; + tensor var_9412_end_mask_0 = const()[name = tensor("op_9412_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9412_cast = slice_by_index(begin = var_9412_begin_0, end = var_9412_end_0, end_mask = var_9412_end_mask_0, x = q_37_cast)[name = tensor("op_9412_cast")]; + tensor var_9416_begin_0 = const()[name = tensor("op_9416_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_9416_end_0 = const()[name = tensor("op_9416_end_0"), val = tensor([2, 1152, 1, 256])]; + tensor var_9416_end_mask_0 = const()[name = tensor("op_9416_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9416_cast = slice_by_index(begin = var_9416_begin_0, end = var_9416_end_0, end_mask = var_9416_end_mask_0, x = q_37_cast)[name = tensor("op_9416_cast")]; + tensor var_9420_begin_0 = const()[name = tensor("op_9420_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_9420_end_0 = const()[name = tensor("op_9420_end_0"), val = tensor([2, 1216, 1, 256])]; + tensor var_9420_end_mask_0 = const()[name = tensor("op_9420_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9420_cast = slice_by_index(begin = var_9420_begin_0, end = var_9420_end_0, end_mask = var_9420_end_mask_0, x = q_37_cast)[name = tensor("op_9420_cast")]; + tensor var_9424_begin_0 = const()[name = tensor("op_9424_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_9424_end_0 = const()[name = tensor("op_9424_end_0"), val = tensor([2, 1280, 1, 256])]; + tensor var_9424_end_mask_0 = const()[name = tensor("op_9424_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9424_cast = slice_by_index(begin = var_9424_begin_0, end = var_9424_end_0, end_mask = var_9424_end_mask_0, x = q_37_cast)[name = tensor("op_9424_cast")]; + tensor k_75_perm_0 = const()[name = tensor("k_75_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_9431_begin_0 = const()[name = tensor("op_9431_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_9431_end_0 = const()[name = tensor("op_9431_end_0"), val = tensor([2, 256, 1, 64])]; + tensor var_9431_end_mask_0 = const()[name = tensor("op_9431_end_mask_0"), val = tensor([true, true, true, false])]; + tensor transpose_13 = transpose(perm = k_75_perm_0, x = k_73_cast)[name = tensor("transpose_13")]; + tensor var_9431_cast = slice_by_index(begin = var_9431_begin_0, end = var_9431_end_0, end_mask = var_9431_end_mask_0, x = transpose_13)[name = tensor("op_9431_cast")]; + tensor var_9435_begin_0 = const()[name = tensor("op_9435_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_9435_end_0 = const()[name = tensor("op_9435_end_0"), val = tensor([2, 256, 1, 128])]; + tensor var_9435_end_mask_0 = const()[name = tensor("op_9435_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9435_cast = slice_by_index(begin = var_9435_begin_0, end = var_9435_end_0, end_mask = var_9435_end_mask_0, x = transpose_13)[name = tensor("op_9435_cast")]; + tensor var_9439_begin_0 = const()[name = tensor("op_9439_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_9439_end_0 = const()[name = tensor("op_9439_end_0"), val = tensor([2, 256, 1, 192])]; + tensor var_9439_end_mask_0 = const()[name = tensor("op_9439_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9439_cast = slice_by_index(begin = var_9439_begin_0, end = var_9439_end_0, end_mask = var_9439_end_mask_0, x = transpose_13)[name = tensor("op_9439_cast")]; + tensor var_9443_begin_0 = const()[name = tensor("op_9443_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_9443_end_0 = const()[name = tensor("op_9443_end_0"), val = tensor([2, 256, 1, 256])]; + tensor var_9443_end_mask_0 = const()[name = tensor("op_9443_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9443_cast = slice_by_index(begin = var_9443_begin_0, end = var_9443_end_0, end_mask = var_9443_end_mask_0, x = transpose_13)[name = tensor("op_9443_cast")]; + tensor var_9447_begin_0 = const()[name = tensor("op_9447_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_9447_end_0 = const()[name = tensor("op_9447_end_0"), val = tensor([2, 256, 1, 320])]; + tensor var_9447_end_mask_0 = const()[name = tensor("op_9447_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9447_cast = slice_by_index(begin = var_9447_begin_0, end = var_9447_end_0, end_mask = var_9447_end_mask_0, x = transpose_13)[name = tensor("op_9447_cast")]; + tensor var_9451_begin_0 = const()[name = tensor("op_9451_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_9451_end_0 = const()[name = tensor("op_9451_end_0"), val = tensor([2, 256, 1, 384])]; + tensor var_9451_end_mask_0 = const()[name = tensor("op_9451_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9451_cast = slice_by_index(begin = var_9451_begin_0, end = var_9451_end_0, end_mask = var_9451_end_mask_0, x = transpose_13)[name = tensor("op_9451_cast")]; + tensor var_9455_begin_0 = const()[name = tensor("op_9455_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_9455_end_0 = const()[name = tensor("op_9455_end_0"), val = tensor([2, 256, 1, 448])]; + tensor var_9455_end_mask_0 = const()[name = tensor("op_9455_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9455_cast = slice_by_index(begin = var_9455_begin_0, end = var_9455_end_0, end_mask = var_9455_end_mask_0, x = transpose_13)[name = tensor("op_9455_cast")]; + tensor var_9459_begin_0 = const()[name = tensor("op_9459_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_9459_end_0 = const()[name = tensor("op_9459_end_0"), val = tensor([2, 256, 1, 512])]; + tensor var_9459_end_mask_0 = const()[name = tensor("op_9459_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9459_cast = slice_by_index(begin = var_9459_begin_0, end = var_9459_end_0, end_mask = var_9459_end_mask_0, x = transpose_13)[name = tensor("op_9459_cast")]; + tensor var_9463_begin_0 = const()[name = tensor("op_9463_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_9463_end_0 = const()[name = tensor("op_9463_end_0"), val = tensor([2, 256, 1, 576])]; + tensor var_9463_end_mask_0 = const()[name = tensor("op_9463_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9463_cast = slice_by_index(begin = var_9463_begin_0, end = var_9463_end_0, end_mask = var_9463_end_mask_0, x = transpose_13)[name = tensor("op_9463_cast")]; + tensor var_9467_begin_0 = const()[name = tensor("op_9467_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_9467_end_0 = const()[name = tensor("op_9467_end_0"), val = tensor([2, 256, 1, 640])]; + tensor var_9467_end_mask_0 = const()[name = tensor("op_9467_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9467_cast = slice_by_index(begin = var_9467_begin_0, end = var_9467_end_0, end_mask = var_9467_end_mask_0, x = transpose_13)[name = tensor("op_9467_cast")]; + tensor var_9471_begin_0 = const()[name = tensor("op_9471_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_9471_end_0 = const()[name = tensor("op_9471_end_0"), val = tensor([2, 256, 1, 704])]; + tensor var_9471_end_mask_0 = const()[name = tensor("op_9471_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9471_cast = slice_by_index(begin = var_9471_begin_0, end = var_9471_end_0, end_mask = var_9471_end_mask_0, x = transpose_13)[name = tensor("op_9471_cast")]; + tensor var_9475_begin_0 = const()[name = tensor("op_9475_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_9475_end_0 = const()[name = tensor("op_9475_end_0"), val = tensor([2, 256, 1, 768])]; + tensor var_9475_end_mask_0 = const()[name = tensor("op_9475_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9475_cast = slice_by_index(begin = var_9475_begin_0, end = var_9475_end_0, end_mask = var_9475_end_mask_0, x = transpose_13)[name = tensor("op_9475_cast")]; + tensor var_9479_begin_0 = const()[name = tensor("op_9479_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_9479_end_0 = const()[name = tensor("op_9479_end_0"), val = tensor([2, 256, 1, 832])]; + tensor var_9479_end_mask_0 = const()[name = tensor("op_9479_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9479_cast = slice_by_index(begin = var_9479_begin_0, end = var_9479_end_0, end_mask = var_9479_end_mask_0, x = transpose_13)[name = tensor("op_9479_cast")]; + tensor var_9483_begin_0 = const()[name = tensor("op_9483_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_9483_end_0 = const()[name = tensor("op_9483_end_0"), val = tensor([2, 256, 1, 896])]; + tensor var_9483_end_mask_0 = const()[name = tensor("op_9483_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9483_cast = slice_by_index(begin = var_9483_begin_0, end = var_9483_end_0, end_mask = var_9483_end_mask_0, x = transpose_13)[name = tensor("op_9483_cast")]; + tensor var_9487_begin_0 = const()[name = tensor("op_9487_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_9487_end_0 = const()[name = tensor("op_9487_end_0"), val = tensor([2, 256, 1, 960])]; + tensor var_9487_end_mask_0 = const()[name = tensor("op_9487_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9487_cast = slice_by_index(begin = var_9487_begin_0, end = var_9487_end_0, end_mask = var_9487_end_mask_0, x = transpose_13)[name = tensor("op_9487_cast")]; + tensor var_9491_begin_0 = const()[name = tensor("op_9491_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_9491_end_0 = const()[name = tensor("op_9491_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_9491_end_mask_0 = const()[name = tensor("op_9491_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9491_cast = slice_by_index(begin = var_9491_begin_0, end = var_9491_end_0, end_mask = var_9491_end_mask_0, x = transpose_13)[name = tensor("op_9491_cast")]; + tensor var_9495_begin_0 = const()[name = tensor("op_9495_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_9495_end_0 = const()[name = tensor("op_9495_end_0"), val = tensor([2, 256, 1, 1088])]; + tensor var_9495_end_mask_0 = const()[name = tensor("op_9495_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9495_cast = slice_by_index(begin = var_9495_begin_0, end = var_9495_end_0, end_mask = var_9495_end_mask_0, x = transpose_13)[name = tensor("op_9495_cast")]; + tensor var_9499_begin_0 = const()[name = tensor("op_9499_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_9499_end_0 = const()[name = tensor("op_9499_end_0"), val = tensor([2, 256, 1, 1152])]; + tensor var_9499_end_mask_0 = const()[name = tensor("op_9499_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9499_cast = slice_by_index(begin = var_9499_begin_0, end = var_9499_end_0, end_mask = var_9499_end_mask_0, x = transpose_13)[name = tensor("op_9499_cast")]; + tensor var_9503_begin_0 = const()[name = tensor("op_9503_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_9503_end_0 = const()[name = tensor("op_9503_end_0"), val = tensor([2, 256, 1, 1216])]; + tensor var_9503_end_mask_0 = const()[name = tensor("op_9503_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9503_cast = slice_by_index(begin = var_9503_begin_0, end = var_9503_end_0, end_mask = var_9503_end_mask_0, x = transpose_13)[name = tensor("op_9503_cast")]; + tensor var_9507_begin_0 = const()[name = tensor("op_9507_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_9507_end_0 = const()[name = tensor("op_9507_end_0"), val = tensor([2, 256, 1, 1280])]; + tensor var_9507_end_mask_0 = const()[name = tensor("op_9507_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9507_cast = slice_by_index(begin = var_9507_begin_0, end = var_9507_end_0, end_mask = var_9507_end_mask_0, x = transpose_13)[name = tensor("op_9507_cast")]; + tensor var_9509_begin_0 = const()[name = tensor("op_9509_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_9509_end_0 = const()[name = tensor("op_9509_end_0"), val = tensor([2, 64, 1, 256])]; + tensor var_9509_end_mask_0 = const()[name = tensor("op_9509_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9509_cast = slice_by_index(begin = var_9509_begin_0, end = var_9509_end_0, end_mask = var_9509_end_mask_0, x = v_37_cast)[name = tensor("op_9509_cast")]; + tensor var_9513_begin_0 = const()[name = tensor("op_9513_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_9513_end_0 = const()[name = tensor("op_9513_end_0"), val = tensor([2, 128, 1, 256])]; + tensor var_9513_end_mask_0 = const()[name = tensor("op_9513_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9513_cast = slice_by_index(begin = var_9513_begin_0, end = var_9513_end_0, end_mask = var_9513_end_mask_0, x = v_37_cast)[name = tensor("op_9513_cast")]; + tensor var_9517_begin_0 = const()[name = tensor("op_9517_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_9517_end_0 = const()[name = tensor("op_9517_end_0"), val = tensor([2, 192, 1, 256])]; + tensor var_9517_end_mask_0 = const()[name = tensor("op_9517_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9517_cast = slice_by_index(begin = var_9517_begin_0, end = var_9517_end_0, end_mask = var_9517_end_mask_0, x = v_37_cast)[name = tensor("op_9517_cast")]; + tensor var_9521_begin_0 = const()[name = tensor("op_9521_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_9521_end_0 = const()[name = tensor("op_9521_end_0"), val = tensor([2, 256, 1, 256])]; + tensor var_9521_end_mask_0 = const()[name = tensor("op_9521_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9521_cast = slice_by_index(begin = var_9521_begin_0, end = var_9521_end_0, end_mask = var_9521_end_mask_0, x = v_37_cast)[name = tensor("op_9521_cast")]; + tensor var_9525_begin_0 = const()[name = tensor("op_9525_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_9525_end_0 = const()[name = tensor("op_9525_end_0"), val = tensor([2, 320, 1, 256])]; + tensor var_9525_end_mask_0 = const()[name = tensor("op_9525_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9525_cast = slice_by_index(begin = var_9525_begin_0, end = var_9525_end_0, end_mask = var_9525_end_mask_0, x = v_37_cast)[name = tensor("op_9525_cast")]; + tensor var_9529_begin_0 = const()[name = tensor("op_9529_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_9529_end_0 = const()[name = tensor("op_9529_end_0"), val = tensor([2, 384, 1, 256])]; + tensor var_9529_end_mask_0 = const()[name = tensor("op_9529_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9529_cast = slice_by_index(begin = var_9529_begin_0, end = var_9529_end_0, end_mask = var_9529_end_mask_0, x = v_37_cast)[name = tensor("op_9529_cast")]; + tensor var_9533_begin_0 = const()[name = tensor("op_9533_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_9533_end_0 = const()[name = tensor("op_9533_end_0"), val = tensor([2, 448, 1, 256])]; + tensor var_9533_end_mask_0 = const()[name = tensor("op_9533_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9533_cast = slice_by_index(begin = var_9533_begin_0, end = var_9533_end_0, end_mask = var_9533_end_mask_0, x = v_37_cast)[name = tensor("op_9533_cast")]; + tensor var_9537_begin_0 = const()[name = tensor("op_9537_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_9537_end_0 = const()[name = tensor("op_9537_end_0"), val = tensor([2, 512, 1, 256])]; + tensor var_9537_end_mask_0 = const()[name = tensor("op_9537_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9537_cast = slice_by_index(begin = var_9537_begin_0, end = var_9537_end_0, end_mask = var_9537_end_mask_0, x = v_37_cast)[name = tensor("op_9537_cast")]; + tensor var_9541_begin_0 = const()[name = tensor("op_9541_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_9541_end_0 = const()[name = tensor("op_9541_end_0"), val = tensor([2, 576, 1, 256])]; + tensor var_9541_end_mask_0 = const()[name = tensor("op_9541_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9541_cast = slice_by_index(begin = var_9541_begin_0, end = var_9541_end_0, end_mask = var_9541_end_mask_0, x = v_37_cast)[name = tensor("op_9541_cast")]; + tensor var_9545_begin_0 = const()[name = tensor("op_9545_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_9545_end_0 = const()[name = tensor("op_9545_end_0"), val = tensor([2, 640, 1, 256])]; + tensor var_9545_end_mask_0 = const()[name = tensor("op_9545_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9545_cast = slice_by_index(begin = var_9545_begin_0, end = var_9545_end_0, end_mask = var_9545_end_mask_0, x = v_37_cast)[name = tensor("op_9545_cast")]; + tensor var_9549_begin_0 = const()[name = tensor("op_9549_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_9549_end_0 = const()[name = tensor("op_9549_end_0"), val = tensor([2, 704, 1, 256])]; + tensor var_9549_end_mask_0 = const()[name = tensor("op_9549_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9549_cast = slice_by_index(begin = var_9549_begin_0, end = var_9549_end_0, end_mask = var_9549_end_mask_0, x = v_37_cast)[name = tensor("op_9549_cast")]; + tensor var_9553_begin_0 = const()[name = tensor("op_9553_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_9553_end_0 = const()[name = tensor("op_9553_end_0"), val = tensor([2, 768, 1, 256])]; + tensor var_9553_end_mask_0 = const()[name = tensor("op_9553_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9553_cast = slice_by_index(begin = var_9553_begin_0, end = var_9553_end_0, end_mask = var_9553_end_mask_0, x = v_37_cast)[name = tensor("op_9553_cast")]; + tensor var_9557_begin_0 = const()[name = tensor("op_9557_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_9557_end_0 = const()[name = tensor("op_9557_end_0"), val = tensor([2, 832, 1, 256])]; + tensor var_9557_end_mask_0 = const()[name = tensor("op_9557_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9557_cast = slice_by_index(begin = var_9557_begin_0, end = var_9557_end_0, end_mask = var_9557_end_mask_0, x = v_37_cast)[name = tensor("op_9557_cast")]; + tensor var_9561_begin_0 = const()[name = tensor("op_9561_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_9561_end_0 = const()[name = tensor("op_9561_end_0"), val = tensor([2, 896, 1, 256])]; + tensor var_9561_end_mask_0 = const()[name = tensor("op_9561_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9561_cast = slice_by_index(begin = var_9561_begin_0, end = var_9561_end_0, end_mask = var_9561_end_mask_0, x = v_37_cast)[name = tensor("op_9561_cast")]; + tensor var_9565_begin_0 = const()[name = tensor("op_9565_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_9565_end_0 = const()[name = tensor("op_9565_end_0"), val = tensor([2, 960, 1, 256])]; + tensor var_9565_end_mask_0 = const()[name = tensor("op_9565_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9565_cast = slice_by_index(begin = var_9565_begin_0, end = var_9565_end_0, end_mask = var_9565_end_mask_0, x = v_37_cast)[name = tensor("op_9565_cast")]; + tensor var_9569_begin_0 = const()[name = tensor("op_9569_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_9569_end_0 = const()[name = tensor("op_9569_end_0"), val = tensor([2, 1024, 1, 256])]; + tensor var_9569_end_mask_0 = const()[name = tensor("op_9569_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9569_cast = slice_by_index(begin = var_9569_begin_0, end = var_9569_end_0, end_mask = var_9569_end_mask_0, x = v_37_cast)[name = tensor("op_9569_cast")]; + tensor var_9573_begin_0 = const()[name = tensor("op_9573_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_9573_end_0 = const()[name = tensor("op_9573_end_0"), val = tensor([2, 1088, 1, 256])]; + tensor var_9573_end_mask_0 = const()[name = tensor("op_9573_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9573_cast = slice_by_index(begin = var_9573_begin_0, end = var_9573_end_0, end_mask = var_9573_end_mask_0, x = v_37_cast)[name = tensor("op_9573_cast")]; + tensor var_9577_begin_0 = const()[name = tensor("op_9577_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_9577_end_0 = const()[name = tensor("op_9577_end_0"), val = tensor([2, 1152, 1, 256])]; + tensor var_9577_end_mask_0 = const()[name = tensor("op_9577_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9577_cast = slice_by_index(begin = var_9577_begin_0, end = var_9577_end_0, end_mask = var_9577_end_mask_0, x = v_37_cast)[name = tensor("op_9577_cast")]; + tensor var_9581_begin_0 = const()[name = tensor("op_9581_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_9581_end_0 = const()[name = tensor("op_9581_end_0"), val = tensor([2, 1216, 1, 256])]; + tensor var_9581_end_mask_0 = const()[name = tensor("op_9581_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9581_cast = slice_by_index(begin = var_9581_begin_0, end = var_9581_end_0, end_mask = var_9581_end_mask_0, x = v_37_cast)[name = tensor("op_9581_cast")]; + tensor var_9585_begin_0 = const()[name = tensor("op_9585_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_9585_end_0 = const()[name = tensor("op_9585_end_0"), val = tensor([2, 1280, 1, 256])]; + tensor var_9585_end_mask_0 = const()[name = tensor("op_9585_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9585_cast = slice_by_index(begin = var_9585_begin_0, end = var_9585_end_0, end_mask = var_9585_end_mask_0, x = v_37_cast)[name = tensor("op_9585_cast")]; + tensor var_9589_equation_0 = const()[name = tensor("op_9589_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9589_cast = einsum(equation = var_9589_equation_0, values = (var_9431_cast, var_9348_cast))[name = tensor("op_9589_cast")]; + tensor var_9590_to_fp16 = const()[name = tensor("op_9590_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_401_cast = mul(x = var_9589_cast, y = var_9590_to_fp16)[name = tensor("aw_401_cast")]; + tensor var_9593_equation_0 = const()[name = tensor("op_9593_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9593_cast = einsum(equation = var_9593_equation_0, values = (var_9435_cast, var_9352_cast))[name = tensor("op_9593_cast")]; + tensor var_9594_to_fp16 = const()[name = tensor("op_9594_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_403_cast = mul(x = var_9593_cast, y = var_9594_to_fp16)[name = tensor("aw_403_cast")]; + tensor var_9597_equation_0 = const()[name = tensor("op_9597_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9597_cast = einsum(equation = var_9597_equation_0, values = (var_9439_cast, var_9356_cast))[name = tensor("op_9597_cast")]; + tensor var_9598_to_fp16 = const()[name = tensor("op_9598_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_405_cast = mul(x = var_9597_cast, y = var_9598_to_fp16)[name = tensor("aw_405_cast")]; + tensor var_9601_equation_0 = const()[name = tensor("op_9601_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9601_cast = einsum(equation = var_9601_equation_0, values = (var_9443_cast, var_9360_cast))[name = tensor("op_9601_cast")]; + tensor var_9602_to_fp16 = const()[name = tensor("op_9602_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_407_cast = mul(x = var_9601_cast, y = var_9602_to_fp16)[name = tensor("aw_407_cast")]; + tensor var_9605_equation_0 = const()[name = tensor("op_9605_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9605_cast = einsum(equation = var_9605_equation_0, values = (var_9447_cast, var_9364_cast))[name = tensor("op_9605_cast")]; + tensor var_9606_to_fp16 = const()[name = tensor("op_9606_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_409_cast = mul(x = var_9605_cast, y = var_9606_to_fp16)[name = tensor("aw_409_cast")]; + tensor var_9609_equation_0 = const()[name = tensor("op_9609_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9609_cast = einsum(equation = var_9609_equation_0, values = (var_9451_cast, var_9368_cast))[name = tensor("op_9609_cast")]; + tensor var_9610_to_fp16 = const()[name = tensor("op_9610_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_411_cast = mul(x = var_9609_cast, y = var_9610_to_fp16)[name = tensor("aw_411_cast")]; + tensor var_9613_equation_0 = const()[name = tensor("op_9613_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9613_cast = einsum(equation = var_9613_equation_0, values = (var_9455_cast, var_9372_cast))[name = tensor("op_9613_cast")]; + tensor var_9614_to_fp16 = const()[name = tensor("op_9614_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_413_cast = mul(x = var_9613_cast, y = var_9614_to_fp16)[name = tensor("aw_413_cast")]; + tensor var_9617_equation_0 = const()[name = tensor("op_9617_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9617_cast = einsum(equation = var_9617_equation_0, values = (var_9459_cast, var_9376_cast))[name = tensor("op_9617_cast")]; + tensor var_9618_to_fp16 = const()[name = tensor("op_9618_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_415_cast = mul(x = var_9617_cast, y = var_9618_to_fp16)[name = tensor("aw_415_cast")]; + tensor var_9621_equation_0 = const()[name = tensor("op_9621_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9621_cast = einsum(equation = var_9621_equation_0, values = (var_9463_cast, var_9380_cast))[name = tensor("op_9621_cast")]; + tensor var_9622_to_fp16 = const()[name = tensor("op_9622_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_417_cast = mul(x = var_9621_cast, y = var_9622_to_fp16)[name = tensor("aw_417_cast")]; + tensor var_9625_equation_0 = const()[name = tensor("op_9625_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9625_cast = einsum(equation = var_9625_equation_0, values = (var_9467_cast, var_9384_cast))[name = tensor("op_9625_cast")]; + tensor var_9626_to_fp16 = const()[name = tensor("op_9626_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_419_cast = mul(x = var_9625_cast, y = var_9626_to_fp16)[name = tensor("aw_419_cast")]; + tensor var_9629_equation_0 = const()[name = tensor("op_9629_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9629_cast = einsum(equation = var_9629_equation_0, values = (var_9471_cast, var_9388_cast))[name = tensor("op_9629_cast")]; + tensor var_9630_to_fp16 = const()[name = tensor("op_9630_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_421_cast = mul(x = var_9629_cast, y = var_9630_to_fp16)[name = tensor("aw_421_cast")]; + tensor var_9633_equation_0 = const()[name = tensor("op_9633_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9633_cast = einsum(equation = var_9633_equation_0, values = (var_9475_cast, var_9392_cast))[name = tensor("op_9633_cast")]; + tensor var_9634_to_fp16 = const()[name = tensor("op_9634_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_423_cast = mul(x = var_9633_cast, y = var_9634_to_fp16)[name = tensor("aw_423_cast")]; + tensor var_9637_equation_0 = const()[name = tensor("op_9637_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9637_cast = einsum(equation = var_9637_equation_0, values = (var_9479_cast, var_9396_cast))[name = tensor("op_9637_cast")]; + tensor var_9638_to_fp16 = const()[name = tensor("op_9638_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_425_cast = mul(x = var_9637_cast, y = var_9638_to_fp16)[name = tensor("aw_425_cast")]; + tensor var_9641_equation_0 = const()[name = tensor("op_9641_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9641_cast = einsum(equation = var_9641_equation_0, values = (var_9483_cast, var_9400_cast))[name = tensor("op_9641_cast")]; + tensor var_9642_to_fp16 = const()[name = tensor("op_9642_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_427_cast = mul(x = var_9641_cast, y = var_9642_to_fp16)[name = tensor("aw_427_cast")]; + tensor var_9645_equation_0 = const()[name = tensor("op_9645_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9645_cast = einsum(equation = var_9645_equation_0, values = (var_9487_cast, var_9404_cast))[name = tensor("op_9645_cast")]; + tensor var_9646_to_fp16 = const()[name = tensor("op_9646_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_429_cast = mul(x = var_9645_cast, y = var_9646_to_fp16)[name = tensor("aw_429_cast")]; + tensor var_9649_equation_0 = const()[name = tensor("op_9649_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9649_cast = einsum(equation = var_9649_equation_0, values = (var_9491_cast, var_9408_cast))[name = tensor("op_9649_cast")]; + tensor var_9650_to_fp16 = const()[name = tensor("op_9650_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_431_cast = mul(x = var_9649_cast, y = var_9650_to_fp16)[name = tensor("aw_431_cast")]; + tensor var_9653_equation_0 = const()[name = tensor("op_9653_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9653_cast = einsum(equation = var_9653_equation_0, values = (var_9495_cast, var_9412_cast))[name = tensor("op_9653_cast")]; + tensor var_9654_to_fp16 = const()[name = tensor("op_9654_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_433_cast = mul(x = var_9653_cast, y = var_9654_to_fp16)[name = tensor("aw_433_cast")]; + tensor var_9657_equation_0 = const()[name = tensor("op_9657_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9657_cast = einsum(equation = var_9657_equation_0, values = (var_9499_cast, var_9416_cast))[name = tensor("op_9657_cast")]; + tensor var_9658_to_fp16 = const()[name = tensor("op_9658_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_435_cast = mul(x = var_9657_cast, y = var_9658_to_fp16)[name = tensor("aw_435_cast")]; + tensor var_9661_equation_0 = const()[name = tensor("op_9661_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9661_cast = einsum(equation = var_9661_equation_0, values = (var_9503_cast, var_9420_cast))[name = tensor("op_9661_cast")]; + tensor var_9662_to_fp16 = const()[name = tensor("op_9662_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_437_cast = mul(x = var_9661_cast, y = var_9662_to_fp16)[name = tensor("aw_437_cast")]; + tensor var_9665_equation_0 = const()[name = tensor("op_9665_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9665_cast = einsum(equation = var_9665_equation_0, values = (var_9507_cast, var_9424_cast))[name = tensor("op_9665_cast")]; + tensor var_9666_to_fp16 = const()[name = tensor("op_9666_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_439_cast = mul(x = var_9665_cast, y = var_9666_to_fp16)[name = tensor("aw_439_cast")]; + tensor var_9668_cast = softmax(axis = var_7228, x = aw_401_cast)[name = tensor("op_9668_cast")]; + tensor var_9669_cast = softmax(axis = var_7228, x = aw_403_cast)[name = tensor("op_9669_cast")]; + tensor var_9670_cast = softmax(axis = var_7228, x = aw_405_cast)[name = tensor("op_9670_cast")]; + tensor var_9671_cast = softmax(axis = var_7228, x = aw_407_cast)[name = tensor("op_9671_cast")]; + tensor var_9672_cast = softmax(axis = var_7228, x = aw_409_cast)[name = tensor("op_9672_cast")]; + tensor var_9673_cast = softmax(axis = var_7228, x = aw_411_cast)[name = tensor("op_9673_cast")]; + tensor var_9674_cast = softmax(axis = var_7228, x = aw_413_cast)[name = tensor("op_9674_cast")]; + tensor var_9675_cast = softmax(axis = var_7228, x = aw_415_cast)[name = tensor("op_9675_cast")]; + tensor var_9676_cast = softmax(axis = var_7228, x = aw_417_cast)[name = tensor("op_9676_cast")]; + tensor var_9677_cast = softmax(axis = var_7228, x = aw_419_cast)[name = tensor("op_9677_cast")]; + tensor var_9678_cast = softmax(axis = var_7228, x = aw_421_cast)[name = tensor("op_9678_cast")]; + tensor var_9679_cast = softmax(axis = var_7228, x = aw_423_cast)[name = tensor("op_9679_cast")]; + tensor var_9680_cast = softmax(axis = var_7228, x = aw_425_cast)[name = tensor("op_9680_cast")]; + tensor var_9681_cast = softmax(axis = var_7228, x = aw_427_cast)[name = tensor("op_9681_cast")]; + tensor var_9682_cast = softmax(axis = var_7228, x = aw_429_cast)[name = tensor("op_9682_cast")]; + tensor var_9683_cast = softmax(axis = var_7228, x = aw_431_cast)[name = tensor("op_9683_cast")]; + tensor var_9684_cast = softmax(axis = var_7228, x = aw_433_cast)[name = tensor("op_9684_cast")]; + tensor var_9685_cast = softmax(axis = var_7228, x = aw_435_cast)[name = tensor("op_9685_cast")]; + tensor var_9686_cast = softmax(axis = var_7228, x = aw_437_cast)[name = tensor("op_9686_cast")]; + tensor var_9687_cast = softmax(axis = var_7228, x = aw_439_cast)[name = tensor("op_9687_cast")]; + tensor var_9689_equation_0 = const()[name = tensor("op_9689_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9689_cast = einsum(equation = var_9689_equation_0, values = (var_9509_cast, var_9668_cast))[name = tensor("op_9689_cast")]; + tensor var_9691_equation_0 = const()[name = tensor("op_9691_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9691_cast = einsum(equation = var_9691_equation_0, values = (var_9513_cast, var_9669_cast))[name = tensor("op_9691_cast")]; + tensor var_9693_equation_0 = const()[name = tensor("op_9693_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9693_cast = einsum(equation = var_9693_equation_0, values = (var_9517_cast, var_9670_cast))[name = tensor("op_9693_cast")]; + tensor var_9695_equation_0 = const()[name = tensor("op_9695_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9695_cast = einsum(equation = var_9695_equation_0, values = (var_9521_cast, var_9671_cast))[name = tensor("op_9695_cast")]; + tensor var_9697_equation_0 = const()[name = tensor("op_9697_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9697_cast = einsum(equation = var_9697_equation_0, values = (var_9525_cast, var_9672_cast))[name = tensor("op_9697_cast")]; + tensor var_9699_equation_0 = const()[name = tensor("op_9699_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9699_cast = einsum(equation = var_9699_equation_0, values = (var_9529_cast, var_9673_cast))[name = tensor("op_9699_cast")]; + tensor var_9701_equation_0 = const()[name = tensor("op_9701_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9701_cast = einsum(equation = var_9701_equation_0, values = (var_9533_cast, var_9674_cast))[name = tensor("op_9701_cast")]; + tensor var_9703_equation_0 = const()[name = tensor("op_9703_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9703_cast = einsum(equation = var_9703_equation_0, values = (var_9537_cast, var_9675_cast))[name = tensor("op_9703_cast")]; + tensor var_9705_equation_0 = const()[name = tensor("op_9705_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9705_cast = einsum(equation = var_9705_equation_0, values = (var_9541_cast, var_9676_cast))[name = tensor("op_9705_cast")]; + tensor var_9707_equation_0 = const()[name = tensor("op_9707_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9707_cast = einsum(equation = var_9707_equation_0, values = (var_9545_cast, var_9677_cast))[name = tensor("op_9707_cast")]; + tensor var_9709_equation_0 = const()[name = tensor("op_9709_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9709_cast = einsum(equation = var_9709_equation_0, values = (var_9549_cast, var_9678_cast))[name = tensor("op_9709_cast")]; + tensor var_9711_equation_0 = const()[name = tensor("op_9711_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9711_cast = einsum(equation = var_9711_equation_0, values = (var_9553_cast, var_9679_cast))[name = tensor("op_9711_cast")]; + tensor var_9713_equation_0 = const()[name = tensor("op_9713_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9713_cast = einsum(equation = var_9713_equation_0, values = (var_9557_cast, var_9680_cast))[name = tensor("op_9713_cast")]; + tensor var_9715_equation_0 = const()[name = tensor("op_9715_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9715_cast = einsum(equation = var_9715_equation_0, values = (var_9561_cast, var_9681_cast))[name = tensor("op_9715_cast")]; + tensor var_9717_equation_0 = const()[name = tensor("op_9717_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9717_cast = einsum(equation = var_9717_equation_0, values = (var_9565_cast, var_9682_cast))[name = tensor("op_9717_cast")]; + tensor var_9719_equation_0 = const()[name = tensor("op_9719_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9719_cast = einsum(equation = var_9719_equation_0, values = (var_9569_cast, var_9683_cast))[name = tensor("op_9719_cast")]; + tensor var_9721_equation_0 = const()[name = tensor("op_9721_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9721_cast = einsum(equation = var_9721_equation_0, values = (var_9573_cast, var_9684_cast))[name = tensor("op_9721_cast")]; + tensor var_9723_equation_0 = const()[name = tensor("op_9723_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9723_cast = einsum(equation = var_9723_equation_0, values = (var_9577_cast, var_9685_cast))[name = tensor("op_9723_cast")]; + tensor var_9725_equation_0 = const()[name = tensor("op_9725_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9725_cast = einsum(equation = var_9725_equation_0, values = (var_9581_cast, var_9686_cast))[name = tensor("op_9725_cast")]; + tensor var_9727_equation_0 = const()[name = tensor("op_9727_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9727_cast = einsum(equation = var_9727_equation_0, values = (var_9585_cast, var_9687_cast))[name = tensor("op_9727_cast")]; + tensor input_353_interleave_0 = const()[name = tensor("input_353_interleave_0"), val = tensor(false)]; + tensor input_353_cast = concat(axis = var_7228, interleave = input_353_interleave_0, values = (var_9689_cast, var_9691_cast, var_9693_cast, var_9695_cast, var_9697_cast, var_9699_cast, var_9701_cast, var_9703_cast, var_9705_cast, var_9707_cast, var_9709_cast, var_9711_cast, var_9713_cast, var_9715_cast, var_9717_cast, var_9719_cast, var_9721_cast, var_9723_cast, var_9725_cast, var_9727_cast))[name = tensor("input_353_cast")]; + tensor var_9733 = const()[name = tensor("op_9733"), val = tensor([1, 1])]; + tensor var_9735 = const()[name = tensor("op_9735"), val = tensor([1, 1])]; + tensor var_9737_pad_type_0 = const()[name = tensor("op_9737_pad_type_0"), val = tensor("custom")]; + tensor var_9737_pad_0 = const()[name = tensor("op_9737_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_1_attentions_2_transformer_blocks_0_attn1_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(548332352))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(549561216))), name = tensor("up_blocks_1_attentions_2_transformer_blocks_0_attn1_to_out_0_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor up_blocks_1_attentions_2_transformer_blocks_0_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("up_blocks_1_attentions_2_transformer_blocks_0_attn1_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(549561408)))]; + tensor var_9737_cast = conv(bias = up_blocks_1_attentions_2_transformer_blocks_0_attn1_to_out_0_bias_to_fp16, dilations = var_9735, groups = var_7228, pad = var_9737_pad_0, pad_type = var_9737_pad_type_0, strides = var_9733, weight = up_blocks_1_attentions_2_transformer_blocks_0_attn1_to_out_0_weight_to_fp16_palettized, x = input_353_cast)[name = tensor("op_9737_cast")]; + tensor inputs_57_cast = add(x = var_9737_cast, y = inputs_55_cast)[name = tensor("inputs_57_cast")]; + tensor var_9741 = const()[name = tensor("op_9741"), val = tensor([1])]; + tensor channels_mean_57_cast = reduce_mean(axes = var_9741, keep_dims = var_7223, x = inputs_57_cast)[name = tensor("channels_mean_57_cast")]; + tensor zero_mean_57_cast = sub(x = inputs_57_cast, y = channels_mean_57_cast)[name = tensor("zero_mean_57_cast")]; + tensor zero_mean_sq_57_cast = mul(x = zero_mean_57_cast, y = zero_mean_57_cast)[name = tensor("zero_mean_sq_57_cast")]; + tensor var_9745 = const()[name = tensor("op_9745"), val = tensor([1])]; + tensor var_9746_cast = reduce_mean(axes = var_9745, keep_dims = var_7223, x = zero_mean_sq_57_cast)[name = tensor("op_9746_cast")]; + tensor var_9747_to_fp16 = const()[name = tensor("op_9747_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_9748_cast = add(x = var_9746_cast, y = var_9747_to_fp16)[name = tensor("op_9748_cast")]; + tensor denom_57_epsilon_0_to_fp16 = const()[name = tensor("denom_57_epsilon_0_to_fp16"), val = tensor(0x1p-24)]; + tensor denom_57_cast = rsqrt(epsilon = denom_57_epsilon_0_to_fp16, x = var_9748_cast)[name = tensor("denom_57_cast")]; + tensor out_57_cast = mul(x = zero_mean_57_cast, y = denom_57_cast)[name = tensor("out_57_cast")]; + tensor var_9752_to_fp16 = const()[name = tensor("op_9752_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(549564032)))]; + tensor var_9753_cast = add(x = out_57_cast, y = var_9752_to_fp16)[name = tensor("op_9753_cast")]; + tensor var_9755_to_fp16 = const()[name = tensor("op_9755_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(549566656)))]; + tensor hidden_states_209_cast = mul(x = var_9753_cast, y = var_9755_to_fp16)[name = tensor("hidden_states_209_cast")]; + tensor var_9762 = const()[name = tensor("op_9762"), val = tensor([1, 1])]; + tensor var_9764 = const()[name = tensor("op_9764"), val = tensor([1, 1])]; + tensor q_39_pad_type_0 = const()[name = tensor("q_39_pad_type_0"), val = tensor("custom")]; + tensor q_39_pad_0 = const()[name = tensor("q_39_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_1_attentions_2_transformer_blocks_0_attn2_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(549569280))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(550798144))), name = tensor("up_blocks_1_attentions_2_transformer_blocks_0_attn2_to_q_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor q_39_cast = conv(dilations = var_9764, groups = var_7228, pad = q_39_pad_0, pad_type = q_39_pad_type_0, strides = var_9762, weight = up_blocks_1_attentions_2_transformer_blocks_0_attn2_to_q_weight_to_fp16_palettized, x = hidden_states_209_cast)[name = tensor("q_39_cast")]; + tensor var_9768 = const()[name = tensor("op_9768"), val = tensor([1, 1])]; + tensor var_9770 = const()[name = tensor("op_9770"), val = tensor([1, 1])]; + tensor k_77_pad_type_0 = const()[name = tensor("k_77_pad_type_0"), val = tensor("custom")]; + tensor k_77_pad_0 = const()[name = tensor("k_77_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_1_attentions_2_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(550798336))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(551781440))), name = tensor("up_blocks_1_attentions_2_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized"), shape = tensor([1280, 1024, 1, 1])]; + tensor k_77_cast = conv(dilations = var_9770, groups = var_7228, pad = k_77_pad_0, pad_type = k_77_pad_type_0, strides = var_9768, weight = up_blocks_1_attentions_2_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor("k_77_cast")]; + tensor var_9774 = const()[name = tensor("op_9774"), val = tensor([1, 1])]; + tensor var_9776 = const()[name = tensor("op_9776"), val = tensor([1, 1])]; + tensor v_39_pad_type_0 = const()[name = tensor("v_39_pad_type_0"), val = tensor("custom")]; + tensor v_39_pad_0 = const()[name = tensor("v_39_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_1_attentions_2_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(551781632))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(552764736))), name = tensor("up_blocks_1_attentions_2_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized"), shape = tensor([1280, 1024, 1, 1])]; + tensor v_39_cast = conv(dilations = var_9776, groups = var_7228, pad = v_39_pad_0, pad_type = v_39_pad_type_0, strides = var_9774, weight = up_blocks_1_attentions_2_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor("v_39_cast")]; + tensor var_9780_begin_0 = const()[name = tensor("op_9780_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_9780_end_0 = const()[name = tensor("op_9780_end_0"), val = tensor([2, 64, 1, 256])]; + tensor var_9780_end_mask_0 = const()[name = tensor("op_9780_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9780_cast = slice_by_index(begin = var_9780_begin_0, end = var_9780_end_0, end_mask = var_9780_end_mask_0, x = q_39_cast)[name = tensor("op_9780_cast")]; + tensor var_9784_begin_0 = const()[name = tensor("op_9784_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_9784_end_0 = const()[name = tensor("op_9784_end_0"), val = tensor([2, 128, 1, 256])]; + tensor var_9784_end_mask_0 = const()[name = tensor("op_9784_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9784_cast = slice_by_index(begin = var_9784_begin_0, end = var_9784_end_0, end_mask = var_9784_end_mask_0, x = q_39_cast)[name = tensor("op_9784_cast")]; + tensor var_9788_begin_0 = const()[name = tensor("op_9788_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_9788_end_0 = const()[name = tensor("op_9788_end_0"), val = tensor([2, 192, 1, 256])]; + tensor var_9788_end_mask_0 = const()[name = tensor("op_9788_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9788_cast = slice_by_index(begin = var_9788_begin_0, end = var_9788_end_0, end_mask = var_9788_end_mask_0, x = q_39_cast)[name = tensor("op_9788_cast")]; + tensor var_9792_begin_0 = const()[name = tensor("op_9792_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_9792_end_0 = const()[name = tensor("op_9792_end_0"), val = tensor([2, 256, 1, 256])]; + tensor var_9792_end_mask_0 = const()[name = tensor("op_9792_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9792_cast = slice_by_index(begin = var_9792_begin_0, end = var_9792_end_0, end_mask = var_9792_end_mask_0, x = q_39_cast)[name = tensor("op_9792_cast")]; + tensor var_9796_begin_0 = const()[name = tensor("op_9796_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_9796_end_0 = const()[name = tensor("op_9796_end_0"), val = tensor([2, 320, 1, 256])]; + tensor var_9796_end_mask_0 = const()[name = tensor("op_9796_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9796_cast = slice_by_index(begin = var_9796_begin_0, end = var_9796_end_0, end_mask = var_9796_end_mask_0, x = q_39_cast)[name = tensor("op_9796_cast")]; + tensor var_9800_begin_0 = const()[name = tensor("op_9800_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_9800_end_0 = const()[name = tensor("op_9800_end_0"), val = tensor([2, 384, 1, 256])]; + tensor var_9800_end_mask_0 = const()[name = tensor("op_9800_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9800_cast = slice_by_index(begin = var_9800_begin_0, end = var_9800_end_0, end_mask = var_9800_end_mask_0, x = q_39_cast)[name = tensor("op_9800_cast")]; + tensor var_9804_begin_0 = const()[name = tensor("op_9804_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_9804_end_0 = const()[name = tensor("op_9804_end_0"), val = tensor([2, 448, 1, 256])]; + tensor var_9804_end_mask_0 = const()[name = tensor("op_9804_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9804_cast = slice_by_index(begin = var_9804_begin_0, end = var_9804_end_0, end_mask = var_9804_end_mask_0, x = q_39_cast)[name = tensor("op_9804_cast")]; + tensor var_9808_begin_0 = const()[name = tensor("op_9808_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_9808_end_0 = const()[name = tensor("op_9808_end_0"), val = tensor([2, 512, 1, 256])]; + tensor var_9808_end_mask_0 = const()[name = tensor("op_9808_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9808_cast = slice_by_index(begin = var_9808_begin_0, end = var_9808_end_0, end_mask = var_9808_end_mask_0, x = q_39_cast)[name = tensor("op_9808_cast")]; + tensor var_9812_begin_0 = const()[name = tensor("op_9812_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_9812_end_0 = const()[name = tensor("op_9812_end_0"), val = tensor([2, 576, 1, 256])]; + tensor var_9812_end_mask_0 = const()[name = tensor("op_9812_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9812_cast = slice_by_index(begin = var_9812_begin_0, end = var_9812_end_0, end_mask = var_9812_end_mask_0, x = q_39_cast)[name = tensor("op_9812_cast")]; + tensor var_9816_begin_0 = const()[name = tensor("op_9816_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_9816_end_0 = const()[name = tensor("op_9816_end_0"), val = tensor([2, 640, 1, 256])]; + tensor var_9816_end_mask_0 = const()[name = tensor("op_9816_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9816_cast = slice_by_index(begin = var_9816_begin_0, end = var_9816_end_0, end_mask = var_9816_end_mask_0, x = q_39_cast)[name = tensor("op_9816_cast")]; + tensor var_9820_begin_0 = const()[name = tensor("op_9820_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_9820_end_0 = const()[name = tensor("op_9820_end_0"), val = tensor([2, 704, 1, 256])]; + tensor var_9820_end_mask_0 = const()[name = tensor("op_9820_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9820_cast = slice_by_index(begin = var_9820_begin_0, end = var_9820_end_0, end_mask = var_9820_end_mask_0, x = q_39_cast)[name = tensor("op_9820_cast")]; + tensor var_9824_begin_0 = const()[name = tensor("op_9824_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_9824_end_0 = const()[name = tensor("op_9824_end_0"), val = tensor([2, 768, 1, 256])]; + tensor var_9824_end_mask_0 = const()[name = tensor("op_9824_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9824_cast = slice_by_index(begin = var_9824_begin_0, end = var_9824_end_0, end_mask = var_9824_end_mask_0, x = q_39_cast)[name = tensor("op_9824_cast")]; + tensor var_9828_begin_0 = const()[name = tensor("op_9828_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_9828_end_0 = const()[name = tensor("op_9828_end_0"), val = tensor([2, 832, 1, 256])]; + tensor var_9828_end_mask_0 = const()[name = tensor("op_9828_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9828_cast = slice_by_index(begin = var_9828_begin_0, end = var_9828_end_0, end_mask = var_9828_end_mask_0, x = q_39_cast)[name = tensor("op_9828_cast")]; + tensor var_9832_begin_0 = const()[name = tensor("op_9832_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_9832_end_0 = const()[name = tensor("op_9832_end_0"), val = tensor([2, 896, 1, 256])]; + tensor var_9832_end_mask_0 = const()[name = tensor("op_9832_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9832_cast = slice_by_index(begin = var_9832_begin_0, end = var_9832_end_0, end_mask = var_9832_end_mask_0, x = q_39_cast)[name = tensor("op_9832_cast")]; + tensor var_9836_begin_0 = const()[name = tensor("op_9836_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_9836_end_0 = const()[name = tensor("op_9836_end_0"), val = tensor([2, 960, 1, 256])]; + tensor var_9836_end_mask_0 = const()[name = tensor("op_9836_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9836_cast = slice_by_index(begin = var_9836_begin_0, end = var_9836_end_0, end_mask = var_9836_end_mask_0, x = q_39_cast)[name = tensor("op_9836_cast")]; + tensor var_9840_begin_0 = const()[name = tensor("op_9840_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_9840_end_0 = const()[name = tensor("op_9840_end_0"), val = tensor([2, 1024, 1, 256])]; + tensor var_9840_end_mask_0 = const()[name = tensor("op_9840_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9840_cast = slice_by_index(begin = var_9840_begin_0, end = var_9840_end_0, end_mask = var_9840_end_mask_0, x = q_39_cast)[name = tensor("op_9840_cast")]; + tensor var_9844_begin_0 = const()[name = tensor("op_9844_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_9844_end_0 = const()[name = tensor("op_9844_end_0"), val = tensor([2, 1088, 1, 256])]; + tensor var_9844_end_mask_0 = const()[name = tensor("op_9844_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9844_cast = slice_by_index(begin = var_9844_begin_0, end = var_9844_end_0, end_mask = var_9844_end_mask_0, x = q_39_cast)[name = tensor("op_9844_cast")]; + tensor var_9848_begin_0 = const()[name = tensor("op_9848_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_9848_end_0 = const()[name = tensor("op_9848_end_0"), val = tensor([2, 1152, 1, 256])]; + tensor var_9848_end_mask_0 = const()[name = tensor("op_9848_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9848_cast = slice_by_index(begin = var_9848_begin_0, end = var_9848_end_0, end_mask = var_9848_end_mask_0, x = q_39_cast)[name = tensor("op_9848_cast")]; + tensor var_9852_begin_0 = const()[name = tensor("op_9852_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_9852_end_0 = const()[name = tensor("op_9852_end_0"), val = tensor([2, 1216, 1, 256])]; + tensor var_9852_end_mask_0 = const()[name = tensor("op_9852_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9852_cast = slice_by_index(begin = var_9852_begin_0, end = var_9852_end_0, end_mask = var_9852_end_mask_0, x = q_39_cast)[name = tensor("op_9852_cast")]; + tensor var_9856_begin_0 = const()[name = tensor("op_9856_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_9856_end_0 = const()[name = tensor("op_9856_end_0"), val = tensor([2, 1280, 1, 256])]; + tensor var_9856_end_mask_0 = const()[name = tensor("op_9856_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9856_cast = slice_by_index(begin = var_9856_begin_0, end = var_9856_end_0, end_mask = var_9856_end_mask_0, x = q_39_cast)[name = tensor("op_9856_cast")]; + tensor k_79_perm_0 = const()[name = tensor("k_79_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_9863_begin_0 = const()[name = tensor("op_9863_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_9863_end_0 = const()[name = tensor("op_9863_end_0"), val = tensor([2, 77, 1, 64])]; + tensor var_9863_end_mask_0 = const()[name = tensor("op_9863_end_mask_0"), val = tensor([true, true, true, false])]; + tensor transpose_12 = transpose(perm = k_79_perm_0, x = k_77_cast)[name = tensor("transpose_12")]; + tensor var_9863_cast = slice_by_index(begin = var_9863_begin_0, end = var_9863_end_0, end_mask = var_9863_end_mask_0, x = transpose_12)[name = tensor("op_9863_cast")]; + tensor var_9867_begin_0 = const()[name = tensor("op_9867_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_9867_end_0 = const()[name = tensor("op_9867_end_0"), val = tensor([2, 77, 1, 128])]; + tensor var_9867_end_mask_0 = const()[name = tensor("op_9867_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9867_cast = slice_by_index(begin = var_9867_begin_0, end = var_9867_end_0, end_mask = var_9867_end_mask_0, x = transpose_12)[name = tensor("op_9867_cast")]; + tensor var_9871_begin_0 = const()[name = tensor("op_9871_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_9871_end_0 = const()[name = tensor("op_9871_end_0"), val = tensor([2, 77, 1, 192])]; + tensor var_9871_end_mask_0 = const()[name = tensor("op_9871_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9871_cast = slice_by_index(begin = var_9871_begin_0, end = var_9871_end_0, end_mask = var_9871_end_mask_0, x = transpose_12)[name = tensor("op_9871_cast")]; + tensor var_9875_begin_0 = const()[name = tensor("op_9875_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_9875_end_0 = const()[name = tensor("op_9875_end_0"), val = tensor([2, 77, 1, 256])]; + tensor var_9875_end_mask_0 = const()[name = tensor("op_9875_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9875_cast = slice_by_index(begin = var_9875_begin_0, end = var_9875_end_0, end_mask = var_9875_end_mask_0, x = transpose_12)[name = tensor("op_9875_cast")]; + tensor var_9879_begin_0 = const()[name = tensor("op_9879_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_9879_end_0 = const()[name = tensor("op_9879_end_0"), val = tensor([2, 77, 1, 320])]; + tensor var_9879_end_mask_0 = const()[name = tensor("op_9879_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9879_cast = slice_by_index(begin = var_9879_begin_0, end = var_9879_end_0, end_mask = var_9879_end_mask_0, x = transpose_12)[name = tensor("op_9879_cast")]; + tensor var_9883_begin_0 = const()[name = tensor("op_9883_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_9883_end_0 = const()[name = tensor("op_9883_end_0"), val = tensor([2, 77, 1, 384])]; + tensor var_9883_end_mask_0 = const()[name = tensor("op_9883_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9883_cast = slice_by_index(begin = var_9883_begin_0, end = var_9883_end_0, end_mask = var_9883_end_mask_0, x = transpose_12)[name = tensor("op_9883_cast")]; + tensor var_9887_begin_0 = const()[name = tensor("op_9887_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_9887_end_0 = const()[name = tensor("op_9887_end_0"), val = tensor([2, 77, 1, 448])]; + tensor var_9887_end_mask_0 = const()[name = tensor("op_9887_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9887_cast = slice_by_index(begin = var_9887_begin_0, end = var_9887_end_0, end_mask = var_9887_end_mask_0, x = transpose_12)[name = tensor("op_9887_cast")]; + tensor var_9891_begin_0 = const()[name = tensor("op_9891_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_9891_end_0 = const()[name = tensor("op_9891_end_0"), val = tensor([2, 77, 1, 512])]; + tensor var_9891_end_mask_0 = const()[name = tensor("op_9891_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9891_cast = slice_by_index(begin = var_9891_begin_0, end = var_9891_end_0, end_mask = var_9891_end_mask_0, x = transpose_12)[name = tensor("op_9891_cast")]; + tensor var_9895_begin_0 = const()[name = tensor("op_9895_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_9895_end_0 = const()[name = tensor("op_9895_end_0"), val = tensor([2, 77, 1, 576])]; + tensor var_9895_end_mask_0 = const()[name = tensor("op_9895_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9895_cast = slice_by_index(begin = var_9895_begin_0, end = var_9895_end_0, end_mask = var_9895_end_mask_0, x = transpose_12)[name = tensor("op_9895_cast")]; + tensor var_9899_begin_0 = const()[name = tensor("op_9899_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_9899_end_0 = const()[name = tensor("op_9899_end_0"), val = tensor([2, 77, 1, 640])]; + tensor var_9899_end_mask_0 = const()[name = tensor("op_9899_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9899_cast = slice_by_index(begin = var_9899_begin_0, end = var_9899_end_0, end_mask = var_9899_end_mask_0, x = transpose_12)[name = tensor("op_9899_cast")]; + tensor var_9903_begin_0 = const()[name = tensor("op_9903_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_9903_end_0 = const()[name = tensor("op_9903_end_0"), val = tensor([2, 77, 1, 704])]; + tensor var_9903_end_mask_0 = const()[name = tensor("op_9903_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9903_cast = slice_by_index(begin = var_9903_begin_0, end = var_9903_end_0, end_mask = var_9903_end_mask_0, x = transpose_12)[name = tensor("op_9903_cast")]; + tensor var_9907_begin_0 = const()[name = tensor("op_9907_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_9907_end_0 = const()[name = tensor("op_9907_end_0"), val = tensor([2, 77, 1, 768])]; + tensor var_9907_end_mask_0 = const()[name = tensor("op_9907_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9907_cast = slice_by_index(begin = var_9907_begin_0, end = var_9907_end_0, end_mask = var_9907_end_mask_0, x = transpose_12)[name = tensor("op_9907_cast")]; + tensor var_9911_begin_0 = const()[name = tensor("op_9911_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_9911_end_0 = const()[name = tensor("op_9911_end_0"), val = tensor([2, 77, 1, 832])]; + tensor var_9911_end_mask_0 = const()[name = tensor("op_9911_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9911_cast = slice_by_index(begin = var_9911_begin_0, end = var_9911_end_0, end_mask = var_9911_end_mask_0, x = transpose_12)[name = tensor("op_9911_cast")]; + tensor var_9915_begin_0 = const()[name = tensor("op_9915_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_9915_end_0 = const()[name = tensor("op_9915_end_0"), val = tensor([2, 77, 1, 896])]; + tensor var_9915_end_mask_0 = const()[name = tensor("op_9915_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9915_cast = slice_by_index(begin = var_9915_begin_0, end = var_9915_end_0, end_mask = var_9915_end_mask_0, x = transpose_12)[name = tensor("op_9915_cast")]; + tensor var_9919_begin_0 = const()[name = tensor("op_9919_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_9919_end_0 = const()[name = tensor("op_9919_end_0"), val = tensor([2, 77, 1, 960])]; + tensor var_9919_end_mask_0 = const()[name = tensor("op_9919_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9919_cast = slice_by_index(begin = var_9919_begin_0, end = var_9919_end_0, end_mask = var_9919_end_mask_0, x = transpose_12)[name = tensor("op_9919_cast")]; + tensor var_9923_begin_0 = const()[name = tensor("op_9923_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_9923_end_0 = const()[name = tensor("op_9923_end_0"), val = tensor([2, 77, 1, 1024])]; + tensor var_9923_end_mask_0 = const()[name = tensor("op_9923_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9923_cast = slice_by_index(begin = var_9923_begin_0, end = var_9923_end_0, end_mask = var_9923_end_mask_0, x = transpose_12)[name = tensor("op_9923_cast")]; + tensor var_9927_begin_0 = const()[name = tensor("op_9927_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_9927_end_0 = const()[name = tensor("op_9927_end_0"), val = tensor([2, 77, 1, 1088])]; + tensor var_9927_end_mask_0 = const()[name = tensor("op_9927_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9927_cast = slice_by_index(begin = var_9927_begin_0, end = var_9927_end_0, end_mask = var_9927_end_mask_0, x = transpose_12)[name = tensor("op_9927_cast")]; + tensor var_9931_begin_0 = const()[name = tensor("op_9931_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_9931_end_0 = const()[name = tensor("op_9931_end_0"), val = tensor([2, 77, 1, 1152])]; + tensor var_9931_end_mask_0 = const()[name = tensor("op_9931_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9931_cast = slice_by_index(begin = var_9931_begin_0, end = var_9931_end_0, end_mask = var_9931_end_mask_0, x = transpose_12)[name = tensor("op_9931_cast")]; + tensor var_9935_begin_0 = const()[name = tensor("op_9935_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_9935_end_0 = const()[name = tensor("op_9935_end_0"), val = tensor([2, 77, 1, 1216])]; + tensor var_9935_end_mask_0 = const()[name = tensor("op_9935_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9935_cast = slice_by_index(begin = var_9935_begin_0, end = var_9935_end_0, end_mask = var_9935_end_mask_0, x = transpose_12)[name = tensor("op_9935_cast")]; + tensor var_9939_begin_0 = const()[name = tensor("op_9939_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_9939_end_0 = const()[name = tensor("op_9939_end_0"), val = tensor([2, 77, 1, 1280])]; + tensor var_9939_end_mask_0 = const()[name = tensor("op_9939_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9939_cast = slice_by_index(begin = var_9939_begin_0, end = var_9939_end_0, end_mask = var_9939_end_mask_0, x = transpose_12)[name = tensor("op_9939_cast")]; + tensor var_9941_begin_0 = const()[name = tensor("op_9941_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_9941_end_0 = const()[name = tensor("op_9941_end_0"), val = tensor([2, 64, 1, 77])]; + tensor var_9941_end_mask_0 = const()[name = tensor("op_9941_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9941_cast = slice_by_index(begin = var_9941_begin_0, end = var_9941_end_0, end_mask = var_9941_end_mask_0, x = v_39_cast)[name = tensor("op_9941_cast")]; + tensor var_9945_begin_0 = const()[name = tensor("op_9945_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_9945_end_0 = const()[name = tensor("op_9945_end_0"), val = tensor([2, 128, 1, 77])]; + tensor var_9945_end_mask_0 = const()[name = tensor("op_9945_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9945_cast = slice_by_index(begin = var_9945_begin_0, end = var_9945_end_0, end_mask = var_9945_end_mask_0, x = v_39_cast)[name = tensor("op_9945_cast")]; + tensor var_9949_begin_0 = const()[name = tensor("op_9949_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_9949_end_0 = const()[name = tensor("op_9949_end_0"), val = tensor([2, 192, 1, 77])]; + tensor var_9949_end_mask_0 = const()[name = tensor("op_9949_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9949_cast = slice_by_index(begin = var_9949_begin_0, end = var_9949_end_0, end_mask = var_9949_end_mask_0, x = v_39_cast)[name = tensor("op_9949_cast")]; + tensor var_9953_begin_0 = const()[name = tensor("op_9953_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_9953_end_0 = const()[name = tensor("op_9953_end_0"), val = tensor([2, 256, 1, 77])]; + tensor var_9953_end_mask_0 = const()[name = tensor("op_9953_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9953_cast = slice_by_index(begin = var_9953_begin_0, end = var_9953_end_0, end_mask = var_9953_end_mask_0, x = v_39_cast)[name = tensor("op_9953_cast")]; + tensor var_9957_begin_0 = const()[name = tensor("op_9957_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_9957_end_0 = const()[name = tensor("op_9957_end_0"), val = tensor([2, 320, 1, 77])]; + tensor var_9957_end_mask_0 = const()[name = tensor("op_9957_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9957_cast = slice_by_index(begin = var_9957_begin_0, end = var_9957_end_0, end_mask = var_9957_end_mask_0, x = v_39_cast)[name = tensor("op_9957_cast")]; + tensor var_9961_begin_0 = const()[name = tensor("op_9961_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_9961_end_0 = const()[name = tensor("op_9961_end_0"), val = tensor([2, 384, 1, 77])]; + tensor var_9961_end_mask_0 = const()[name = tensor("op_9961_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9961_cast = slice_by_index(begin = var_9961_begin_0, end = var_9961_end_0, end_mask = var_9961_end_mask_0, x = v_39_cast)[name = tensor("op_9961_cast")]; + tensor var_9965_begin_0 = const()[name = tensor("op_9965_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_9965_end_0 = const()[name = tensor("op_9965_end_0"), val = tensor([2, 448, 1, 77])]; + tensor var_9965_end_mask_0 = const()[name = tensor("op_9965_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9965_cast = slice_by_index(begin = var_9965_begin_0, end = var_9965_end_0, end_mask = var_9965_end_mask_0, x = v_39_cast)[name = tensor("op_9965_cast")]; + tensor var_9969_begin_0 = const()[name = tensor("op_9969_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_9969_end_0 = const()[name = tensor("op_9969_end_0"), val = tensor([2, 512, 1, 77])]; + tensor var_9969_end_mask_0 = const()[name = tensor("op_9969_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9969_cast = slice_by_index(begin = var_9969_begin_0, end = var_9969_end_0, end_mask = var_9969_end_mask_0, x = v_39_cast)[name = tensor("op_9969_cast")]; + tensor var_9973_begin_0 = const()[name = tensor("op_9973_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_9973_end_0 = const()[name = tensor("op_9973_end_0"), val = tensor([2, 576, 1, 77])]; + tensor var_9973_end_mask_0 = const()[name = tensor("op_9973_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9973_cast = slice_by_index(begin = var_9973_begin_0, end = var_9973_end_0, end_mask = var_9973_end_mask_0, x = v_39_cast)[name = tensor("op_9973_cast")]; + tensor var_9977_begin_0 = const()[name = tensor("op_9977_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_9977_end_0 = const()[name = tensor("op_9977_end_0"), val = tensor([2, 640, 1, 77])]; + tensor var_9977_end_mask_0 = const()[name = tensor("op_9977_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9977_cast = slice_by_index(begin = var_9977_begin_0, end = var_9977_end_0, end_mask = var_9977_end_mask_0, x = v_39_cast)[name = tensor("op_9977_cast")]; + tensor var_9981_begin_0 = const()[name = tensor("op_9981_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_9981_end_0 = const()[name = tensor("op_9981_end_0"), val = tensor([2, 704, 1, 77])]; + tensor var_9981_end_mask_0 = const()[name = tensor("op_9981_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9981_cast = slice_by_index(begin = var_9981_begin_0, end = var_9981_end_0, end_mask = var_9981_end_mask_0, x = v_39_cast)[name = tensor("op_9981_cast")]; + tensor var_9985_begin_0 = const()[name = tensor("op_9985_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_9985_end_0 = const()[name = tensor("op_9985_end_0"), val = tensor([2, 768, 1, 77])]; + tensor var_9985_end_mask_0 = const()[name = tensor("op_9985_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9985_cast = slice_by_index(begin = var_9985_begin_0, end = var_9985_end_0, end_mask = var_9985_end_mask_0, x = v_39_cast)[name = tensor("op_9985_cast")]; + tensor var_9989_begin_0 = const()[name = tensor("op_9989_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_9989_end_0 = const()[name = tensor("op_9989_end_0"), val = tensor([2, 832, 1, 77])]; + tensor var_9989_end_mask_0 = const()[name = tensor("op_9989_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9989_cast = slice_by_index(begin = var_9989_begin_0, end = var_9989_end_0, end_mask = var_9989_end_mask_0, x = v_39_cast)[name = tensor("op_9989_cast")]; + tensor var_9993_begin_0 = const()[name = tensor("op_9993_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_9993_end_0 = const()[name = tensor("op_9993_end_0"), val = tensor([2, 896, 1, 77])]; + tensor var_9993_end_mask_0 = const()[name = tensor("op_9993_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9993_cast = slice_by_index(begin = var_9993_begin_0, end = var_9993_end_0, end_mask = var_9993_end_mask_0, x = v_39_cast)[name = tensor("op_9993_cast")]; + tensor var_9997_begin_0 = const()[name = tensor("op_9997_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_9997_end_0 = const()[name = tensor("op_9997_end_0"), val = tensor([2, 960, 1, 77])]; + tensor var_9997_end_mask_0 = const()[name = tensor("op_9997_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9997_cast = slice_by_index(begin = var_9997_begin_0, end = var_9997_end_0, end_mask = var_9997_end_mask_0, x = v_39_cast)[name = tensor("op_9997_cast")]; + tensor var_10001_begin_0 = const()[name = tensor("op_10001_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_10001_end_0 = const()[name = tensor("op_10001_end_0"), val = tensor([2, 1024, 1, 77])]; + tensor var_10001_end_mask_0 = const()[name = tensor("op_10001_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10001_cast = slice_by_index(begin = var_10001_begin_0, end = var_10001_end_0, end_mask = var_10001_end_mask_0, x = v_39_cast)[name = tensor("op_10001_cast")]; + tensor var_10005_begin_0 = const()[name = tensor("op_10005_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_10005_end_0 = const()[name = tensor("op_10005_end_0"), val = tensor([2, 1088, 1, 77])]; + tensor var_10005_end_mask_0 = const()[name = tensor("op_10005_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10005_cast = slice_by_index(begin = var_10005_begin_0, end = var_10005_end_0, end_mask = var_10005_end_mask_0, x = v_39_cast)[name = tensor("op_10005_cast")]; + tensor var_10009_begin_0 = const()[name = tensor("op_10009_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_10009_end_0 = const()[name = tensor("op_10009_end_0"), val = tensor([2, 1152, 1, 77])]; + tensor var_10009_end_mask_0 = const()[name = tensor("op_10009_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10009_cast = slice_by_index(begin = var_10009_begin_0, end = var_10009_end_0, end_mask = var_10009_end_mask_0, x = v_39_cast)[name = tensor("op_10009_cast")]; + tensor var_10013_begin_0 = const()[name = tensor("op_10013_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_10013_end_0 = const()[name = tensor("op_10013_end_0"), val = tensor([2, 1216, 1, 77])]; + tensor var_10013_end_mask_0 = const()[name = tensor("op_10013_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10013_cast = slice_by_index(begin = var_10013_begin_0, end = var_10013_end_0, end_mask = var_10013_end_mask_0, x = v_39_cast)[name = tensor("op_10013_cast")]; + tensor var_10017_begin_0 = const()[name = tensor("op_10017_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_10017_end_0 = const()[name = tensor("op_10017_end_0"), val = tensor([2, 1280, 1, 77])]; + tensor var_10017_end_mask_0 = const()[name = tensor("op_10017_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10017_cast = slice_by_index(begin = var_10017_begin_0, end = var_10017_end_0, end_mask = var_10017_end_mask_0, x = v_39_cast)[name = tensor("op_10017_cast")]; + tensor var_10021_equation_0 = const()[name = tensor("op_10021_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10021_cast = einsum(equation = var_10021_equation_0, values = (var_9863_cast, var_9780_cast))[name = tensor("op_10021_cast")]; + tensor var_10022_to_fp16 = const()[name = tensor("op_10022_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_441_cast = mul(x = var_10021_cast, y = var_10022_to_fp16)[name = tensor("aw_441_cast")]; + tensor var_10025_equation_0 = const()[name = tensor("op_10025_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10025_cast = einsum(equation = var_10025_equation_0, values = (var_9867_cast, var_9784_cast))[name = tensor("op_10025_cast")]; + tensor var_10026_to_fp16 = const()[name = tensor("op_10026_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_443_cast = mul(x = var_10025_cast, y = var_10026_to_fp16)[name = tensor("aw_443_cast")]; + tensor var_10029_equation_0 = const()[name = tensor("op_10029_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10029_cast = einsum(equation = var_10029_equation_0, values = (var_9871_cast, var_9788_cast))[name = tensor("op_10029_cast")]; + tensor var_10030_to_fp16 = const()[name = tensor("op_10030_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_445_cast = mul(x = var_10029_cast, y = var_10030_to_fp16)[name = tensor("aw_445_cast")]; + tensor var_10033_equation_0 = const()[name = tensor("op_10033_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10033_cast = einsum(equation = var_10033_equation_0, values = (var_9875_cast, var_9792_cast))[name = tensor("op_10033_cast")]; + tensor var_10034_to_fp16 = const()[name = tensor("op_10034_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_447_cast = mul(x = var_10033_cast, y = var_10034_to_fp16)[name = tensor("aw_447_cast")]; + tensor var_10037_equation_0 = const()[name = tensor("op_10037_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10037_cast = einsum(equation = var_10037_equation_0, values = (var_9879_cast, var_9796_cast))[name = tensor("op_10037_cast")]; + tensor var_10038_to_fp16 = const()[name = tensor("op_10038_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_449_cast = mul(x = var_10037_cast, y = var_10038_to_fp16)[name = tensor("aw_449_cast")]; + tensor var_10041_equation_0 = const()[name = tensor("op_10041_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10041_cast = einsum(equation = var_10041_equation_0, values = (var_9883_cast, var_9800_cast))[name = tensor("op_10041_cast")]; + tensor var_10042_to_fp16 = const()[name = tensor("op_10042_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_451_cast = mul(x = var_10041_cast, y = var_10042_to_fp16)[name = tensor("aw_451_cast")]; + tensor var_10045_equation_0 = const()[name = tensor("op_10045_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10045_cast = einsum(equation = var_10045_equation_0, values = (var_9887_cast, var_9804_cast))[name = tensor("op_10045_cast")]; + tensor var_10046_to_fp16 = const()[name = tensor("op_10046_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_453_cast = mul(x = var_10045_cast, y = var_10046_to_fp16)[name = tensor("aw_453_cast")]; + tensor var_10049_equation_0 = const()[name = tensor("op_10049_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10049_cast = einsum(equation = var_10049_equation_0, values = (var_9891_cast, var_9808_cast))[name = tensor("op_10049_cast")]; + tensor var_10050_to_fp16 = const()[name = tensor("op_10050_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_455_cast = mul(x = var_10049_cast, y = var_10050_to_fp16)[name = tensor("aw_455_cast")]; + tensor var_10053_equation_0 = const()[name = tensor("op_10053_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10053_cast = einsum(equation = var_10053_equation_0, values = (var_9895_cast, var_9812_cast))[name = tensor("op_10053_cast")]; + tensor var_10054_to_fp16 = const()[name = tensor("op_10054_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_457_cast = mul(x = var_10053_cast, y = var_10054_to_fp16)[name = tensor("aw_457_cast")]; + tensor var_10057_equation_0 = const()[name = tensor("op_10057_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10057_cast = einsum(equation = var_10057_equation_0, values = (var_9899_cast, var_9816_cast))[name = tensor("op_10057_cast")]; + tensor var_10058_to_fp16 = const()[name = tensor("op_10058_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_459_cast = mul(x = var_10057_cast, y = var_10058_to_fp16)[name = tensor("aw_459_cast")]; + tensor var_10061_equation_0 = const()[name = tensor("op_10061_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10061_cast = einsum(equation = var_10061_equation_0, values = (var_9903_cast, var_9820_cast))[name = tensor("op_10061_cast")]; + tensor var_10062_to_fp16 = const()[name = tensor("op_10062_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_461_cast = mul(x = var_10061_cast, y = var_10062_to_fp16)[name = tensor("aw_461_cast")]; + tensor var_10065_equation_0 = const()[name = tensor("op_10065_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10065_cast = einsum(equation = var_10065_equation_0, values = (var_9907_cast, var_9824_cast))[name = tensor("op_10065_cast")]; + tensor var_10066_to_fp16 = const()[name = tensor("op_10066_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_463_cast = mul(x = var_10065_cast, y = var_10066_to_fp16)[name = tensor("aw_463_cast")]; + tensor var_10069_equation_0 = const()[name = tensor("op_10069_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10069_cast = einsum(equation = var_10069_equation_0, values = (var_9911_cast, var_9828_cast))[name = tensor("op_10069_cast")]; + tensor var_10070_to_fp16 = const()[name = tensor("op_10070_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_465_cast = mul(x = var_10069_cast, y = var_10070_to_fp16)[name = tensor("aw_465_cast")]; + tensor var_10073_equation_0 = const()[name = tensor("op_10073_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10073_cast = einsum(equation = var_10073_equation_0, values = (var_9915_cast, var_9832_cast))[name = tensor("op_10073_cast")]; + tensor var_10074_to_fp16 = const()[name = tensor("op_10074_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_467_cast = mul(x = var_10073_cast, y = var_10074_to_fp16)[name = tensor("aw_467_cast")]; + tensor var_10077_equation_0 = const()[name = tensor("op_10077_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10077_cast = einsum(equation = var_10077_equation_0, values = (var_9919_cast, var_9836_cast))[name = tensor("op_10077_cast")]; + tensor var_10078_to_fp16 = const()[name = tensor("op_10078_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_469_cast = mul(x = var_10077_cast, y = var_10078_to_fp16)[name = tensor("aw_469_cast")]; + tensor var_10081_equation_0 = const()[name = tensor("op_10081_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10081_cast = einsum(equation = var_10081_equation_0, values = (var_9923_cast, var_9840_cast))[name = tensor("op_10081_cast")]; + tensor var_10082_to_fp16 = const()[name = tensor("op_10082_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_471_cast = mul(x = var_10081_cast, y = var_10082_to_fp16)[name = tensor("aw_471_cast")]; + tensor var_10085_equation_0 = const()[name = tensor("op_10085_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10085_cast = einsum(equation = var_10085_equation_0, values = (var_9927_cast, var_9844_cast))[name = tensor("op_10085_cast")]; + tensor var_10086_to_fp16 = const()[name = tensor("op_10086_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_473_cast = mul(x = var_10085_cast, y = var_10086_to_fp16)[name = tensor("aw_473_cast")]; + tensor var_10089_equation_0 = const()[name = tensor("op_10089_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10089_cast = einsum(equation = var_10089_equation_0, values = (var_9931_cast, var_9848_cast))[name = tensor("op_10089_cast")]; + tensor var_10090_to_fp16 = const()[name = tensor("op_10090_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_475_cast = mul(x = var_10089_cast, y = var_10090_to_fp16)[name = tensor("aw_475_cast")]; + tensor var_10093_equation_0 = const()[name = tensor("op_10093_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10093_cast = einsum(equation = var_10093_equation_0, values = (var_9935_cast, var_9852_cast))[name = tensor("op_10093_cast")]; + tensor var_10094_to_fp16 = const()[name = tensor("op_10094_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_477_cast = mul(x = var_10093_cast, y = var_10094_to_fp16)[name = tensor("aw_477_cast")]; + tensor var_10097_equation_0 = const()[name = tensor("op_10097_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10097_cast = einsum(equation = var_10097_equation_0, values = (var_9939_cast, var_9856_cast))[name = tensor("op_10097_cast")]; + tensor var_10098_to_fp16 = const()[name = tensor("op_10098_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_cast = mul(x = var_10097_cast, y = var_10098_to_fp16)[name = tensor("aw_cast")]; + tensor var_10100_cast = softmax(axis = var_7228, x = aw_441_cast)[name = tensor("op_10100_cast")]; + tensor var_10101_cast = softmax(axis = var_7228, x = aw_443_cast)[name = tensor("op_10101_cast")]; + tensor var_10102_cast = softmax(axis = var_7228, x = aw_445_cast)[name = tensor("op_10102_cast")]; + tensor var_10103_cast = softmax(axis = var_7228, x = aw_447_cast)[name = tensor("op_10103_cast")]; + tensor var_10104_cast = softmax(axis = var_7228, x = aw_449_cast)[name = tensor("op_10104_cast")]; + tensor var_10105_cast = softmax(axis = var_7228, x = aw_451_cast)[name = tensor("op_10105_cast")]; + tensor var_10106_cast = softmax(axis = var_7228, x = aw_453_cast)[name = tensor("op_10106_cast")]; + tensor var_10107_cast = softmax(axis = var_7228, x = aw_455_cast)[name = tensor("op_10107_cast")]; + tensor var_10108_cast = softmax(axis = var_7228, x = aw_457_cast)[name = tensor("op_10108_cast")]; + tensor var_10109_cast = softmax(axis = var_7228, x = aw_459_cast)[name = tensor("op_10109_cast")]; + tensor var_10110_cast = softmax(axis = var_7228, x = aw_461_cast)[name = tensor("op_10110_cast")]; + tensor var_10111_cast = softmax(axis = var_7228, x = aw_463_cast)[name = tensor("op_10111_cast")]; + tensor var_10112_cast = softmax(axis = var_7228, x = aw_465_cast)[name = tensor("op_10112_cast")]; + tensor var_10113_cast = softmax(axis = var_7228, x = aw_467_cast)[name = tensor("op_10113_cast")]; + tensor var_10114_cast = softmax(axis = var_7228, x = aw_469_cast)[name = tensor("op_10114_cast")]; + tensor var_10115_cast = softmax(axis = var_7228, x = aw_471_cast)[name = tensor("op_10115_cast")]; + tensor var_10116_cast = softmax(axis = var_7228, x = aw_473_cast)[name = tensor("op_10116_cast")]; + tensor var_10117_cast = softmax(axis = var_7228, x = aw_475_cast)[name = tensor("op_10117_cast")]; + tensor var_10118_cast = softmax(axis = var_7228, x = aw_477_cast)[name = tensor("op_10118_cast")]; + tensor var_10119_cast = softmax(axis = var_7228, x = aw_cast)[name = tensor("op_10119_cast")]; + tensor var_10121_equation_0 = const()[name = tensor("op_10121_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_10121_cast = einsum(equation = var_10121_equation_0, values = (var_9941_cast, var_10100_cast))[name = tensor("op_10121_cast")]; + tensor var_10123_equation_0 = const()[name = tensor("op_10123_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_10123_cast = einsum(equation = var_10123_equation_0, values = (var_9945_cast, var_10101_cast))[name = tensor("op_10123_cast")]; + tensor var_10125_equation_0 = const()[name = tensor("op_10125_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_10125_cast = einsum(equation = var_10125_equation_0, values = (var_9949_cast, var_10102_cast))[name = tensor("op_10125_cast")]; + tensor var_10127_equation_0 = const()[name = tensor("op_10127_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_10127_cast = einsum(equation = var_10127_equation_0, values = (var_9953_cast, var_10103_cast))[name = tensor("op_10127_cast")]; + tensor var_10129_equation_0 = const()[name = tensor("op_10129_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_10129_cast = einsum(equation = var_10129_equation_0, values = (var_9957_cast, var_10104_cast))[name = tensor("op_10129_cast")]; + tensor var_10131_equation_0 = const()[name = tensor("op_10131_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_10131_cast = einsum(equation = var_10131_equation_0, values = (var_9961_cast, var_10105_cast))[name = tensor("op_10131_cast")]; + tensor var_10133_equation_0 = const()[name = tensor("op_10133_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_10133_cast = einsum(equation = var_10133_equation_0, values = (var_9965_cast, var_10106_cast))[name = tensor("op_10133_cast")]; + tensor var_10135_equation_0 = const()[name = tensor("op_10135_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_10135_cast = einsum(equation = var_10135_equation_0, values = (var_9969_cast, var_10107_cast))[name = tensor("op_10135_cast")]; + tensor var_10137_equation_0 = const()[name = tensor("op_10137_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_10137_cast = einsum(equation = var_10137_equation_0, values = (var_9973_cast, var_10108_cast))[name = tensor("op_10137_cast")]; + tensor var_10139_equation_0 = const()[name = tensor("op_10139_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_10139_cast = einsum(equation = var_10139_equation_0, values = (var_9977_cast, var_10109_cast))[name = tensor("op_10139_cast")]; + tensor var_10141_equation_0 = const()[name = tensor("op_10141_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_10141_cast = einsum(equation = var_10141_equation_0, values = (var_9981_cast, var_10110_cast))[name = tensor("op_10141_cast")]; + tensor var_10143_equation_0 = const()[name = tensor("op_10143_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_10143_cast = einsum(equation = var_10143_equation_0, values = (var_9985_cast, var_10111_cast))[name = tensor("op_10143_cast")]; + tensor var_10145_equation_0 = const()[name = tensor("op_10145_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_10145_cast = einsum(equation = var_10145_equation_0, values = (var_9989_cast, var_10112_cast))[name = tensor("op_10145_cast")]; + tensor var_10147_equation_0 = const()[name = tensor("op_10147_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_10147_cast = einsum(equation = var_10147_equation_0, values = (var_9993_cast, var_10113_cast))[name = tensor("op_10147_cast")]; + tensor var_10149_equation_0 = const()[name = tensor("op_10149_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_10149_cast = einsum(equation = var_10149_equation_0, values = (var_9997_cast, var_10114_cast))[name = tensor("op_10149_cast")]; + tensor var_10151_equation_0 = const()[name = tensor("op_10151_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_10151_cast = einsum(equation = var_10151_equation_0, values = (var_10001_cast, var_10115_cast))[name = tensor("op_10151_cast")]; + tensor var_10153_equation_0 = const()[name = tensor("op_10153_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_10153_cast = einsum(equation = var_10153_equation_0, values = (var_10005_cast, var_10116_cast))[name = tensor("op_10153_cast")]; + tensor var_10155_equation_0 = const()[name = tensor("op_10155_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_10155_cast = einsum(equation = var_10155_equation_0, values = (var_10009_cast, var_10117_cast))[name = tensor("op_10155_cast")]; + tensor var_10157_equation_0 = const()[name = tensor("op_10157_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_10157_cast = einsum(equation = var_10157_equation_0, values = (var_10013_cast, var_10118_cast))[name = tensor("op_10157_cast")]; + tensor var_10159_equation_0 = const()[name = tensor("op_10159_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_10159_cast = einsum(equation = var_10159_equation_0, values = (var_10017_cast, var_10119_cast))[name = tensor("op_10159_cast")]; + tensor input_355_interleave_0 = const()[name = tensor("input_355_interleave_0"), val = tensor(false)]; + tensor input_355_cast = concat(axis = var_7228, interleave = input_355_interleave_0, values = (var_10121_cast, var_10123_cast, var_10125_cast, var_10127_cast, var_10129_cast, var_10131_cast, var_10133_cast, var_10135_cast, var_10137_cast, var_10139_cast, var_10141_cast, var_10143_cast, var_10145_cast, var_10147_cast, var_10149_cast, var_10151_cast, var_10153_cast, var_10155_cast, var_10157_cast, var_10159_cast))[name = tensor("input_355_cast")]; + tensor var_10165 = const()[name = tensor("op_10165"), val = tensor([1, 1])]; + tensor var_10167 = const()[name = tensor("op_10167"), val = tensor([1, 1])]; + tensor var_10169_pad_type_0 = const()[name = tensor("op_10169_pad_type_0"), val = tensor("custom")]; + tensor var_10169_pad_0 = const()[name = tensor("op_10169_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_1_attentions_2_transformer_blocks_0_attn2_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(552764928))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(553993792))), name = tensor("up_blocks_1_attentions_2_transformer_blocks_0_attn2_to_out_0_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor up_blocks_1_attentions_2_transformer_blocks_0_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("up_blocks_1_attentions_2_transformer_blocks_0_attn2_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(553993984)))]; + tensor var_10169_cast = conv(bias = up_blocks_1_attentions_2_transformer_blocks_0_attn2_to_out_0_bias_to_fp16, dilations = var_10167, groups = var_7228, pad = var_10169_pad_0, pad_type = var_10169_pad_type_0, strides = var_10165, weight = up_blocks_1_attentions_2_transformer_blocks_0_attn2_to_out_0_weight_to_fp16_palettized, x = input_355_cast)[name = tensor("op_10169_cast")]; + tensor inputs_59_cast = add(x = var_10169_cast, y = inputs_57_cast)[name = tensor("inputs_59_cast")]; + tensor var_10173 = const()[name = tensor("op_10173"), val = tensor([1])]; + tensor channels_mean_59_cast = reduce_mean(axes = var_10173, keep_dims = var_7223, x = inputs_59_cast)[name = tensor("channels_mean_59_cast")]; + tensor zero_mean_59_cast = sub(x = inputs_59_cast, y = channels_mean_59_cast)[name = tensor("zero_mean_59_cast")]; + tensor zero_mean_sq_59_cast = mul(x = zero_mean_59_cast, y = zero_mean_59_cast)[name = tensor("zero_mean_sq_59_cast")]; + tensor var_10177 = const()[name = tensor("op_10177"), val = tensor([1])]; + tensor var_10178_cast = reduce_mean(axes = var_10177, keep_dims = var_7223, x = zero_mean_sq_59_cast)[name = tensor("op_10178_cast")]; + tensor var_10179_to_fp16 = const()[name = tensor("op_10179_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_10180_cast = add(x = var_10178_cast, y = var_10179_to_fp16)[name = tensor("op_10180_cast")]; + tensor denom_59_epsilon_0_to_fp16 = const()[name = tensor("denom_59_epsilon_0_to_fp16"), val = tensor(0x1p-24)]; + tensor denom_59_cast = rsqrt(epsilon = denom_59_epsilon_0_to_fp16, x = var_10180_cast)[name = tensor("denom_59_cast")]; + tensor out_59_cast = mul(x = zero_mean_59_cast, y = denom_59_cast)[name = tensor("out_59_cast")]; + tensor var_10184_to_fp16 = const()[name = tensor("op_10184_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(553996608)))]; + tensor var_10185_cast = add(x = out_59_cast, y = var_10184_to_fp16)[name = tensor("op_10185_cast")]; + tensor var_10187_to_fp16 = const()[name = tensor("op_10187_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(553999232)))]; + tensor input_357_cast = mul(x = var_10185_cast, y = var_10187_to_fp16)[name = tensor("input_357_cast")]; + tensor var_10195 = const()[name = tensor("op_10195"), val = tensor([1, 1])]; + tensor var_10197 = const()[name = tensor("op_10197"), val = tensor([1, 1])]; + tensor var_10199_pad_type_0 = const()[name = tensor("op_10199_pad_type_0"), val = tensor("custom")]; + tensor var_10199_pad_0 = const()[name = tensor("op_10199_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_1_attentions_2_transformer_blocks_0_ff_net_0_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(554001856))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(563832320))), name = tensor("up_blocks_1_attentions_2_transformer_blocks_0_ff_net_0_proj_weight_to_fp16_palettized"), shape = tensor([10240, 1280, 1, 1])]; + tensor up_blocks_1_attentions_2_transformer_blocks_0_ff_net_0_proj_bias_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(563832512))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(563840256))), name = tensor("up_blocks_1_attentions_2_transformer_blocks_0_ff_net_0_proj_bias_to_fp16_palettized"), shape = tensor([10240])]; + tensor var_10199_cast = conv(bias = up_blocks_1_attentions_2_transformer_blocks_0_ff_net_0_proj_bias_to_fp16_palettized, dilations = var_10197, groups = var_7228, pad = var_10199_pad_0, pad_type = var_10199_pad_type_0, strides = var_10195, weight = up_blocks_1_attentions_2_transformer_blocks_0_ff_net_0_proj_weight_to_fp16_palettized, x = input_357_cast)[name = tensor("op_10199_cast")]; + tensor var_10200_split_sizes_0 = const()[name = tensor("op_10200_split_sizes_0"), val = tensor([5120, 5120])]; + tensor var_10200_axis_0 = const()[name = tensor("op_10200_axis_0"), val = tensor(1)]; + tensor var_10200_cast_0, tensor var_10200_cast_1 = split(axis = var_10200_axis_0, split_sizes = var_10200_split_sizes_0, x = var_10199_cast)[name = tensor("op_10200_cast")]; + tensor var_10202_mode_0 = const()[name = tensor("op_10202_mode_0"), val = tensor("EXACT")]; + tensor var_10202_cast = gelu(mode = var_10202_mode_0, x = var_10200_cast_1)[name = tensor("op_10202_cast")]; + tensor input_359_cast = mul(x = var_10200_cast_0, y = var_10202_cast)[name = tensor("input_359_cast")]; + tensor var_10206 = const()[name = tensor("op_10206"), val = tensor([1, 1])]; + tensor var_10208 = const()[name = tensor("op_10208"), val = tensor([1, 1])]; + tensor var_10210_pad_type_0 = const()[name = tensor("op_10210_pad_type_0"), val = tensor("custom")]; + tensor var_10210_pad_0 = const()[name = tensor("op_10210_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_1_attentions_2_transformer_blocks_0_ff_net_2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(563840448))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(568755712))), name = tensor("up_blocks_1_attentions_2_transformer_blocks_0_ff_net_2_weight_to_fp16_palettized"), shape = tensor([1280, 5120, 1, 1])]; + tensor up_blocks_1_attentions_2_transformer_blocks_0_ff_net_2_bias_to_fp16 = const()[name = tensor("up_blocks_1_attentions_2_transformer_blocks_0_ff_net_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(568755904)))]; + tensor var_10210_cast = conv(bias = up_blocks_1_attentions_2_transformer_blocks_0_ff_net_2_bias_to_fp16, dilations = var_10208, groups = var_7228, pad = var_10210_pad_0, pad_type = var_10210_pad_type_0, strides = var_10206, weight = up_blocks_1_attentions_2_transformer_blocks_0_ff_net_2_weight_to_fp16_palettized, x = input_359_cast)[name = tensor("op_10210_cast")]; + tensor hidden_states_213_cast = add(x = var_10210_cast, y = inputs_59_cast)[name = tensor("hidden_states_213_cast")]; + tensor var_10212 = const()[name = tensor("op_10212"), val = tensor([2, 1280, 16, 16])]; + tensor input_361_cast = reshape(shape = var_10212, x = hidden_states_213_cast)[name = tensor("input_361_cast")]; + tensor var_10216 = const()[name = tensor("op_10216"), val = tensor([1, 1])]; + tensor var_10218 = const()[name = tensor("op_10218"), val = tensor([1, 1])]; + tensor hidden_states_215_pad_type_0 = const()[name = tensor("hidden_states_215_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_215_pad_0 = const()[name = tensor("hidden_states_215_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_1_attentions_2_proj_out_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(568758528))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(569987392))), name = tensor("up_blocks_1_attentions_2_proj_out_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor up_blocks_1_attentions_2_proj_out_bias_to_fp16 = const()[name = tensor("up_blocks_1_attentions_2_proj_out_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(569987584)))]; + tensor hidden_states_215_cast = conv(bias = up_blocks_1_attentions_2_proj_out_bias_to_fp16, dilations = var_10218, groups = var_7228, pad = hidden_states_215_pad_0, pad_type = hidden_states_215_pad_type_0, strides = var_10216, weight = up_blocks_1_attentions_2_proj_out_weight_to_fp16_palettized, x = input_361_cast)[name = tensor("hidden_states_215_cast")]; + tensor input_363_cast = add(x = hidden_states_215_cast, y = hidden_states_203_cast)[name = tensor("input_363_cast")]; + tensor input_365_scale_factor_height_0 = const()[name = tensor("input_365_scale_factor_height_0"), val = tensor(0x1p+1)]; + tensor input_365_scale_factor_width_0 = const()[name = tensor("input_365_scale_factor_width_0"), val = tensor(0x1p+1)]; + tensor input_365_cast = upsample_nearest_neighbor(scale_factor_height = input_365_scale_factor_height_0, scale_factor_width = input_365_scale_factor_width_0, x = input_363_cast)[name = tensor("input_365_cast")]; + tensor var_10227 = const()[name = tensor("op_10227"), val = tensor([1, 1])]; + tensor var_10229 = const()[name = tensor("op_10229"), val = tensor([1, 1])]; + tensor hidden_states_217_pad_type_0 = const()[name = tensor("hidden_states_217_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_217_pad_0 = const()[name = tensor("hidden_states_217_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor up_blocks_1_upsamplers_0_conv_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(569990208))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(581049472))), name = tensor("up_blocks_1_upsamplers_0_conv_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 3, 3])]; + tensor up_blocks_1_upsamplers_0_conv_bias_to_fp16 = const()[name = tensor("up_blocks_1_upsamplers_0_conv_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(581049664)))]; + tensor hidden_states_217_cast = conv(bias = up_blocks_1_upsamplers_0_conv_bias_to_fp16, dilations = var_10229, groups = var_7228, pad = hidden_states_217_pad_0, pad_type = hidden_states_217_pad_type_0, strides = var_10227, weight = up_blocks_1_upsamplers_0_conv_weight_to_fp16_palettized, x = input_365_cast)[name = tensor("hidden_states_217_cast")]; + tensor var_10234 = const()[name = tensor("op_10234"), val = tensor(3)]; + tensor var_10252 = const()[name = tensor("op_10252"), val = tensor(true)]; + tensor var_10257 = const()[name = tensor("op_10257"), val = tensor(1)]; + tensor input_367_interleave_0 = const()[name = tensor("input_367_interleave_0"), val = tensor(false)]; + tensor input_367_cast = concat(axis = var_10257, interleave = input_367_interleave_0, values = (hidden_states_217_cast, input_115_cast))[name = tensor("input_367_cast")]; + tensor reshape_168_shape_0 = const()[name = tensor("reshape_168_shape_0"), val = tensor([2, 32, 60, 32, 32])]; + tensor reshape_168_cast = reshape(shape = reshape_168_shape_0, x = input_367_cast)[name = tensor("reshape_168_cast")]; + tensor reduce_mean_126_axes_0 = const()[name = tensor("reduce_mean_126_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_126_keep_dims_0 = const()[name = tensor("reduce_mean_126_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_126_cast = reduce_mean(axes = reduce_mean_126_axes_0, keep_dims = reduce_mean_126_keep_dims_0, x = reshape_168_cast)[name = tensor("reduce_mean_126_cast")]; + tensor sub_84_cast = sub(x = reshape_168_cast, y = reduce_mean_126_cast)[name = tensor("sub_84_cast")]; + tensor square_42_cast = square(x = sub_84_cast)[name = tensor("square_42_cast")]; + tensor reduce_mean_128_axes_0 = const()[name = tensor("reduce_mean_128_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_128_keep_dims_0 = const()[name = tensor("reduce_mean_128_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_128_cast = reduce_mean(axes = reduce_mean_128_axes_0, keep_dims = reduce_mean_128_keep_dims_0, x = square_42_cast)[name = tensor("reduce_mean_128_cast")]; + tensor add_84_y_0_to_fp16 = const()[name = tensor("add_84_y_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_84_cast = add(x = reduce_mean_128_cast, y = add_84_y_0_to_fp16)[name = tensor("add_84_cast")]; + tensor sqrt_42_cast = sqrt(x = add_84_cast)[name = tensor("sqrt_42_cast")]; + tensor real_div_42_cast = real_div(x = sub_84_cast, y = sqrt_42_cast)[name = tensor("real_div_42_cast")]; + tensor reshape_169_shape_0 = const()[name = tensor("reshape_169_shape_0"), val = tensor([2, 1920, 32, 32])]; + tensor reshape_169_cast = reshape(shape = reshape_169_shape_0, x = real_div_42_cast)[name = tensor("reshape_169_cast")]; + tensor add_85_gamma_0_to_fp16 = const()[name = tensor("add_85_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(581052288)))]; + tensor add_85_beta_0_to_fp16 = const()[name = tensor("add_85_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(581056192)))]; + tensor add_85_epsilon_0_to_fp16 = const()[name = tensor("add_85_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_85_cast = batch_norm(beta = add_85_beta_0_to_fp16, epsilon = add_85_epsilon_0_to_fp16, gamma = add_85_gamma_0_to_fp16, mean = add_79_mean_0_to_fp16, variance = add_79_variance_0_to_fp16, x = reshape_169_cast)[name = tensor("add_85_cast")]; + tensor input_371_cast = silu(x = add_85_cast)[name = tensor("input_371_cast")]; + tensor var_10286 = const()[name = tensor("op_10286"), val = tensor([1, 1])]; + tensor var_10288 = const()[name = tensor("op_10288"), val = tensor([1, 1])]; + tensor hidden_states_219_pad_type_0 = const()[name = tensor("hidden_states_219_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_219_pad_0 = const()[name = tensor("hidden_states_219_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor up_blocks_2_resnets_0_conv1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(581060096))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(589354560))), name = tensor("up_blocks_2_resnets_0_conv1_weight_to_fp16_palettized"), shape = tensor([640, 1920, 3, 3])]; + tensor up_blocks_2_resnets_0_conv1_bias_to_fp16 = const()[name = tensor("up_blocks_2_resnets_0_conv1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(589354752)))]; + tensor hidden_states_219_cast = conv(bias = up_blocks_2_resnets_0_conv1_bias_to_fp16, dilations = var_10288, groups = var_10257, pad = hidden_states_219_pad_0, pad_type = hidden_states_219_pad_type_0, strides = var_10286, weight = up_blocks_2_resnets_0_conv1_weight_to_fp16_palettized, x = input_371_cast)[name = tensor("hidden_states_219_cast")]; + tensor var_10294 = const()[name = tensor("op_10294"), val = tensor([1, 1])]; + tensor var_10296 = const()[name = tensor("op_10296"), val = tensor([1, 1])]; + tensor temb_33_pad_type_0 = const()[name = tensor("temb_33_pad_type_0"), val = tensor("custom")]; + tensor temb_33_pad_0 = const()[name = tensor("temb_33_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_2_resnets_0_time_emb_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(589356096))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(589970560))), name = tensor("up_blocks_2_resnets_0_time_emb_proj_weight_to_fp16_palettized"), shape = tensor([640, 1280, 1, 1])]; + tensor up_blocks_2_resnets_0_time_emb_proj_bias_to_fp16 = const()[name = tensor("up_blocks_2_resnets_0_time_emb_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(589970752)))]; + tensor temb_33_cast = conv(bias = up_blocks_2_resnets_0_time_emb_proj_bias_to_fp16, dilations = var_10296, groups = var_10257, pad = temb_33_pad_0, pad_type = temb_33_pad_type_0, strides = var_10294, weight = up_blocks_2_resnets_0_time_emb_proj_weight_to_fp16_palettized, x = input_15_cast)[name = tensor("temb_33_cast")]; + tensor input_375_cast = add(x = hidden_states_219_cast, y = temb_33_cast)[name = tensor("input_375_cast")]; + tensor reshape_172_shape_0 = const()[name = tensor("reshape_172_shape_0"), val = tensor([2, 32, 20, 32, 32])]; + tensor reshape_172_cast = reshape(shape = reshape_172_shape_0, x = input_375_cast)[name = tensor("reshape_172_cast")]; + tensor reduce_mean_129_axes_0 = const()[name = tensor("reduce_mean_129_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_129_keep_dims_0 = const()[name = tensor("reduce_mean_129_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_129_cast = reduce_mean(axes = reduce_mean_129_axes_0, keep_dims = reduce_mean_129_keep_dims_0, x = reshape_172_cast)[name = tensor("reduce_mean_129_cast")]; + tensor sub_86_cast = sub(x = reshape_172_cast, y = reduce_mean_129_cast)[name = tensor("sub_86_cast")]; + tensor square_43_cast = square(x = sub_86_cast)[name = tensor("square_43_cast")]; + tensor reduce_mean_131_axes_0 = const()[name = tensor("reduce_mean_131_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_131_keep_dims_0 = const()[name = tensor("reduce_mean_131_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_131_cast = reduce_mean(axes = reduce_mean_131_axes_0, keep_dims = reduce_mean_131_keep_dims_0, x = square_43_cast)[name = tensor("reduce_mean_131_cast")]; + tensor add_86_y_0_to_fp16 = const()[name = tensor("add_86_y_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_86_cast = add(x = reduce_mean_131_cast, y = add_86_y_0_to_fp16)[name = tensor("add_86_cast")]; + tensor sqrt_43_cast = sqrt(x = add_86_cast)[name = tensor("sqrt_43_cast")]; + tensor real_div_43_cast = real_div(x = sub_86_cast, y = sqrt_43_cast)[name = tensor("real_div_43_cast")]; + tensor reshape_173_shape_0 = const()[name = tensor("reshape_173_shape_0"), val = tensor([2, 640, 32, 32])]; + tensor reshape_173_cast = reshape(shape = reshape_173_shape_0, x = real_div_43_cast)[name = tensor("reshape_173_cast")]; + tensor add_87_gamma_0_to_fp16 = const()[name = tensor("add_87_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(589972096)))]; + tensor add_87_beta_0_to_fp16 = const()[name = tensor("add_87_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(589973440)))]; + tensor add_87_epsilon_0_to_fp16 = const()[name = tensor("add_87_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_87_cast = batch_norm(beta = add_87_beta_0_to_fp16, epsilon = add_87_epsilon_0_to_fp16, gamma = add_87_gamma_0_to_fp16, mean = add_15_mean_0_to_fp16, variance = add_15_variance_0_to_fp16, x = reshape_173_cast)[name = tensor("add_87_cast")]; + tensor input_379_cast = silu(x = add_87_cast)[name = tensor("input_379_cast")]; + tensor var_10306 = const()[name = tensor("op_10306"), val = tensor([1, 1])]; + tensor var_10308 = const()[name = tensor("op_10308"), val = tensor([1, 1])]; + tensor hidden_states_221_pad_type_0 = const()[name = tensor("hidden_states_221_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_221_pad_0 = const()[name = tensor("hidden_states_221_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor up_blocks_2_resnets_0_conv2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(589974784))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(592739648))), name = tensor("up_blocks_2_resnets_0_conv2_weight_to_fp16_palettized"), shape = tensor([640, 640, 3, 3])]; + tensor up_blocks_2_resnets_0_conv2_bias_to_fp16 = const()[name = tensor("up_blocks_2_resnets_0_conv2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(592739840)))]; + tensor hidden_states_221_cast = conv(bias = up_blocks_2_resnets_0_conv2_bias_to_fp16, dilations = var_10308, groups = var_10257, pad = hidden_states_221_pad_0, pad_type = hidden_states_221_pad_type_0, strides = var_10306, weight = up_blocks_2_resnets_0_conv2_weight_to_fp16_palettized, x = input_379_cast)[name = tensor("hidden_states_221_cast")]; + tensor var_10313 = const()[name = tensor("op_10313"), val = tensor([1, 1])]; + tensor var_10315 = const()[name = tensor("op_10315"), val = tensor([1, 1])]; + tensor x_17_pad_type_0 = const()[name = tensor("x_17_pad_type_0"), val = tensor("custom")]; + tensor x_17_pad_0 = const()[name = tensor("x_17_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_2_resnets_0_conv_shortcut_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(592741184))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(593662848))), name = tensor("up_blocks_2_resnets_0_conv_shortcut_weight_to_fp16_palettized"), shape = tensor([640, 1920, 1, 1])]; + tensor up_blocks_2_resnets_0_conv_shortcut_bias_to_fp16 = const()[name = tensor("up_blocks_2_resnets_0_conv_shortcut_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(593663040)))]; + tensor x_17_cast = conv(bias = up_blocks_2_resnets_0_conv_shortcut_bias_to_fp16, dilations = var_10315, groups = var_10257, pad = x_17_pad_0, pad_type = x_17_pad_type_0, strides = var_10313, weight = up_blocks_2_resnets_0_conv_shortcut_weight_to_fp16_palettized, x = input_367_cast)[name = tensor("x_17_cast")]; + tensor hidden_states_223_cast = add(x = x_17_cast, y = hidden_states_221_cast)[name = tensor("hidden_states_223_cast")]; + tensor reshape_176_shape_0 = const()[name = tensor("reshape_176_shape_0"), val = tensor([2, 32, 20, 32, 32])]; + tensor reshape_176_cast = reshape(shape = reshape_176_shape_0, x = hidden_states_223_cast)[name = tensor("reshape_176_cast")]; + tensor reduce_mean_132_axes_0 = const()[name = tensor("reduce_mean_132_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_132_keep_dims_0 = const()[name = tensor("reduce_mean_132_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_132_cast = reduce_mean(axes = reduce_mean_132_axes_0, keep_dims = reduce_mean_132_keep_dims_0, x = reshape_176_cast)[name = tensor("reduce_mean_132_cast")]; + tensor sub_88_cast = sub(x = reshape_176_cast, y = reduce_mean_132_cast)[name = tensor("sub_88_cast")]; + tensor square_44_cast = square(x = sub_88_cast)[name = tensor("square_44_cast")]; + tensor reduce_mean_134_axes_0 = const()[name = tensor("reduce_mean_134_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_134_keep_dims_0 = const()[name = tensor("reduce_mean_134_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_134_cast = reduce_mean(axes = reduce_mean_134_axes_0, keep_dims = reduce_mean_134_keep_dims_0, x = square_44_cast)[name = tensor("reduce_mean_134_cast")]; + tensor add_88_y_0_to_fp16 = const()[name = tensor("add_88_y_0_to_fp16"), val = tensor(0x1.1p-20)]; + tensor add_88_cast = add(x = reduce_mean_134_cast, y = add_88_y_0_to_fp16)[name = tensor("add_88_cast")]; + tensor sqrt_44_cast = sqrt(x = add_88_cast)[name = tensor("sqrt_44_cast")]; + tensor real_div_44_cast = real_div(x = sub_88_cast, y = sqrt_44_cast)[name = tensor("real_div_44_cast")]; + tensor reshape_177_shape_0 = const()[name = tensor("reshape_177_shape_0"), val = tensor([2, 640, 32, 32])]; + tensor reshape_177_cast = reshape(shape = reshape_177_shape_0, x = real_div_44_cast)[name = tensor("reshape_177_cast")]; + tensor add_89_gamma_0_to_fp16 = const()[name = tensor("add_89_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(593664384)))]; + tensor add_89_beta_0_to_fp16 = const()[name = tensor("add_89_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(593665728)))]; + tensor add_89_epsilon_0_to_fp16 = const()[name = tensor("add_89_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_89_cast = batch_norm(beta = add_89_beta_0_to_fp16, epsilon = add_89_epsilon_0_to_fp16, gamma = add_89_gamma_0_to_fp16, mean = add_15_mean_0_to_fp16, variance = add_15_variance_0_to_fp16, x = reshape_177_cast)[name = tensor("add_89_cast")]; + tensor var_10335 = const()[name = tensor("op_10335"), val = tensor([1, 1])]; + tensor var_10337 = const()[name = tensor("op_10337"), val = tensor([1, 1])]; + tensor hidden_states_225_pad_type_0 = const()[name = tensor("hidden_states_225_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_225_pad_0 = const()[name = tensor("hidden_states_225_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_2_attentions_0_proj_in_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(593667072))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(593974336))), name = tensor("up_blocks_2_attentions_0_proj_in_weight_to_fp16_palettized"), shape = tensor([640, 640, 1, 1])]; + tensor up_blocks_2_attentions_0_proj_in_bias_to_fp16 = const()[name = tensor("up_blocks_2_attentions_0_proj_in_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(593974528)))]; + tensor hidden_states_225_cast = conv(bias = up_blocks_2_attentions_0_proj_in_bias_to_fp16, dilations = var_10337, groups = var_10257, pad = hidden_states_225_pad_0, pad_type = hidden_states_225_pad_type_0, strides = var_10335, weight = up_blocks_2_attentions_0_proj_in_weight_to_fp16_palettized, x = add_89_cast)[name = tensor("hidden_states_225_cast")]; + tensor var_10342 = const()[name = tensor("op_10342"), val = tensor([2, 640, 1, 1024])]; + tensor inputs_61_cast = reshape(shape = var_10342, x = hidden_states_225_cast)[name = tensor("inputs_61_cast")]; + tensor var_10352 = const()[name = tensor("op_10352"), val = tensor([1])]; + tensor channels_mean_61_cast = reduce_mean(axes = var_10352, keep_dims = var_10252, x = inputs_61_cast)[name = tensor("channels_mean_61_cast")]; + tensor zero_mean_61_cast = sub(x = inputs_61_cast, y = channels_mean_61_cast)[name = tensor("zero_mean_61_cast")]; + tensor zero_mean_sq_61_cast = mul(x = zero_mean_61_cast, y = zero_mean_61_cast)[name = tensor("zero_mean_sq_61_cast")]; + tensor var_10356 = const()[name = tensor("op_10356"), val = tensor([1])]; + tensor var_10357_cast = reduce_mean(axes = var_10356, keep_dims = var_10252, x = zero_mean_sq_61_cast)[name = tensor("op_10357_cast")]; + tensor var_10358_to_fp16 = const()[name = tensor("op_10358_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_10359_cast = add(x = var_10357_cast, y = var_10358_to_fp16)[name = tensor("op_10359_cast")]; + tensor denom_61_epsilon_0_to_fp16 = const()[name = tensor("denom_61_epsilon_0_to_fp16"), val = tensor(0x1p-24)]; + tensor denom_61_cast = rsqrt(epsilon = denom_61_epsilon_0_to_fp16, x = var_10359_cast)[name = tensor("denom_61_cast")]; + tensor out_61_cast = mul(x = zero_mean_61_cast, y = denom_61_cast)[name = tensor("out_61_cast")]; + tensor var_10363_to_fp16 = const()[name = tensor("op_10363_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(593975872)))]; + tensor var_10364_cast = add(x = out_61_cast, y = var_10363_to_fp16)[name = tensor("op_10364_cast")]; + tensor var_10366_to_fp16 = const()[name = tensor("op_10366_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(593977216)))]; + tensor hidden_states_227_cast = mul(x = var_10364_cast, y = var_10366_to_fp16)[name = tensor("hidden_states_227_cast")]; + tensor var_10373 = const()[name = tensor("op_10373"), val = tensor([1, 1])]; + tensor var_10375 = const()[name = tensor("op_10375"), val = tensor([1, 1])]; + tensor q_41_pad_type_0 = const()[name = tensor("q_41_pad_type_0"), val = tensor("custom")]; + tensor q_41_pad_0 = const()[name = tensor("q_41_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_2_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(593978560))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(594285824))), name = tensor("up_blocks_2_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16_palettized"), shape = tensor([640, 640, 1, 1])]; + tensor q_41_cast = conv(dilations = var_10375, groups = var_10257, pad = q_41_pad_0, pad_type = q_41_pad_type_0, strides = var_10373, weight = up_blocks_2_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16_palettized, x = hidden_states_227_cast)[name = tensor("q_41_cast")]; + tensor var_10379 = const()[name = tensor("op_10379"), val = tensor([1, 1])]; + tensor var_10381 = const()[name = tensor("op_10381"), val = tensor([1, 1])]; + tensor k_81_pad_type_0 = const()[name = tensor("k_81_pad_type_0"), val = tensor("custom")]; + tensor k_81_pad_0 = const()[name = tensor("k_81_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_2_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(594286016))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(594593280))), name = tensor("up_blocks_2_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16_palettized"), shape = tensor([640, 640, 1, 1])]; + tensor k_81_cast = conv(dilations = var_10381, groups = var_10257, pad = k_81_pad_0, pad_type = k_81_pad_type_0, strides = var_10379, weight = up_blocks_2_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16_palettized, x = hidden_states_227_cast)[name = tensor("k_81_cast")]; + tensor var_10385 = const()[name = tensor("op_10385"), val = tensor([1, 1])]; + tensor var_10387 = const()[name = tensor("op_10387"), val = tensor([1, 1])]; + tensor v_41_pad_type_0 = const()[name = tensor("v_41_pad_type_0"), val = tensor("custom")]; + tensor v_41_pad_0 = const()[name = tensor("v_41_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_2_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(594593472))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(594900736))), name = tensor("up_blocks_2_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16_palettized"), shape = tensor([640, 640, 1, 1])]; + tensor v_41_cast = conv(dilations = var_10387, groups = var_10257, pad = v_41_pad_0, pad_type = v_41_pad_type_0, strides = var_10385, weight = up_blocks_2_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16_palettized, x = hidden_states_227_cast)[name = tensor("v_41_cast")]; + tensor var_10391_begin_0 = const()[name = tensor("op_10391_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_10391_end_0 = const()[name = tensor("op_10391_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_10391_end_mask_0 = const()[name = tensor("op_10391_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10391_cast = slice_by_index(begin = var_10391_begin_0, end = var_10391_end_0, end_mask = var_10391_end_mask_0, x = q_41_cast)[name = tensor("op_10391_cast")]; + tensor var_10395_begin_0 = const()[name = tensor("op_10395_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_10395_end_0 = const()[name = tensor("op_10395_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_10395_end_mask_0 = const()[name = tensor("op_10395_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10395_cast = slice_by_index(begin = var_10395_begin_0, end = var_10395_end_0, end_mask = var_10395_end_mask_0, x = q_41_cast)[name = tensor("op_10395_cast")]; + tensor var_10399_begin_0 = const()[name = tensor("op_10399_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_10399_end_0 = const()[name = tensor("op_10399_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_10399_end_mask_0 = const()[name = tensor("op_10399_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10399_cast = slice_by_index(begin = var_10399_begin_0, end = var_10399_end_0, end_mask = var_10399_end_mask_0, x = q_41_cast)[name = tensor("op_10399_cast")]; + tensor var_10403_begin_0 = const()[name = tensor("op_10403_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_10403_end_0 = const()[name = tensor("op_10403_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_10403_end_mask_0 = const()[name = tensor("op_10403_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10403_cast = slice_by_index(begin = var_10403_begin_0, end = var_10403_end_0, end_mask = var_10403_end_mask_0, x = q_41_cast)[name = tensor("op_10403_cast")]; + tensor var_10407_begin_0 = const()[name = tensor("op_10407_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_10407_end_0 = const()[name = tensor("op_10407_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_10407_end_mask_0 = const()[name = tensor("op_10407_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10407_cast = slice_by_index(begin = var_10407_begin_0, end = var_10407_end_0, end_mask = var_10407_end_mask_0, x = q_41_cast)[name = tensor("op_10407_cast")]; + tensor var_10411_begin_0 = const()[name = tensor("op_10411_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_10411_end_0 = const()[name = tensor("op_10411_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_10411_end_mask_0 = const()[name = tensor("op_10411_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10411_cast = slice_by_index(begin = var_10411_begin_0, end = var_10411_end_0, end_mask = var_10411_end_mask_0, x = q_41_cast)[name = tensor("op_10411_cast")]; + tensor var_10415_begin_0 = const()[name = tensor("op_10415_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_10415_end_0 = const()[name = tensor("op_10415_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_10415_end_mask_0 = const()[name = tensor("op_10415_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10415_cast = slice_by_index(begin = var_10415_begin_0, end = var_10415_end_0, end_mask = var_10415_end_mask_0, x = q_41_cast)[name = tensor("op_10415_cast")]; + tensor var_10419_begin_0 = const()[name = tensor("op_10419_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_10419_end_0 = const()[name = tensor("op_10419_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_10419_end_mask_0 = const()[name = tensor("op_10419_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10419_cast = slice_by_index(begin = var_10419_begin_0, end = var_10419_end_0, end_mask = var_10419_end_mask_0, x = q_41_cast)[name = tensor("op_10419_cast")]; + tensor var_10423_begin_0 = const()[name = tensor("op_10423_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_10423_end_0 = const()[name = tensor("op_10423_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_10423_end_mask_0 = const()[name = tensor("op_10423_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10423_cast = slice_by_index(begin = var_10423_begin_0, end = var_10423_end_0, end_mask = var_10423_end_mask_0, x = q_41_cast)[name = tensor("op_10423_cast")]; + tensor var_10427_begin_0 = const()[name = tensor("op_10427_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_10427_end_0 = const()[name = tensor("op_10427_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_10427_end_mask_0 = const()[name = tensor("op_10427_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10427_cast = slice_by_index(begin = var_10427_begin_0, end = var_10427_end_0, end_mask = var_10427_end_mask_0, x = q_41_cast)[name = tensor("op_10427_cast")]; + tensor var_10430_begin_0 = const()[name = tensor("op_10430_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_10430_end_0 = const()[name = tensor("op_10430_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_10430_end_mask_0 = const()[name = tensor("op_10430_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10430_cast = slice_by_index(begin = var_10430_begin_0, end = var_10430_end_0, end_mask = var_10430_end_mask_0, x = var_10391_cast)[name = tensor("op_10430_cast")]; + tensor var_10431_begin_0 = const()[name = tensor("op_10431_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_10431_end_0 = const()[name = tensor("op_10431_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_10431_end_mask_0 = const()[name = tensor("op_10431_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10431_cast = slice_by_index(begin = var_10431_begin_0, end = var_10431_end_0, end_mask = var_10431_end_mask_0, x = var_10391_cast)[name = tensor("op_10431_cast")]; + tensor var_10432_begin_0 = const()[name = tensor("op_10432_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_10432_end_0 = const()[name = tensor("op_10432_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_10432_end_mask_0 = const()[name = tensor("op_10432_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10432_cast = slice_by_index(begin = var_10432_begin_0, end = var_10432_end_0, end_mask = var_10432_end_mask_0, x = var_10395_cast)[name = tensor("op_10432_cast")]; + tensor var_10433_begin_0 = const()[name = tensor("op_10433_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_10433_end_0 = const()[name = tensor("op_10433_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_10433_end_mask_0 = const()[name = tensor("op_10433_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10433_cast = slice_by_index(begin = var_10433_begin_0, end = var_10433_end_0, end_mask = var_10433_end_mask_0, x = var_10395_cast)[name = tensor("op_10433_cast")]; + tensor var_10434_begin_0 = const()[name = tensor("op_10434_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_10434_end_0 = const()[name = tensor("op_10434_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_10434_end_mask_0 = const()[name = tensor("op_10434_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10434_cast = slice_by_index(begin = var_10434_begin_0, end = var_10434_end_0, end_mask = var_10434_end_mask_0, x = var_10399_cast)[name = tensor("op_10434_cast")]; + tensor var_10435_begin_0 = const()[name = tensor("op_10435_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_10435_end_0 = const()[name = tensor("op_10435_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_10435_end_mask_0 = const()[name = tensor("op_10435_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10435_cast = slice_by_index(begin = var_10435_begin_0, end = var_10435_end_0, end_mask = var_10435_end_mask_0, x = var_10399_cast)[name = tensor("op_10435_cast")]; + tensor var_10436_begin_0 = const()[name = tensor("op_10436_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_10436_end_0 = const()[name = tensor("op_10436_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_10436_end_mask_0 = const()[name = tensor("op_10436_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10436_cast = slice_by_index(begin = var_10436_begin_0, end = var_10436_end_0, end_mask = var_10436_end_mask_0, x = var_10403_cast)[name = tensor("op_10436_cast")]; + tensor var_10437_begin_0 = const()[name = tensor("op_10437_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_10437_end_0 = const()[name = tensor("op_10437_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_10437_end_mask_0 = const()[name = tensor("op_10437_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10437_cast = slice_by_index(begin = var_10437_begin_0, end = var_10437_end_0, end_mask = var_10437_end_mask_0, x = var_10403_cast)[name = tensor("op_10437_cast")]; + tensor var_10438_begin_0 = const()[name = tensor("op_10438_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_10438_end_0 = const()[name = tensor("op_10438_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_10438_end_mask_0 = const()[name = tensor("op_10438_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10438_cast = slice_by_index(begin = var_10438_begin_0, end = var_10438_end_0, end_mask = var_10438_end_mask_0, x = var_10407_cast)[name = tensor("op_10438_cast")]; + tensor var_10439_begin_0 = const()[name = tensor("op_10439_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_10439_end_0 = const()[name = tensor("op_10439_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_10439_end_mask_0 = const()[name = tensor("op_10439_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10439_cast = slice_by_index(begin = var_10439_begin_0, end = var_10439_end_0, end_mask = var_10439_end_mask_0, x = var_10407_cast)[name = tensor("op_10439_cast")]; + tensor var_10440_begin_0 = const()[name = tensor("op_10440_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_10440_end_0 = const()[name = tensor("op_10440_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_10440_end_mask_0 = const()[name = tensor("op_10440_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10440_cast = slice_by_index(begin = var_10440_begin_0, end = var_10440_end_0, end_mask = var_10440_end_mask_0, x = var_10411_cast)[name = tensor("op_10440_cast")]; + tensor var_10441_begin_0 = const()[name = tensor("op_10441_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_10441_end_0 = const()[name = tensor("op_10441_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_10441_end_mask_0 = const()[name = tensor("op_10441_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10441_cast = slice_by_index(begin = var_10441_begin_0, end = var_10441_end_0, end_mask = var_10441_end_mask_0, x = var_10411_cast)[name = tensor("op_10441_cast")]; + tensor var_10442_begin_0 = const()[name = tensor("op_10442_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_10442_end_0 = const()[name = tensor("op_10442_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_10442_end_mask_0 = const()[name = tensor("op_10442_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10442_cast = slice_by_index(begin = var_10442_begin_0, end = var_10442_end_0, end_mask = var_10442_end_mask_0, x = var_10415_cast)[name = tensor("op_10442_cast")]; + tensor var_10443_begin_0 = const()[name = tensor("op_10443_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_10443_end_0 = const()[name = tensor("op_10443_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_10443_end_mask_0 = const()[name = tensor("op_10443_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10443_cast = slice_by_index(begin = var_10443_begin_0, end = var_10443_end_0, end_mask = var_10443_end_mask_0, x = var_10415_cast)[name = tensor("op_10443_cast")]; + tensor var_10444_begin_0 = const()[name = tensor("op_10444_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_10444_end_0 = const()[name = tensor("op_10444_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_10444_end_mask_0 = const()[name = tensor("op_10444_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10444_cast = slice_by_index(begin = var_10444_begin_0, end = var_10444_end_0, end_mask = var_10444_end_mask_0, x = var_10419_cast)[name = tensor("op_10444_cast")]; + tensor var_10445_begin_0 = const()[name = tensor("op_10445_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_10445_end_0 = const()[name = tensor("op_10445_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_10445_end_mask_0 = const()[name = tensor("op_10445_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10445_cast = slice_by_index(begin = var_10445_begin_0, end = var_10445_end_0, end_mask = var_10445_end_mask_0, x = var_10419_cast)[name = tensor("op_10445_cast")]; + tensor var_10446_begin_0 = const()[name = tensor("op_10446_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_10446_end_0 = const()[name = tensor("op_10446_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_10446_end_mask_0 = const()[name = tensor("op_10446_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10446_cast = slice_by_index(begin = var_10446_begin_0, end = var_10446_end_0, end_mask = var_10446_end_mask_0, x = var_10423_cast)[name = tensor("op_10446_cast")]; + tensor var_10447_begin_0 = const()[name = tensor("op_10447_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_10447_end_0 = const()[name = tensor("op_10447_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_10447_end_mask_0 = const()[name = tensor("op_10447_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10447_cast = slice_by_index(begin = var_10447_begin_0, end = var_10447_end_0, end_mask = var_10447_end_mask_0, x = var_10423_cast)[name = tensor("op_10447_cast")]; + tensor var_10448_begin_0 = const()[name = tensor("op_10448_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_10448_end_0 = const()[name = tensor("op_10448_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_10448_end_mask_0 = const()[name = tensor("op_10448_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10448_cast = slice_by_index(begin = var_10448_begin_0, end = var_10448_end_0, end_mask = var_10448_end_mask_0, x = var_10427_cast)[name = tensor("op_10448_cast")]; + tensor var_10449_begin_0 = const()[name = tensor("op_10449_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_10449_end_0 = const()[name = tensor("op_10449_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_10449_end_mask_0 = const()[name = tensor("op_10449_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10449_cast = slice_by_index(begin = var_10449_begin_0, end = var_10449_end_0, end_mask = var_10449_end_mask_0, x = var_10427_cast)[name = tensor("op_10449_cast")]; + tensor k_83_perm_0 = const()[name = tensor("k_83_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_10454_begin_0 = const()[name = tensor("op_10454_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_10454_end_0 = const()[name = tensor("op_10454_end_0"), val = tensor([2, 1024, 1, 64])]; + tensor var_10454_end_mask_0 = const()[name = tensor("op_10454_end_mask_0"), val = tensor([true, true, true, false])]; + tensor transpose_11 = transpose(perm = k_83_perm_0, x = k_81_cast)[name = tensor("transpose_11")]; + tensor var_10454_cast = slice_by_index(begin = var_10454_begin_0, end = var_10454_end_0, end_mask = var_10454_end_mask_0, x = transpose_11)[name = tensor("op_10454_cast")]; + tensor var_10458_begin_0 = const()[name = tensor("op_10458_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_10458_end_0 = const()[name = tensor("op_10458_end_0"), val = tensor([2, 1024, 1, 128])]; + tensor var_10458_end_mask_0 = const()[name = tensor("op_10458_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10458_cast = slice_by_index(begin = var_10458_begin_0, end = var_10458_end_0, end_mask = var_10458_end_mask_0, x = transpose_11)[name = tensor("op_10458_cast")]; + tensor var_10462_begin_0 = const()[name = tensor("op_10462_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_10462_end_0 = const()[name = tensor("op_10462_end_0"), val = tensor([2, 1024, 1, 192])]; + tensor var_10462_end_mask_0 = const()[name = tensor("op_10462_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10462_cast = slice_by_index(begin = var_10462_begin_0, end = var_10462_end_0, end_mask = var_10462_end_mask_0, x = transpose_11)[name = tensor("op_10462_cast")]; + tensor var_10466_begin_0 = const()[name = tensor("op_10466_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_10466_end_0 = const()[name = tensor("op_10466_end_0"), val = tensor([2, 1024, 1, 256])]; + tensor var_10466_end_mask_0 = const()[name = tensor("op_10466_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10466_cast = slice_by_index(begin = var_10466_begin_0, end = var_10466_end_0, end_mask = var_10466_end_mask_0, x = transpose_11)[name = tensor("op_10466_cast")]; + tensor var_10470_begin_0 = const()[name = tensor("op_10470_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_10470_end_0 = const()[name = tensor("op_10470_end_0"), val = tensor([2, 1024, 1, 320])]; + tensor var_10470_end_mask_0 = const()[name = tensor("op_10470_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10470_cast = slice_by_index(begin = var_10470_begin_0, end = var_10470_end_0, end_mask = var_10470_end_mask_0, x = transpose_11)[name = tensor("op_10470_cast")]; + tensor var_10474_begin_0 = const()[name = tensor("op_10474_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_10474_end_0 = const()[name = tensor("op_10474_end_0"), val = tensor([2, 1024, 1, 384])]; + tensor var_10474_end_mask_0 = const()[name = tensor("op_10474_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10474_cast = slice_by_index(begin = var_10474_begin_0, end = var_10474_end_0, end_mask = var_10474_end_mask_0, x = transpose_11)[name = tensor("op_10474_cast")]; + tensor var_10478_begin_0 = const()[name = tensor("op_10478_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_10478_end_0 = const()[name = tensor("op_10478_end_0"), val = tensor([2, 1024, 1, 448])]; + tensor var_10478_end_mask_0 = const()[name = tensor("op_10478_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10478_cast = slice_by_index(begin = var_10478_begin_0, end = var_10478_end_0, end_mask = var_10478_end_mask_0, x = transpose_11)[name = tensor("op_10478_cast")]; + tensor var_10482_begin_0 = const()[name = tensor("op_10482_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_10482_end_0 = const()[name = tensor("op_10482_end_0"), val = tensor([2, 1024, 1, 512])]; + tensor var_10482_end_mask_0 = const()[name = tensor("op_10482_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10482_cast = slice_by_index(begin = var_10482_begin_0, end = var_10482_end_0, end_mask = var_10482_end_mask_0, x = transpose_11)[name = tensor("op_10482_cast")]; + tensor var_10486_begin_0 = const()[name = tensor("op_10486_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_10486_end_0 = const()[name = tensor("op_10486_end_0"), val = tensor([2, 1024, 1, 576])]; + tensor var_10486_end_mask_0 = const()[name = tensor("op_10486_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10486_cast = slice_by_index(begin = var_10486_begin_0, end = var_10486_end_0, end_mask = var_10486_end_mask_0, x = transpose_11)[name = tensor("op_10486_cast")]; + tensor var_10490_begin_0 = const()[name = tensor("op_10490_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_10490_end_0 = const()[name = tensor("op_10490_end_0"), val = tensor([2, 1024, 1, 640])]; + tensor var_10490_end_mask_0 = const()[name = tensor("op_10490_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10490_cast = slice_by_index(begin = var_10490_begin_0, end = var_10490_end_0, end_mask = var_10490_end_mask_0, x = transpose_11)[name = tensor("op_10490_cast")]; + tensor var_10492_begin_0 = const()[name = tensor("op_10492_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_10492_end_0 = const()[name = tensor("op_10492_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_10492_end_mask_0 = const()[name = tensor("op_10492_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10492_cast = slice_by_index(begin = var_10492_begin_0, end = var_10492_end_0, end_mask = var_10492_end_mask_0, x = v_41_cast)[name = tensor("op_10492_cast")]; + tensor var_10496_begin_0 = const()[name = tensor("op_10496_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_10496_end_0 = const()[name = tensor("op_10496_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_10496_end_mask_0 = const()[name = tensor("op_10496_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10496_cast = slice_by_index(begin = var_10496_begin_0, end = var_10496_end_0, end_mask = var_10496_end_mask_0, x = v_41_cast)[name = tensor("op_10496_cast")]; + tensor var_10500_begin_0 = const()[name = tensor("op_10500_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_10500_end_0 = const()[name = tensor("op_10500_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_10500_end_mask_0 = const()[name = tensor("op_10500_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10500_cast = slice_by_index(begin = var_10500_begin_0, end = var_10500_end_0, end_mask = var_10500_end_mask_0, x = v_41_cast)[name = tensor("op_10500_cast")]; + tensor var_10504_begin_0 = const()[name = tensor("op_10504_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_10504_end_0 = const()[name = tensor("op_10504_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_10504_end_mask_0 = const()[name = tensor("op_10504_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10504_cast = slice_by_index(begin = var_10504_begin_0, end = var_10504_end_0, end_mask = var_10504_end_mask_0, x = v_41_cast)[name = tensor("op_10504_cast")]; + tensor var_10508_begin_0 = const()[name = tensor("op_10508_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_10508_end_0 = const()[name = tensor("op_10508_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_10508_end_mask_0 = const()[name = tensor("op_10508_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10508_cast = slice_by_index(begin = var_10508_begin_0, end = var_10508_end_0, end_mask = var_10508_end_mask_0, x = v_41_cast)[name = tensor("op_10508_cast")]; + tensor var_10512_begin_0 = const()[name = tensor("op_10512_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_10512_end_0 = const()[name = tensor("op_10512_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_10512_end_mask_0 = const()[name = tensor("op_10512_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10512_cast = slice_by_index(begin = var_10512_begin_0, end = var_10512_end_0, end_mask = var_10512_end_mask_0, x = v_41_cast)[name = tensor("op_10512_cast")]; + tensor var_10516_begin_0 = const()[name = tensor("op_10516_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_10516_end_0 = const()[name = tensor("op_10516_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_10516_end_mask_0 = const()[name = tensor("op_10516_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10516_cast = slice_by_index(begin = var_10516_begin_0, end = var_10516_end_0, end_mask = var_10516_end_mask_0, x = v_41_cast)[name = tensor("op_10516_cast")]; + tensor var_10520_begin_0 = const()[name = tensor("op_10520_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_10520_end_0 = const()[name = tensor("op_10520_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_10520_end_mask_0 = const()[name = tensor("op_10520_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10520_cast = slice_by_index(begin = var_10520_begin_0, end = var_10520_end_0, end_mask = var_10520_end_mask_0, x = v_41_cast)[name = tensor("op_10520_cast")]; + tensor var_10524_begin_0 = const()[name = tensor("op_10524_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_10524_end_0 = const()[name = tensor("op_10524_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_10524_end_mask_0 = const()[name = tensor("op_10524_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10524_cast = slice_by_index(begin = var_10524_begin_0, end = var_10524_end_0, end_mask = var_10524_end_mask_0, x = v_41_cast)[name = tensor("op_10524_cast")]; + tensor var_10528_begin_0 = const()[name = tensor("op_10528_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_10528_end_0 = const()[name = tensor("op_10528_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_10528_end_mask_0 = const()[name = tensor("op_10528_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10528_cast = slice_by_index(begin = var_10528_begin_0, end = var_10528_end_0, end_mask = var_10528_end_mask_0, x = v_41_cast)[name = tensor("op_10528_cast")]; + tensor var_10532_equation_0 = const()[name = tensor("op_10532_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10532_cast = einsum(equation = var_10532_equation_0, values = (var_10454_cast, var_10430_cast))[name = tensor("op_10532_cast")]; + tensor var_10533_to_fp16 = const()[name = tensor("op_10533_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_481_cast = mul(x = var_10532_cast, y = var_10533_to_fp16)[name = tensor("aw_chunk_481_cast")]; + tensor var_10536_equation_0 = const()[name = tensor("op_10536_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10536_cast = einsum(equation = var_10536_equation_0, values = (var_10454_cast, var_10431_cast))[name = tensor("op_10536_cast")]; + tensor var_10537_to_fp16 = const()[name = tensor("op_10537_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_483_cast = mul(x = var_10536_cast, y = var_10537_to_fp16)[name = tensor("aw_chunk_483_cast")]; + tensor var_10540_equation_0 = const()[name = tensor("op_10540_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10540_cast = einsum(equation = var_10540_equation_0, values = (var_10458_cast, var_10432_cast))[name = tensor("op_10540_cast")]; + tensor var_10541_to_fp16 = const()[name = tensor("op_10541_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_485_cast = mul(x = var_10540_cast, y = var_10541_to_fp16)[name = tensor("aw_chunk_485_cast")]; + tensor var_10544_equation_0 = const()[name = tensor("op_10544_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10544_cast = einsum(equation = var_10544_equation_0, values = (var_10458_cast, var_10433_cast))[name = tensor("op_10544_cast")]; + tensor var_10545_to_fp16 = const()[name = tensor("op_10545_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_487_cast = mul(x = var_10544_cast, y = var_10545_to_fp16)[name = tensor("aw_chunk_487_cast")]; + tensor var_10548_equation_0 = const()[name = tensor("op_10548_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10548_cast = einsum(equation = var_10548_equation_0, values = (var_10462_cast, var_10434_cast))[name = tensor("op_10548_cast")]; + tensor var_10549_to_fp16 = const()[name = tensor("op_10549_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_489_cast = mul(x = var_10548_cast, y = var_10549_to_fp16)[name = tensor("aw_chunk_489_cast")]; + tensor var_10552_equation_0 = const()[name = tensor("op_10552_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10552_cast = einsum(equation = var_10552_equation_0, values = (var_10462_cast, var_10435_cast))[name = tensor("op_10552_cast")]; + tensor var_10553_to_fp16 = const()[name = tensor("op_10553_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_491_cast = mul(x = var_10552_cast, y = var_10553_to_fp16)[name = tensor("aw_chunk_491_cast")]; + tensor var_10556_equation_0 = const()[name = tensor("op_10556_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10556_cast = einsum(equation = var_10556_equation_0, values = (var_10466_cast, var_10436_cast))[name = tensor("op_10556_cast")]; + tensor var_10557_to_fp16 = const()[name = tensor("op_10557_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_493_cast = mul(x = var_10556_cast, y = var_10557_to_fp16)[name = tensor("aw_chunk_493_cast")]; + tensor var_10560_equation_0 = const()[name = tensor("op_10560_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10560_cast = einsum(equation = var_10560_equation_0, values = (var_10466_cast, var_10437_cast))[name = tensor("op_10560_cast")]; + tensor var_10561_to_fp16 = const()[name = tensor("op_10561_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_495_cast = mul(x = var_10560_cast, y = var_10561_to_fp16)[name = tensor("aw_chunk_495_cast")]; + tensor var_10564_equation_0 = const()[name = tensor("op_10564_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10564_cast = einsum(equation = var_10564_equation_0, values = (var_10470_cast, var_10438_cast))[name = tensor("op_10564_cast")]; + tensor var_10565_to_fp16 = const()[name = tensor("op_10565_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_497_cast = mul(x = var_10564_cast, y = var_10565_to_fp16)[name = tensor("aw_chunk_497_cast")]; + tensor var_10568_equation_0 = const()[name = tensor("op_10568_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10568_cast = einsum(equation = var_10568_equation_0, values = (var_10470_cast, var_10439_cast))[name = tensor("op_10568_cast")]; + tensor var_10569_to_fp16 = const()[name = tensor("op_10569_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_499_cast = mul(x = var_10568_cast, y = var_10569_to_fp16)[name = tensor("aw_chunk_499_cast")]; + tensor var_10572_equation_0 = const()[name = tensor("op_10572_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10572_cast = einsum(equation = var_10572_equation_0, values = (var_10474_cast, var_10440_cast))[name = tensor("op_10572_cast")]; + tensor var_10573_to_fp16 = const()[name = tensor("op_10573_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_501_cast = mul(x = var_10572_cast, y = var_10573_to_fp16)[name = tensor("aw_chunk_501_cast")]; + tensor var_10576_equation_0 = const()[name = tensor("op_10576_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10576_cast = einsum(equation = var_10576_equation_0, values = (var_10474_cast, var_10441_cast))[name = tensor("op_10576_cast")]; + tensor var_10577_to_fp16 = const()[name = tensor("op_10577_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_503_cast = mul(x = var_10576_cast, y = var_10577_to_fp16)[name = tensor("aw_chunk_503_cast")]; + tensor var_10580_equation_0 = const()[name = tensor("op_10580_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10580_cast = einsum(equation = var_10580_equation_0, values = (var_10478_cast, var_10442_cast))[name = tensor("op_10580_cast")]; + tensor var_10581_to_fp16 = const()[name = tensor("op_10581_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_505_cast = mul(x = var_10580_cast, y = var_10581_to_fp16)[name = tensor("aw_chunk_505_cast")]; + tensor var_10584_equation_0 = const()[name = tensor("op_10584_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10584_cast = einsum(equation = var_10584_equation_0, values = (var_10478_cast, var_10443_cast))[name = tensor("op_10584_cast")]; + tensor var_10585_to_fp16 = const()[name = tensor("op_10585_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_507_cast = mul(x = var_10584_cast, y = var_10585_to_fp16)[name = tensor("aw_chunk_507_cast")]; + tensor var_10588_equation_0 = const()[name = tensor("op_10588_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10588_cast = einsum(equation = var_10588_equation_0, values = (var_10482_cast, var_10444_cast))[name = tensor("op_10588_cast")]; + tensor var_10589_to_fp16 = const()[name = tensor("op_10589_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_509_cast = mul(x = var_10588_cast, y = var_10589_to_fp16)[name = tensor("aw_chunk_509_cast")]; + tensor var_10592_equation_0 = const()[name = tensor("op_10592_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10592_cast = einsum(equation = var_10592_equation_0, values = (var_10482_cast, var_10445_cast))[name = tensor("op_10592_cast")]; + tensor var_10593_to_fp16 = const()[name = tensor("op_10593_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_511_cast = mul(x = var_10592_cast, y = var_10593_to_fp16)[name = tensor("aw_chunk_511_cast")]; + tensor var_10596_equation_0 = const()[name = tensor("op_10596_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10596_cast = einsum(equation = var_10596_equation_0, values = (var_10486_cast, var_10446_cast))[name = tensor("op_10596_cast")]; + tensor var_10597_to_fp16 = const()[name = tensor("op_10597_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_513_cast = mul(x = var_10596_cast, y = var_10597_to_fp16)[name = tensor("aw_chunk_513_cast")]; + tensor var_10600_equation_0 = const()[name = tensor("op_10600_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10600_cast = einsum(equation = var_10600_equation_0, values = (var_10486_cast, var_10447_cast))[name = tensor("op_10600_cast")]; + tensor var_10601_to_fp16 = const()[name = tensor("op_10601_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_515_cast = mul(x = var_10600_cast, y = var_10601_to_fp16)[name = tensor("aw_chunk_515_cast")]; + tensor var_10604_equation_0 = const()[name = tensor("op_10604_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10604_cast = einsum(equation = var_10604_equation_0, values = (var_10490_cast, var_10448_cast))[name = tensor("op_10604_cast")]; + tensor var_10605_to_fp16 = const()[name = tensor("op_10605_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_517_cast = mul(x = var_10604_cast, y = var_10605_to_fp16)[name = tensor("aw_chunk_517_cast")]; + tensor var_10608_equation_0 = const()[name = tensor("op_10608_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10608_cast = einsum(equation = var_10608_equation_0, values = (var_10490_cast, var_10449_cast))[name = tensor("op_10608_cast")]; + tensor var_10609_to_fp16 = const()[name = tensor("op_10609_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_519_cast = mul(x = var_10608_cast, y = var_10609_to_fp16)[name = tensor("aw_chunk_519_cast")]; + tensor var_10611_cast = softmax(axis = var_10257, x = aw_chunk_481_cast)[name = tensor("op_10611_cast")]; + tensor var_10612_cast = softmax(axis = var_10257, x = aw_chunk_483_cast)[name = tensor("op_10612_cast")]; + tensor var_10613_cast = softmax(axis = var_10257, x = aw_chunk_485_cast)[name = tensor("op_10613_cast")]; + tensor var_10614_cast = softmax(axis = var_10257, x = aw_chunk_487_cast)[name = tensor("op_10614_cast")]; + tensor var_10615_cast = softmax(axis = var_10257, x = aw_chunk_489_cast)[name = tensor("op_10615_cast")]; + tensor var_10616_cast = softmax(axis = var_10257, x = aw_chunk_491_cast)[name = tensor("op_10616_cast")]; + tensor var_10617_cast = softmax(axis = var_10257, x = aw_chunk_493_cast)[name = tensor("op_10617_cast")]; + tensor var_10618_cast = softmax(axis = var_10257, x = aw_chunk_495_cast)[name = tensor("op_10618_cast")]; + tensor var_10619_cast = softmax(axis = var_10257, x = aw_chunk_497_cast)[name = tensor("op_10619_cast")]; + tensor var_10620_cast = softmax(axis = var_10257, x = aw_chunk_499_cast)[name = tensor("op_10620_cast")]; + tensor var_10621_cast = softmax(axis = var_10257, x = aw_chunk_501_cast)[name = tensor("op_10621_cast")]; + tensor var_10622_cast = softmax(axis = var_10257, x = aw_chunk_503_cast)[name = tensor("op_10622_cast")]; + tensor var_10623_cast = softmax(axis = var_10257, x = aw_chunk_505_cast)[name = tensor("op_10623_cast")]; + tensor var_10624_cast = softmax(axis = var_10257, x = aw_chunk_507_cast)[name = tensor("op_10624_cast")]; + tensor var_10625_cast = softmax(axis = var_10257, x = aw_chunk_509_cast)[name = tensor("op_10625_cast")]; + tensor var_10626_cast = softmax(axis = var_10257, x = aw_chunk_511_cast)[name = tensor("op_10626_cast")]; + tensor var_10627_cast = softmax(axis = var_10257, x = aw_chunk_513_cast)[name = tensor("op_10627_cast")]; + tensor var_10628_cast = softmax(axis = var_10257, x = aw_chunk_515_cast)[name = tensor("op_10628_cast")]; + tensor var_10629_cast = softmax(axis = var_10257, x = aw_chunk_517_cast)[name = tensor("op_10629_cast")]; + tensor var_10630_cast = softmax(axis = var_10257, x = aw_chunk_519_cast)[name = tensor("op_10630_cast")]; + tensor var_10632_equation_0 = const()[name = tensor("op_10632_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_10632_cast = einsum(equation = var_10632_equation_0, values = (var_10492_cast, var_10611_cast))[name = tensor("op_10632_cast")]; + tensor var_10634_equation_0 = const()[name = tensor("op_10634_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_10634_cast = einsum(equation = var_10634_equation_0, values = (var_10492_cast, var_10612_cast))[name = tensor("op_10634_cast")]; + tensor var_10636_equation_0 = const()[name = tensor("op_10636_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_10636_cast = einsum(equation = var_10636_equation_0, values = (var_10496_cast, var_10613_cast))[name = tensor("op_10636_cast")]; + tensor var_10638_equation_0 = const()[name = tensor("op_10638_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_10638_cast = einsum(equation = var_10638_equation_0, values = (var_10496_cast, var_10614_cast))[name = tensor("op_10638_cast")]; + tensor var_10640_equation_0 = const()[name = tensor("op_10640_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_10640_cast = einsum(equation = var_10640_equation_0, values = (var_10500_cast, var_10615_cast))[name = tensor("op_10640_cast")]; + tensor var_10642_equation_0 = const()[name = tensor("op_10642_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_10642_cast = einsum(equation = var_10642_equation_0, values = (var_10500_cast, var_10616_cast))[name = tensor("op_10642_cast")]; + tensor var_10644_equation_0 = const()[name = tensor("op_10644_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_10644_cast = einsum(equation = var_10644_equation_0, values = (var_10504_cast, var_10617_cast))[name = tensor("op_10644_cast")]; + tensor var_10646_equation_0 = const()[name = tensor("op_10646_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_10646_cast = einsum(equation = var_10646_equation_0, values = (var_10504_cast, var_10618_cast))[name = tensor("op_10646_cast")]; + tensor var_10648_equation_0 = const()[name = tensor("op_10648_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_10648_cast = einsum(equation = var_10648_equation_0, values = (var_10508_cast, var_10619_cast))[name = tensor("op_10648_cast")]; + tensor var_10650_equation_0 = const()[name = tensor("op_10650_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_10650_cast = einsum(equation = var_10650_equation_0, values = (var_10508_cast, var_10620_cast))[name = tensor("op_10650_cast")]; + tensor var_10652_equation_0 = const()[name = tensor("op_10652_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_10652_cast = einsum(equation = var_10652_equation_0, values = (var_10512_cast, var_10621_cast))[name = tensor("op_10652_cast")]; + tensor var_10654_equation_0 = const()[name = tensor("op_10654_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_10654_cast = einsum(equation = var_10654_equation_0, values = (var_10512_cast, var_10622_cast))[name = tensor("op_10654_cast")]; + tensor var_10656_equation_0 = const()[name = tensor("op_10656_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_10656_cast = einsum(equation = var_10656_equation_0, values = (var_10516_cast, var_10623_cast))[name = tensor("op_10656_cast")]; + tensor var_10658_equation_0 = const()[name = tensor("op_10658_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_10658_cast = einsum(equation = var_10658_equation_0, values = (var_10516_cast, var_10624_cast))[name = tensor("op_10658_cast")]; + tensor var_10660_equation_0 = const()[name = tensor("op_10660_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_10660_cast = einsum(equation = var_10660_equation_0, values = (var_10520_cast, var_10625_cast))[name = tensor("op_10660_cast")]; + tensor var_10662_equation_0 = const()[name = tensor("op_10662_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_10662_cast = einsum(equation = var_10662_equation_0, values = (var_10520_cast, var_10626_cast))[name = tensor("op_10662_cast")]; + tensor var_10664_equation_0 = const()[name = tensor("op_10664_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_10664_cast = einsum(equation = var_10664_equation_0, values = (var_10524_cast, var_10627_cast))[name = tensor("op_10664_cast")]; + tensor var_10666_equation_0 = const()[name = tensor("op_10666_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_10666_cast = einsum(equation = var_10666_equation_0, values = (var_10524_cast, var_10628_cast))[name = tensor("op_10666_cast")]; + tensor var_10668_equation_0 = const()[name = tensor("op_10668_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_10668_cast = einsum(equation = var_10668_equation_0, values = (var_10528_cast, var_10629_cast))[name = tensor("op_10668_cast")]; + tensor var_10670_equation_0 = const()[name = tensor("op_10670_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_10670_cast = einsum(equation = var_10670_equation_0, values = (var_10528_cast, var_10630_cast))[name = tensor("op_10670_cast")]; + tensor var_10672_interleave_0 = const()[name = tensor("op_10672_interleave_0"), val = tensor(false)]; + tensor var_10672_cast = concat(axis = var_10234, interleave = var_10672_interleave_0, values = (var_10632_cast, var_10634_cast))[name = tensor("op_10672_cast")]; + tensor var_10674_interleave_0 = const()[name = tensor("op_10674_interleave_0"), val = tensor(false)]; + tensor var_10674_cast = concat(axis = var_10234, interleave = var_10674_interleave_0, values = (var_10636_cast, var_10638_cast))[name = tensor("op_10674_cast")]; + tensor var_10676_interleave_0 = const()[name = tensor("op_10676_interleave_0"), val = tensor(false)]; + tensor var_10676_cast = concat(axis = var_10234, interleave = var_10676_interleave_0, values = (var_10640_cast, var_10642_cast))[name = tensor("op_10676_cast")]; + tensor var_10678_interleave_0 = const()[name = tensor("op_10678_interleave_0"), val = tensor(false)]; + tensor var_10678_cast = concat(axis = var_10234, interleave = var_10678_interleave_0, values = (var_10644_cast, var_10646_cast))[name = tensor("op_10678_cast")]; + tensor var_10680_interleave_0 = const()[name = tensor("op_10680_interleave_0"), val = tensor(false)]; + tensor var_10680_cast = concat(axis = var_10234, interleave = var_10680_interleave_0, values = (var_10648_cast, var_10650_cast))[name = tensor("op_10680_cast")]; + tensor var_10682_interleave_0 = const()[name = tensor("op_10682_interleave_0"), val = tensor(false)]; + tensor var_10682_cast = concat(axis = var_10234, interleave = var_10682_interleave_0, values = (var_10652_cast, var_10654_cast))[name = tensor("op_10682_cast")]; + tensor var_10684_interleave_0 = const()[name = tensor("op_10684_interleave_0"), val = tensor(false)]; + tensor var_10684_cast = concat(axis = var_10234, interleave = var_10684_interleave_0, values = (var_10656_cast, var_10658_cast))[name = tensor("op_10684_cast")]; + tensor var_10686_interleave_0 = const()[name = tensor("op_10686_interleave_0"), val = tensor(false)]; + tensor var_10686_cast = concat(axis = var_10234, interleave = var_10686_interleave_0, values = (var_10660_cast, var_10662_cast))[name = tensor("op_10686_cast")]; + tensor var_10688_interleave_0 = const()[name = tensor("op_10688_interleave_0"), val = tensor(false)]; + tensor var_10688_cast = concat(axis = var_10234, interleave = var_10688_interleave_0, values = (var_10664_cast, var_10666_cast))[name = tensor("op_10688_cast")]; + tensor var_10690_interleave_0 = const()[name = tensor("op_10690_interleave_0"), val = tensor(false)]; + tensor var_10690_cast = concat(axis = var_10234, interleave = var_10690_interleave_0, values = (var_10668_cast, var_10670_cast))[name = tensor("op_10690_cast")]; + tensor input_383_interleave_0 = const()[name = tensor("input_383_interleave_0"), val = tensor(false)]; + tensor input_383_cast = concat(axis = var_10257, interleave = input_383_interleave_0, values = (var_10672_cast, var_10674_cast, var_10676_cast, var_10678_cast, var_10680_cast, var_10682_cast, var_10684_cast, var_10686_cast, var_10688_cast, var_10690_cast))[name = tensor("input_383_cast")]; + tensor var_10696 = const()[name = tensor("op_10696"), val = tensor([1, 1])]; + tensor var_10698 = const()[name = tensor("op_10698"), val = tensor([1, 1])]; + tensor var_10700_pad_type_0 = const()[name = tensor("op_10700_pad_type_0"), val = tensor("custom")]; + tensor var_10700_pad_0 = const()[name = tensor("op_10700_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_2_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(594900928))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(595208192))), name = tensor("up_blocks_2_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16_palettized"), shape = tensor([640, 640, 1, 1])]; + tensor up_blocks_2_attentions_0_transformer_blocks_0_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("up_blocks_2_attentions_0_transformer_blocks_0_attn1_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(595208384)))]; + tensor var_10700_cast = conv(bias = up_blocks_2_attentions_0_transformer_blocks_0_attn1_to_out_0_bias_to_fp16, dilations = var_10698, groups = var_10257, pad = var_10700_pad_0, pad_type = var_10700_pad_type_0, strides = var_10696, weight = up_blocks_2_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16_palettized, x = input_383_cast)[name = tensor("op_10700_cast")]; + tensor inputs_63_cast = add(x = var_10700_cast, y = inputs_61_cast)[name = tensor("inputs_63_cast")]; + tensor var_10704 = const()[name = tensor("op_10704"), val = tensor([1])]; + tensor channels_mean_63_cast = reduce_mean(axes = var_10704, keep_dims = var_10252, x = inputs_63_cast)[name = tensor("channels_mean_63_cast")]; + tensor zero_mean_63_cast = sub(x = inputs_63_cast, y = channels_mean_63_cast)[name = tensor("zero_mean_63_cast")]; + tensor zero_mean_sq_63_cast = mul(x = zero_mean_63_cast, y = zero_mean_63_cast)[name = tensor("zero_mean_sq_63_cast")]; + tensor var_10708 = const()[name = tensor("op_10708"), val = tensor([1])]; + tensor var_10709_cast = reduce_mean(axes = var_10708, keep_dims = var_10252, x = zero_mean_sq_63_cast)[name = tensor("op_10709_cast")]; + tensor var_10710_to_fp16 = const()[name = tensor("op_10710_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_10711_cast = add(x = var_10709_cast, y = var_10710_to_fp16)[name = tensor("op_10711_cast")]; + tensor denom_63_epsilon_0_to_fp16 = const()[name = tensor("denom_63_epsilon_0_to_fp16"), val = tensor(0x1p-24)]; + tensor denom_63_cast = rsqrt(epsilon = denom_63_epsilon_0_to_fp16, x = var_10711_cast)[name = tensor("denom_63_cast")]; + tensor out_63_cast = mul(x = zero_mean_63_cast, y = denom_63_cast)[name = tensor("out_63_cast")]; + tensor var_10715_to_fp16 = const()[name = tensor("op_10715_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(595209728)))]; + tensor var_10716_cast = add(x = out_63_cast, y = var_10715_to_fp16)[name = tensor("op_10716_cast")]; + tensor var_10718_to_fp16 = const()[name = tensor("op_10718_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(595211072)))]; + tensor hidden_states_229_cast = mul(x = var_10716_cast, y = var_10718_to_fp16)[name = tensor("hidden_states_229_cast")]; + tensor var_10725 = const()[name = tensor("op_10725"), val = tensor([1, 1])]; + tensor var_10727 = const()[name = tensor("op_10727"), val = tensor([1, 1])]; + tensor q_43_pad_type_0 = const()[name = tensor("q_43_pad_type_0"), val = tensor("custom")]; + tensor q_43_pad_0 = const()[name = tensor("q_43_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_2_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(595212416))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(595519680))), name = tensor("up_blocks_2_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16_palettized"), shape = tensor([640, 640, 1, 1])]; + tensor q_43_cast = conv(dilations = var_10727, groups = var_10257, pad = q_43_pad_0, pad_type = q_43_pad_type_0, strides = var_10725, weight = up_blocks_2_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16_palettized, x = hidden_states_229_cast)[name = tensor("q_43_cast")]; + tensor var_10731 = const()[name = tensor("op_10731"), val = tensor([1, 1])]; + tensor var_10733 = const()[name = tensor("op_10733"), val = tensor([1, 1])]; + tensor k_85_pad_type_0 = const()[name = tensor("k_85_pad_type_0"), val = tensor("custom")]; + tensor k_85_pad_0 = const()[name = tensor("k_85_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_2_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(595519872))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(596011456))), name = tensor("up_blocks_2_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized"), shape = tensor([640, 1024, 1, 1])]; + tensor k_85_cast = conv(dilations = var_10733, groups = var_10257, pad = k_85_pad_0, pad_type = k_85_pad_type_0, strides = var_10731, weight = up_blocks_2_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor("k_85_cast")]; + tensor var_10737 = const()[name = tensor("op_10737"), val = tensor([1, 1])]; + tensor var_10739 = const()[name = tensor("op_10739"), val = tensor([1, 1])]; + tensor v_43_pad_type_0 = const()[name = tensor("v_43_pad_type_0"), val = tensor("custom")]; + tensor v_43_pad_0 = const()[name = tensor("v_43_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_2_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(596011648))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(596503232))), name = tensor("up_blocks_2_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized"), shape = tensor([640, 1024, 1, 1])]; + tensor v_43_cast = conv(dilations = var_10739, groups = var_10257, pad = v_43_pad_0, pad_type = v_43_pad_type_0, strides = var_10737, weight = up_blocks_2_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor("v_43_cast")]; + tensor var_10743_begin_0 = const()[name = tensor("op_10743_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_10743_end_0 = const()[name = tensor("op_10743_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_10743_end_mask_0 = const()[name = tensor("op_10743_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10743_cast = slice_by_index(begin = var_10743_begin_0, end = var_10743_end_0, end_mask = var_10743_end_mask_0, x = q_43_cast)[name = tensor("op_10743_cast")]; + tensor var_10747_begin_0 = const()[name = tensor("op_10747_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_10747_end_0 = const()[name = tensor("op_10747_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_10747_end_mask_0 = const()[name = tensor("op_10747_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10747_cast = slice_by_index(begin = var_10747_begin_0, end = var_10747_end_0, end_mask = var_10747_end_mask_0, x = q_43_cast)[name = tensor("op_10747_cast")]; + tensor var_10751_begin_0 = const()[name = tensor("op_10751_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_10751_end_0 = const()[name = tensor("op_10751_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_10751_end_mask_0 = const()[name = tensor("op_10751_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10751_cast = slice_by_index(begin = var_10751_begin_0, end = var_10751_end_0, end_mask = var_10751_end_mask_0, x = q_43_cast)[name = tensor("op_10751_cast")]; + tensor var_10755_begin_0 = const()[name = tensor("op_10755_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_10755_end_0 = const()[name = tensor("op_10755_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_10755_end_mask_0 = const()[name = tensor("op_10755_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10755_cast = slice_by_index(begin = var_10755_begin_0, end = var_10755_end_0, end_mask = var_10755_end_mask_0, x = q_43_cast)[name = tensor("op_10755_cast")]; + tensor var_10759_begin_0 = const()[name = tensor("op_10759_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_10759_end_0 = const()[name = tensor("op_10759_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_10759_end_mask_0 = const()[name = tensor("op_10759_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10759_cast = slice_by_index(begin = var_10759_begin_0, end = var_10759_end_0, end_mask = var_10759_end_mask_0, x = q_43_cast)[name = tensor("op_10759_cast")]; + tensor var_10763_begin_0 = const()[name = tensor("op_10763_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_10763_end_0 = const()[name = tensor("op_10763_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_10763_end_mask_0 = const()[name = tensor("op_10763_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10763_cast = slice_by_index(begin = var_10763_begin_0, end = var_10763_end_0, end_mask = var_10763_end_mask_0, x = q_43_cast)[name = tensor("op_10763_cast")]; + tensor var_10767_begin_0 = const()[name = tensor("op_10767_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_10767_end_0 = const()[name = tensor("op_10767_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_10767_end_mask_0 = const()[name = tensor("op_10767_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10767_cast = slice_by_index(begin = var_10767_begin_0, end = var_10767_end_0, end_mask = var_10767_end_mask_0, x = q_43_cast)[name = tensor("op_10767_cast")]; + tensor var_10771_begin_0 = const()[name = tensor("op_10771_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_10771_end_0 = const()[name = tensor("op_10771_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_10771_end_mask_0 = const()[name = tensor("op_10771_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10771_cast = slice_by_index(begin = var_10771_begin_0, end = var_10771_end_0, end_mask = var_10771_end_mask_0, x = q_43_cast)[name = tensor("op_10771_cast")]; + tensor var_10775_begin_0 = const()[name = tensor("op_10775_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_10775_end_0 = const()[name = tensor("op_10775_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_10775_end_mask_0 = const()[name = tensor("op_10775_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10775_cast = slice_by_index(begin = var_10775_begin_0, end = var_10775_end_0, end_mask = var_10775_end_mask_0, x = q_43_cast)[name = tensor("op_10775_cast")]; + tensor var_10779_begin_0 = const()[name = tensor("op_10779_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_10779_end_0 = const()[name = tensor("op_10779_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_10779_end_mask_0 = const()[name = tensor("op_10779_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10779_cast = slice_by_index(begin = var_10779_begin_0, end = var_10779_end_0, end_mask = var_10779_end_mask_0, x = q_43_cast)[name = tensor("op_10779_cast")]; + tensor var_10782_begin_0 = const()[name = tensor("op_10782_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_10782_end_0 = const()[name = tensor("op_10782_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_10782_end_mask_0 = const()[name = tensor("op_10782_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10782_cast = slice_by_index(begin = var_10782_begin_0, end = var_10782_end_0, end_mask = var_10782_end_mask_0, x = var_10743_cast)[name = tensor("op_10782_cast")]; + tensor var_10783_begin_0 = const()[name = tensor("op_10783_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_10783_end_0 = const()[name = tensor("op_10783_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_10783_end_mask_0 = const()[name = tensor("op_10783_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10783_cast = slice_by_index(begin = var_10783_begin_0, end = var_10783_end_0, end_mask = var_10783_end_mask_0, x = var_10743_cast)[name = tensor("op_10783_cast")]; + tensor var_10784_begin_0 = const()[name = tensor("op_10784_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_10784_end_0 = const()[name = tensor("op_10784_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_10784_end_mask_0 = const()[name = tensor("op_10784_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10784_cast = slice_by_index(begin = var_10784_begin_0, end = var_10784_end_0, end_mask = var_10784_end_mask_0, x = var_10747_cast)[name = tensor("op_10784_cast")]; + tensor var_10785_begin_0 = const()[name = tensor("op_10785_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_10785_end_0 = const()[name = tensor("op_10785_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_10785_end_mask_0 = const()[name = tensor("op_10785_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10785_cast = slice_by_index(begin = var_10785_begin_0, end = var_10785_end_0, end_mask = var_10785_end_mask_0, x = var_10747_cast)[name = tensor("op_10785_cast")]; + tensor var_10786_begin_0 = const()[name = tensor("op_10786_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_10786_end_0 = const()[name = tensor("op_10786_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_10786_end_mask_0 = const()[name = tensor("op_10786_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10786_cast = slice_by_index(begin = var_10786_begin_0, end = var_10786_end_0, end_mask = var_10786_end_mask_0, x = var_10751_cast)[name = tensor("op_10786_cast")]; + tensor var_10787_begin_0 = const()[name = tensor("op_10787_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_10787_end_0 = const()[name = tensor("op_10787_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_10787_end_mask_0 = const()[name = tensor("op_10787_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10787_cast = slice_by_index(begin = var_10787_begin_0, end = var_10787_end_0, end_mask = var_10787_end_mask_0, x = var_10751_cast)[name = tensor("op_10787_cast")]; + tensor var_10788_begin_0 = const()[name = tensor("op_10788_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_10788_end_0 = const()[name = tensor("op_10788_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_10788_end_mask_0 = const()[name = tensor("op_10788_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10788_cast = slice_by_index(begin = var_10788_begin_0, end = var_10788_end_0, end_mask = var_10788_end_mask_0, x = var_10755_cast)[name = tensor("op_10788_cast")]; + tensor var_10789_begin_0 = const()[name = tensor("op_10789_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_10789_end_0 = const()[name = tensor("op_10789_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_10789_end_mask_0 = const()[name = tensor("op_10789_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10789_cast = slice_by_index(begin = var_10789_begin_0, end = var_10789_end_0, end_mask = var_10789_end_mask_0, x = var_10755_cast)[name = tensor("op_10789_cast")]; + tensor var_10790_begin_0 = const()[name = tensor("op_10790_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_10790_end_0 = const()[name = tensor("op_10790_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_10790_end_mask_0 = const()[name = tensor("op_10790_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10790_cast = slice_by_index(begin = var_10790_begin_0, end = var_10790_end_0, end_mask = var_10790_end_mask_0, x = var_10759_cast)[name = tensor("op_10790_cast")]; + tensor var_10791_begin_0 = const()[name = tensor("op_10791_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_10791_end_0 = const()[name = tensor("op_10791_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_10791_end_mask_0 = const()[name = tensor("op_10791_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10791_cast = slice_by_index(begin = var_10791_begin_0, end = var_10791_end_0, end_mask = var_10791_end_mask_0, x = var_10759_cast)[name = tensor("op_10791_cast")]; + tensor var_10792_begin_0 = const()[name = tensor("op_10792_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_10792_end_0 = const()[name = tensor("op_10792_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_10792_end_mask_0 = const()[name = tensor("op_10792_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10792_cast = slice_by_index(begin = var_10792_begin_0, end = var_10792_end_0, end_mask = var_10792_end_mask_0, x = var_10763_cast)[name = tensor("op_10792_cast")]; + tensor var_10793_begin_0 = const()[name = tensor("op_10793_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_10793_end_0 = const()[name = tensor("op_10793_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_10793_end_mask_0 = const()[name = tensor("op_10793_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10793_cast = slice_by_index(begin = var_10793_begin_0, end = var_10793_end_0, end_mask = var_10793_end_mask_0, x = var_10763_cast)[name = tensor("op_10793_cast")]; + tensor var_10794_begin_0 = const()[name = tensor("op_10794_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_10794_end_0 = const()[name = tensor("op_10794_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_10794_end_mask_0 = const()[name = tensor("op_10794_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10794_cast = slice_by_index(begin = var_10794_begin_0, end = var_10794_end_0, end_mask = var_10794_end_mask_0, x = var_10767_cast)[name = tensor("op_10794_cast")]; + tensor var_10795_begin_0 = const()[name = tensor("op_10795_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_10795_end_0 = const()[name = tensor("op_10795_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_10795_end_mask_0 = const()[name = tensor("op_10795_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10795_cast = slice_by_index(begin = var_10795_begin_0, end = var_10795_end_0, end_mask = var_10795_end_mask_0, x = var_10767_cast)[name = tensor("op_10795_cast")]; + tensor var_10796_begin_0 = const()[name = tensor("op_10796_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_10796_end_0 = const()[name = tensor("op_10796_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_10796_end_mask_0 = const()[name = tensor("op_10796_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10796_cast = slice_by_index(begin = var_10796_begin_0, end = var_10796_end_0, end_mask = var_10796_end_mask_0, x = var_10771_cast)[name = tensor("op_10796_cast")]; + tensor var_10797_begin_0 = const()[name = tensor("op_10797_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_10797_end_0 = const()[name = tensor("op_10797_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_10797_end_mask_0 = const()[name = tensor("op_10797_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10797_cast = slice_by_index(begin = var_10797_begin_0, end = var_10797_end_0, end_mask = var_10797_end_mask_0, x = var_10771_cast)[name = tensor("op_10797_cast")]; + tensor var_10798_begin_0 = const()[name = tensor("op_10798_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_10798_end_0 = const()[name = tensor("op_10798_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_10798_end_mask_0 = const()[name = tensor("op_10798_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10798_cast = slice_by_index(begin = var_10798_begin_0, end = var_10798_end_0, end_mask = var_10798_end_mask_0, x = var_10775_cast)[name = tensor("op_10798_cast")]; + tensor var_10799_begin_0 = const()[name = tensor("op_10799_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_10799_end_0 = const()[name = tensor("op_10799_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_10799_end_mask_0 = const()[name = tensor("op_10799_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10799_cast = slice_by_index(begin = var_10799_begin_0, end = var_10799_end_0, end_mask = var_10799_end_mask_0, x = var_10775_cast)[name = tensor("op_10799_cast")]; + tensor var_10800_begin_0 = const()[name = tensor("op_10800_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_10800_end_0 = const()[name = tensor("op_10800_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_10800_end_mask_0 = const()[name = tensor("op_10800_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10800_cast = slice_by_index(begin = var_10800_begin_0, end = var_10800_end_0, end_mask = var_10800_end_mask_0, x = var_10779_cast)[name = tensor("op_10800_cast")]; + tensor var_10801_begin_0 = const()[name = tensor("op_10801_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_10801_end_0 = const()[name = tensor("op_10801_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_10801_end_mask_0 = const()[name = tensor("op_10801_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10801_cast = slice_by_index(begin = var_10801_begin_0, end = var_10801_end_0, end_mask = var_10801_end_mask_0, x = var_10779_cast)[name = tensor("op_10801_cast")]; + tensor k_87_perm_0 = const()[name = tensor("k_87_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_10806_begin_0 = const()[name = tensor("op_10806_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_10806_end_0 = const()[name = tensor("op_10806_end_0"), val = tensor([2, 77, 1, 64])]; + tensor var_10806_end_mask_0 = const()[name = tensor("op_10806_end_mask_0"), val = tensor([true, true, true, false])]; + tensor transpose_10 = transpose(perm = k_87_perm_0, x = k_85_cast)[name = tensor("transpose_10")]; + tensor var_10806_cast = slice_by_index(begin = var_10806_begin_0, end = var_10806_end_0, end_mask = var_10806_end_mask_0, x = transpose_10)[name = tensor("op_10806_cast")]; + tensor var_10810_begin_0 = const()[name = tensor("op_10810_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_10810_end_0 = const()[name = tensor("op_10810_end_0"), val = tensor([2, 77, 1, 128])]; + tensor var_10810_end_mask_0 = const()[name = tensor("op_10810_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10810_cast = slice_by_index(begin = var_10810_begin_0, end = var_10810_end_0, end_mask = var_10810_end_mask_0, x = transpose_10)[name = tensor("op_10810_cast")]; + tensor var_10814_begin_0 = const()[name = tensor("op_10814_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_10814_end_0 = const()[name = tensor("op_10814_end_0"), val = tensor([2, 77, 1, 192])]; + tensor var_10814_end_mask_0 = const()[name = tensor("op_10814_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10814_cast = slice_by_index(begin = var_10814_begin_0, end = var_10814_end_0, end_mask = var_10814_end_mask_0, x = transpose_10)[name = tensor("op_10814_cast")]; + tensor var_10818_begin_0 = const()[name = tensor("op_10818_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_10818_end_0 = const()[name = tensor("op_10818_end_0"), val = tensor([2, 77, 1, 256])]; + tensor var_10818_end_mask_0 = const()[name = tensor("op_10818_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10818_cast = slice_by_index(begin = var_10818_begin_0, end = var_10818_end_0, end_mask = var_10818_end_mask_0, x = transpose_10)[name = tensor("op_10818_cast")]; + tensor var_10822_begin_0 = const()[name = tensor("op_10822_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_10822_end_0 = const()[name = tensor("op_10822_end_0"), val = tensor([2, 77, 1, 320])]; + tensor var_10822_end_mask_0 = const()[name = tensor("op_10822_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10822_cast = slice_by_index(begin = var_10822_begin_0, end = var_10822_end_0, end_mask = var_10822_end_mask_0, x = transpose_10)[name = tensor("op_10822_cast")]; + tensor var_10826_begin_0 = const()[name = tensor("op_10826_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_10826_end_0 = const()[name = tensor("op_10826_end_0"), val = tensor([2, 77, 1, 384])]; + tensor var_10826_end_mask_0 = const()[name = tensor("op_10826_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10826_cast = slice_by_index(begin = var_10826_begin_0, end = var_10826_end_0, end_mask = var_10826_end_mask_0, x = transpose_10)[name = tensor("op_10826_cast")]; + tensor var_10830_begin_0 = const()[name = tensor("op_10830_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_10830_end_0 = const()[name = tensor("op_10830_end_0"), val = tensor([2, 77, 1, 448])]; + tensor var_10830_end_mask_0 = const()[name = tensor("op_10830_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10830_cast = slice_by_index(begin = var_10830_begin_0, end = var_10830_end_0, end_mask = var_10830_end_mask_0, x = transpose_10)[name = tensor("op_10830_cast")]; + tensor var_10834_begin_0 = const()[name = tensor("op_10834_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_10834_end_0 = const()[name = tensor("op_10834_end_0"), val = tensor([2, 77, 1, 512])]; + tensor var_10834_end_mask_0 = const()[name = tensor("op_10834_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10834_cast = slice_by_index(begin = var_10834_begin_0, end = var_10834_end_0, end_mask = var_10834_end_mask_0, x = transpose_10)[name = tensor("op_10834_cast")]; + tensor var_10838_begin_0 = const()[name = tensor("op_10838_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_10838_end_0 = const()[name = tensor("op_10838_end_0"), val = tensor([2, 77, 1, 576])]; + tensor var_10838_end_mask_0 = const()[name = tensor("op_10838_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10838_cast = slice_by_index(begin = var_10838_begin_0, end = var_10838_end_0, end_mask = var_10838_end_mask_0, x = transpose_10)[name = tensor("op_10838_cast")]; + tensor var_10842_begin_0 = const()[name = tensor("op_10842_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_10842_end_0 = const()[name = tensor("op_10842_end_0"), val = tensor([2, 77, 1, 640])]; + tensor var_10842_end_mask_0 = const()[name = tensor("op_10842_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10842_cast = slice_by_index(begin = var_10842_begin_0, end = var_10842_end_0, end_mask = var_10842_end_mask_0, x = transpose_10)[name = tensor("op_10842_cast")]; + tensor var_10844_begin_0 = const()[name = tensor("op_10844_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_10844_end_0 = const()[name = tensor("op_10844_end_0"), val = tensor([2, 64, 1, 77])]; + tensor var_10844_end_mask_0 = const()[name = tensor("op_10844_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10844_cast = slice_by_index(begin = var_10844_begin_0, end = var_10844_end_0, end_mask = var_10844_end_mask_0, x = v_43_cast)[name = tensor("op_10844_cast")]; + tensor var_10848_begin_0 = const()[name = tensor("op_10848_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_10848_end_0 = const()[name = tensor("op_10848_end_0"), val = tensor([2, 128, 1, 77])]; + tensor var_10848_end_mask_0 = const()[name = tensor("op_10848_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10848_cast = slice_by_index(begin = var_10848_begin_0, end = var_10848_end_0, end_mask = var_10848_end_mask_0, x = v_43_cast)[name = tensor("op_10848_cast")]; + tensor var_10852_begin_0 = const()[name = tensor("op_10852_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_10852_end_0 = const()[name = tensor("op_10852_end_0"), val = tensor([2, 192, 1, 77])]; + tensor var_10852_end_mask_0 = const()[name = tensor("op_10852_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10852_cast = slice_by_index(begin = var_10852_begin_0, end = var_10852_end_0, end_mask = var_10852_end_mask_0, x = v_43_cast)[name = tensor("op_10852_cast")]; + tensor var_10856_begin_0 = const()[name = tensor("op_10856_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_10856_end_0 = const()[name = tensor("op_10856_end_0"), val = tensor([2, 256, 1, 77])]; + tensor var_10856_end_mask_0 = const()[name = tensor("op_10856_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10856_cast = slice_by_index(begin = var_10856_begin_0, end = var_10856_end_0, end_mask = var_10856_end_mask_0, x = v_43_cast)[name = tensor("op_10856_cast")]; + tensor var_10860_begin_0 = const()[name = tensor("op_10860_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_10860_end_0 = const()[name = tensor("op_10860_end_0"), val = tensor([2, 320, 1, 77])]; + tensor var_10860_end_mask_0 = const()[name = tensor("op_10860_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10860_cast = slice_by_index(begin = var_10860_begin_0, end = var_10860_end_0, end_mask = var_10860_end_mask_0, x = v_43_cast)[name = tensor("op_10860_cast")]; + tensor var_10864_begin_0 = const()[name = tensor("op_10864_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_10864_end_0 = const()[name = tensor("op_10864_end_0"), val = tensor([2, 384, 1, 77])]; + tensor var_10864_end_mask_0 = const()[name = tensor("op_10864_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10864_cast = slice_by_index(begin = var_10864_begin_0, end = var_10864_end_0, end_mask = var_10864_end_mask_0, x = v_43_cast)[name = tensor("op_10864_cast")]; + tensor var_10868_begin_0 = const()[name = tensor("op_10868_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_10868_end_0 = const()[name = tensor("op_10868_end_0"), val = tensor([2, 448, 1, 77])]; + tensor var_10868_end_mask_0 = const()[name = tensor("op_10868_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10868_cast = slice_by_index(begin = var_10868_begin_0, end = var_10868_end_0, end_mask = var_10868_end_mask_0, x = v_43_cast)[name = tensor("op_10868_cast")]; + tensor var_10872_begin_0 = const()[name = tensor("op_10872_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_10872_end_0 = const()[name = tensor("op_10872_end_0"), val = tensor([2, 512, 1, 77])]; + tensor var_10872_end_mask_0 = const()[name = tensor("op_10872_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10872_cast = slice_by_index(begin = var_10872_begin_0, end = var_10872_end_0, end_mask = var_10872_end_mask_0, x = v_43_cast)[name = tensor("op_10872_cast")]; + tensor var_10876_begin_0 = const()[name = tensor("op_10876_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_10876_end_0 = const()[name = tensor("op_10876_end_0"), val = tensor([2, 576, 1, 77])]; + tensor var_10876_end_mask_0 = const()[name = tensor("op_10876_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10876_cast = slice_by_index(begin = var_10876_begin_0, end = var_10876_end_0, end_mask = var_10876_end_mask_0, x = v_43_cast)[name = tensor("op_10876_cast")]; + tensor var_10880_begin_0 = const()[name = tensor("op_10880_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_10880_end_0 = const()[name = tensor("op_10880_end_0"), val = tensor([2, 640, 1, 77])]; + tensor var_10880_end_mask_0 = const()[name = tensor("op_10880_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10880_cast = slice_by_index(begin = var_10880_begin_0, end = var_10880_end_0, end_mask = var_10880_end_mask_0, x = v_43_cast)[name = tensor("op_10880_cast")]; + tensor var_10884_equation_0 = const()[name = tensor("op_10884_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10884_cast = einsum(equation = var_10884_equation_0, values = (var_10806_cast, var_10782_cast))[name = tensor("op_10884_cast")]; + tensor var_10885_to_fp16 = const()[name = tensor("op_10885_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_521_cast = mul(x = var_10884_cast, y = var_10885_to_fp16)[name = tensor("aw_chunk_521_cast")]; + tensor var_10888_equation_0 = const()[name = tensor("op_10888_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10888_cast = einsum(equation = var_10888_equation_0, values = (var_10806_cast, var_10783_cast))[name = tensor("op_10888_cast")]; + tensor var_10889_to_fp16 = const()[name = tensor("op_10889_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_523_cast = mul(x = var_10888_cast, y = var_10889_to_fp16)[name = tensor("aw_chunk_523_cast")]; + tensor var_10892_equation_0 = const()[name = tensor("op_10892_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10892_cast = einsum(equation = var_10892_equation_0, values = (var_10810_cast, var_10784_cast))[name = tensor("op_10892_cast")]; + tensor var_10893_to_fp16 = const()[name = tensor("op_10893_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_525_cast = mul(x = var_10892_cast, y = var_10893_to_fp16)[name = tensor("aw_chunk_525_cast")]; + tensor var_10896_equation_0 = const()[name = tensor("op_10896_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10896_cast = einsum(equation = var_10896_equation_0, values = (var_10810_cast, var_10785_cast))[name = tensor("op_10896_cast")]; + tensor var_10897_to_fp16 = const()[name = tensor("op_10897_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_527_cast = mul(x = var_10896_cast, y = var_10897_to_fp16)[name = tensor("aw_chunk_527_cast")]; + tensor var_10900_equation_0 = const()[name = tensor("op_10900_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10900_cast = einsum(equation = var_10900_equation_0, values = (var_10814_cast, var_10786_cast))[name = tensor("op_10900_cast")]; + tensor var_10901_to_fp16 = const()[name = tensor("op_10901_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_529_cast = mul(x = var_10900_cast, y = var_10901_to_fp16)[name = tensor("aw_chunk_529_cast")]; + tensor var_10904_equation_0 = const()[name = tensor("op_10904_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10904_cast = einsum(equation = var_10904_equation_0, values = (var_10814_cast, var_10787_cast))[name = tensor("op_10904_cast")]; + tensor var_10905_to_fp16 = const()[name = tensor("op_10905_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_531_cast = mul(x = var_10904_cast, y = var_10905_to_fp16)[name = tensor("aw_chunk_531_cast")]; + tensor var_10908_equation_0 = const()[name = tensor("op_10908_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10908_cast = einsum(equation = var_10908_equation_0, values = (var_10818_cast, var_10788_cast))[name = tensor("op_10908_cast")]; + tensor var_10909_to_fp16 = const()[name = tensor("op_10909_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_533_cast = mul(x = var_10908_cast, y = var_10909_to_fp16)[name = tensor("aw_chunk_533_cast")]; + tensor var_10912_equation_0 = const()[name = tensor("op_10912_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10912_cast = einsum(equation = var_10912_equation_0, values = (var_10818_cast, var_10789_cast))[name = tensor("op_10912_cast")]; + tensor var_10913_to_fp16 = const()[name = tensor("op_10913_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_535_cast = mul(x = var_10912_cast, y = var_10913_to_fp16)[name = tensor("aw_chunk_535_cast")]; + tensor var_10916_equation_0 = const()[name = tensor("op_10916_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10916_cast = einsum(equation = var_10916_equation_0, values = (var_10822_cast, var_10790_cast))[name = tensor("op_10916_cast")]; + tensor var_10917_to_fp16 = const()[name = tensor("op_10917_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_537_cast = mul(x = var_10916_cast, y = var_10917_to_fp16)[name = tensor("aw_chunk_537_cast")]; + tensor var_10920_equation_0 = const()[name = tensor("op_10920_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10920_cast = einsum(equation = var_10920_equation_0, values = (var_10822_cast, var_10791_cast))[name = tensor("op_10920_cast")]; + tensor var_10921_to_fp16 = const()[name = tensor("op_10921_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_539_cast = mul(x = var_10920_cast, y = var_10921_to_fp16)[name = tensor("aw_chunk_539_cast")]; + tensor var_10924_equation_0 = const()[name = tensor("op_10924_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10924_cast = einsum(equation = var_10924_equation_0, values = (var_10826_cast, var_10792_cast))[name = tensor("op_10924_cast")]; + tensor var_10925_to_fp16 = const()[name = tensor("op_10925_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_541_cast = mul(x = var_10924_cast, y = var_10925_to_fp16)[name = tensor("aw_chunk_541_cast")]; + tensor var_10928_equation_0 = const()[name = tensor("op_10928_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10928_cast = einsum(equation = var_10928_equation_0, values = (var_10826_cast, var_10793_cast))[name = tensor("op_10928_cast")]; + tensor var_10929_to_fp16 = const()[name = tensor("op_10929_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_543_cast = mul(x = var_10928_cast, y = var_10929_to_fp16)[name = tensor("aw_chunk_543_cast")]; + tensor var_10932_equation_0 = const()[name = tensor("op_10932_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10932_cast = einsum(equation = var_10932_equation_0, values = (var_10830_cast, var_10794_cast))[name = tensor("op_10932_cast")]; + tensor var_10933_to_fp16 = const()[name = tensor("op_10933_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_545_cast = mul(x = var_10932_cast, y = var_10933_to_fp16)[name = tensor("aw_chunk_545_cast")]; + tensor var_10936_equation_0 = const()[name = tensor("op_10936_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10936_cast = einsum(equation = var_10936_equation_0, values = (var_10830_cast, var_10795_cast))[name = tensor("op_10936_cast")]; + tensor var_10937_to_fp16 = const()[name = tensor("op_10937_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_547_cast = mul(x = var_10936_cast, y = var_10937_to_fp16)[name = tensor("aw_chunk_547_cast")]; + tensor var_10940_equation_0 = const()[name = tensor("op_10940_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10940_cast = einsum(equation = var_10940_equation_0, values = (var_10834_cast, var_10796_cast))[name = tensor("op_10940_cast")]; + tensor var_10941_to_fp16 = const()[name = tensor("op_10941_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_549_cast = mul(x = var_10940_cast, y = var_10941_to_fp16)[name = tensor("aw_chunk_549_cast")]; + tensor var_10944_equation_0 = const()[name = tensor("op_10944_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10944_cast = einsum(equation = var_10944_equation_0, values = (var_10834_cast, var_10797_cast))[name = tensor("op_10944_cast")]; + tensor var_10945_to_fp16 = const()[name = tensor("op_10945_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_551_cast = mul(x = var_10944_cast, y = var_10945_to_fp16)[name = tensor("aw_chunk_551_cast")]; + tensor var_10948_equation_0 = const()[name = tensor("op_10948_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10948_cast = einsum(equation = var_10948_equation_0, values = (var_10838_cast, var_10798_cast))[name = tensor("op_10948_cast")]; + tensor var_10949_to_fp16 = const()[name = tensor("op_10949_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_553_cast = mul(x = var_10948_cast, y = var_10949_to_fp16)[name = tensor("aw_chunk_553_cast")]; + tensor var_10952_equation_0 = const()[name = tensor("op_10952_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10952_cast = einsum(equation = var_10952_equation_0, values = (var_10838_cast, var_10799_cast))[name = tensor("op_10952_cast")]; + tensor var_10953_to_fp16 = const()[name = tensor("op_10953_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_555_cast = mul(x = var_10952_cast, y = var_10953_to_fp16)[name = tensor("aw_chunk_555_cast")]; + tensor var_10956_equation_0 = const()[name = tensor("op_10956_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10956_cast = einsum(equation = var_10956_equation_0, values = (var_10842_cast, var_10800_cast))[name = tensor("op_10956_cast")]; + tensor var_10957_to_fp16 = const()[name = tensor("op_10957_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_557_cast = mul(x = var_10956_cast, y = var_10957_to_fp16)[name = tensor("aw_chunk_557_cast")]; + tensor var_10960_equation_0 = const()[name = tensor("op_10960_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10960_cast = einsum(equation = var_10960_equation_0, values = (var_10842_cast, var_10801_cast))[name = tensor("op_10960_cast")]; + tensor var_10961_to_fp16 = const()[name = tensor("op_10961_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_559_cast = mul(x = var_10960_cast, y = var_10961_to_fp16)[name = tensor("aw_chunk_559_cast")]; + tensor var_10963_cast = softmax(axis = var_10257, x = aw_chunk_521_cast)[name = tensor("op_10963_cast")]; + tensor var_10964_cast = softmax(axis = var_10257, x = aw_chunk_523_cast)[name = tensor("op_10964_cast")]; + tensor var_10965_cast = softmax(axis = var_10257, x = aw_chunk_525_cast)[name = tensor("op_10965_cast")]; + tensor var_10966_cast = softmax(axis = var_10257, x = aw_chunk_527_cast)[name = tensor("op_10966_cast")]; + tensor var_10967_cast = softmax(axis = var_10257, x = aw_chunk_529_cast)[name = tensor("op_10967_cast")]; + tensor var_10968_cast = softmax(axis = var_10257, x = aw_chunk_531_cast)[name = tensor("op_10968_cast")]; + tensor var_10969_cast = softmax(axis = var_10257, x = aw_chunk_533_cast)[name = tensor("op_10969_cast")]; + tensor var_10970_cast = softmax(axis = var_10257, x = aw_chunk_535_cast)[name = tensor("op_10970_cast")]; + tensor var_10971_cast = softmax(axis = var_10257, x = aw_chunk_537_cast)[name = tensor("op_10971_cast")]; + tensor var_10972_cast = softmax(axis = var_10257, x = aw_chunk_539_cast)[name = tensor("op_10972_cast")]; + tensor var_10973_cast = softmax(axis = var_10257, x = aw_chunk_541_cast)[name = tensor("op_10973_cast")]; + tensor var_10974_cast = softmax(axis = var_10257, x = aw_chunk_543_cast)[name = tensor("op_10974_cast")]; + tensor var_10975_cast = softmax(axis = var_10257, x = aw_chunk_545_cast)[name = tensor("op_10975_cast")]; + tensor var_10976_cast = softmax(axis = var_10257, x = aw_chunk_547_cast)[name = tensor("op_10976_cast")]; + tensor var_10977_cast = softmax(axis = var_10257, x = aw_chunk_549_cast)[name = tensor("op_10977_cast")]; + tensor var_10978_cast = softmax(axis = var_10257, x = aw_chunk_551_cast)[name = tensor("op_10978_cast")]; + tensor var_10979_cast = softmax(axis = var_10257, x = aw_chunk_553_cast)[name = tensor("op_10979_cast")]; + tensor var_10980_cast = softmax(axis = var_10257, x = aw_chunk_555_cast)[name = tensor("op_10980_cast")]; + tensor var_10981_cast = softmax(axis = var_10257, x = aw_chunk_557_cast)[name = tensor("op_10981_cast")]; + tensor var_10982_cast = softmax(axis = var_10257, x = aw_chunk_559_cast)[name = tensor("op_10982_cast")]; + tensor var_10984_equation_0 = const()[name = tensor("op_10984_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_10984_cast = einsum(equation = var_10984_equation_0, values = (var_10844_cast, var_10963_cast))[name = tensor("op_10984_cast")]; + tensor var_10986_equation_0 = const()[name = tensor("op_10986_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_10986_cast = einsum(equation = var_10986_equation_0, values = (var_10844_cast, var_10964_cast))[name = tensor("op_10986_cast")]; + tensor var_10988_equation_0 = const()[name = tensor("op_10988_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_10988_cast = einsum(equation = var_10988_equation_0, values = (var_10848_cast, var_10965_cast))[name = tensor("op_10988_cast")]; + tensor var_10990_equation_0 = const()[name = tensor("op_10990_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_10990_cast = einsum(equation = var_10990_equation_0, values = (var_10848_cast, var_10966_cast))[name = tensor("op_10990_cast")]; + tensor var_10992_equation_0 = const()[name = tensor("op_10992_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_10992_cast = einsum(equation = var_10992_equation_0, values = (var_10852_cast, var_10967_cast))[name = tensor("op_10992_cast")]; + tensor var_10994_equation_0 = const()[name = tensor("op_10994_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_10994_cast = einsum(equation = var_10994_equation_0, values = (var_10852_cast, var_10968_cast))[name = tensor("op_10994_cast")]; + tensor var_10996_equation_0 = const()[name = tensor("op_10996_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_10996_cast = einsum(equation = var_10996_equation_0, values = (var_10856_cast, var_10969_cast))[name = tensor("op_10996_cast")]; + tensor var_10998_equation_0 = const()[name = tensor("op_10998_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_10998_cast = einsum(equation = var_10998_equation_0, values = (var_10856_cast, var_10970_cast))[name = tensor("op_10998_cast")]; + tensor var_11000_equation_0 = const()[name = tensor("op_11000_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11000_cast = einsum(equation = var_11000_equation_0, values = (var_10860_cast, var_10971_cast))[name = tensor("op_11000_cast")]; + tensor var_11002_equation_0 = const()[name = tensor("op_11002_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11002_cast = einsum(equation = var_11002_equation_0, values = (var_10860_cast, var_10972_cast))[name = tensor("op_11002_cast")]; + tensor var_11004_equation_0 = const()[name = tensor("op_11004_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11004_cast = einsum(equation = var_11004_equation_0, values = (var_10864_cast, var_10973_cast))[name = tensor("op_11004_cast")]; + tensor var_11006_equation_0 = const()[name = tensor("op_11006_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11006_cast = einsum(equation = var_11006_equation_0, values = (var_10864_cast, var_10974_cast))[name = tensor("op_11006_cast")]; + tensor var_11008_equation_0 = const()[name = tensor("op_11008_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11008_cast = einsum(equation = var_11008_equation_0, values = (var_10868_cast, var_10975_cast))[name = tensor("op_11008_cast")]; + tensor var_11010_equation_0 = const()[name = tensor("op_11010_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11010_cast = einsum(equation = var_11010_equation_0, values = (var_10868_cast, var_10976_cast))[name = tensor("op_11010_cast")]; + tensor var_11012_equation_0 = const()[name = tensor("op_11012_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11012_cast = einsum(equation = var_11012_equation_0, values = (var_10872_cast, var_10977_cast))[name = tensor("op_11012_cast")]; + tensor var_11014_equation_0 = const()[name = tensor("op_11014_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11014_cast = einsum(equation = var_11014_equation_0, values = (var_10872_cast, var_10978_cast))[name = tensor("op_11014_cast")]; + tensor var_11016_equation_0 = const()[name = tensor("op_11016_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11016_cast = einsum(equation = var_11016_equation_0, values = (var_10876_cast, var_10979_cast))[name = tensor("op_11016_cast")]; + tensor var_11018_equation_0 = const()[name = tensor("op_11018_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11018_cast = einsum(equation = var_11018_equation_0, values = (var_10876_cast, var_10980_cast))[name = tensor("op_11018_cast")]; + tensor var_11020_equation_0 = const()[name = tensor("op_11020_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11020_cast = einsum(equation = var_11020_equation_0, values = (var_10880_cast, var_10981_cast))[name = tensor("op_11020_cast")]; + tensor var_11022_equation_0 = const()[name = tensor("op_11022_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11022_cast = einsum(equation = var_11022_equation_0, values = (var_10880_cast, var_10982_cast))[name = tensor("op_11022_cast")]; + tensor var_11024_interleave_0 = const()[name = tensor("op_11024_interleave_0"), val = tensor(false)]; + tensor var_11024_cast = concat(axis = var_10234, interleave = var_11024_interleave_0, values = (var_10984_cast, var_10986_cast))[name = tensor("op_11024_cast")]; + tensor var_11026_interleave_0 = const()[name = tensor("op_11026_interleave_0"), val = tensor(false)]; + tensor var_11026_cast = concat(axis = var_10234, interleave = var_11026_interleave_0, values = (var_10988_cast, var_10990_cast))[name = tensor("op_11026_cast")]; + tensor var_11028_interleave_0 = const()[name = tensor("op_11028_interleave_0"), val = tensor(false)]; + tensor var_11028_cast = concat(axis = var_10234, interleave = var_11028_interleave_0, values = (var_10992_cast, var_10994_cast))[name = tensor("op_11028_cast")]; + tensor var_11030_interleave_0 = const()[name = tensor("op_11030_interleave_0"), val = tensor(false)]; + tensor var_11030_cast = concat(axis = var_10234, interleave = var_11030_interleave_0, values = (var_10996_cast, var_10998_cast))[name = tensor("op_11030_cast")]; + tensor var_11032_interleave_0 = const()[name = tensor("op_11032_interleave_0"), val = tensor(false)]; + tensor var_11032_cast = concat(axis = var_10234, interleave = var_11032_interleave_0, values = (var_11000_cast, var_11002_cast))[name = tensor("op_11032_cast")]; + tensor var_11034_interleave_0 = const()[name = tensor("op_11034_interleave_0"), val = tensor(false)]; + tensor var_11034_cast = concat(axis = var_10234, interleave = var_11034_interleave_0, values = (var_11004_cast, var_11006_cast))[name = tensor("op_11034_cast")]; + tensor var_11036_interleave_0 = const()[name = tensor("op_11036_interleave_0"), val = tensor(false)]; + tensor var_11036_cast = concat(axis = var_10234, interleave = var_11036_interleave_0, values = (var_11008_cast, var_11010_cast))[name = tensor("op_11036_cast")]; + tensor var_11038_interleave_0 = const()[name = tensor("op_11038_interleave_0"), val = tensor(false)]; + tensor var_11038_cast = concat(axis = var_10234, interleave = var_11038_interleave_0, values = (var_11012_cast, var_11014_cast))[name = tensor("op_11038_cast")]; + tensor var_11040_interleave_0 = const()[name = tensor("op_11040_interleave_0"), val = tensor(false)]; + tensor var_11040_cast = concat(axis = var_10234, interleave = var_11040_interleave_0, values = (var_11016_cast, var_11018_cast))[name = tensor("op_11040_cast")]; + tensor var_11042_interleave_0 = const()[name = tensor("op_11042_interleave_0"), val = tensor(false)]; + tensor var_11042_cast = concat(axis = var_10234, interleave = var_11042_interleave_0, values = (var_11020_cast, var_11022_cast))[name = tensor("op_11042_cast")]; + tensor input_385_interleave_0 = const()[name = tensor("input_385_interleave_0"), val = tensor(false)]; + tensor input_385_cast = concat(axis = var_10257, interleave = input_385_interleave_0, values = (var_11024_cast, var_11026_cast, var_11028_cast, var_11030_cast, var_11032_cast, var_11034_cast, var_11036_cast, var_11038_cast, var_11040_cast, var_11042_cast))[name = tensor("input_385_cast")]; + tensor var_11048 = const()[name = tensor("op_11048"), val = tensor([1, 1])]; + tensor var_11050 = const()[name = tensor("op_11050"), val = tensor([1, 1])]; + tensor var_11052_pad_type_0 = const()[name = tensor("op_11052_pad_type_0"), val = tensor("custom")]; + tensor var_11052_pad_0 = const()[name = tensor("op_11052_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_2_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(596503424))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(596810688))), name = tensor("up_blocks_2_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16_palettized"), shape = tensor([640, 640, 1, 1])]; + tensor up_blocks_2_attentions_0_transformer_blocks_0_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("up_blocks_2_attentions_0_transformer_blocks_0_attn2_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(596810880)))]; + tensor var_11052_cast = conv(bias = up_blocks_2_attentions_0_transformer_blocks_0_attn2_to_out_0_bias_to_fp16, dilations = var_11050, groups = var_10257, pad = var_11052_pad_0, pad_type = var_11052_pad_type_0, strides = var_11048, weight = up_blocks_2_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16_palettized, x = input_385_cast)[name = tensor("op_11052_cast")]; + tensor inputs_65_cast = add(x = var_11052_cast, y = inputs_63_cast)[name = tensor("inputs_65_cast")]; + tensor var_11056 = const()[name = tensor("op_11056"), val = tensor([1])]; + tensor channels_mean_65_cast = reduce_mean(axes = var_11056, keep_dims = var_10252, x = inputs_65_cast)[name = tensor("channels_mean_65_cast")]; + tensor zero_mean_65_cast = sub(x = inputs_65_cast, y = channels_mean_65_cast)[name = tensor("zero_mean_65_cast")]; + tensor zero_mean_sq_65_cast = mul(x = zero_mean_65_cast, y = zero_mean_65_cast)[name = tensor("zero_mean_sq_65_cast")]; + tensor var_11060 = const()[name = tensor("op_11060"), val = tensor([1])]; + tensor var_11061_cast = reduce_mean(axes = var_11060, keep_dims = var_10252, x = zero_mean_sq_65_cast)[name = tensor("op_11061_cast")]; + tensor var_11062_to_fp16 = const()[name = tensor("op_11062_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_11063_cast = add(x = var_11061_cast, y = var_11062_to_fp16)[name = tensor("op_11063_cast")]; + tensor denom_65_epsilon_0_to_fp16 = const()[name = tensor("denom_65_epsilon_0_to_fp16"), val = tensor(0x1p-24)]; + tensor denom_65_cast = rsqrt(epsilon = denom_65_epsilon_0_to_fp16, x = var_11063_cast)[name = tensor("denom_65_cast")]; + tensor out_65_cast = mul(x = zero_mean_65_cast, y = denom_65_cast)[name = tensor("out_65_cast")]; + tensor var_11067_to_fp16 = const()[name = tensor("op_11067_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(596812224)))]; + tensor var_11068_cast = add(x = out_65_cast, y = var_11067_to_fp16)[name = tensor("op_11068_cast")]; + tensor var_11070_to_fp16 = const()[name = tensor("op_11070_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(596813568)))]; + tensor input_387_cast = mul(x = var_11068_cast, y = var_11070_to_fp16)[name = tensor("input_387_cast")]; + tensor var_11078 = const()[name = tensor("op_11078"), val = tensor([1, 1])]; + tensor var_11080 = const()[name = tensor("op_11080"), val = tensor([1, 1])]; + tensor var_11082_pad_type_0 = const()[name = tensor("op_11082_pad_type_0"), val = tensor("custom")]; + tensor var_11082_pad_0 = const()[name = tensor("op_11082_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_2_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(596814912))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(599272576))), name = tensor("up_blocks_2_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16_palettized"), shape = tensor([5120, 640, 1, 1])]; + tensor up_blocks_2_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(599272768))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(599276672))), name = tensor("up_blocks_2_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16_palettized"), shape = tensor([5120])]; + tensor var_11082_cast = conv(bias = up_blocks_2_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16_palettized, dilations = var_11080, groups = var_10257, pad = var_11082_pad_0, pad_type = var_11082_pad_type_0, strides = var_11078, weight = up_blocks_2_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16_palettized, x = input_387_cast)[name = tensor("op_11082_cast")]; + tensor var_11083_split_sizes_0 = const()[name = tensor("op_11083_split_sizes_0"), val = tensor([2560, 2560])]; + tensor var_11083_axis_0 = const()[name = tensor("op_11083_axis_0"), val = tensor(1)]; + tensor var_11083_cast_0, tensor var_11083_cast_1 = split(axis = var_11083_axis_0, split_sizes = var_11083_split_sizes_0, x = var_11082_cast)[name = tensor("op_11083_cast")]; + tensor var_11085_mode_0 = const()[name = tensor("op_11085_mode_0"), val = tensor("EXACT")]; + tensor var_11085_cast = gelu(mode = var_11085_mode_0, x = var_11083_cast_1)[name = tensor("op_11085_cast")]; + tensor input_389_cast = mul(x = var_11083_cast_0, y = var_11085_cast)[name = tensor("input_389_cast")]; + tensor var_11089 = const()[name = tensor("op_11089"), val = tensor([1, 1])]; + tensor var_11091 = const()[name = tensor("op_11091"), val = tensor([1, 1])]; + tensor var_11093_pad_type_0 = const()[name = tensor("op_11093_pad_type_0"), val = tensor("custom")]; + tensor var_11093_pad_0 = const()[name = tensor("op_11093_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_2_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(599276864))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(600505728))), name = tensor("up_blocks_2_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16_palettized"), shape = tensor([640, 2560, 1, 1])]; + tensor up_blocks_2_attentions_0_transformer_blocks_0_ff_net_2_bias_to_fp16 = const()[name = tensor("up_blocks_2_attentions_0_transformer_blocks_0_ff_net_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(600505920)))]; + tensor var_11093_cast = conv(bias = up_blocks_2_attentions_0_transformer_blocks_0_ff_net_2_bias_to_fp16, dilations = var_11091, groups = var_10257, pad = var_11093_pad_0, pad_type = var_11093_pad_type_0, strides = var_11089, weight = up_blocks_2_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16_palettized, x = input_389_cast)[name = tensor("op_11093_cast")]; + tensor hidden_states_233_cast = add(x = var_11093_cast, y = inputs_65_cast)[name = tensor("hidden_states_233_cast")]; + tensor var_11095 = const()[name = tensor("op_11095"), val = tensor([2, 640, 32, 32])]; + tensor input_391_cast = reshape(shape = var_11095, x = hidden_states_233_cast)[name = tensor("input_391_cast")]; + tensor var_11099 = const()[name = tensor("op_11099"), val = tensor([1, 1])]; + tensor var_11101 = const()[name = tensor("op_11101"), val = tensor([1, 1])]; + tensor hidden_states_235_pad_type_0 = const()[name = tensor("hidden_states_235_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_235_pad_0 = const()[name = tensor("hidden_states_235_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_2_attentions_0_proj_out_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(600507264))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(600814528))), name = tensor("up_blocks_2_attentions_0_proj_out_weight_to_fp16_palettized"), shape = tensor([640, 640, 1, 1])]; + tensor up_blocks_2_attentions_0_proj_out_bias_to_fp16 = const()[name = tensor("up_blocks_2_attentions_0_proj_out_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(600814720)))]; + tensor hidden_states_235_cast = conv(bias = up_blocks_2_attentions_0_proj_out_bias_to_fp16, dilations = var_11101, groups = var_10257, pad = hidden_states_235_pad_0, pad_type = hidden_states_235_pad_type_0, strides = var_11099, weight = up_blocks_2_attentions_0_proj_out_weight_to_fp16_palettized, x = input_391_cast)[name = tensor("hidden_states_235_cast")]; + tensor hidden_states_237_cast = add(x = hidden_states_235_cast, y = hidden_states_223_cast)[name = tensor("hidden_states_237_cast")]; + tensor input_393_interleave_0 = const()[name = tensor("input_393_interleave_0"), val = tensor(false)]; + tensor input_393_cast = concat(axis = var_10257, interleave = input_393_interleave_0, values = (hidden_states_237_cast, input_89_cast))[name = tensor("input_393_cast")]; + tensor reshape_180_shape_0 = const()[name = tensor("reshape_180_shape_0"), val = tensor([2, 32, 40, 32, 32])]; + tensor reshape_180_cast = reshape(shape = reshape_180_shape_0, x = input_393_cast)[name = tensor("reshape_180_cast")]; + tensor reduce_mean_135_axes_0 = const()[name = tensor("reduce_mean_135_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_135_keep_dims_0 = const()[name = tensor("reduce_mean_135_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_135_cast = reduce_mean(axes = reduce_mean_135_axes_0, keep_dims = reduce_mean_135_keep_dims_0, x = reshape_180_cast)[name = tensor("reduce_mean_135_cast")]; + tensor sub_90_cast = sub(x = reshape_180_cast, y = reduce_mean_135_cast)[name = tensor("sub_90_cast")]; + tensor square_45_cast = square(x = sub_90_cast)[name = tensor("square_45_cast")]; + tensor reduce_mean_137_axes_0 = const()[name = tensor("reduce_mean_137_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_137_keep_dims_0 = const()[name = tensor("reduce_mean_137_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_137_cast = reduce_mean(axes = reduce_mean_137_axes_0, keep_dims = reduce_mean_137_keep_dims_0, x = square_45_cast)[name = tensor("reduce_mean_137_cast")]; + tensor add_90_y_0_to_fp16 = const()[name = tensor("add_90_y_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_90_cast = add(x = reduce_mean_137_cast, y = add_90_y_0_to_fp16)[name = tensor("add_90_cast")]; + tensor sqrt_45_cast = sqrt(x = add_90_cast)[name = tensor("sqrt_45_cast")]; + tensor real_div_45_cast = real_div(x = sub_90_cast, y = sqrt_45_cast)[name = tensor("real_div_45_cast")]; + tensor reshape_181_shape_0 = const()[name = tensor("reshape_181_shape_0"), val = tensor([2, 1280, 32, 32])]; + tensor reshape_181_cast = reshape(shape = reshape_181_shape_0, x = real_div_45_cast)[name = tensor("reshape_181_cast")]; + tensor add_91_gamma_0_to_fp16 = const()[name = tensor("add_91_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(600816064)))]; + tensor add_91_beta_0_to_fp16 = const()[name = tensor("add_91_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(600818688)))]; + tensor add_91_epsilon_0_to_fp16 = const()[name = tensor("add_91_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_91_cast = batch_norm(beta = add_91_beta_0_to_fp16, epsilon = add_91_epsilon_0_to_fp16, gamma = add_91_gamma_0_to_fp16, mean = add_27_mean_0_to_fp16, variance = add_27_variance_0_to_fp16, x = reshape_181_cast)[name = tensor("add_91_cast")]; + tensor input_397_cast = silu(x = add_91_cast)[name = tensor("input_397_cast")]; + tensor var_11119 = const()[name = tensor("op_11119"), val = tensor([1, 1])]; + tensor var_11121 = const()[name = tensor("op_11121"), val = tensor([1, 1])]; + tensor hidden_states_239_pad_type_0 = const()[name = tensor("hidden_states_239_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_239_pad_0 = const()[name = tensor("hidden_states_239_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor up_blocks_2_resnets_1_conv1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(600821312))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(606350976))), name = tensor("up_blocks_2_resnets_1_conv1_weight_to_fp16_palettized"), shape = tensor([640, 1280, 3, 3])]; + tensor up_blocks_2_resnets_1_conv1_bias_to_fp16 = const()[name = tensor("up_blocks_2_resnets_1_conv1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(606351168)))]; + tensor hidden_states_239_cast = conv(bias = up_blocks_2_resnets_1_conv1_bias_to_fp16, dilations = var_11121, groups = var_10257, pad = hidden_states_239_pad_0, pad_type = hidden_states_239_pad_type_0, strides = var_11119, weight = up_blocks_2_resnets_1_conv1_weight_to_fp16_palettized, x = input_397_cast)[name = tensor("hidden_states_239_cast")]; + tensor var_11127 = const()[name = tensor("op_11127"), val = tensor([1, 1])]; + tensor var_11129 = const()[name = tensor("op_11129"), val = tensor([1, 1])]; + tensor temb_35_pad_type_0 = const()[name = tensor("temb_35_pad_type_0"), val = tensor("custom")]; + tensor temb_35_pad_0 = const()[name = tensor("temb_35_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_2_resnets_1_time_emb_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(606352512))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(606966976))), name = tensor("up_blocks_2_resnets_1_time_emb_proj_weight_to_fp16_palettized"), shape = tensor([640, 1280, 1, 1])]; + tensor up_blocks_2_resnets_1_time_emb_proj_bias_to_fp16 = const()[name = tensor("up_blocks_2_resnets_1_time_emb_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(606967168)))]; + tensor temb_35_cast = conv(bias = up_blocks_2_resnets_1_time_emb_proj_bias_to_fp16, dilations = var_11129, groups = var_10257, pad = temb_35_pad_0, pad_type = temb_35_pad_type_0, strides = var_11127, weight = up_blocks_2_resnets_1_time_emb_proj_weight_to_fp16_palettized, x = input_15_cast)[name = tensor("temb_35_cast")]; + tensor input_401_cast = add(x = hidden_states_239_cast, y = temb_35_cast)[name = tensor("input_401_cast")]; + tensor reshape_184_shape_0 = const()[name = tensor("reshape_184_shape_0"), val = tensor([2, 32, 20, 32, 32])]; + tensor reshape_184_cast = reshape(shape = reshape_184_shape_0, x = input_401_cast)[name = tensor("reshape_184_cast")]; + tensor reduce_mean_138_axes_0 = const()[name = tensor("reduce_mean_138_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_138_keep_dims_0 = const()[name = tensor("reduce_mean_138_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_138_cast = reduce_mean(axes = reduce_mean_138_axes_0, keep_dims = reduce_mean_138_keep_dims_0, x = reshape_184_cast)[name = tensor("reduce_mean_138_cast")]; + tensor sub_92_cast = sub(x = reshape_184_cast, y = reduce_mean_138_cast)[name = tensor("sub_92_cast")]; + tensor square_46_cast = square(x = sub_92_cast)[name = tensor("square_46_cast")]; + tensor reduce_mean_140_axes_0 = const()[name = tensor("reduce_mean_140_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_140_keep_dims_0 = const()[name = tensor("reduce_mean_140_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_140_cast = reduce_mean(axes = reduce_mean_140_axes_0, keep_dims = reduce_mean_140_keep_dims_0, x = square_46_cast)[name = tensor("reduce_mean_140_cast")]; + tensor add_92_y_0_to_fp16 = const()[name = tensor("add_92_y_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_92_cast = add(x = reduce_mean_140_cast, y = add_92_y_0_to_fp16)[name = tensor("add_92_cast")]; + tensor sqrt_46_cast = sqrt(x = add_92_cast)[name = tensor("sqrt_46_cast")]; + tensor real_div_46_cast = real_div(x = sub_92_cast, y = sqrt_46_cast)[name = tensor("real_div_46_cast")]; + tensor reshape_185_shape_0 = const()[name = tensor("reshape_185_shape_0"), val = tensor([2, 640, 32, 32])]; + tensor reshape_185_cast = reshape(shape = reshape_185_shape_0, x = real_div_46_cast)[name = tensor("reshape_185_cast")]; + tensor add_93_gamma_0_to_fp16 = const()[name = tensor("add_93_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(606968512)))]; + tensor add_93_beta_0_to_fp16 = const()[name = tensor("add_93_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(606969856)))]; + tensor add_93_epsilon_0_to_fp16 = const()[name = tensor("add_93_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_93_cast = batch_norm(beta = add_93_beta_0_to_fp16, epsilon = add_93_epsilon_0_to_fp16, gamma = add_93_gamma_0_to_fp16, mean = add_15_mean_0_to_fp16, variance = add_15_variance_0_to_fp16, x = reshape_185_cast)[name = tensor("add_93_cast")]; + tensor input_405_cast = silu(x = add_93_cast)[name = tensor("input_405_cast")]; + tensor var_11139 = const()[name = tensor("op_11139"), val = tensor([1, 1])]; + tensor var_11141 = const()[name = tensor("op_11141"), val = tensor([1, 1])]; + tensor hidden_states_241_pad_type_0 = const()[name = tensor("hidden_states_241_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_241_pad_0 = const()[name = tensor("hidden_states_241_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor up_blocks_2_resnets_1_conv2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(606971200))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(609736064))), name = tensor("up_blocks_2_resnets_1_conv2_weight_to_fp16_palettized"), shape = tensor([640, 640, 3, 3])]; + tensor up_blocks_2_resnets_1_conv2_bias_to_fp16 = const()[name = tensor("up_blocks_2_resnets_1_conv2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(609736256)))]; + tensor hidden_states_241_cast = conv(bias = up_blocks_2_resnets_1_conv2_bias_to_fp16, dilations = var_11141, groups = var_10257, pad = hidden_states_241_pad_0, pad_type = hidden_states_241_pad_type_0, strides = var_11139, weight = up_blocks_2_resnets_1_conv2_weight_to_fp16_palettized, x = input_405_cast)[name = tensor("hidden_states_241_cast")]; + tensor var_11146 = const()[name = tensor("op_11146"), val = tensor([1, 1])]; + tensor var_11148 = const()[name = tensor("op_11148"), val = tensor([1, 1])]; + tensor x_19_pad_type_0 = const()[name = tensor("x_19_pad_type_0"), val = tensor("custom")]; + tensor x_19_pad_0 = const()[name = tensor("x_19_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_2_resnets_1_conv_shortcut_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(609737600))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(610352064))), name = tensor("up_blocks_2_resnets_1_conv_shortcut_weight_to_fp16_palettized"), shape = tensor([640, 1280, 1, 1])]; + tensor up_blocks_2_resnets_1_conv_shortcut_bias_to_fp16 = const()[name = tensor("up_blocks_2_resnets_1_conv_shortcut_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(610352256)))]; + tensor x_19_cast = conv(bias = up_blocks_2_resnets_1_conv_shortcut_bias_to_fp16, dilations = var_11148, groups = var_10257, pad = x_19_pad_0, pad_type = x_19_pad_type_0, strides = var_11146, weight = up_blocks_2_resnets_1_conv_shortcut_weight_to_fp16_palettized, x = input_393_cast)[name = tensor("x_19_cast")]; + tensor hidden_states_243_cast = add(x = x_19_cast, y = hidden_states_241_cast)[name = tensor("hidden_states_243_cast")]; + tensor reshape_188_shape_0 = const()[name = tensor("reshape_188_shape_0"), val = tensor([2, 32, 20, 32, 32])]; + tensor reshape_188_cast = reshape(shape = reshape_188_shape_0, x = hidden_states_243_cast)[name = tensor("reshape_188_cast")]; + tensor reduce_mean_141_axes_0 = const()[name = tensor("reduce_mean_141_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_141_keep_dims_0 = const()[name = tensor("reduce_mean_141_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_141_cast = reduce_mean(axes = reduce_mean_141_axes_0, keep_dims = reduce_mean_141_keep_dims_0, x = reshape_188_cast)[name = tensor("reduce_mean_141_cast")]; + tensor sub_94_cast = sub(x = reshape_188_cast, y = reduce_mean_141_cast)[name = tensor("sub_94_cast")]; + tensor square_47_cast = square(x = sub_94_cast)[name = tensor("square_47_cast")]; + tensor reduce_mean_143_axes_0 = const()[name = tensor("reduce_mean_143_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_143_keep_dims_0 = const()[name = tensor("reduce_mean_143_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_143_cast = reduce_mean(axes = reduce_mean_143_axes_0, keep_dims = reduce_mean_143_keep_dims_0, x = square_47_cast)[name = tensor("reduce_mean_143_cast")]; + tensor add_94_y_0_to_fp16 = const()[name = tensor("add_94_y_0_to_fp16"), val = tensor(0x1.1p-20)]; + tensor add_94_cast = add(x = reduce_mean_143_cast, y = add_94_y_0_to_fp16)[name = tensor("add_94_cast")]; + tensor sqrt_47_cast = sqrt(x = add_94_cast)[name = tensor("sqrt_47_cast")]; + tensor real_div_47_cast = real_div(x = sub_94_cast, y = sqrt_47_cast)[name = tensor("real_div_47_cast")]; + tensor reshape_189_shape_0 = const()[name = tensor("reshape_189_shape_0"), val = tensor([2, 640, 32, 32])]; + tensor reshape_189_cast = reshape(shape = reshape_189_shape_0, x = real_div_47_cast)[name = tensor("reshape_189_cast")]; + tensor add_95_gamma_0_to_fp16 = const()[name = tensor("add_95_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(610353600)))]; + tensor add_95_beta_0_to_fp16 = const()[name = tensor("add_95_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(610354944)))]; + tensor add_95_epsilon_0_to_fp16 = const()[name = tensor("add_95_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_95_cast = batch_norm(beta = add_95_beta_0_to_fp16, epsilon = add_95_epsilon_0_to_fp16, gamma = add_95_gamma_0_to_fp16, mean = add_15_mean_0_to_fp16, variance = add_15_variance_0_to_fp16, x = reshape_189_cast)[name = tensor("add_95_cast")]; + tensor var_11168 = const()[name = tensor("op_11168"), val = tensor([1, 1])]; + tensor var_11170 = const()[name = tensor("op_11170"), val = tensor([1, 1])]; + tensor hidden_states_245_pad_type_0 = const()[name = tensor("hidden_states_245_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_245_pad_0 = const()[name = tensor("hidden_states_245_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_2_attentions_1_proj_in_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(610356288))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(610663552))), name = tensor("up_blocks_2_attentions_1_proj_in_weight_to_fp16_palettized"), shape = tensor([640, 640, 1, 1])]; + tensor up_blocks_2_attentions_1_proj_in_bias_to_fp16 = const()[name = tensor("up_blocks_2_attentions_1_proj_in_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(610663744)))]; + tensor hidden_states_245_cast = conv(bias = up_blocks_2_attentions_1_proj_in_bias_to_fp16, dilations = var_11170, groups = var_10257, pad = hidden_states_245_pad_0, pad_type = hidden_states_245_pad_type_0, strides = var_11168, weight = up_blocks_2_attentions_1_proj_in_weight_to_fp16_palettized, x = add_95_cast)[name = tensor("hidden_states_245_cast")]; + tensor var_11175 = const()[name = tensor("op_11175"), val = tensor([2, 640, 1, 1024])]; + tensor inputs_67_cast = reshape(shape = var_11175, x = hidden_states_245_cast)[name = tensor("inputs_67_cast")]; + tensor var_11185 = const()[name = tensor("op_11185"), val = tensor([1])]; + tensor channels_mean_67_cast = reduce_mean(axes = var_11185, keep_dims = var_10252, x = inputs_67_cast)[name = tensor("channels_mean_67_cast")]; + tensor zero_mean_67_cast = sub(x = inputs_67_cast, y = channels_mean_67_cast)[name = tensor("zero_mean_67_cast")]; + tensor zero_mean_sq_67_cast = mul(x = zero_mean_67_cast, y = zero_mean_67_cast)[name = tensor("zero_mean_sq_67_cast")]; + tensor var_11189 = const()[name = tensor("op_11189"), val = tensor([1])]; + tensor var_11190_cast = reduce_mean(axes = var_11189, keep_dims = var_10252, x = zero_mean_sq_67_cast)[name = tensor("op_11190_cast")]; + tensor var_11191_to_fp16 = const()[name = tensor("op_11191_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_11192_cast = add(x = var_11190_cast, y = var_11191_to_fp16)[name = tensor("op_11192_cast")]; + tensor denom_67_epsilon_0_to_fp16 = const()[name = tensor("denom_67_epsilon_0_to_fp16"), val = tensor(0x1p-24)]; + tensor denom_67_cast = rsqrt(epsilon = denom_67_epsilon_0_to_fp16, x = var_11192_cast)[name = tensor("denom_67_cast")]; + tensor out_67_cast = mul(x = zero_mean_67_cast, y = denom_67_cast)[name = tensor("out_67_cast")]; + tensor var_11196_to_fp16 = const()[name = tensor("op_11196_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(610665088)))]; + tensor var_11197_cast = add(x = out_67_cast, y = var_11196_to_fp16)[name = tensor("op_11197_cast")]; + tensor var_11199_to_fp16 = const()[name = tensor("op_11199_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(610666432)))]; + tensor hidden_states_247_cast = mul(x = var_11197_cast, y = var_11199_to_fp16)[name = tensor("hidden_states_247_cast")]; + tensor var_11206 = const()[name = tensor("op_11206"), val = tensor([1, 1])]; + tensor var_11208 = const()[name = tensor("op_11208"), val = tensor([1, 1])]; + tensor q_45_pad_type_0 = const()[name = tensor("q_45_pad_type_0"), val = tensor("custom")]; + tensor q_45_pad_0 = const()[name = tensor("q_45_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_2_attentions_1_transformer_blocks_0_attn1_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(610667776))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(610975040))), name = tensor("up_blocks_2_attentions_1_transformer_blocks_0_attn1_to_q_weight_to_fp16_palettized"), shape = tensor([640, 640, 1, 1])]; + tensor q_45_cast = conv(dilations = var_11208, groups = var_10257, pad = q_45_pad_0, pad_type = q_45_pad_type_0, strides = var_11206, weight = up_blocks_2_attentions_1_transformer_blocks_0_attn1_to_q_weight_to_fp16_palettized, x = hidden_states_247_cast)[name = tensor("q_45_cast")]; + tensor var_11212 = const()[name = tensor("op_11212"), val = tensor([1, 1])]; + tensor var_11214 = const()[name = tensor("op_11214"), val = tensor([1, 1])]; + tensor k_89_pad_type_0 = const()[name = tensor("k_89_pad_type_0"), val = tensor("custom")]; + tensor k_89_pad_0 = const()[name = tensor("k_89_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_2_attentions_1_transformer_blocks_0_attn1_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(610975232))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(611282496))), name = tensor("up_blocks_2_attentions_1_transformer_blocks_0_attn1_to_k_weight_to_fp16_palettized"), shape = tensor([640, 640, 1, 1])]; + tensor k_89_cast = conv(dilations = var_11214, groups = var_10257, pad = k_89_pad_0, pad_type = k_89_pad_type_0, strides = var_11212, weight = up_blocks_2_attentions_1_transformer_blocks_0_attn1_to_k_weight_to_fp16_palettized, x = hidden_states_247_cast)[name = tensor("k_89_cast")]; + tensor var_11218 = const()[name = tensor("op_11218"), val = tensor([1, 1])]; + tensor var_11220 = const()[name = tensor("op_11220"), val = tensor([1, 1])]; + tensor v_45_pad_type_0 = const()[name = tensor("v_45_pad_type_0"), val = tensor("custom")]; + tensor v_45_pad_0 = const()[name = tensor("v_45_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_2_attentions_1_transformer_blocks_0_attn1_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(611282688))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(611589952))), name = tensor("up_blocks_2_attentions_1_transformer_blocks_0_attn1_to_v_weight_to_fp16_palettized"), shape = tensor([640, 640, 1, 1])]; + tensor v_45_cast = conv(dilations = var_11220, groups = var_10257, pad = v_45_pad_0, pad_type = v_45_pad_type_0, strides = var_11218, weight = up_blocks_2_attentions_1_transformer_blocks_0_attn1_to_v_weight_to_fp16_palettized, x = hidden_states_247_cast)[name = tensor("v_45_cast")]; + tensor var_11224_begin_0 = const()[name = tensor("op_11224_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11224_end_0 = const()[name = tensor("op_11224_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_11224_end_mask_0 = const()[name = tensor("op_11224_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11224_cast = slice_by_index(begin = var_11224_begin_0, end = var_11224_end_0, end_mask = var_11224_end_mask_0, x = q_45_cast)[name = tensor("op_11224_cast")]; + tensor var_11228_begin_0 = const()[name = tensor("op_11228_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_11228_end_0 = const()[name = tensor("op_11228_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_11228_end_mask_0 = const()[name = tensor("op_11228_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11228_cast = slice_by_index(begin = var_11228_begin_0, end = var_11228_end_0, end_mask = var_11228_end_mask_0, x = q_45_cast)[name = tensor("op_11228_cast")]; + tensor var_11232_begin_0 = const()[name = tensor("op_11232_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_11232_end_0 = const()[name = tensor("op_11232_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_11232_end_mask_0 = const()[name = tensor("op_11232_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11232_cast = slice_by_index(begin = var_11232_begin_0, end = var_11232_end_0, end_mask = var_11232_end_mask_0, x = q_45_cast)[name = tensor("op_11232_cast")]; + tensor var_11236_begin_0 = const()[name = tensor("op_11236_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_11236_end_0 = const()[name = tensor("op_11236_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_11236_end_mask_0 = const()[name = tensor("op_11236_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11236_cast = slice_by_index(begin = var_11236_begin_0, end = var_11236_end_0, end_mask = var_11236_end_mask_0, x = q_45_cast)[name = tensor("op_11236_cast")]; + tensor var_11240_begin_0 = const()[name = tensor("op_11240_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_11240_end_0 = const()[name = tensor("op_11240_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_11240_end_mask_0 = const()[name = tensor("op_11240_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11240_cast = slice_by_index(begin = var_11240_begin_0, end = var_11240_end_0, end_mask = var_11240_end_mask_0, x = q_45_cast)[name = tensor("op_11240_cast")]; + tensor var_11244_begin_0 = const()[name = tensor("op_11244_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_11244_end_0 = const()[name = tensor("op_11244_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_11244_end_mask_0 = const()[name = tensor("op_11244_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11244_cast = slice_by_index(begin = var_11244_begin_0, end = var_11244_end_0, end_mask = var_11244_end_mask_0, x = q_45_cast)[name = tensor("op_11244_cast")]; + tensor var_11248_begin_0 = const()[name = tensor("op_11248_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_11248_end_0 = const()[name = tensor("op_11248_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_11248_end_mask_0 = const()[name = tensor("op_11248_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11248_cast = slice_by_index(begin = var_11248_begin_0, end = var_11248_end_0, end_mask = var_11248_end_mask_0, x = q_45_cast)[name = tensor("op_11248_cast")]; + tensor var_11252_begin_0 = const()[name = tensor("op_11252_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_11252_end_0 = const()[name = tensor("op_11252_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_11252_end_mask_0 = const()[name = tensor("op_11252_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11252_cast = slice_by_index(begin = var_11252_begin_0, end = var_11252_end_0, end_mask = var_11252_end_mask_0, x = q_45_cast)[name = tensor("op_11252_cast")]; + tensor var_11256_begin_0 = const()[name = tensor("op_11256_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_11256_end_0 = const()[name = tensor("op_11256_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_11256_end_mask_0 = const()[name = tensor("op_11256_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11256_cast = slice_by_index(begin = var_11256_begin_0, end = var_11256_end_0, end_mask = var_11256_end_mask_0, x = q_45_cast)[name = tensor("op_11256_cast")]; + tensor var_11260_begin_0 = const()[name = tensor("op_11260_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_11260_end_0 = const()[name = tensor("op_11260_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_11260_end_mask_0 = const()[name = tensor("op_11260_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11260_cast = slice_by_index(begin = var_11260_begin_0, end = var_11260_end_0, end_mask = var_11260_end_mask_0, x = q_45_cast)[name = tensor("op_11260_cast")]; + tensor var_11263_begin_0 = const()[name = tensor("op_11263_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11263_end_0 = const()[name = tensor("op_11263_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_11263_end_mask_0 = const()[name = tensor("op_11263_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11263_cast = slice_by_index(begin = var_11263_begin_0, end = var_11263_end_0, end_mask = var_11263_end_mask_0, x = var_11224_cast)[name = tensor("op_11263_cast")]; + tensor var_11264_begin_0 = const()[name = tensor("op_11264_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_11264_end_0 = const()[name = tensor("op_11264_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_11264_end_mask_0 = const()[name = tensor("op_11264_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11264_cast = slice_by_index(begin = var_11264_begin_0, end = var_11264_end_0, end_mask = var_11264_end_mask_0, x = var_11224_cast)[name = tensor("op_11264_cast")]; + tensor var_11265_begin_0 = const()[name = tensor("op_11265_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11265_end_0 = const()[name = tensor("op_11265_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_11265_end_mask_0 = const()[name = tensor("op_11265_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11265_cast = slice_by_index(begin = var_11265_begin_0, end = var_11265_end_0, end_mask = var_11265_end_mask_0, x = var_11228_cast)[name = tensor("op_11265_cast")]; + tensor var_11266_begin_0 = const()[name = tensor("op_11266_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_11266_end_0 = const()[name = tensor("op_11266_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_11266_end_mask_0 = const()[name = tensor("op_11266_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11266_cast = slice_by_index(begin = var_11266_begin_0, end = var_11266_end_0, end_mask = var_11266_end_mask_0, x = var_11228_cast)[name = tensor("op_11266_cast")]; + tensor var_11267_begin_0 = const()[name = tensor("op_11267_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11267_end_0 = const()[name = tensor("op_11267_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_11267_end_mask_0 = const()[name = tensor("op_11267_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11267_cast = slice_by_index(begin = var_11267_begin_0, end = var_11267_end_0, end_mask = var_11267_end_mask_0, x = var_11232_cast)[name = tensor("op_11267_cast")]; + tensor var_11268_begin_0 = const()[name = tensor("op_11268_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_11268_end_0 = const()[name = tensor("op_11268_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_11268_end_mask_0 = const()[name = tensor("op_11268_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11268_cast = slice_by_index(begin = var_11268_begin_0, end = var_11268_end_0, end_mask = var_11268_end_mask_0, x = var_11232_cast)[name = tensor("op_11268_cast")]; + tensor var_11269_begin_0 = const()[name = tensor("op_11269_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11269_end_0 = const()[name = tensor("op_11269_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_11269_end_mask_0 = const()[name = tensor("op_11269_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11269_cast = slice_by_index(begin = var_11269_begin_0, end = var_11269_end_0, end_mask = var_11269_end_mask_0, x = var_11236_cast)[name = tensor("op_11269_cast")]; + tensor var_11270_begin_0 = const()[name = tensor("op_11270_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_11270_end_0 = const()[name = tensor("op_11270_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_11270_end_mask_0 = const()[name = tensor("op_11270_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11270_cast = slice_by_index(begin = var_11270_begin_0, end = var_11270_end_0, end_mask = var_11270_end_mask_0, x = var_11236_cast)[name = tensor("op_11270_cast")]; + tensor var_11271_begin_0 = const()[name = tensor("op_11271_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11271_end_0 = const()[name = tensor("op_11271_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_11271_end_mask_0 = const()[name = tensor("op_11271_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11271_cast = slice_by_index(begin = var_11271_begin_0, end = var_11271_end_0, end_mask = var_11271_end_mask_0, x = var_11240_cast)[name = tensor("op_11271_cast")]; + tensor var_11272_begin_0 = const()[name = tensor("op_11272_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_11272_end_0 = const()[name = tensor("op_11272_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_11272_end_mask_0 = const()[name = tensor("op_11272_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11272_cast = slice_by_index(begin = var_11272_begin_0, end = var_11272_end_0, end_mask = var_11272_end_mask_0, x = var_11240_cast)[name = tensor("op_11272_cast")]; + tensor var_11273_begin_0 = const()[name = tensor("op_11273_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11273_end_0 = const()[name = tensor("op_11273_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_11273_end_mask_0 = const()[name = tensor("op_11273_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11273_cast = slice_by_index(begin = var_11273_begin_0, end = var_11273_end_0, end_mask = var_11273_end_mask_0, x = var_11244_cast)[name = tensor("op_11273_cast")]; + tensor var_11274_begin_0 = const()[name = tensor("op_11274_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_11274_end_0 = const()[name = tensor("op_11274_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_11274_end_mask_0 = const()[name = tensor("op_11274_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11274_cast = slice_by_index(begin = var_11274_begin_0, end = var_11274_end_0, end_mask = var_11274_end_mask_0, x = var_11244_cast)[name = tensor("op_11274_cast")]; + tensor var_11275_begin_0 = const()[name = tensor("op_11275_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11275_end_0 = const()[name = tensor("op_11275_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_11275_end_mask_0 = const()[name = tensor("op_11275_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11275_cast = slice_by_index(begin = var_11275_begin_0, end = var_11275_end_0, end_mask = var_11275_end_mask_0, x = var_11248_cast)[name = tensor("op_11275_cast")]; + tensor var_11276_begin_0 = const()[name = tensor("op_11276_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_11276_end_0 = const()[name = tensor("op_11276_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_11276_end_mask_0 = const()[name = tensor("op_11276_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11276_cast = slice_by_index(begin = var_11276_begin_0, end = var_11276_end_0, end_mask = var_11276_end_mask_0, x = var_11248_cast)[name = tensor("op_11276_cast")]; + tensor var_11277_begin_0 = const()[name = tensor("op_11277_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11277_end_0 = const()[name = tensor("op_11277_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_11277_end_mask_0 = const()[name = tensor("op_11277_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11277_cast = slice_by_index(begin = var_11277_begin_0, end = var_11277_end_0, end_mask = var_11277_end_mask_0, x = var_11252_cast)[name = tensor("op_11277_cast")]; + tensor var_11278_begin_0 = const()[name = tensor("op_11278_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_11278_end_0 = const()[name = tensor("op_11278_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_11278_end_mask_0 = const()[name = tensor("op_11278_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11278_cast = slice_by_index(begin = var_11278_begin_0, end = var_11278_end_0, end_mask = var_11278_end_mask_0, x = var_11252_cast)[name = tensor("op_11278_cast")]; + tensor var_11279_begin_0 = const()[name = tensor("op_11279_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11279_end_0 = const()[name = tensor("op_11279_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_11279_end_mask_0 = const()[name = tensor("op_11279_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11279_cast = slice_by_index(begin = var_11279_begin_0, end = var_11279_end_0, end_mask = var_11279_end_mask_0, x = var_11256_cast)[name = tensor("op_11279_cast")]; + tensor var_11280_begin_0 = const()[name = tensor("op_11280_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_11280_end_0 = const()[name = tensor("op_11280_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_11280_end_mask_0 = const()[name = tensor("op_11280_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11280_cast = slice_by_index(begin = var_11280_begin_0, end = var_11280_end_0, end_mask = var_11280_end_mask_0, x = var_11256_cast)[name = tensor("op_11280_cast")]; + tensor var_11281_begin_0 = const()[name = tensor("op_11281_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11281_end_0 = const()[name = tensor("op_11281_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_11281_end_mask_0 = const()[name = tensor("op_11281_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11281_cast = slice_by_index(begin = var_11281_begin_0, end = var_11281_end_0, end_mask = var_11281_end_mask_0, x = var_11260_cast)[name = tensor("op_11281_cast")]; + tensor var_11282_begin_0 = const()[name = tensor("op_11282_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_11282_end_0 = const()[name = tensor("op_11282_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_11282_end_mask_0 = const()[name = tensor("op_11282_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11282_cast = slice_by_index(begin = var_11282_begin_0, end = var_11282_end_0, end_mask = var_11282_end_mask_0, x = var_11260_cast)[name = tensor("op_11282_cast")]; + tensor k_91_perm_0 = const()[name = tensor("k_91_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_11287_begin_0 = const()[name = tensor("op_11287_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11287_end_0 = const()[name = tensor("op_11287_end_0"), val = tensor([2, 1024, 1, 64])]; + tensor var_11287_end_mask_0 = const()[name = tensor("op_11287_end_mask_0"), val = tensor([true, true, true, false])]; + tensor transpose_9 = transpose(perm = k_91_perm_0, x = k_89_cast)[name = tensor("transpose_9")]; + tensor var_11287_cast = slice_by_index(begin = var_11287_begin_0, end = var_11287_end_0, end_mask = var_11287_end_mask_0, x = transpose_9)[name = tensor("op_11287_cast")]; + tensor var_11291_begin_0 = const()[name = tensor("op_11291_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_11291_end_0 = const()[name = tensor("op_11291_end_0"), val = tensor([2, 1024, 1, 128])]; + tensor var_11291_end_mask_0 = const()[name = tensor("op_11291_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11291_cast = slice_by_index(begin = var_11291_begin_0, end = var_11291_end_0, end_mask = var_11291_end_mask_0, x = transpose_9)[name = tensor("op_11291_cast")]; + tensor var_11295_begin_0 = const()[name = tensor("op_11295_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_11295_end_0 = const()[name = tensor("op_11295_end_0"), val = tensor([2, 1024, 1, 192])]; + tensor var_11295_end_mask_0 = const()[name = tensor("op_11295_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11295_cast = slice_by_index(begin = var_11295_begin_0, end = var_11295_end_0, end_mask = var_11295_end_mask_0, x = transpose_9)[name = tensor("op_11295_cast")]; + tensor var_11299_begin_0 = const()[name = tensor("op_11299_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_11299_end_0 = const()[name = tensor("op_11299_end_0"), val = tensor([2, 1024, 1, 256])]; + tensor var_11299_end_mask_0 = const()[name = tensor("op_11299_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11299_cast = slice_by_index(begin = var_11299_begin_0, end = var_11299_end_0, end_mask = var_11299_end_mask_0, x = transpose_9)[name = tensor("op_11299_cast")]; + tensor var_11303_begin_0 = const()[name = tensor("op_11303_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_11303_end_0 = const()[name = tensor("op_11303_end_0"), val = tensor([2, 1024, 1, 320])]; + tensor var_11303_end_mask_0 = const()[name = tensor("op_11303_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11303_cast = slice_by_index(begin = var_11303_begin_0, end = var_11303_end_0, end_mask = var_11303_end_mask_0, x = transpose_9)[name = tensor("op_11303_cast")]; + tensor var_11307_begin_0 = const()[name = tensor("op_11307_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_11307_end_0 = const()[name = tensor("op_11307_end_0"), val = tensor([2, 1024, 1, 384])]; + tensor var_11307_end_mask_0 = const()[name = tensor("op_11307_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11307_cast = slice_by_index(begin = var_11307_begin_0, end = var_11307_end_0, end_mask = var_11307_end_mask_0, x = transpose_9)[name = tensor("op_11307_cast")]; + tensor var_11311_begin_0 = const()[name = tensor("op_11311_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_11311_end_0 = const()[name = tensor("op_11311_end_0"), val = tensor([2, 1024, 1, 448])]; + tensor var_11311_end_mask_0 = const()[name = tensor("op_11311_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11311_cast = slice_by_index(begin = var_11311_begin_0, end = var_11311_end_0, end_mask = var_11311_end_mask_0, x = transpose_9)[name = tensor("op_11311_cast")]; + tensor var_11315_begin_0 = const()[name = tensor("op_11315_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_11315_end_0 = const()[name = tensor("op_11315_end_0"), val = tensor([2, 1024, 1, 512])]; + tensor var_11315_end_mask_0 = const()[name = tensor("op_11315_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11315_cast = slice_by_index(begin = var_11315_begin_0, end = var_11315_end_0, end_mask = var_11315_end_mask_0, x = transpose_9)[name = tensor("op_11315_cast")]; + tensor var_11319_begin_0 = const()[name = tensor("op_11319_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_11319_end_0 = const()[name = tensor("op_11319_end_0"), val = tensor([2, 1024, 1, 576])]; + tensor var_11319_end_mask_0 = const()[name = tensor("op_11319_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11319_cast = slice_by_index(begin = var_11319_begin_0, end = var_11319_end_0, end_mask = var_11319_end_mask_0, x = transpose_9)[name = tensor("op_11319_cast")]; + tensor var_11323_begin_0 = const()[name = tensor("op_11323_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_11323_end_0 = const()[name = tensor("op_11323_end_0"), val = tensor([2, 1024, 1, 640])]; + tensor var_11323_end_mask_0 = const()[name = tensor("op_11323_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11323_cast = slice_by_index(begin = var_11323_begin_0, end = var_11323_end_0, end_mask = var_11323_end_mask_0, x = transpose_9)[name = tensor("op_11323_cast")]; + tensor var_11325_begin_0 = const()[name = tensor("op_11325_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11325_end_0 = const()[name = tensor("op_11325_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_11325_end_mask_0 = const()[name = tensor("op_11325_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11325_cast = slice_by_index(begin = var_11325_begin_0, end = var_11325_end_0, end_mask = var_11325_end_mask_0, x = v_45_cast)[name = tensor("op_11325_cast")]; + tensor var_11329_begin_0 = const()[name = tensor("op_11329_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_11329_end_0 = const()[name = tensor("op_11329_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_11329_end_mask_0 = const()[name = tensor("op_11329_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11329_cast = slice_by_index(begin = var_11329_begin_0, end = var_11329_end_0, end_mask = var_11329_end_mask_0, x = v_45_cast)[name = tensor("op_11329_cast")]; + tensor var_11333_begin_0 = const()[name = tensor("op_11333_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_11333_end_0 = const()[name = tensor("op_11333_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_11333_end_mask_0 = const()[name = tensor("op_11333_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11333_cast = slice_by_index(begin = var_11333_begin_0, end = var_11333_end_0, end_mask = var_11333_end_mask_0, x = v_45_cast)[name = tensor("op_11333_cast")]; + tensor var_11337_begin_0 = const()[name = tensor("op_11337_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_11337_end_0 = const()[name = tensor("op_11337_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_11337_end_mask_0 = const()[name = tensor("op_11337_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11337_cast = slice_by_index(begin = var_11337_begin_0, end = var_11337_end_0, end_mask = var_11337_end_mask_0, x = v_45_cast)[name = tensor("op_11337_cast")]; + tensor var_11341_begin_0 = const()[name = tensor("op_11341_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_11341_end_0 = const()[name = tensor("op_11341_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_11341_end_mask_0 = const()[name = tensor("op_11341_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11341_cast = slice_by_index(begin = var_11341_begin_0, end = var_11341_end_0, end_mask = var_11341_end_mask_0, x = v_45_cast)[name = tensor("op_11341_cast")]; + tensor var_11345_begin_0 = const()[name = tensor("op_11345_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_11345_end_0 = const()[name = tensor("op_11345_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_11345_end_mask_0 = const()[name = tensor("op_11345_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11345_cast = slice_by_index(begin = var_11345_begin_0, end = var_11345_end_0, end_mask = var_11345_end_mask_0, x = v_45_cast)[name = tensor("op_11345_cast")]; + tensor var_11349_begin_0 = const()[name = tensor("op_11349_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_11349_end_0 = const()[name = tensor("op_11349_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_11349_end_mask_0 = const()[name = tensor("op_11349_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11349_cast = slice_by_index(begin = var_11349_begin_0, end = var_11349_end_0, end_mask = var_11349_end_mask_0, x = v_45_cast)[name = tensor("op_11349_cast")]; + tensor var_11353_begin_0 = const()[name = tensor("op_11353_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_11353_end_0 = const()[name = tensor("op_11353_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_11353_end_mask_0 = const()[name = tensor("op_11353_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11353_cast = slice_by_index(begin = var_11353_begin_0, end = var_11353_end_0, end_mask = var_11353_end_mask_0, x = v_45_cast)[name = tensor("op_11353_cast")]; + tensor var_11357_begin_0 = const()[name = tensor("op_11357_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_11357_end_0 = const()[name = tensor("op_11357_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_11357_end_mask_0 = const()[name = tensor("op_11357_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11357_cast = slice_by_index(begin = var_11357_begin_0, end = var_11357_end_0, end_mask = var_11357_end_mask_0, x = v_45_cast)[name = tensor("op_11357_cast")]; + tensor var_11361_begin_0 = const()[name = tensor("op_11361_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_11361_end_0 = const()[name = tensor("op_11361_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_11361_end_mask_0 = const()[name = tensor("op_11361_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11361_cast = slice_by_index(begin = var_11361_begin_0, end = var_11361_end_0, end_mask = var_11361_end_mask_0, x = v_45_cast)[name = tensor("op_11361_cast")]; + tensor var_11365_equation_0 = const()[name = tensor("op_11365_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11365_cast = einsum(equation = var_11365_equation_0, values = (var_11287_cast, var_11263_cast))[name = tensor("op_11365_cast")]; + tensor var_11366_to_fp16 = const()[name = tensor("op_11366_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_561_cast = mul(x = var_11365_cast, y = var_11366_to_fp16)[name = tensor("aw_chunk_561_cast")]; + tensor var_11369_equation_0 = const()[name = tensor("op_11369_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11369_cast = einsum(equation = var_11369_equation_0, values = (var_11287_cast, var_11264_cast))[name = tensor("op_11369_cast")]; + tensor var_11370_to_fp16 = const()[name = tensor("op_11370_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_563_cast = mul(x = var_11369_cast, y = var_11370_to_fp16)[name = tensor("aw_chunk_563_cast")]; + tensor var_11373_equation_0 = const()[name = tensor("op_11373_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11373_cast = einsum(equation = var_11373_equation_0, values = (var_11291_cast, var_11265_cast))[name = tensor("op_11373_cast")]; + tensor var_11374_to_fp16 = const()[name = tensor("op_11374_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_565_cast = mul(x = var_11373_cast, y = var_11374_to_fp16)[name = tensor("aw_chunk_565_cast")]; + tensor var_11377_equation_0 = const()[name = tensor("op_11377_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11377_cast = einsum(equation = var_11377_equation_0, values = (var_11291_cast, var_11266_cast))[name = tensor("op_11377_cast")]; + tensor var_11378_to_fp16 = const()[name = tensor("op_11378_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_567_cast = mul(x = var_11377_cast, y = var_11378_to_fp16)[name = tensor("aw_chunk_567_cast")]; + tensor var_11381_equation_0 = const()[name = tensor("op_11381_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11381_cast = einsum(equation = var_11381_equation_0, values = (var_11295_cast, var_11267_cast))[name = tensor("op_11381_cast")]; + tensor var_11382_to_fp16 = const()[name = tensor("op_11382_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_569_cast = mul(x = var_11381_cast, y = var_11382_to_fp16)[name = tensor("aw_chunk_569_cast")]; + tensor var_11385_equation_0 = const()[name = tensor("op_11385_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11385_cast = einsum(equation = var_11385_equation_0, values = (var_11295_cast, var_11268_cast))[name = tensor("op_11385_cast")]; + tensor var_11386_to_fp16 = const()[name = tensor("op_11386_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_571_cast = mul(x = var_11385_cast, y = var_11386_to_fp16)[name = tensor("aw_chunk_571_cast")]; + tensor var_11389_equation_0 = const()[name = tensor("op_11389_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11389_cast = einsum(equation = var_11389_equation_0, values = (var_11299_cast, var_11269_cast))[name = tensor("op_11389_cast")]; + tensor var_11390_to_fp16 = const()[name = tensor("op_11390_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_573_cast = mul(x = var_11389_cast, y = var_11390_to_fp16)[name = tensor("aw_chunk_573_cast")]; + tensor var_11393_equation_0 = const()[name = tensor("op_11393_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11393_cast = einsum(equation = var_11393_equation_0, values = (var_11299_cast, var_11270_cast))[name = tensor("op_11393_cast")]; + tensor var_11394_to_fp16 = const()[name = tensor("op_11394_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_575_cast = mul(x = var_11393_cast, y = var_11394_to_fp16)[name = tensor("aw_chunk_575_cast")]; + tensor var_11397_equation_0 = const()[name = tensor("op_11397_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11397_cast = einsum(equation = var_11397_equation_0, values = (var_11303_cast, var_11271_cast))[name = tensor("op_11397_cast")]; + tensor var_11398_to_fp16 = const()[name = tensor("op_11398_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_577_cast = mul(x = var_11397_cast, y = var_11398_to_fp16)[name = tensor("aw_chunk_577_cast")]; + tensor var_11401_equation_0 = const()[name = tensor("op_11401_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11401_cast = einsum(equation = var_11401_equation_0, values = (var_11303_cast, var_11272_cast))[name = tensor("op_11401_cast")]; + tensor var_11402_to_fp16 = const()[name = tensor("op_11402_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_579_cast = mul(x = var_11401_cast, y = var_11402_to_fp16)[name = tensor("aw_chunk_579_cast")]; + tensor var_11405_equation_0 = const()[name = tensor("op_11405_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11405_cast = einsum(equation = var_11405_equation_0, values = (var_11307_cast, var_11273_cast))[name = tensor("op_11405_cast")]; + tensor var_11406_to_fp16 = const()[name = tensor("op_11406_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_581_cast = mul(x = var_11405_cast, y = var_11406_to_fp16)[name = tensor("aw_chunk_581_cast")]; + tensor var_11409_equation_0 = const()[name = tensor("op_11409_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11409_cast = einsum(equation = var_11409_equation_0, values = (var_11307_cast, var_11274_cast))[name = tensor("op_11409_cast")]; + tensor var_11410_to_fp16 = const()[name = tensor("op_11410_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_583_cast = mul(x = var_11409_cast, y = var_11410_to_fp16)[name = tensor("aw_chunk_583_cast")]; + tensor var_11413_equation_0 = const()[name = tensor("op_11413_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11413_cast = einsum(equation = var_11413_equation_0, values = (var_11311_cast, var_11275_cast))[name = tensor("op_11413_cast")]; + tensor var_11414_to_fp16 = const()[name = tensor("op_11414_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_585_cast = mul(x = var_11413_cast, y = var_11414_to_fp16)[name = tensor("aw_chunk_585_cast")]; + tensor var_11417_equation_0 = const()[name = tensor("op_11417_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11417_cast = einsum(equation = var_11417_equation_0, values = (var_11311_cast, var_11276_cast))[name = tensor("op_11417_cast")]; + tensor var_11418_to_fp16 = const()[name = tensor("op_11418_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_587_cast = mul(x = var_11417_cast, y = var_11418_to_fp16)[name = tensor("aw_chunk_587_cast")]; + tensor var_11421_equation_0 = const()[name = tensor("op_11421_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11421_cast = einsum(equation = var_11421_equation_0, values = (var_11315_cast, var_11277_cast))[name = tensor("op_11421_cast")]; + tensor var_11422_to_fp16 = const()[name = tensor("op_11422_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_589_cast = mul(x = var_11421_cast, y = var_11422_to_fp16)[name = tensor("aw_chunk_589_cast")]; + tensor var_11425_equation_0 = const()[name = tensor("op_11425_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11425_cast = einsum(equation = var_11425_equation_0, values = (var_11315_cast, var_11278_cast))[name = tensor("op_11425_cast")]; + tensor var_11426_to_fp16 = const()[name = tensor("op_11426_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_591_cast = mul(x = var_11425_cast, y = var_11426_to_fp16)[name = tensor("aw_chunk_591_cast")]; + tensor var_11429_equation_0 = const()[name = tensor("op_11429_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11429_cast = einsum(equation = var_11429_equation_0, values = (var_11319_cast, var_11279_cast))[name = tensor("op_11429_cast")]; + tensor var_11430_to_fp16 = const()[name = tensor("op_11430_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_593_cast = mul(x = var_11429_cast, y = var_11430_to_fp16)[name = tensor("aw_chunk_593_cast")]; + tensor var_11433_equation_0 = const()[name = tensor("op_11433_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11433_cast = einsum(equation = var_11433_equation_0, values = (var_11319_cast, var_11280_cast))[name = tensor("op_11433_cast")]; + tensor var_11434_to_fp16 = const()[name = tensor("op_11434_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_595_cast = mul(x = var_11433_cast, y = var_11434_to_fp16)[name = tensor("aw_chunk_595_cast")]; + tensor var_11437_equation_0 = const()[name = tensor("op_11437_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11437_cast = einsum(equation = var_11437_equation_0, values = (var_11323_cast, var_11281_cast))[name = tensor("op_11437_cast")]; + tensor var_11438_to_fp16 = const()[name = tensor("op_11438_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_597_cast = mul(x = var_11437_cast, y = var_11438_to_fp16)[name = tensor("aw_chunk_597_cast")]; + tensor var_11441_equation_0 = const()[name = tensor("op_11441_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11441_cast = einsum(equation = var_11441_equation_0, values = (var_11323_cast, var_11282_cast))[name = tensor("op_11441_cast")]; + tensor var_11442_to_fp16 = const()[name = tensor("op_11442_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_599_cast = mul(x = var_11441_cast, y = var_11442_to_fp16)[name = tensor("aw_chunk_599_cast")]; + tensor var_11444_cast = softmax(axis = var_10257, x = aw_chunk_561_cast)[name = tensor("op_11444_cast")]; + tensor var_11445_cast = softmax(axis = var_10257, x = aw_chunk_563_cast)[name = tensor("op_11445_cast")]; + tensor var_11446_cast = softmax(axis = var_10257, x = aw_chunk_565_cast)[name = tensor("op_11446_cast")]; + tensor var_11447_cast = softmax(axis = var_10257, x = aw_chunk_567_cast)[name = tensor("op_11447_cast")]; + tensor var_11448_cast = softmax(axis = var_10257, x = aw_chunk_569_cast)[name = tensor("op_11448_cast")]; + tensor var_11449_cast = softmax(axis = var_10257, x = aw_chunk_571_cast)[name = tensor("op_11449_cast")]; + tensor var_11450_cast = softmax(axis = var_10257, x = aw_chunk_573_cast)[name = tensor("op_11450_cast")]; + tensor var_11451_cast = softmax(axis = var_10257, x = aw_chunk_575_cast)[name = tensor("op_11451_cast")]; + tensor var_11452_cast = softmax(axis = var_10257, x = aw_chunk_577_cast)[name = tensor("op_11452_cast")]; + tensor var_11453_cast = softmax(axis = var_10257, x = aw_chunk_579_cast)[name = tensor("op_11453_cast")]; + tensor var_11454_cast = softmax(axis = var_10257, x = aw_chunk_581_cast)[name = tensor("op_11454_cast")]; + tensor var_11455_cast = softmax(axis = var_10257, x = aw_chunk_583_cast)[name = tensor("op_11455_cast")]; + tensor var_11456_cast = softmax(axis = var_10257, x = aw_chunk_585_cast)[name = tensor("op_11456_cast")]; + tensor var_11457_cast = softmax(axis = var_10257, x = aw_chunk_587_cast)[name = tensor("op_11457_cast")]; + tensor var_11458_cast = softmax(axis = var_10257, x = aw_chunk_589_cast)[name = tensor("op_11458_cast")]; + tensor var_11459_cast = softmax(axis = var_10257, x = aw_chunk_591_cast)[name = tensor("op_11459_cast")]; + tensor var_11460_cast = softmax(axis = var_10257, x = aw_chunk_593_cast)[name = tensor("op_11460_cast")]; + tensor var_11461_cast = softmax(axis = var_10257, x = aw_chunk_595_cast)[name = tensor("op_11461_cast")]; + tensor var_11462_cast = softmax(axis = var_10257, x = aw_chunk_597_cast)[name = tensor("op_11462_cast")]; + tensor var_11463_cast = softmax(axis = var_10257, x = aw_chunk_599_cast)[name = tensor("op_11463_cast")]; + tensor var_11465_equation_0 = const()[name = tensor("op_11465_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11465_cast = einsum(equation = var_11465_equation_0, values = (var_11325_cast, var_11444_cast))[name = tensor("op_11465_cast")]; + tensor var_11467_equation_0 = const()[name = tensor("op_11467_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11467_cast = einsum(equation = var_11467_equation_0, values = (var_11325_cast, var_11445_cast))[name = tensor("op_11467_cast")]; + tensor var_11469_equation_0 = const()[name = tensor("op_11469_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11469_cast = einsum(equation = var_11469_equation_0, values = (var_11329_cast, var_11446_cast))[name = tensor("op_11469_cast")]; + tensor var_11471_equation_0 = const()[name = tensor("op_11471_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11471_cast = einsum(equation = var_11471_equation_0, values = (var_11329_cast, var_11447_cast))[name = tensor("op_11471_cast")]; + tensor var_11473_equation_0 = const()[name = tensor("op_11473_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11473_cast = einsum(equation = var_11473_equation_0, values = (var_11333_cast, var_11448_cast))[name = tensor("op_11473_cast")]; + tensor var_11475_equation_0 = const()[name = tensor("op_11475_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11475_cast = einsum(equation = var_11475_equation_0, values = (var_11333_cast, var_11449_cast))[name = tensor("op_11475_cast")]; + tensor var_11477_equation_0 = const()[name = tensor("op_11477_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11477_cast = einsum(equation = var_11477_equation_0, values = (var_11337_cast, var_11450_cast))[name = tensor("op_11477_cast")]; + tensor var_11479_equation_0 = const()[name = tensor("op_11479_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11479_cast = einsum(equation = var_11479_equation_0, values = (var_11337_cast, var_11451_cast))[name = tensor("op_11479_cast")]; + tensor var_11481_equation_0 = const()[name = tensor("op_11481_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11481_cast = einsum(equation = var_11481_equation_0, values = (var_11341_cast, var_11452_cast))[name = tensor("op_11481_cast")]; + tensor var_11483_equation_0 = const()[name = tensor("op_11483_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11483_cast = einsum(equation = var_11483_equation_0, values = (var_11341_cast, var_11453_cast))[name = tensor("op_11483_cast")]; + tensor var_11485_equation_0 = const()[name = tensor("op_11485_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11485_cast = einsum(equation = var_11485_equation_0, values = (var_11345_cast, var_11454_cast))[name = tensor("op_11485_cast")]; + tensor var_11487_equation_0 = const()[name = tensor("op_11487_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11487_cast = einsum(equation = var_11487_equation_0, values = (var_11345_cast, var_11455_cast))[name = tensor("op_11487_cast")]; + tensor var_11489_equation_0 = const()[name = tensor("op_11489_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11489_cast = einsum(equation = var_11489_equation_0, values = (var_11349_cast, var_11456_cast))[name = tensor("op_11489_cast")]; + tensor var_11491_equation_0 = const()[name = tensor("op_11491_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11491_cast = einsum(equation = var_11491_equation_0, values = (var_11349_cast, var_11457_cast))[name = tensor("op_11491_cast")]; + tensor var_11493_equation_0 = const()[name = tensor("op_11493_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11493_cast = einsum(equation = var_11493_equation_0, values = (var_11353_cast, var_11458_cast))[name = tensor("op_11493_cast")]; + tensor var_11495_equation_0 = const()[name = tensor("op_11495_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11495_cast = einsum(equation = var_11495_equation_0, values = (var_11353_cast, var_11459_cast))[name = tensor("op_11495_cast")]; + tensor var_11497_equation_0 = const()[name = tensor("op_11497_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11497_cast = einsum(equation = var_11497_equation_0, values = (var_11357_cast, var_11460_cast))[name = tensor("op_11497_cast")]; + tensor var_11499_equation_0 = const()[name = tensor("op_11499_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11499_cast = einsum(equation = var_11499_equation_0, values = (var_11357_cast, var_11461_cast))[name = tensor("op_11499_cast")]; + tensor var_11501_equation_0 = const()[name = tensor("op_11501_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11501_cast = einsum(equation = var_11501_equation_0, values = (var_11361_cast, var_11462_cast))[name = tensor("op_11501_cast")]; + tensor var_11503_equation_0 = const()[name = tensor("op_11503_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11503_cast = einsum(equation = var_11503_equation_0, values = (var_11361_cast, var_11463_cast))[name = tensor("op_11503_cast")]; + tensor var_11505_interleave_0 = const()[name = tensor("op_11505_interleave_0"), val = tensor(false)]; + tensor var_11505_cast = concat(axis = var_10234, interleave = var_11505_interleave_0, values = (var_11465_cast, var_11467_cast))[name = tensor("op_11505_cast")]; + tensor var_11507_interleave_0 = const()[name = tensor("op_11507_interleave_0"), val = tensor(false)]; + tensor var_11507_cast = concat(axis = var_10234, interleave = var_11507_interleave_0, values = (var_11469_cast, var_11471_cast))[name = tensor("op_11507_cast")]; + tensor var_11509_interleave_0 = const()[name = tensor("op_11509_interleave_0"), val = tensor(false)]; + tensor var_11509_cast = concat(axis = var_10234, interleave = var_11509_interleave_0, values = (var_11473_cast, var_11475_cast))[name = tensor("op_11509_cast")]; + tensor var_11511_interleave_0 = const()[name = tensor("op_11511_interleave_0"), val = tensor(false)]; + tensor var_11511_cast = concat(axis = var_10234, interleave = var_11511_interleave_0, values = (var_11477_cast, var_11479_cast))[name = tensor("op_11511_cast")]; + tensor var_11513_interleave_0 = const()[name = tensor("op_11513_interleave_0"), val = tensor(false)]; + tensor var_11513_cast = concat(axis = var_10234, interleave = var_11513_interleave_0, values = (var_11481_cast, var_11483_cast))[name = tensor("op_11513_cast")]; + tensor var_11515_interleave_0 = const()[name = tensor("op_11515_interleave_0"), val = tensor(false)]; + tensor var_11515_cast = concat(axis = var_10234, interleave = var_11515_interleave_0, values = (var_11485_cast, var_11487_cast))[name = tensor("op_11515_cast")]; + tensor var_11517_interleave_0 = const()[name = tensor("op_11517_interleave_0"), val = tensor(false)]; + tensor var_11517_cast = concat(axis = var_10234, interleave = var_11517_interleave_0, values = (var_11489_cast, var_11491_cast))[name = tensor("op_11517_cast")]; + tensor var_11519_interleave_0 = const()[name = tensor("op_11519_interleave_0"), val = tensor(false)]; + tensor var_11519_cast = concat(axis = var_10234, interleave = var_11519_interleave_0, values = (var_11493_cast, var_11495_cast))[name = tensor("op_11519_cast")]; + tensor var_11521_interleave_0 = const()[name = tensor("op_11521_interleave_0"), val = tensor(false)]; + tensor var_11521_cast = concat(axis = var_10234, interleave = var_11521_interleave_0, values = (var_11497_cast, var_11499_cast))[name = tensor("op_11521_cast")]; + tensor var_11523_interleave_0 = const()[name = tensor("op_11523_interleave_0"), val = tensor(false)]; + tensor var_11523_cast = concat(axis = var_10234, interleave = var_11523_interleave_0, values = (var_11501_cast, var_11503_cast))[name = tensor("op_11523_cast")]; + tensor input_409_interleave_0 = const()[name = tensor("input_409_interleave_0"), val = tensor(false)]; + tensor input_409_cast = concat(axis = var_10257, interleave = input_409_interleave_0, values = (var_11505_cast, var_11507_cast, var_11509_cast, var_11511_cast, var_11513_cast, var_11515_cast, var_11517_cast, var_11519_cast, var_11521_cast, var_11523_cast))[name = tensor("input_409_cast")]; + tensor var_11529 = const()[name = tensor("op_11529"), val = tensor([1, 1])]; + tensor var_11531 = const()[name = tensor("op_11531"), val = tensor([1, 1])]; + tensor var_11533_pad_type_0 = const()[name = tensor("op_11533_pad_type_0"), val = tensor("custom")]; + tensor var_11533_pad_0 = const()[name = tensor("op_11533_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_2_attentions_1_transformer_blocks_0_attn1_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(611590144))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(611897408))), name = tensor("up_blocks_2_attentions_1_transformer_blocks_0_attn1_to_out_0_weight_to_fp16_palettized"), shape = tensor([640, 640, 1, 1])]; + tensor up_blocks_2_attentions_1_transformer_blocks_0_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("up_blocks_2_attentions_1_transformer_blocks_0_attn1_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(611897600)))]; + tensor var_11533_cast = conv(bias = up_blocks_2_attentions_1_transformer_blocks_0_attn1_to_out_0_bias_to_fp16, dilations = var_11531, groups = var_10257, pad = var_11533_pad_0, pad_type = var_11533_pad_type_0, strides = var_11529, weight = up_blocks_2_attentions_1_transformer_blocks_0_attn1_to_out_0_weight_to_fp16_palettized, x = input_409_cast)[name = tensor("op_11533_cast")]; + tensor inputs_69_cast = add(x = var_11533_cast, y = inputs_67_cast)[name = tensor("inputs_69_cast")]; + tensor var_11537 = const()[name = tensor("op_11537"), val = tensor([1])]; + tensor channels_mean_69_cast = reduce_mean(axes = var_11537, keep_dims = var_10252, x = inputs_69_cast)[name = tensor("channels_mean_69_cast")]; + tensor zero_mean_69_cast = sub(x = inputs_69_cast, y = channels_mean_69_cast)[name = tensor("zero_mean_69_cast")]; + tensor zero_mean_sq_69_cast = mul(x = zero_mean_69_cast, y = zero_mean_69_cast)[name = tensor("zero_mean_sq_69_cast")]; + tensor var_11541 = const()[name = tensor("op_11541"), val = tensor([1])]; + tensor var_11542_cast = reduce_mean(axes = var_11541, keep_dims = var_10252, x = zero_mean_sq_69_cast)[name = tensor("op_11542_cast")]; + tensor var_11543_to_fp16 = const()[name = tensor("op_11543_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_11544_cast = add(x = var_11542_cast, y = var_11543_to_fp16)[name = tensor("op_11544_cast")]; + tensor denom_69_epsilon_0_to_fp16 = const()[name = tensor("denom_69_epsilon_0_to_fp16"), val = tensor(0x1p-24)]; + tensor denom_69_cast = rsqrt(epsilon = denom_69_epsilon_0_to_fp16, x = var_11544_cast)[name = tensor("denom_69_cast")]; + tensor out_69_cast = mul(x = zero_mean_69_cast, y = denom_69_cast)[name = tensor("out_69_cast")]; + tensor var_11548_to_fp16 = const()[name = tensor("op_11548_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(611898944)))]; + tensor var_11549_cast = add(x = out_69_cast, y = var_11548_to_fp16)[name = tensor("op_11549_cast")]; + tensor var_11551_to_fp16 = const()[name = tensor("op_11551_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(611900288)))]; + tensor hidden_states_249_cast = mul(x = var_11549_cast, y = var_11551_to_fp16)[name = tensor("hidden_states_249_cast")]; + tensor var_11558 = const()[name = tensor("op_11558"), val = tensor([1, 1])]; + tensor var_11560 = const()[name = tensor("op_11560"), val = tensor([1, 1])]; + tensor q_47_pad_type_0 = const()[name = tensor("q_47_pad_type_0"), val = tensor("custom")]; + tensor q_47_pad_0 = const()[name = tensor("q_47_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_2_attentions_1_transformer_blocks_0_attn2_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(611901632))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(612208896))), name = tensor("up_blocks_2_attentions_1_transformer_blocks_0_attn2_to_q_weight_to_fp16_palettized"), shape = tensor([640, 640, 1, 1])]; + tensor q_47_cast = conv(dilations = var_11560, groups = var_10257, pad = q_47_pad_0, pad_type = q_47_pad_type_0, strides = var_11558, weight = up_blocks_2_attentions_1_transformer_blocks_0_attn2_to_q_weight_to_fp16_palettized, x = hidden_states_249_cast)[name = tensor("q_47_cast")]; + tensor var_11564 = const()[name = tensor("op_11564"), val = tensor([1, 1])]; + tensor var_11566 = const()[name = tensor("op_11566"), val = tensor([1, 1])]; + tensor k_93_pad_type_0 = const()[name = tensor("k_93_pad_type_0"), val = tensor("custom")]; + tensor k_93_pad_0 = const()[name = tensor("k_93_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_2_attentions_1_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(612209088))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(612700672))), name = tensor("up_blocks_2_attentions_1_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized"), shape = tensor([640, 1024, 1, 1])]; + tensor k_93_cast = conv(dilations = var_11566, groups = var_10257, pad = k_93_pad_0, pad_type = k_93_pad_type_0, strides = var_11564, weight = up_blocks_2_attentions_1_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor("k_93_cast")]; + tensor var_11570 = const()[name = tensor("op_11570"), val = tensor([1, 1])]; + tensor var_11572 = const()[name = tensor("op_11572"), val = tensor([1, 1])]; + tensor v_47_pad_type_0 = const()[name = tensor("v_47_pad_type_0"), val = tensor("custom")]; + tensor v_47_pad_0 = const()[name = tensor("v_47_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_2_attentions_1_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(612700864))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(613192448))), name = tensor("up_blocks_2_attentions_1_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized"), shape = tensor([640, 1024, 1, 1])]; + tensor v_47_cast = conv(dilations = var_11572, groups = var_10257, pad = v_47_pad_0, pad_type = v_47_pad_type_0, strides = var_11570, weight = up_blocks_2_attentions_1_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor("v_47_cast")]; + tensor var_11576_begin_0 = const()[name = tensor("op_11576_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11576_end_0 = const()[name = tensor("op_11576_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_11576_end_mask_0 = const()[name = tensor("op_11576_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11576_cast = slice_by_index(begin = var_11576_begin_0, end = var_11576_end_0, end_mask = var_11576_end_mask_0, x = q_47_cast)[name = tensor("op_11576_cast")]; + tensor var_11580_begin_0 = const()[name = tensor("op_11580_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_11580_end_0 = const()[name = tensor("op_11580_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_11580_end_mask_0 = const()[name = tensor("op_11580_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11580_cast = slice_by_index(begin = var_11580_begin_0, end = var_11580_end_0, end_mask = var_11580_end_mask_0, x = q_47_cast)[name = tensor("op_11580_cast")]; + tensor var_11584_begin_0 = const()[name = tensor("op_11584_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_11584_end_0 = const()[name = tensor("op_11584_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_11584_end_mask_0 = const()[name = tensor("op_11584_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11584_cast = slice_by_index(begin = var_11584_begin_0, end = var_11584_end_0, end_mask = var_11584_end_mask_0, x = q_47_cast)[name = tensor("op_11584_cast")]; + tensor var_11588_begin_0 = const()[name = tensor("op_11588_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_11588_end_0 = const()[name = tensor("op_11588_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_11588_end_mask_0 = const()[name = tensor("op_11588_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11588_cast = slice_by_index(begin = var_11588_begin_0, end = var_11588_end_0, end_mask = var_11588_end_mask_0, x = q_47_cast)[name = tensor("op_11588_cast")]; + tensor var_11592_begin_0 = const()[name = tensor("op_11592_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_11592_end_0 = const()[name = tensor("op_11592_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_11592_end_mask_0 = const()[name = tensor("op_11592_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11592_cast = slice_by_index(begin = var_11592_begin_0, end = var_11592_end_0, end_mask = var_11592_end_mask_0, x = q_47_cast)[name = tensor("op_11592_cast")]; + tensor var_11596_begin_0 = const()[name = tensor("op_11596_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_11596_end_0 = const()[name = tensor("op_11596_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_11596_end_mask_0 = const()[name = tensor("op_11596_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11596_cast = slice_by_index(begin = var_11596_begin_0, end = var_11596_end_0, end_mask = var_11596_end_mask_0, x = q_47_cast)[name = tensor("op_11596_cast")]; + tensor var_11600_begin_0 = const()[name = tensor("op_11600_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_11600_end_0 = const()[name = tensor("op_11600_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_11600_end_mask_0 = const()[name = tensor("op_11600_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11600_cast = slice_by_index(begin = var_11600_begin_0, end = var_11600_end_0, end_mask = var_11600_end_mask_0, x = q_47_cast)[name = tensor("op_11600_cast")]; + tensor var_11604_begin_0 = const()[name = tensor("op_11604_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_11604_end_0 = const()[name = tensor("op_11604_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_11604_end_mask_0 = const()[name = tensor("op_11604_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11604_cast = slice_by_index(begin = var_11604_begin_0, end = var_11604_end_0, end_mask = var_11604_end_mask_0, x = q_47_cast)[name = tensor("op_11604_cast")]; + tensor var_11608_begin_0 = const()[name = tensor("op_11608_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_11608_end_0 = const()[name = tensor("op_11608_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_11608_end_mask_0 = const()[name = tensor("op_11608_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11608_cast = slice_by_index(begin = var_11608_begin_0, end = var_11608_end_0, end_mask = var_11608_end_mask_0, x = q_47_cast)[name = tensor("op_11608_cast")]; + tensor var_11612_begin_0 = const()[name = tensor("op_11612_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_11612_end_0 = const()[name = tensor("op_11612_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_11612_end_mask_0 = const()[name = tensor("op_11612_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11612_cast = slice_by_index(begin = var_11612_begin_0, end = var_11612_end_0, end_mask = var_11612_end_mask_0, x = q_47_cast)[name = tensor("op_11612_cast")]; + tensor var_11615_begin_0 = const()[name = tensor("op_11615_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11615_end_0 = const()[name = tensor("op_11615_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_11615_end_mask_0 = const()[name = tensor("op_11615_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11615_cast = slice_by_index(begin = var_11615_begin_0, end = var_11615_end_0, end_mask = var_11615_end_mask_0, x = var_11576_cast)[name = tensor("op_11615_cast")]; + tensor var_11616_begin_0 = const()[name = tensor("op_11616_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_11616_end_0 = const()[name = tensor("op_11616_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_11616_end_mask_0 = const()[name = tensor("op_11616_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11616_cast = slice_by_index(begin = var_11616_begin_0, end = var_11616_end_0, end_mask = var_11616_end_mask_0, x = var_11576_cast)[name = tensor("op_11616_cast")]; + tensor var_11617_begin_0 = const()[name = tensor("op_11617_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11617_end_0 = const()[name = tensor("op_11617_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_11617_end_mask_0 = const()[name = tensor("op_11617_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11617_cast = slice_by_index(begin = var_11617_begin_0, end = var_11617_end_0, end_mask = var_11617_end_mask_0, x = var_11580_cast)[name = tensor("op_11617_cast")]; + tensor var_11618_begin_0 = const()[name = tensor("op_11618_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_11618_end_0 = const()[name = tensor("op_11618_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_11618_end_mask_0 = const()[name = tensor("op_11618_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11618_cast = slice_by_index(begin = var_11618_begin_0, end = var_11618_end_0, end_mask = var_11618_end_mask_0, x = var_11580_cast)[name = tensor("op_11618_cast")]; + tensor var_11619_begin_0 = const()[name = tensor("op_11619_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11619_end_0 = const()[name = tensor("op_11619_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_11619_end_mask_0 = const()[name = tensor("op_11619_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11619_cast = slice_by_index(begin = var_11619_begin_0, end = var_11619_end_0, end_mask = var_11619_end_mask_0, x = var_11584_cast)[name = tensor("op_11619_cast")]; + tensor var_11620_begin_0 = const()[name = tensor("op_11620_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_11620_end_0 = const()[name = tensor("op_11620_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_11620_end_mask_0 = const()[name = tensor("op_11620_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11620_cast = slice_by_index(begin = var_11620_begin_0, end = var_11620_end_0, end_mask = var_11620_end_mask_0, x = var_11584_cast)[name = tensor("op_11620_cast")]; + tensor var_11621_begin_0 = const()[name = tensor("op_11621_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11621_end_0 = const()[name = tensor("op_11621_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_11621_end_mask_0 = const()[name = tensor("op_11621_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11621_cast = slice_by_index(begin = var_11621_begin_0, end = var_11621_end_0, end_mask = var_11621_end_mask_0, x = var_11588_cast)[name = tensor("op_11621_cast")]; + tensor var_11622_begin_0 = const()[name = tensor("op_11622_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_11622_end_0 = const()[name = tensor("op_11622_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_11622_end_mask_0 = const()[name = tensor("op_11622_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11622_cast = slice_by_index(begin = var_11622_begin_0, end = var_11622_end_0, end_mask = var_11622_end_mask_0, x = var_11588_cast)[name = tensor("op_11622_cast")]; + tensor var_11623_begin_0 = const()[name = tensor("op_11623_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11623_end_0 = const()[name = tensor("op_11623_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_11623_end_mask_0 = const()[name = tensor("op_11623_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11623_cast = slice_by_index(begin = var_11623_begin_0, end = var_11623_end_0, end_mask = var_11623_end_mask_0, x = var_11592_cast)[name = tensor("op_11623_cast")]; + tensor var_11624_begin_0 = const()[name = tensor("op_11624_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_11624_end_0 = const()[name = tensor("op_11624_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_11624_end_mask_0 = const()[name = tensor("op_11624_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11624_cast = slice_by_index(begin = var_11624_begin_0, end = var_11624_end_0, end_mask = var_11624_end_mask_0, x = var_11592_cast)[name = tensor("op_11624_cast")]; + tensor var_11625_begin_0 = const()[name = tensor("op_11625_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11625_end_0 = const()[name = tensor("op_11625_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_11625_end_mask_0 = const()[name = tensor("op_11625_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11625_cast = slice_by_index(begin = var_11625_begin_0, end = var_11625_end_0, end_mask = var_11625_end_mask_0, x = var_11596_cast)[name = tensor("op_11625_cast")]; + tensor var_11626_begin_0 = const()[name = tensor("op_11626_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_11626_end_0 = const()[name = tensor("op_11626_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_11626_end_mask_0 = const()[name = tensor("op_11626_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11626_cast = slice_by_index(begin = var_11626_begin_0, end = var_11626_end_0, end_mask = var_11626_end_mask_0, x = var_11596_cast)[name = tensor("op_11626_cast")]; + tensor var_11627_begin_0 = const()[name = tensor("op_11627_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11627_end_0 = const()[name = tensor("op_11627_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_11627_end_mask_0 = const()[name = tensor("op_11627_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11627_cast = slice_by_index(begin = var_11627_begin_0, end = var_11627_end_0, end_mask = var_11627_end_mask_0, x = var_11600_cast)[name = tensor("op_11627_cast")]; + tensor var_11628_begin_0 = const()[name = tensor("op_11628_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_11628_end_0 = const()[name = tensor("op_11628_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_11628_end_mask_0 = const()[name = tensor("op_11628_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11628_cast = slice_by_index(begin = var_11628_begin_0, end = var_11628_end_0, end_mask = var_11628_end_mask_0, x = var_11600_cast)[name = tensor("op_11628_cast")]; + tensor var_11629_begin_0 = const()[name = tensor("op_11629_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11629_end_0 = const()[name = tensor("op_11629_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_11629_end_mask_0 = const()[name = tensor("op_11629_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11629_cast = slice_by_index(begin = var_11629_begin_0, end = var_11629_end_0, end_mask = var_11629_end_mask_0, x = var_11604_cast)[name = tensor("op_11629_cast")]; + tensor var_11630_begin_0 = const()[name = tensor("op_11630_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_11630_end_0 = const()[name = tensor("op_11630_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_11630_end_mask_0 = const()[name = tensor("op_11630_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11630_cast = slice_by_index(begin = var_11630_begin_0, end = var_11630_end_0, end_mask = var_11630_end_mask_0, x = var_11604_cast)[name = tensor("op_11630_cast")]; + tensor var_11631_begin_0 = const()[name = tensor("op_11631_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11631_end_0 = const()[name = tensor("op_11631_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_11631_end_mask_0 = const()[name = tensor("op_11631_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11631_cast = slice_by_index(begin = var_11631_begin_0, end = var_11631_end_0, end_mask = var_11631_end_mask_0, x = var_11608_cast)[name = tensor("op_11631_cast")]; + tensor var_11632_begin_0 = const()[name = tensor("op_11632_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_11632_end_0 = const()[name = tensor("op_11632_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_11632_end_mask_0 = const()[name = tensor("op_11632_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11632_cast = slice_by_index(begin = var_11632_begin_0, end = var_11632_end_0, end_mask = var_11632_end_mask_0, x = var_11608_cast)[name = tensor("op_11632_cast")]; + tensor var_11633_begin_0 = const()[name = tensor("op_11633_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11633_end_0 = const()[name = tensor("op_11633_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_11633_end_mask_0 = const()[name = tensor("op_11633_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11633_cast = slice_by_index(begin = var_11633_begin_0, end = var_11633_end_0, end_mask = var_11633_end_mask_0, x = var_11612_cast)[name = tensor("op_11633_cast")]; + tensor var_11634_begin_0 = const()[name = tensor("op_11634_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_11634_end_0 = const()[name = tensor("op_11634_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_11634_end_mask_0 = const()[name = tensor("op_11634_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11634_cast = slice_by_index(begin = var_11634_begin_0, end = var_11634_end_0, end_mask = var_11634_end_mask_0, x = var_11612_cast)[name = tensor("op_11634_cast")]; + tensor k_95_perm_0 = const()[name = tensor("k_95_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_11639_begin_0 = const()[name = tensor("op_11639_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11639_end_0 = const()[name = tensor("op_11639_end_0"), val = tensor([2, 77, 1, 64])]; + tensor var_11639_end_mask_0 = const()[name = tensor("op_11639_end_mask_0"), val = tensor([true, true, true, false])]; + tensor transpose_8 = transpose(perm = k_95_perm_0, x = k_93_cast)[name = tensor("transpose_8")]; + tensor var_11639_cast = slice_by_index(begin = var_11639_begin_0, end = var_11639_end_0, end_mask = var_11639_end_mask_0, x = transpose_8)[name = tensor("op_11639_cast")]; + tensor var_11643_begin_0 = const()[name = tensor("op_11643_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_11643_end_0 = const()[name = tensor("op_11643_end_0"), val = tensor([2, 77, 1, 128])]; + tensor var_11643_end_mask_0 = const()[name = tensor("op_11643_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11643_cast = slice_by_index(begin = var_11643_begin_0, end = var_11643_end_0, end_mask = var_11643_end_mask_0, x = transpose_8)[name = tensor("op_11643_cast")]; + tensor var_11647_begin_0 = const()[name = tensor("op_11647_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_11647_end_0 = const()[name = tensor("op_11647_end_0"), val = tensor([2, 77, 1, 192])]; + tensor var_11647_end_mask_0 = const()[name = tensor("op_11647_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11647_cast = slice_by_index(begin = var_11647_begin_0, end = var_11647_end_0, end_mask = var_11647_end_mask_0, x = transpose_8)[name = tensor("op_11647_cast")]; + tensor var_11651_begin_0 = const()[name = tensor("op_11651_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_11651_end_0 = const()[name = tensor("op_11651_end_0"), val = tensor([2, 77, 1, 256])]; + tensor var_11651_end_mask_0 = const()[name = tensor("op_11651_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11651_cast = slice_by_index(begin = var_11651_begin_0, end = var_11651_end_0, end_mask = var_11651_end_mask_0, x = transpose_8)[name = tensor("op_11651_cast")]; + tensor var_11655_begin_0 = const()[name = tensor("op_11655_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_11655_end_0 = const()[name = tensor("op_11655_end_0"), val = tensor([2, 77, 1, 320])]; + tensor var_11655_end_mask_0 = const()[name = tensor("op_11655_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11655_cast = slice_by_index(begin = var_11655_begin_0, end = var_11655_end_0, end_mask = var_11655_end_mask_0, x = transpose_8)[name = tensor("op_11655_cast")]; + tensor var_11659_begin_0 = const()[name = tensor("op_11659_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_11659_end_0 = const()[name = tensor("op_11659_end_0"), val = tensor([2, 77, 1, 384])]; + tensor var_11659_end_mask_0 = const()[name = tensor("op_11659_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11659_cast = slice_by_index(begin = var_11659_begin_0, end = var_11659_end_0, end_mask = var_11659_end_mask_0, x = transpose_8)[name = tensor("op_11659_cast")]; + tensor var_11663_begin_0 = const()[name = tensor("op_11663_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_11663_end_0 = const()[name = tensor("op_11663_end_0"), val = tensor([2, 77, 1, 448])]; + tensor var_11663_end_mask_0 = const()[name = tensor("op_11663_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11663_cast = slice_by_index(begin = var_11663_begin_0, end = var_11663_end_0, end_mask = var_11663_end_mask_0, x = transpose_8)[name = tensor("op_11663_cast")]; + tensor var_11667_begin_0 = const()[name = tensor("op_11667_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_11667_end_0 = const()[name = tensor("op_11667_end_0"), val = tensor([2, 77, 1, 512])]; + tensor var_11667_end_mask_0 = const()[name = tensor("op_11667_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11667_cast = slice_by_index(begin = var_11667_begin_0, end = var_11667_end_0, end_mask = var_11667_end_mask_0, x = transpose_8)[name = tensor("op_11667_cast")]; + tensor var_11671_begin_0 = const()[name = tensor("op_11671_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_11671_end_0 = const()[name = tensor("op_11671_end_0"), val = tensor([2, 77, 1, 576])]; + tensor var_11671_end_mask_0 = const()[name = tensor("op_11671_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11671_cast = slice_by_index(begin = var_11671_begin_0, end = var_11671_end_0, end_mask = var_11671_end_mask_0, x = transpose_8)[name = tensor("op_11671_cast")]; + tensor var_11675_begin_0 = const()[name = tensor("op_11675_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_11675_end_0 = const()[name = tensor("op_11675_end_0"), val = tensor([2, 77, 1, 640])]; + tensor var_11675_end_mask_0 = const()[name = tensor("op_11675_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11675_cast = slice_by_index(begin = var_11675_begin_0, end = var_11675_end_0, end_mask = var_11675_end_mask_0, x = transpose_8)[name = tensor("op_11675_cast")]; + tensor var_11677_begin_0 = const()[name = tensor("op_11677_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11677_end_0 = const()[name = tensor("op_11677_end_0"), val = tensor([2, 64, 1, 77])]; + tensor var_11677_end_mask_0 = const()[name = tensor("op_11677_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11677_cast = slice_by_index(begin = var_11677_begin_0, end = var_11677_end_0, end_mask = var_11677_end_mask_0, x = v_47_cast)[name = tensor("op_11677_cast")]; + tensor var_11681_begin_0 = const()[name = tensor("op_11681_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_11681_end_0 = const()[name = tensor("op_11681_end_0"), val = tensor([2, 128, 1, 77])]; + tensor var_11681_end_mask_0 = const()[name = tensor("op_11681_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11681_cast = slice_by_index(begin = var_11681_begin_0, end = var_11681_end_0, end_mask = var_11681_end_mask_0, x = v_47_cast)[name = tensor("op_11681_cast")]; + tensor var_11685_begin_0 = const()[name = tensor("op_11685_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_11685_end_0 = const()[name = tensor("op_11685_end_0"), val = tensor([2, 192, 1, 77])]; + tensor var_11685_end_mask_0 = const()[name = tensor("op_11685_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11685_cast = slice_by_index(begin = var_11685_begin_0, end = var_11685_end_0, end_mask = var_11685_end_mask_0, x = v_47_cast)[name = tensor("op_11685_cast")]; + tensor var_11689_begin_0 = const()[name = tensor("op_11689_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_11689_end_0 = const()[name = tensor("op_11689_end_0"), val = tensor([2, 256, 1, 77])]; + tensor var_11689_end_mask_0 = const()[name = tensor("op_11689_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11689_cast = slice_by_index(begin = var_11689_begin_0, end = var_11689_end_0, end_mask = var_11689_end_mask_0, x = v_47_cast)[name = tensor("op_11689_cast")]; + tensor var_11693_begin_0 = const()[name = tensor("op_11693_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_11693_end_0 = const()[name = tensor("op_11693_end_0"), val = tensor([2, 320, 1, 77])]; + tensor var_11693_end_mask_0 = const()[name = tensor("op_11693_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11693_cast = slice_by_index(begin = var_11693_begin_0, end = var_11693_end_0, end_mask = var_11693_end_mask_0, x = v_47_cast)[name = tensor("op_11693_cast")]; + tensor var_11697_begin_0 = const()[name = tensor("op_11697_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_11697_end_0 = const()[name = tensor("op_11697_end_0"), val = tensor([2, 384, 1, 77])]; + tensor var_11697_end_mask_0 = const()[name = tensor("op_11697_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11697_cast = slice_by_index(begin = var_11697_begin_0, end = var_11697_end_0, end_mask = var_11697_end_mask_0, x = v_47_cast)[name = tensor("op_11697_cast")]; + tensor var_11701_begin_0 = const()[name = tensor("op_11701_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_11701_end_0 = const()[name = tensor("op_11701_end_0"), val = tensor([2, 448, 1, 77])]; + tensor var_11701_end_mask_0 = const()[name = tensor("op_11701_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11701_cast = slice_by_index(begin = var_11701_begin_0, end = var_11701_end_0, end_mask = var_11701_end_mask_0, x = v_47_cast)[name = tensor("op_11701_cast")]; + tensor var_11705_begin_0 = const()[name = tensor("op_11705_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_11705_end_0 = const()[name = tensor("op_11705_end_0"), val = tensor([2, 512, 1, 77])]; + tensor var_11705_end_mask_0 = const()[name = tensor("op_11705_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11705_cast = slice_by_index(begin = var_11705_begin_0, end = var_11705_end_0, end_mask = var_11705_end_mask_0, x = v_47_cast)[name = tensor("op_11705_cast")]; + tensor var_11709_begin_0 = const()[name = tensor("op_11709_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_11709_end_0 = const()[name = tensor("op_11709_end_0"), val = tensor([2, 576, 1, 77])]; + tensor var_11709_end_mask_0 = const()[name = tensor("op_11709_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11709_cast = slice_by_index(begin = var_11709_begin_0, end = var_11709_end_0, end_mask = var_11709_end_mask_0, x = v_47_cast)[name = tensor("op_11709_cast")]; + tensor var_11713_begin_0 = const()[name = tensor("op_11713_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_11713_end_0 = const()[name = tensor("op_11713_end_0"), val = tensor([2, 640, 1, 77])]; + tensor var_11713_end_mask_0 = const()[name = tensor("op_11713_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11713_cast = slice_by_index(begin = var_11713_begin_0, end = var_11713_end_0, end_mask = var_11713_end_mask_0, x = v_47_cast)[name = tensor("op_11713_cast")]; + tensor var_11717_equation_0 = const()[name = tensor("op_11717_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11717_cast = einsum(equation = var_11717_equation_0, values = (var_11639_cast, var_11615_cast))[name = tensor("op_11717_cast")]; + tensor var_11718_to_fp16 = const()[name = tensor("op_11718_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_601_cast = mul(x = var_11717_cast, y = var_11718_to_fp16)[name = tensor("aw_chunk_601_cast")]; + tensor var_11721_equation_0 = const()[name = tensor("op_11721_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11721_cast = einsum(equation = var_11721_equation_0, values = (var_11639_cast, var_11616_cast))[name = tensor("op_11721_cast")]; + tensor var_11722_to_fp16 = const()[name = tensor("op_11722_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_603_cast = mul(x = var_11721_cast, y = var_11722_to_fp16)[name = tensor("aw_chunk_603_cast")]; + tensor var_11725_equation_0 = const()[name = tensor("op_11725_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11725_cast = einsum(equation = var_11725_equation_0, values = (var_11643_cast, var_11617_cast))[name = tensor("op_11725_cast")]; + tensor var_11726_to_fp16 = const()[name = tensor("op_11726_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_605_cast = mul(x = var_11725_cast, y = var_11726_to_fp16)[name = tensor("aw_chunk_605_cast")]; + tensor var_11729_equation_0 = const()[name = tensor("op_11729_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11729_cast = einsum(equation = var_11729_equation_0, values = (var_11643_cast, var_11618_cast))[name = tensor("op_11729_cast")]; + tensor var_11730_to_fp16 = const()[name = tensor("op_11730_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_607_cast = mul(x = var_11729_cast, y = var_11730_to_fp16)[name = tensor("aw_chunk_607_cast")]; + tensor var_11733_equation_0 = const()[name = tensor("op_11733_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11733_cast = einsum(equation = var_11733_equation_0, values = (var_11647_cast, var_11619_cast))[name = tensor("op_11733_cast")]; + tensor var_11734_to_fp16 = const()[name = tensor("op_11734_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_609_cast = mul(x = var_11733_cast, y = var_11734_to_fp16)[name = tensor("aw_chunk_609_cast")]; + tensor var_11737_equation_0 = const()[name = tensor("op_11737_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11737_cast = einsum(equation = var_11737_equation_0, values = (var_11647_cast, var_11620_cast))[name = tensor("op_11737_cast")]; + tensor var_11738_to_fp16 = const()[name = tensor("op_11738_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_611_cast = mul(x = var_11737_cast, y = var_11738_to_fp16)[name = tensor("aw_chunk_611_cast")]; + tensor var_11741_equation_0 = const()[name = tensor("op_11741_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11741_cast = einsum(equation = var_11741_equation_0, values = (var_11651_cast, var_11621_cast))[name = tensor("op_11741_cast")]; + tensor var_11742_to_fp16 = const()[name = tensor("op_11742_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_613_cast = mul(x = var_11741_cast, y = var_11742_to_fp16)[name = tensor("aw_chunk_613_cast")]; + tensor var_11745_equation_0 = const()[name = tensor("op_11745_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11745_cast = einsum(equation = var_11745_equation_0, values = (var_11651_cast, var_11622_cast))[name = tensor("op_11745_cast")]; + tensor var_11746_to_fp16 = const()[name = tensor("op_11746_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_615_cast = mul(x = var_11745_cast, y = var_11746_to_fp16)[name = tensor("aw_chunk_615_cast")]; + tensor var_11749_equation_0 = const()[name = tensor("op_11749_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11749_cast = einsum(equation = var_11749_equation_0, values = (var_11655_cast, var_11623_cast))[name = tensor("op_11749_cast")]; + tensor var_11750_to_fp16 = const()[name = tensor("op_11750_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_617_cast = mul(x = var_11749_cast, y = var_11750_to_fp16)[name = tensor("aw_chunk_617_cast")]; + tensor var_11753_equation_0 = const()[name = tensor("op_11753_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11753_cast = einsum(equation = var_11753_equation_0, values = (var_11655_cast, var_11624_cast))[name = tensor("op_11753_cast")]; + tensor var_11754_to_fp16 = const()[name = tensor("op_11754_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_619_cast = mul(x = var_11753_cast, y = var_11754_to_fp16)[name = tensor("aw_chunk_619_cast")]; + tensor var_11757_equation_0 = const()[name = tensor("op_11757_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11757_cast = einsum(equation = var_11757_equation_0, values = (var_11659_cast, var_11625_cast))[name = tensor("op_11757_cast")]; + tensor var_11758_to_fp16 = const()[name = tensor("op_11758_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_621_cast = mul(x = var_11757_cast, y = var_11758_to_fp16)[name = tensor("aw_chunk_621_cast")]; + tensor var_11761_equation_0 = const()[name = tensor("op_11761_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11761_cast = einsum(equation = var_11761_equation_0, values = (var_11659_cast, var_11626_cast))[name = tensor("op_11761_cast")]; + tensor var_11762_to_fp16 = const()[name = tensor("op_11762_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_623_cast = mul(x = var_11761_cast, y = var_11762_to_fp16)[name = tensor("aw_chunk_623_cast")]; + tensor var_11765_equation_0 = const()[name = tensor("op_11765_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11765_cast = einsum(equation = var_11765_equation_0, values = (var_11663_cast, var_11627_cast))[name = tensor("op_11765_cast")]; + tensor var_11766_to_fp16 = const()[name = tensor("op_11766_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_625_cast = mul(x = var_11765_cast, y = var_11766_to_fp16)[name = tensor("aw_chunk_625_cast")]; + tensor var_11769_equation_0 = const()[name = tensor("op_11769_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11769_cast = einsum(equation = var_11769_equation_0, values = (var_11663_cast, var_11628_cast))[name = tensor("op_11769_cast")]; + tensor var_11770_to_fp16 = const()[name = tensor("op_11770_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_627_cast = mul(x = var_11769_cast, y = var_11770_to_fp16)[name = tensor("aw_chunk_627_cast")]; + tensor var_11773_equation_0 = const()[name = tensor("op_11773_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11773_cast = einsum(equation = var_11773_equation_0, values = (var_11667_cast, var_11629_cast))[name = tensor("op_11773_cast")]; + tensor var_11774_to_fp16 = const()[name = tensor("op_11774_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_629_cast = mul(x = var_11773_cast, y = var_11774_to_fp16)[name = tensor("aw_chunk_629_cast")]; + tensor var_11777_equation_0 = const()[name = tensor("op_11777_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11777_cast = einsum(equation = var_11777_equation_0, values = (var_11667_cast, var_11630_cast))[name = tensor("op_11777_cast")]; + tensor var_11778_to_fp16 = const()[name = tensor("op_11778_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_631_cast = mul(x = var_11777_cast, y = var_11778_to_fp16)[name = tensor("aw_chunk_631_cast")]; + tensor var_11781_equation_0 = const()[name = tensor("op_11781_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11781_cast = einsum(equation = var_11781_equation_0, values = (var_11671_cast, var_11631_cast))[name = tensor("op_11781_cast")]; + tensor var_11782_to_fp16 = const()[name = tensor("op_11782_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_633_cast = mul(x = var_11781_cast, y = var_11782_to_fp16)[name = tensor("aw_chunk_633_cast")]; + tensor var_11785_equation_0 = const()[name = tensor("op_11785_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11785_cast = einsum(equation = var_11785_equation_0, values = (var_11671_cast, var_11632_cast))[name = tensor("op_11785_cast")]; + tensor var_11786_to_fp16 = const()[name = tensor("op_11786_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_635_cast = mul(x = var_11785_cast, y = var_11786_to_fp16)[name = tensor("aw_chunk_635_cast")]; + tensor var_11789_equation_0 = const()[name = tensor("op_11789_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11789_cast = einsum(equation = var_11789_equation_0, values = (var_11675_cast, var_11633_cast))[name = tensor("op_11789_cast")]; + tensor var_11790_to_fp16 = const()[name = tensor("op_11790_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_637_cast = mul(x = var_11789_cast, y = var_11790_to_fp16)[name = tensor("aw_chunk_637_cast")]; + tensor var_11793_equation_0 = const()[name = tensor("op_11793_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11793_cast = einsum(equation = var_11793_equation_0, values = (var_11675_cast, var_11634_cast))[name = tensor("op_11793_cast")]; + tensor var_11794_to_fp16 = const()[name = tensor("op_11794_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_639_cast = mul(x = var_11793_cast, y = var_11794_to_fp16)[name = tensor("aw_chunk_639_cast")]; + tensor var_11796_cast = softmax(axis = var_10257, x = aw_chunk_601_cast)[name = tensor("op_11796_cast")]; + tensor var_11797_cast = softmax(axis = var_10257, x = aw_chunk_603_cast)[name = tensor("op_11797_cast")]; + tensor var_11798_cast = softmax(axis = var_10257, x = aw_chunk_605_cast)[name = tensor("op_11798_cast")]; + tensor var_11799_cast = softmax(axis = var_10257, x = aw_chunk_607_cast)[name = tensor("op_11799_cast")]; + tensor var_11800_cast = softmax(axis = var_10257, x = aw_chunk_609_cast)[name = tensor("op_11800_cast")]; + tensor var_11801_cast = softmax(axis = var_10257, x = aw_chunk_611_cast)[name = tensor("op_11801_cast")]; + tensor var_11802_cast = softmax(axis = var_10257, x = aw_chunk_613_cast)[name = tensor("op_11802_cast")]; + tensor var_11803_cast = softmax(axis = var_10257, x = aw_chunk_615_cast)[name = tensor("op_11803_cast")]; + tensor var_11804_cast = softmax(axis = var_10257, x = aw_chunk_617_cast)[name = tensor("op_11804_cast")]; + tensor var_11805_cast = softmax(axis = var_10257, x = aw_chunk_619_cast)[name = tensor("op_11805_cast")]; + tensor var_11806_cast = softmax(axis = var_10257, x = aw_chunk_621_cast)[name = tensor("op_11806_cast")]; + tensor var_11807_cast = softmax(axis = var_10257, x = aw_chunk_623_cast)[name = tensor("op_11807_cast")]; + tensor var_11808_cast = softmax(axis = var_10257, x = aw_chunk_625_cast)[name = tensor("op_11808_cast")]; + tensor var_11809_cast = softmax(axis = var_10257, x = aw_chunk_627_cast)[name = tensor("op_11809_cast")]; + tensor var_11810_cast = softmax(axis = var_10257, x = aw_chunk_629_cast)[name = tensor("op_11810_cast")]; + tensor var_11811_cast = softmax(axis = var_10257, x = aw_chunk_631_cast)[name = tensor("op_11811_cast")]; + tensor var_11812_cast = softmax(axis = var_10257, x = aw_chunk_633_cast)[name = tensor("op_11812_cast")]; + tensor var_11813_cast = softmax(axis = var_10257, x = aw_chunk_635_cast)[name = tensor("op_11813_cast")]; + tensor var_11814_cast = softmax(axis = var_10257, x = aw_chunk_637_cast)[name = tensor("op_11814_cast")]; + tensor var_11815_cast = softmax(axis = var_10257, x = aw_chunk_639_cast)[name = tensor("op_11815_cast")]; + tensor var_11817_equation_0 = const()[name = tensor("op_11817_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11817_cast = einsum(equation = var_11817_equation_0, values = (var_11677_cast, var_11796_cast))[name = tensor("op_11817_cast")]; + tensor var_11819_equation_0 = const()[name = tensor("op_11819_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11819_cast = einsum(equation = var_11819_equation_0, values = (var_11677_cast, var_11797_cast))[name = tensor("op_11819_cast")]; + tensor var_11821_equation_0 = const()[name = tensor("op_11821_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11821_cast = einsum(equation = var_11821_equation_0, values = (var_11681_cast, var_11798_cast))[name = tensor("op_11821_cast")]; + tensor var_11823_equation_0 = const()[name = tensor("op_11823_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11823_cast = einsum(equation = var_11823_equation_0, values = (var_11681_cast, var_11799_cast))[name = tensor("op_11823_cast")]; + tensor var_11825_equation_0 = const()[name = tensor("op_11825_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11825_cast = einsum(equation = var_11825_equation_0, values = (var_11685_cast, var_11800_cast))[name = tensor("op_11825_cast")]; + tensor var_11827_equation_0 = const()[name = tensor("op_11827_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11827_cast = einsum(equation = var_11827_equation_0, values = (var_11685_cast, var_11801_cast))[name = tensor("op_11827_cast")]; + tensor var_11829_equation_0 = const()[name = tensor("op_11829_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11829_cast = einsum(equation = var_11829_equation_0, values = (var_11689_cast, var_11802_cast))[name = tensor("op_11829_cast")]; + tensor var_11831_equation_0 = const()[name = tensor("op_11831_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11831_cast = einsum(equation = var_11831_equation_0, values = (var_11689_cast, var_11803_cast))[name = tensor("op_11831_cast")]; + tensor var_11833_equation_0 = const()[name = tensor("op_11833_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11833_cast = einsum(equation = var_11833_equation_0, values = (var_11693_cast, var_11804_cast))[name = tensor("op_11833_cast")]; + tensor var_11835_equation_0 = const()[name = tensor("op_11835_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11835_cast = einsum(equation = var_11835_equation_0, values = (var_11693_cast, var_11805_cast))[name = tensor("op_11835_cast")]; + tensor var_11837_equation_0 = const()[name = tensor("op_11837_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11837_cast = einsum(equation = var_11837_equation_0, values = (var_11697_cast, var_11806_cast))[name = tensor("op_11837_cast")]; + tensor var_11839_equation_0 = const()[name = tensor("op_11839_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11839_cast = einsum(equation = var_11839_equation_0, values = (var_11697_cast, var_11807_cast))[name = tensor("op_11839_cast")]; + tensor var_11841_equation_0 = const()[name = tensor("op_11841_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11841_cast = einsum(equation = var_11841_equation_0, values = (var_11701_cast, var_11808_cast))[name = tensor("op_11841_cast")]; + tensor var_11843_equation_0 = const()[name = tensor("op_11843_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11843_cast = einsum(equation = var_11843_equation_0, values = (var_11701_cast, var_11809_cast))[name = tensor("op_11843_cast")]; + tensor var_11845_equation_0 = const()[name = tensor("op_11845_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11845_cast = einsum(equation = var_11845_equation_0, values = (var_11705_cast, var_11810_cast))[name = tensor("op_11845_cast")]; + tensor var_11847_equation_0 = const()[name = tensor("op_11847_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11847_cast = einsum(equation = var_11847_equation_0, values = (var_11705_cast, var_11811_cast))[name = tensor("op_11847_cast")]; + tensor var_11849_equation_0 = const()[name = tensor("op_11849_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11849_cast = einsum(equation = var_11849_equation_0, values = (var_11709_cast, var_11812_cast))[name = tensor("op_11849_cast")]; + tensor var_11851_equation_0 = const()[name = tensor("op_11851_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11851_cast = einsum(equation = var_11851_equation_0, values = (var_11709_cast, var_11813_cast))[name = tensor("op_11851_cast")]; + tensor var_11853_equation_0 = const()[name = tensor("op_11853_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11853_cast = einsum(equation = var_11853_equation_0, values = (var_11713_cast, var_11814_cast))[name = tensor("op_11853_cast")]; + tensor var_11855_equation_0 = const()[name = tensor("op_11855_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11855_cast = einsum(equation = var_11855_equation_0, values = (var_11713_cast, var_11815_cast))[name = tensor("op_11855_cast")]; + tensor var_11857_interleave_0 = const()[name = tensor("op_11857_interleave_0"), val = tensor(false)]; + tensor var_11857_cast = concat(axis = var_10234, interleave = var_11857_interleave_0, values = (var_11817_cast, var_11819_cast))[name = tensor("op_11857_cast")]; + tensor var_11859_interleave_0 = const()[name = tensor("op_11859_interleave_0"), val = tensor(false)]; + tensor var_11859_cast = concat(axis = var_10234, interleave = var_11859_interleave_0, values = (var_11821_cast, var_11823_cast))[name = tensor("op_11859_cast")]; + tensor var_11861_interleave_0 = const()[name = tensor("op_11861_interleave_0"), val = tensor(false)]; + tensor var_11861_cast = concat(axis = var_10234, interleave = var_11861_interleave_0, values = (var_11825_cast, var_11827_cast))[name = tensor("op_11861_cast")]; + tensor var_11863_interleave_0 = const()[name = tensor("op_11863_interleave_0"), val = tensor(false)]; + tensor var_11863_cast = concat(axis = var_10234, interleave = var_11863_interleave_0, values = (var_11829_cast, var_11831_cast))[name = tensor("op_11863_cast")]; + tensor var_11865_interleave_0 = const()[name = tensor("op_11865_interleave_0"), val = tensor(false)]; + tensor var_11865_cast = concat(axis = var_10234, interleave = var_11865_interleave_0, values = (var_11833_cast, var_11835_cast))[name = tensor("op_11865_cast")]; + tensor var_11867_interleave_0 = const()[name = tensor("op_11867_interleave_0"), val = tensor(false)]; + tensor var_11867_cast = concat(axis = var_10234, interleave = var_11867_interleave_0, values = (var_11837_cast, var_11839_cast))[name = tensor("op_11867_cast")]; + tensor var_11869_interleave_0 = const()[name = tensor("op_11869_interleave_0"), val = tensor(false)]; + tensor var_11869_cast = concat(axis = var_10234, interleave = var_11869_interleave_0, values = (var_11841_cast, var_11843_cast))[name = tensor("op_11869_cast")]; + tensor var_11871_interleave_0 = const()[name = tensor("op_11871_interleave_0"), val = tensor(false)]; + tensor var_11871_cast = concat(axis = var_10234, interleave = var_11871_interleave_0, values = (var_11845_cast, var_11847_cast))[name = tensor("op_11871_cast")]; + tensor var_11873_interleave_0 = const()[name = tensor("op_11873_interleave_0"), val = tensor(false)]; + tensor var_11873_cast = concat(axis = var_10234, interleave = var_11873_interleave_0, values = (var_11849_cast, var_11851_cast))[name = tensor("op_11873_cast")]; + tensor var_11875_interleave_0 = const()[name = tensor("op_11875_interleave_0"), val = tensor(false)]; + tensor var_11875_cast = concat(axis = var_10234, interleave = var_11875_interleave_0, values = (var_11853_cast, var_11855_cast))[name = tensor("op_11875_cast")]; + tensor input_411_interleave_0 = const()[name = tensor("input_411_interleave_0"), val = tensor(false)]; + tensor input_411_cast = concat(axis = var_10257, interleave = input_411_interleave_0, values = (var_11857_cast, var_11859_cast, var_11861_cast, var_11863_cast, var_11865_cast, var_11867_cast, var_11869_cast, var_11871_cast, var_11873_cast, var_11875_cast))[name = tensor("input_411_cast")]; + tensor var_11881 = const()[name = tensor("op_11881"), val = tensor([1, 1])]; + tensor var_11883 = const()[name = tensor("op_11883"), val = tensor([1, 1])]; + tensor var_11885_pad_type_0 = const()[name = tensor("op_11885_pad_type_0"), val = tensor("custom")]; + tensor var_11885_pad_0 = const()[name = tensor("op_11885_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_2_attentions_1_transformer_blocks_0_attn2_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(613192640))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(613499904))), name = tensor("up_blocks_2_attentions_1_transformer_blocks_0_attn2_to_out_0_weight_to_fp16_palettized"), shape = tensor([640, 640, 1, 1])]; + tensor up_blocks_2_attentions_1_transformer_blocks_0_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("up_blocks_2_attentions_1_transformer_blocks_0_attn2_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(613500096)))]; + tensor var_11885_cast = conv(bias = up_blocks_2_attentions_1_transformer_blocks_0_attn2_to_out_0_bias_to_fp16, dilations = var_11883, groups = var_10257, pad = var_11885_pad_0, pad_type = var_11885_pad_type_0, strides = var_11881, weight = up_blocks_2_attentions_1_transformer_blocks_0_attn2_to_out_0_weight_to_fp16_palettized, x = input_411_cast)[name = tensor("op_11885_cast")]; + tensor inputs_71_cast = add(x = var_11885_cast, y = inputs_69_cast)[name = tensor("inputs_71_cast")]; + tensor var_11889 = const()[name = tensor("op_11889"), val = tensor([1])]; + tensor channels_mean_71_cast = reduce_mean(axes = var_11889, keep_dims = var_10252, x = inputs_71_cast)[name = tensor("channels_mean_71_cast")]; + tensor zero_mean_71_cast = sub(x = inputs_71_cast, y = channels_mean_71_cast)[name = tensor("zero_mean_71_cast")]; + tensor zero_mean_sq_71_cast = mul(x = zero_mean_71_cast, y = zero_mean_71_cast)[name = tensor("zero_mean_sq_71_cast")]; + tensor var_11893 = const()[name = tensor("op_11893"), val = tensor([1])]; + tensor var_11894_cast = reduce_mean(axes = var_11893, keep_dims = var_10252, x = zero_mean_sq_71_cast)[name = tensor("op_11894_cast")]; + tensor var_11895_to_fp16 = const()[name = tensor("op_11895_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_11896_cast = add(x = var_11894_cast, y = var_11895_to_fp16)[name = tensor("op_11896_cast")]; + tensor denom_71_epsilon_0_to_fp16 = const()[name = tensor("denom_71_epsilon_0_to_fp16"), val = tensor(0x1p-24)]; + tensor denom_71_cast = rsqrt(epsilon = denom_71_epsilon_0_to_fp16, x = var_11896_cast)[name = tensor("denom_71_cast")]; + tensor out_71_cast = mul(x = zero_mean_71_cast, y = denom_71_cast)[name = tensor("out_71_cast")]; + tensor var_11900_to_fp16 = const()[name = tensor("op_11900_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(613501440)))]; + tensor var_11901_cast = add(x = out_71_cast, y = var_11900_to_fp16)[name = tensor("op_11901_cast")]; + tensor var_11903_to_fp16 = const()[name = tensor("op_11903_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(613502784)))]; + tensor input_413_cast = mul(x = var_11901_cast, y = var_11903_to_fp16)[name = tensor("input_413_cast")]; + tensor var_11911 = const()[name = tensor("op_11911"), val = tensor([1, 1])]; + tensor var_11913 = const()[name = tensor("op_11913"), val = tensor([1, 1])]; + tensor var_11915_pad_type_0 = const()[name = tensor("op_11915_pad_type_0"), val = tensor("custom")]; + tensor var_11915_pad_0 = const()[name = tensor("op_11915_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_2_attentions_1_transformer_blocks_0_ff_net_0_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(613504128))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(615961792))), name = tensor("up_blocks_2_attentions_1_transformer_blocks_0_ff_net_0_proj_weight_to_fp16_palettized"), shape = tensor([5120, 640, 1, 1])]; + tensor up_blocks_2_attentions_1_transformer_blocks_0_ff_net_0_proj_bias_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(615961984))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(615965888))), name = tensor("up_blocks_2_attentions_1_transformer_blocks_0_ff_net_0_proj_bias_to_fp16_palettized"), shape = tensor([5120])]; + tensor var_11915_cast = conv(bias = up_blocks_2_attentions_1_transformer_blocks_0_ff_net_0_proj_bias_to_fp16_palettized, dilations = var_11913, groups = var_10257, pad = var_11915_pad_0, pad_type = var_11915_pad_type_0, strides = var_11911, weight = up_blocks_2_attentions_1_transformer_blocks_0_ff_net_0_proj_weight_to_fp16_palettized, x = input_413_cast)[name = tensor("op_11915_cast")]; + tensor var_11916_split_sizes_0 = const()[name = tensor("op_11916_split_sizes_0"), val = tensor([2560, 2560])]; + tensor var_11916_axis_0 = const()[name = tensor("op_11916_axis_0"), val = tensor(1)]; + tensor var_11916_cast_0, tensor var_11916_cast_1 = split(axis = var_11916_axis_0, split_sizes = var_11916_split_sizes_0, x = var_11915_cast)[name = tensor("op_11916_cast")]; + tensor var_11918_mode_0 = const()[name = tensor("op_11918_mode_0"), val = tensor("EXACT")]; + tensor var_11918_cast = gelu(mode = var_11918_mode_0, x = var_11916_cast_1)[name = tensor("op_11918_cast")]; + tensor input_415_cast = mul(x = var_11916_cast_0, y = var_11918_cast)[name = tensor("input_415_cast")]; + tensor var_11922 = const()[name = tensor("op_11922"), val = tensor([1, 1])]; + tensor var_11924 = const()[name = tensor("op_11924"), val = tensor([1, 1])]; + tensor var_11926_pad_type_0 = const()[name = tensor("op_11926_pad_type_0"), val = tensor("custom")]; + tensor var_11926_pad_0 = const()[name = tensor("op_11926_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_2_attentions_1_transformer_blocks_0_ff_net_2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(615966080))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(617194944))), name = tensor("up_blocks_2_attentions_1_transformer_blocks_0_ff_net_2_weight_to_fp16_palettized"), shape = tensor([640, 2560, 1, 1])]; + tensor up_blocks_2_attentions_1_transformer_blocks_0_ff_net_2_bias_to_fp16 = const()[name = tensor("up_blocks_2_attentions_1_transformer_blocks_0_ff_net_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(617195136)))]; + tensor var_11926_cast = conv(bias = up_blocks_2_attentions_1_transformer_blocks_0_ff_net_2_bias_to_fp16, dilations = var_11924, groups = var_10257, pad = var_11926_pad_0, pad_type = var_11926_pad_type_0, strides = var_11922, weight = up_blocks_2_attentions_1_transformer_blocks_0_ff_net_2_weight_to_fp16_palettized, x = input_415_cast)[name = tensor("op_11926_cast")]; + tensor hidden_states_253_cast = add(x = var_11926_cast, y = inputs_71_cast)[name = tensor("hidden_states_253_cast")]; + tensor var_11928 = const()[name = tensor("op_11928"), val = tensor([2, 640, 32, 32])]; + tensor input_417_cast = reshape(shape = var_11928, x = hidden_states_253_cast)[name = tensor("input_417_cast")]; + tensor var_11932 = const()[name = tensor("op_11932"), val = tensor([1, 1])]; + tensor var_11934 = const()[name = tensor("op_11934"), val = tensor([1, 1])]; + tensor hidden_states_255_pad_type_0 = const()[name = tensor("hidden_states_255_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_255_pad_0 = const()[name = tensor("hidden_states_255_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_2_attentions_1_proj_out_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(617196480))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(617503744))), name = tensor("up_blocks_2_attentions_1_proj_out_weight_to_fp16_palettized"), shape = tensor([640, 640, 1, 1])]; + tensor up_blocks_2_attentions_1_proj_out_bias_to_fp16 = const()[name = tensor("up_blocks_2_attentions_1_proj_out_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(617503936)))]; + tensor hidden_states_255_cast = conv(bias = up_blocks_2_attentions_1_proj_out_bias_to_fp16, dilations = var_11934, groups = var_10257, pad = hidden_states_255_pad_0, pad_type = hidden_states_255_pad_type_0, strides = var_11932, weight = up_blocks_2_attentions_1_proj_out_weight_to_fp16_palettized, x = input_417_cast)[name = tensor("hidden_states_255_cast")]; + tensor hidden_states_257_cast = add(x = hidden_states_255_cast, y = hidden_states_243_cast)[name = tensor("hidden_states_257_cast")]; + tensor input_419_interleave_0 = const()[name = tensor("input_419_interleave_0"), val = tensor(false)]; + tensor input_419_cast = concat(axis = var_10257, interleave = input_419_interleave_0, values = (hidden_states_257_cast, input_63_cast))[name = tensor("input_419_cast")]; + tensor reshape_192_shape_0 = const()[name = tensor("reshape_192_shape_0"), val = tensor([2, 32, 30, 32, 32])]; + tensor reshape_192_cast = reshape(shape = reshape_192_shape_0, x = input_419_cast)[name = tensor("reshape_192_cast")]; + tensor reduce_mean_144_axes_0 = const()[name = tensor("reduce_mean_144_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_144_keep_dims_0 = const()[name = tensor("reduce_mean_144_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_144_cast = reduce_mean(axes = reduce_mean_144_axes_0, keep_dims = reduce_mean_144_keep_dims_0, x = reshape_192_cast)[name = tensor("reduce_mean_144_cast")]; + tensor sub_96_cast = sub(x = reshape_192_cast, y = reduce_mean_144_cast)[name = tensor("sub_96_cast")]; + tensor square_48_cast = square(x = sub_96_cast)[name = tensor("square_48_cast")]; + tensor reduce_mean_146_axes_0 = const()[name = tensor("reduce_mean_146_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_146_keep_dims_0 = const()[name = tensor("reduce_mean_146_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_146_cast = reduce_mean(axes = reduce_mean_146_axes_0, keep_dims = reduce_mean_146_keep_dims_0, x = square_48_cast)[name = tensor("reduce_mean_146_cast")]; + tensor add_96_y_0_to_fp16 = const()[name = tensor("add_96_y_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_96_cast = add(x = reduce_mean_146_cast, y = add_96_y_0_to_fp16)[name = tensor("add_96_cast")]; + tensor sqrt_48_cast = sqrt(x = add_96_cast)[name = tensor("sqrt_48_cast")]; + tensor real_div_48_cast = real_div(x = sub_96_cast, y = sqrt_48_cast)[name = tensor("real_div_48_cast")]; + tensor reshape_193_shape_0 = const()[name = tensor("reshape_193_shape_0"), val = tensor([2, 960, 32, 32])]; + tensor reshape_193_cast = reshape(shape = reshape_193_shape_0, x = real_div_48_cast)[name = tensor("reshape_193_cast")]; + tensor add_97_mean_0_to_fp16 = const()[name = tensor("add_97_mean_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(617505280)))]; + tensor add_97_variance_0_to_fp16 = const()[name = tensor("add_97_variance_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(617507264)))]; + tensor add_97_gamma_0_to_fp16 = const()[name = tensor("add_97_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(617509248)))]; + tensor add_97_beta_0_to_fp16 = const()[name = tensor("add_97_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(617511232)))]; + tensor add_97_epsilon_0_to_fp16 = const()[name = tensor("add_97_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_97_cast = batch_norm(beta = add_97_beta_0_to_fp16, epsilon = add_97_epsilon_0_to_fp16, gamma = add_97_gamma_0_to_fp16, mean = add_97_mean_0_to_fp16, variance = add_97_variance_0_to_fp16, x = reshape_193_cast)[name = tensor("add_97_cast")]; + tensor input_423_cast = silu(x = add_97_cast)[name = tensor("input_423_cast")]; + tensor var_11952 = const()[name = tensor("op_11952"), val = tensor([1, 1])]; + tensor var_11954 = const()[name = tensor("op_11954"), val = tensor([1, 1])]; + tensor hidden_states_259_pad_type_0 = const()[name = tensor("hidden_states_259_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_259_pad_0 = const()[name = tensor("hidden_states_259_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor up_blocks_2_resnets_2_conv1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(617513216))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(621660480))), name = tensor("up_blocks_2_resnets_2_conv1_weight_to_fp16_palettized"), shape = tensor([640, 960, 3, 3])]; + tensor up_blocks_2_resnets_2_conv1_bias_to_fp16 = const()[name = tensor("up_blocks_2_resnets_2_conv1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(621660672)))]; + tensor hidden_states_259_cast = conv(bias = up_blocks_2_resnets_2_conv1_bias_to_fp16, dilations = var_11954, groups = var_10257, pad = hidden_states_259_pad_0, pad_type = hidden_states_259_pad_type_0, strides = var_11952, weight = up_blocks_2_resnets_2_conv1_weight_to_fp16_palettized, x = input_423_cast)[name = tensor("hidden_states_259_cast")]; + tensor var_11960 = const()[name = tensor("op_11960"), val = tensor([1, 1])]; + tensor var_11962 = const()[name = tensor("op_11962"), val = tensor([1, 1])]; + tensor temb_37_pad_type_0 = const()[name = tensor("temb_37_pad_type_0"), val = tensor("custom")]; + tensor temb_37_pad_0 = const()[name = tensor("temb_37_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_2_resnets_2_time_emb_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(621662016))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(622276480))), name = tensor("up_blocks_2_resnets_2_time_emb_proj_weight_to_fp16_palettized"), shape = tensor([640, 1280, 1, 1])]; + tensor up_blocks_2_resnets_2_time_emb_proj_bias_to_fp16 = const()[name = tensor("up_blocks_2_resnets_2_time_emb_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(622276672)))]; + tensor temb_37_cast = conv(bias = up_blocks_2_resnets_2_time_emb_proj_bias_to_fp16, dilations = var_11962, groups = var_10257, pad = temb_37_pad_0, pad_type = temb_37_pad_type_0, strides = var_11960, weight = up_blocks_2_resnets_2_time_emb_proj_weight_to_fp16_palettized, x = input_15_cast)[name = tensor("temb_37_cast")]; + tensor input_427_cast = add(x = hidden_states_259_cast, y = temb_37_cast)[name = tensor("input_427_cast")]; + tensor reshape_196_shape_0 = const()[name = tensor("reshape_196_shape_0"), val = tensor([2, 32, 20, 32, 32])]; + tensor reshape_196_cast = reshape(shape = reshape_196_shape_0, x = input_427_cast)[name = tensor("reshape_196_cast")]; + tensor reduce_mean_147_axes_0 = const()[name = tensor("reduce_mean_147_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_147_keep_dims_0 = const()[name = tensor("reduce_mean_147_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_147_cast = reduce_mean(axes = reduce_mean_147_axes_0, keep_dims = reduce_mean_147_keep_dims_0, x = reshape_196_cast)[name = tensor("reduce_mean_147_cast")]; + tensor sub_98_cast = sub(x = reshape_196_cast, y = reduce_mean_147_cast)[name = tensor("sub_98_cast")]; + tensor square_49_cast = square(x = sub_98_cast)[name = tensor("square_49_cast")]; + tensor reduce_mean_149_axes_0 = const()[name = tensor("reduce_mean_149_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_149_keep_dims_0 = const()[name = tensor("reduce_mean_149_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_149_cast = reduce_mean(axes = reduce_mean_149_axes_0, keep_dims = reduce_mean_149_keep_dims_0, x = square_49_cast)[name = tensor("reduce_mean_149_cast")]; + tensor add_98_y_0_to_fp16 = const()[name = tensor("add_98_y_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_98_cast = add(x = reduce_mean_149_cast, y = add_98_y_0_to_fp16)[name = tensor("add_98_cast")]; + tensor sqrt_49_cast = sqrt(x = add_98_cast)[name = tensor("sqrt_49_cast")]; + tensor real_div_49_cast = real_div(x = sub_98_cast, y = sqrt_49_cast)[name = tensor("real_div_49_cast")]; + tensor reshape_197_shape_0 = const()[name = tensor("reshape_197_shape_0"), val = tensor([2, 640, 32, 32])]; + tensor reshape_197_cast = reshape(shape = reshape_197_shape_0, x = real_div_49_cast)[name = tensor("reshape_197_cast")]; + tensor add_99_gamma_0_to_fp16 = const()[name = tensor("add_99_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(622278016)))]; + tensor add_99_beta_0_to_fp16 = const()[name = tensor("add_99_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(622279360)))]; + tensor add_99_epsilon_0_to_fp16 = const()[name = tensor("add_99_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_99_cast = batch_norm(beta = add_99_beta_0_to_fp16, epsilon = add_99_epsilon_0_to_fp16, gamma = add_99_gamma_0_to_fp16, mean = add_15_mean_0_to_fp16, variance = add_15_variance_0_to_fp16, x = reshape_197_cast)[name = tensor("add_99_cast")]; + tensor input_431_cast = silu(x = add_99_cast)[name = tensor("input_431_cast")]; + tensor var_11972 = const()[name = tensor("op_11972"), val = tensor([1, 1])]; + tensor var_11974 = const()[name = tensor("op_11974"), val = tensor([1, 1])]; + tensor hidden_states_261_pad_type_0 = const()[name = tensor("hidden_states_261_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_261_pad_0 = const()[name = tensor("hidden_states_261_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor up_blocks_2_resnets_2_conv2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(622280704))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(625045568))), name = tensor("up_blocks_2_resnets_2_conv2_weight_to_fp16_palettized"), shape = tensor([640, 640, 3, 3])]; + tensor up_blocks_2_resnets_2_conv2_bias_to_fp16 = const()[name = tensor("up_blocks_2_resnets_2_conv2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(625045760)))]; + tensor hidden_states_261_cast = conv(bias = up_blocks_2_resnets_2_conv2_bias_to_fp16, dilations = var_11974, groups = var_10257, pad = hidden_states_261_pad_0, pad_type = hidden_states_261_pad_type_0, strides = var_11972, weight = up_blocks_2_resnets_2_conv2_weight_to_fp16_palettized, x = input_431_cast)[name = tensor("hidden_states_261_cast")]; + tensor var_11979 = const()[name = tensor("op_11979"), val = tensor([1, 1])]; + tensor var_11981 = const()[name = tensor("op_11981"), val = tensor([1, 1])]; + tensor x_21_pad_type_0 = const()[name = tensor("x_21_pad_type_0"), val = tensor("custom")]; + tensor x_21_pad_0 = const()[name = tensor("x_21_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_2_resnets_2_conv_shortcut_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(625047104))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(625507968))), name = tensor("up_blocks_2_resnets_2_conv_shortcut_weight_to_fp16_palettized"), shape = tensor([640, 960, 1, 1])]; + tensor up_blocks_2_resnets_2_conv_shortcut_bias_to_fp16 = const()[name = tensor("up_blocks_2_resnets_2_conv_shortcut_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(625508160)))]; + tensor x_21_cast = conv(bias = up_blocks_2_resnets_2_conv_shortcut_bias_to_fp16, dilations = var_11981, groups = var_10257, pad = x_21_pad_0, pad_type = x_21_pad_type_0, strides = var_11979, weight = up_blocks_2_resnets_2_conv_shortcut_weight_to_fp16_palettized, x = input_419_cast)[name = tensor("x_21_cast")]; + tensor hidden_states_263_cast = add(x = x_21_cast, y = hidden_states_261_cast)[name = tensor("hidden_states_263_cast")]; + tensor reshape_200_shape_0 = const()[name = tensor("reshape_200_shape_0"), val = tensor([2, 32, 20, 32, 32])]; + tensor reshape_200_cast = reshape(shape = reshape_200_shape_0, x = hidden_states_263_cast)[name = tensor("reshape_200_cast")]; + tensor reduce_mean_150_axes_0 = const()[name = tensor("reduce_mean_150_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_150_keep_dims_0 = const()[name = tensor("reduce_mean_150_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_150_cast = reduce_mean(axes = reduce_mean_150_axes_0, keep_dims = reduce_mean_150_keep_dims_0, x = reshape_200_cast)[name = tensor("reduce_mean_150_cast")]; + tensor sub_100_cast = sub(x = reshape_200_cast, y = reduce_mean_150_cast)[name = tensor("sub_100_cast")]; + tensor square_50_cast = square(x = sub_100_cast)[name = tensor("square_50_cast")]; + tensor reduce_mean_152_axes_0 = const()[name = tensor("reduce_mean_152_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_152_keep_dims_0 = const()[name = tensor("reduce_mean_152_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_152_cast = reduce_mean(axes = reduce_mean_152_axes_0, keep_dims = reduce_mean_152_keep_dims_0, x = square_50_cast)[name = tensor("reduce_mean_152_cast")]; + tensor add_100_y_0_to_fp16 = const()[name = tensor("add_100_y_0_to_fp16"), val = tensor(0x1.1p-20)]; + tensor add_100_cast = add(x = reduce_mean_152_cast, y = add_100_y_0_to_fp16)[name = tensor("add_100_cast")]; + tensor sqrt_50_cast = sqrt(x = add_100_cast)[name = tensor("sqrt_50_cast")]; + tensor real_div_50_cast = real_div(x = sub_100_cast, y = sqrt_50_cast)[name = tensor("real_div_50_cast")]; + tensor reshape_201_shape_0 = const()[name = tensor("reshape_201_shape_0"), val = tensor([2, 640, 32, 32])]; + tensor reshape_201_cast = reshape(shape = reshape_201_shape_0, x = real_div_50_cast)[name = tensor("reshape_201_cast")]; + tensor add_101_gamma_0_to_fp16 = const()[name = tensor("add_101_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(625509504)))]; + tensor add_101_beta_0_to_fp16 = const()[name = tensor("add_101_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(625510848)))]; + tensor add_101_epsilon_0_to_fp16 = const()[name = tensor("add_101_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_101_cast = batch_norm(beta = add_101_beta_0_to_fp16, epsilon = add_101_epsilon_0_to_fp16, gamma = add_101_gamma_0_to_fp16, mean = add_15_mean_0_to_fp16, variance = add_15_variance_0_to_fp16, x = reshape_201_cast)[name = tensor("add_101_cast")]; + tensor var_12001 = const()[name = tensor("op_12001"), val = tensor([1, 1])]; + tensor var_12003 = const()[name = tensor("op_12003"), val = tensor([1, 1])]; + tensor hidden_states_265_pad_type_0 = const()[name = tensor("hidden_states_265_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_265_pad_0 = const()[name = tensor("hidden_states_265_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_2_attentions_2_proj_in_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(625512192))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(625819456))), name = tensor("up_blocks_2_attentions_2_proj_in_weight_to_fp16_palettized"), shape = tensor([640, 640, 1, 1])]; + tensor up_blocks_2_attentions_2_proj_in_bias_to_fp16 = const()[name = tensor("up_blocks_2_attentions_2_proj_in_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(625819648)))]; + tensor hidden_states_265_cast = conv(bias = up_blocks_2_attentions_2_proj_in_bias_to_fp16, dilations = var_12003, groups = var_10257, pad = hidden_states_265_pad_0, pad_type = hidden_states_265_pad_type_0, strides = var_12001, weight = up_blocks_2_attentions_2_proj_in_weight_to_fp16_palettized, x = add_101_cast)[name = tensor("hidden_states_265_cast")]; + tensor var_12008 = const()[name = tensor("op_12008"), val = tensor([2, 640, 1, 1024])]; + tensor inputs_73_cast = reshape(shape = var_12008, x = hidden_states_265_cast)[name = tensor("inputs_73_cast")]; + tensor var_12018 = const()[name = tensor("op_12018"), val = tensor([1])]; + tensor channels_mean_73_cast = reduce_mean(axes = var_12018, keep_dims = var_10252, x = inputs_73_cast)[name = tensor("channels_mean_73_cast")]; + tensor zero_mean_73_cast = sub(x = inputs_73_cast, y = channels_mean_73_cast)[name = tensor("zero_mean_73_cast")]; + tensor zero_mean_sq_73_cast = mul(x = zero_mean_73_cast, y = zero_mean_73_cast)[name = tensor("zero_mean_sq_73_cast")]; + tensor var_12022 = const()[name = tensor("op_12022"), val = tensor([1])]; + tensor var_12023_cast = reduce_mean(axes = var_12022, keep_dims = var_10252, x = zero_mean_sq_73_cast)[name = tensor("op_12023_cast")]; + tensor var_12024_to_fp16 = const()[name = tensor("op_12024_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_12025_cast = add(x = var_12023_cast, y = var_12024_to_fp16)[name = tensor("op_12025_cast")]; + tensor denom_73_epsilon_0_to_fp16 = const()[name = tensor("denom_73_epsilon_0_to_fp16"), val = tensor(0x1p-24)]; + tensor denom_73_cast = rsqrt(epsilon = denom_73_epsilon_0_to_fp16, x = var_12025_cast)[name = tensor("denom_73_cast")]; + tensor out_73_cast = mul(x = zero_mean_73_cast, y = denom_73_cast)[name = tensor("out_73_cast")]; + tensor var_12029_to_fp16 = const()[name = tensor("op_12029_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(625820992)))]; + tensor var_12030_cast = add(x = out_73_cast, y = var_12029_to_fp16)[name = tensor("op_12030_cast")]; + tensor var_12032_to_fp16 = const()[name = tensor("op_12032_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(625822336)))]; + tensor hidden_states_267_cast = mul(x = var_12030_cast, y = var_12032_to_fp16)[name = tensor("hidden_states_267_cast")]; + tensor var_12039 = const()[name = tensor("op_12039"), val = tensor([1, 1])]; + tensor var_12041 = const()[name = tensor("op_12041"), val = tensor([1, 1])]; + tensor q_49_pad_type_0 = const()[name = tensor("q_49_pad_type_0"), val = tensor("custom")]; + tensor q_49_pad_0 = const()[name = tensor("q_49_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_2_attentions_2_transformer_blocks_0_attn1_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(625823680))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(626130944))), name = tensor("up_blocks_2_attentions_2_transformer_blocks_0_attn1_to_q_weight_to_fp16_palettized"), shape = tensor([640, 640, 1, 1])]; + tensor q_49_cast = conv(dilations = var_12041, groups = var_10257, pad = q_49_pad_0, pad_type = q_49_pad_type_0, strides = var_12039, weight = up_blocks_2_attentions_2_transformer_blocks_0_attn1_to_q_weight_to_fp16_palettized, x = hidden_states_267_cast)[name = tensor("q_49_cast")]; + tensor var_12045 = const()[name = tensor("op_12045"), val = tensor([1, 1])]; + tensor var_12047 = const()[name = tensor("op_12047"), val = tensor([1, 1])]; + tensor k_97_pad_type_0 = const()[name = tensor("k_97_pad_type_0"), val = tensor("custom")]; + tensor k_97_pad_0 = const()[name = tensor("k_97_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_2_attentions_2_transformer_blocks_0_attn1_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(626131136))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(626438400))), name = tensor("up_blocks_2_attentions_2_transformer_blocks_0_attn1_to_k_weight_to_fp16_palettized"), shape = tensor([640, 640, 1, 1])]; + tensor k_97_cast = conv(dilations = var_12047, groups = var_10257, pad = k_97_pad_0, pad_type = k_97_pad_type_0, strides = var_12045, weight = up_blocks_2_attentions_2_transformer_blocks_0_attn1_to_k_weight_to_fp16_palettized, x = hidden_states_267_cast)[name = tensor("k_97_cast")]; + tensor var_12051 = const()[name = tensor("op_12051"), val = tensor([1, 1])]; + tensor var_12053 = const()[name = tensor("op_12053"), val = tensor([1, 1])]; + tensor v_49_pad_type_0 = const()[name = tensor("v_49_pad_type_0"), val = tensor("custom")]; + tensor v_49_pad_0 = const()[name = tensor("v_49_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_2_attentions_2_transformer_blocks_0_attn1_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(626438592))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(626745856))), name = tensor("up_blocks_2_attentions_2_transformer_blocks_0_attn1_to_v_weight_to_fp16_palettized"), shape = tensor([640, 640, 1, 1])]; + tensor v_49_cast = conv(dilations = var_12053, groups = var_10257, pad = v_49_pad_0, pad_type = v_49_pad_type_0, strides = var_12051, weight = up_blocks_2_attentions_2_transformer_blocks_0_attn1_to_v_weight_to_fp16_palettized, x = hidden_states_267_cast)[name = tensor("v_49_cast")]; + tensor var_12057_begin_0 = const()[name = tensor("op_12057_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_12057_end_0 = const()[name = tensor("op_12057_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_12057_end_mask_0 = const()[name = tensor("op_12057_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12057_cast = slice_by_index(begin = var_12057_begin_0, end = var_12057_end_0, end_mask = var_12057_end_mask_0, x = q_49_cast)[name = tensor("op_12057_cast")]; + tensor var_12061_begin_0 = const()[name = tensor("op_12061_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_12061_end_0 = const()[name = tensor("op_12061_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_12061_end_mask_0 = const()[name = tensor("op_12061_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12061_cast = slice_by_index(begin = var_12061_begin_0, end = var_12061_end_0, end_mask = var_12061_end_mask_0, x = q_49_cast)[name = tensor("op_12061_cast")]; + tensor var_12065_begin_0 = const()[name = tensor("op_12065_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_12065_end_0 = const()[name = tensor("op_12065_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_12065_end_mask_0 = const()[name = tensor("op_12065_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12065_cast = slice_by_index(begin = var_12065_begin_0, end = var_12065_end_0, end_mask = var_12065_end_mask_0, x = q_49_cast)[name = tensor("op_12065_cast")]; + tensor var_12069_begin_0 = const()[name = tensor("op_12069_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_12069_end_0 = const()[name = tensor("op_12069_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_12069_end_mask_0 = const()[name = tensor("op_12069_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12069_cast = slice_by_index(begin = var_12069_begin_0, end = var_12069_end_0, end_mask = var_12069_end_mask_0, x = q_49_cast)[name = tensor("op_12069_cast")]; + tensor var_12073_begin_0 = const()[name = tensor("op_12073_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_12073_end_0 = const()[name = tensor("op_12073_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_12073_end_mask_0 = const()[name = tensor("op_12073_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12073_cast = slice_by_index(begin = var_12073_begin_0, end = var_12073_end_0, end_mask = var_12073_end_mask_0, x = q_49_cast)[name = tensor("op_12073_cast")]; + tensor var_12077_begin_0 = const()[name = tensor("op_12077_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_12077_end_0 = const()[name = tensor("op_12077_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_12077_end_mask_0 = const()[name = tensor("op_12077_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12077_cast = slice_by_index(begin = var_12077_begin_0, end = var_12077_end_0, end_mask = var_12077_end_mask_0, x = q_49_cast)[name = tensor("op_12077_cast")]; + tensor var_12081_begin_0 = const()[name = tensor("op_12081_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_12081_end_0 = const()[name = tensor("op_12081_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_12081_end_mask_0 = const()[name = tensor("op_12081_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12081_cast = slice_by_index(begin = var_12081_begin_0, end = var_12081_end_0, end_mask = var_12081_end_mask_0, x = q_49_cast)[name = tensor("op_12081_cast")]; + tensor var_12085_begin_0 = const()[name = tensor("op_12085_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_12085_end_0 = const()[name = tensor("op_12085_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_12085_end_mask_0 = const()[name = tensor("op_12085_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12085_cast = slice_by_index(begin = var_12085_begin_0, end = var_12085_end_0, end_mask = var_12085_end_mask_0, x = q_49_cast)[name = tensor("op_12085_cast")]; + tensor var_12089_begin_0 = const()[name = tensor("op_12089_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_12089_end_0 = const()[name = tensor("op_12089_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_12089_end_mask_0 = const()[name = tensor("op_12089_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12089_cast = slice_by_index(begin = var_12089_begin_0, end = var_12089_end_0, end_mask = var_12089_end_mask_0, x = q_49_cast)[name = tensor("op_12089_cast")]; + tensor var_12093_begin_0 = const()[name = tensor("op_12093_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_12093_end_0 = const()[name = tensor("op_12093_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_12093_end_mask_0 = const()[name = tensor("op_12093_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12093_cast = slice_by_index(begin = var_12093_begin_0, end = var_12093_end_0, end_mask = var_12093_end_mask_0, x = q_49_cast)[name = tensor("op_12093_cast")]; + tensor var_12096_begin_0 = const()[name = tensor("op_12096_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_12096_end_0 = const()[name = tensor("op_12096_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_12096_end_mask_0 = const()[name = tensor("op_12096_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12096_cast = slice_by_index(begin = var_12096_begin_0, end = var_12096_end_0, end_mask = var_12096_end_mask_0, x = var_12057_cast)[name = tensor("op_12096_cast")]; + tensor var_12097_begin_0 = const()[name = tensor("op_12097_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_12097_end_0 = const()[name = tensor("op_12097_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_12097_end_mask_0 = const()[name = tensor("op_12097_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12097_cast = slice_by_index(begin = var_12097_begin_0, end = var_12097_end_0, end_mask = var_12097_end_mask_0, x = var_12057_cast)[name = tensor("op_12097_cast")]; + tensor var_12098_begin_0 = const()[name = tensor("op_12098_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_12098_end_0 = const()[name = tensor("op_12098_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_12098_end_mask_0 = const()[name = tensor("op_12098_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12098_cast = slice_by_index(begin = var_12098_begin_0, end = var_12098_end_0, end_mask = var_12098_end_mask_0, x = var_12061_cast)[name = tensor("op_12098_cast")]; + tensor var_12099_begin_0 = const()[name = tensor("op_12099_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_12099_end_0 = const()[name = tensor("op_12099_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_12099_end_mask_0 = const()[name = tensor("op_12099_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12099_cast = slice_by_index(begin = var_12099_begin_0, end = var_12099_end_0, end_mask = var_12099_end_mask_0, x = var_12061_cast)[name = tensor("op_12099_cast")]; + tensor var_12100_begin_0 = const()[name = tensor("op_12100_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_12100_end_0 = const()[name = tensor("op_12100_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_12100_end_mask_0 = const()[name = tensor("op_12100_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12100_cast = slice_by_index(begin = var_12100_begin_0, end = var_12100_end_0, end_mask = var_12100_end_mask_0, x = var_12065_cast)[name = tensor("op_12100_cast")]; + tensor var_12101_begin_0 = const()[name = tensor("op_12101_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_12101_end_0 = const()[name = tensor("op_12101_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_12101_end_mask_0 = const()[name = tensor("op_12101_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12101_cast = slice_by_index(begin = var_12101_begin_0, end = var_12101_end_0, end_mask = var_12101_end_mask_0, x = var_12065_cast)[name = tensor("op_12101_cast")]; + tensor var_12102_begin_0 = const()[name = tensor("op_12102_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_12102_end_0 = const()[name = tensor("op_12102_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_12102_end_mask_0 = const()[name = tensor("op_12102_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12102_cast = slice_by_index(begin = var_12102_begin_0, end = var_12102_end_0, end_mask = var_12102_end_mask_0, x = var_12069_cast)[name = tensor("op_12102_cast")]; + tensor var_12103_begin_0 = const()[name = tensor("op_12103_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_12103_end_0 = const()[name = tensor("op_12103_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_12103_end_mask_0 = const()[name = tensor("op_12103_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12103_cast = slice_by_index(begin = var_12103_begin_0, end = var_12103_end_0, end_mask = var_12103_end_mask_0, x = var_12069_cast)[name = tensor("op_12103_cast")]; + tensor var_12104_begin_0 = const()[name = tensor("op_12104_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_12104_end_0 = const()[name = tensor("op_12104_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_12104_end_mask_0 = const()[name = tensor("op_12104_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12104_cast = slice_by_index(begin = var_12104_begin_0, end = var_12104_end_0, end_mask = var_12104_end_mask_0, x = var_12073_cast)[name = tensor("op_12104_cast")]; + tensor var_12105_begin_0 = const()[name = tensor("op_12105_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_12105_end_0 = const()[name = tensor("op_12105_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_12105_end_mask_0 = const()[name = tensor("op_12105_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12105_cast = slice_by_index(begin = var_12105_begin_0, end = var_12105_end_0, end_mask = var_12105_end_mask_0, x = var_12073_cast)[name = tensor("op_12105_cast")]; + tensor var_12106_begin_0 = const()[name = tensor("op_12106_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_12106_end_0 = const()[name = tensor("op_12106_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_12106_end_mask_0 = const()[name = tensor("op_12106_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12106_cast = slice_by_index(begin = var_12106_begin_0, end = var_12106_end_0, end_mask = var_12106_end_mask_0, x = var_12077_cast)[name = tensor("op_12106_cast")]; + tensor var_12107_begin_0 = const()[name = tensor("op_12107_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_12107_end_0 = const()[name = tensor("op_12107_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_12107_end_mask_0 = const()[name = tensor("op_12107_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12107_cast = slice_by_index(begin = var_12107_begin_0, end = var_12107_end_0, end_mask = var_12107_end_mask_0, x = var_12077_cast)[name = tensor("op_12107_cast")]; + tensor var_12108_begin_0 = const()[name = tensor("op_12108_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_12108_end_0 = const()[name = tensor("op_12108_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_12108_end_mask_0 = const()[name = tensor("op_12108_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12108_cast = slice_by_index(begin = var_12108_begin_0, end = var_12108_end_0, end_mask = var_12108_end_mask_0, x = var_12081_cast)[name = tensor("op_12108_cast")]; + tensor var_12109_begin_0 = const()[name = tensor("op_12109_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_12109_end_0 = const()[name = tensor("op_12109_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_12109_end_mask_0 = const()[name = tensor("op_12109_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12109_cast = slice_by_index(begin = var_12109_begin_0, end = var_12109_end_0, end_mask = var_12109_end_mask_0, x = var_12081_cast)[name = tensor("op_12109_cast")]; + tensor var_12110_begin_0 = const()[name = tensor("op_12110_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_12110_end_0 = const()[name = tensor("op_12110_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_12110_end_mask_0 = const()[name = tensor("op_12110_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12110_cast = slice_by_index(begin = var_12110_begin_0, end = var_12110_end_0, end_mask = var_12110_end_mask_0, x = var_12085_cast)[name = tensor("op_12110_cast")]; + tensor var_12111_begin_0 = const()[name = tensor("op_12111_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_12111_end_0 = const()[name = tensor("op_12111_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_12111_end_mask_0 = const()[name = tensor("op_12111_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12111_cast = slice_by_index(begin = var_12111_begin_0, end = var_12111_end_0, end_mask = var_12111_end_mask_0, x = var_12085_cast)[name = tensor("op_12111_cast")]; + tensor var_12112_begin_0 = const()[name = tensor("op_12112_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_12112_end_0 = const()[name = tensor("op_12112_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_12112_end_mask_0 = const()[name = tensor("op_12112_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12112_cast = slice_by_index(begin = var_12112_begin_0, end = var_12112_end_0, end_mask = var_12112_end_mask_0, x = var_12089_cast)[name = tensor("op_12112_cast")]; + tensor var_12113_begin_0 = const()[name = tensor("op_12113_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_12113_end_0 = const()[name = tensor("op_12113_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_12113_end_mask_0 = const()[name = tensor("op_12113_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12113_cast = slice_by_index(begin = var_12113_begin_0, end = var_12113_end_0, end_mask = var_12113_end_mask_0, x = var_12089_cast)[name = tensor("op_12113_cast")]; + tensor var_12114_begin_0 = const()[name = tensor("op_12114_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_12114_end_0 = const()[name = tensor("op_12114_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_12114_end_mask_0 = const()[name = tensor("op_12114_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12114_cast = slice_by_index(begin = var_12114_begin_0, end = var_12114_end_0, end_mask = var_12114_end_mask_0, x = var_12093_cast)[name = tensor("op_12114_cast")]; + tensor var_12115_begin_0 = const()[name = tensor("op_12115_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_12115_end_0 = const()[name = tensor("op_12115_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_12115_end_mask_0 = const()[name = tensor("op_12115_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12115_cast = slice_by_index(begin = var_12115_begin_0, end = var_12115_end_0, end_mask = var_12115_end_mask_0, x = var_12093_cast)[name = tensor("op_12115_cast")]; + tensor k_99_perm_0 = const()[name = tensor("k_99_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_12120_begin_0 = const()[name = tensor("op_12120_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_12120_end_0 = const()[name = tensor("op_12120_end_0"), val = tensor([2, 1024, 1, 64])]; + tensor var_12120_end_mask_0 = const()[name = tensor("op_12120_end_mask_0"), val = tensor([true, true, true, false])]; + tensor transpose_7 = transpose(perm = k_99_perm_0, x = k_97_cast)[name = tensor("transpose_7")]; + tensor var_12120_cast = slice_by_index(begin = var_12120_begin_0, end = var_12120_end_0, end_mask = var_12120_end_mask_0, x = transpose_7)[name = tensor("op_12120_cast")]; + tensor var_12124_begin_0 = const()[name = tensor("op_12124_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_12124_end_0 = const()[name = tensor("op_12124_end_0"), val = tensor([2, 1024, 1, 128])]; + tensor var_12124_end_mask_0 = const()[name = tensor("op_12124_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12124_cast = slice_by_index(begin = var_12124_begin_0, end = var_12124_end_0, end_mask = var_12124_end_mask_0, x = transpose_7)[name = tensor("op_12124_cast")]; + tensor var_12128_begin_0 = const()[name = tensor("op_12128_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_12128_end_0 = const()[name = tensor("op_12128_end_0"), val = tensor([2, 1024, 1, 192])]; + tensor var_12128_end_mask_0 = const()[name = tensor("op_12128_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12128_cast = slice_by_index(begin = var_12128_begin_0, end = var_12128_end_0, end_mask = var_12128_end_mask_0, x = transpose_7)[name = tensor("op_12128_cast")]; + tensor var_12132_begin_0 = const()[name = tensor("op_12132_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_12132_end_0 = const()[name = tensor("op_12132_end_0"), val = tensor([2, 1024, 1, 256])]; + tensor var_12132_end_mask_0 = const()[name = tensor("op_12132_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12132_cast = slice_by_index(begin = var_12132_begin_0, end = var_12132_end_0, end_mask = var_12132_end_mask_0, x = transpose_7)[name = tensor("op_12132_cast")]; + tensor var_12136_begin_0 = const()[name = tensor("op_12136_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_12136_end_0 = const()[name = tensor("op_12136_end_0"), val = tensor([2, 1024, 1, 320])]; + tensor var_12136_end_mask_0 = const()[name = tensor("op_12136_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12136_cast = slice_by_index(begin = var_12136_begin_0, end = var_12136_end_0, end_mask = var_12136_end_mask_0, x = transpose_7)[name = tensor("op_12136_cast")]; + tensor var_12140_begin_0 = const()[name = tensor("op_12140_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_12140_end_0 = const()[name = tensor("op_12140_end_0"), val = tensor([2, 1024, 1, 384])]; + tensor var_12140_end_mask_0 = const()[name = tensor("op_12140_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12140_cast = slice_by_index(begin = var_12140_begin_0, end = var_12140_end_0, end_mask = var_12140_end_mask_0, x = transpose_7)[name = tensor("op_12140_cast")]; + tensor var_12144_begin_0 = const()[name = tensor("op_12144_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_12144_end_0 = const()[name = tensor("op_12144_end_0"), val = tensor([2, 1024, 1, 448])]; + tensor var_12144_end_mask_0 = const()[name = tensor("op_12144_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12144_cast = slice_by_index(begin = var_12144_begin_0, end = var_12144_end_0, end_mask = var_12144_end_mask_0, x = transpose_7)[name = tensor("op_12144_cast")]; + tensor var_12148_begin_0 = const()[name = tensor("op_12148_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_12148_end_0 = const()[name = tensor("op_12148_end_0"), val = tensor([2, 1024, 1, 512])]; + tensor var_12148_end_mask_0 = const()[name = tensor("op_12148_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12148_cast = slice_by_index(begin = var_12148_begin_0, end = var_12148_end_0, end_mask = var_12148_end_mask_0, x = transpose_7)[name = tensor("op_12148_cast")]; + tensor var_12152_begin_0 = const()[name = tensor("op_12152_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_12152_end_0 = const()[name = tensor("op_12152_end_0"), val = tensor([2, 1024, 1, 576])]; + tensor var_12152_end_mask_0 = const()[name = tensor("op_12152_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12152_cast = slice_by_index(begin = var_12152_begin_0, end = var_12152_end_0, end_mask = var_12152_end_mask_0, x = transpose_7)[name = tensor("op_12152_cast")]; + tensor var_12156_begin_0 = const()[name = tensor("op_12156_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_12156_end_0 = const()[name = tensor("op_12156_end_0"), val = tensor([2, 1024, 1, 640])]; + tensor var_12156_end_mask_0 = const()[name = tensor("op_12156_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12156_cast = slice_by_index(begin = var_12156_begin_0, end = var_12156_end_0, end_mask = var_12156_end_mask_0, x = transpose_7)[name = tensor("op_12156_cast")]; + tensor var_12158_begin_0 = const()[name = tensor("op_12158_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_12158_end_0 = const()[name = tensor("op_12158_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_12158_end_mask_0 = const()[name = tensor("op_12158_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12158_cast = slice_by_index(begin = var_12158_begin_0, end = var_12158_end_0, end_mask = var_12158_end_mask_0, x = v_49_cast)[name = tensor("op_12158_cast")]; + tensor var_12162_begin_0 = const()[name = tensor("op_12162_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_12162_end_0 = const()[name = tensor("op_12162_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_12162_end_mask_0 = const()[name = tensor("op_12162_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12162_cast = slice_by_index(begin = var_12162_begin_0, end = var_12162_end_0, end_mask = var_12162_end_mask_0, x = v_49_cast)[name = tensor("op_12162_cast")]; + tensor var_12166_begin_0 = const()[name = tensor("op_12166_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_12166_end_0 = const()[name = tensor("op_12166_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_12166_end_mask_0 = const()[name = tensor("op_12166_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12166_cast = slice_by_index(begin = var_12166_begin_0, end = var_12166_end_0, end_mask = var_12166_end_mask_0, x = v_49_cast)[name = tensor("op_12166_cast")]; + tensor var_12170_begin_0 = const()[name = tensor("op_12170_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_12170_end_0 = const()[name = tensor("op_12170_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_12170_end_mask_0 = const()[name = tensor("op_12170_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12170_cast = slice_by_index(begin = var_12170_begin_0, end = var_12170_end_0, end_mask = var_12170_end_mask_0, x = v_49_cast)[name = tensor("op_12170_cast")]; + tensor var_12174_begin_0 = const()[name = tensor("op_12174_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_12174_end_0 = const()[name = tensor("op_12174_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_12174_end_mask_0 = const()[name = tensor("op_12174_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12174_cast = slice_by_index(begin = var_12174_begin_0, end = var_12174_end_0, end_mask = var_12174_end_mask_0, x = v_49_cast)[name = tensor("op_12174_cast")]; + tensor var_12178_begin_0 = const()[name = tensor("op_12178_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_12178_end_0 = const()[name = tensor("op_12178_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_12178_end_mask_0 = const()[name = tensor("op_12178_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12178_cast = slice_by_index(begin = var_12178_begin_0, end = var_12178_end_0, end_mask = var_12178_end_mask_0, x = v_49_cast)[name = tensor("op_12178_cast")]; + tensor var_12182_begin_0 = const()[name = tensor("op_12182_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_12182_end_0 = const()[name = tensor("op_12182_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_12182_end_mask_0 = const()[name = tensor("op_12182_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12182_cast = slice_by_index(begin = var_12182_begin_0, end = var_12182_end_0, end_mask = var_12182_end_mask_0, x = v_49_cast)[name = tensor("op_12182_cast")]; + tensor var_12186_begin_0 = const()[name = tensor("op_12186_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_12186_end_0 = const()[name = tensor("op_12186_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_12186_end_mask_0 = const()[name = tensor("op_12186_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12186_cast = slice_by_index(begin = var_12186_begin_0, end = var_12186_end_0, end_mask = var_12186_end_mask_0, x = v_49_cast)[name = tensor("op_12186_cast")]; + tensor var_12190_begin_0 = const()[name = tensor("op_12190_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_12190_end_0 = const()[name = tensor("op_12190_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_12190_end_mask_0 = const()[name = tensor("op_12190_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12190_cast = slice_by_index(begin = var_12190_begin_0, end = var_12190_end_0, end_mask = var_12190_end_mask_0, x = v_49_cast)[name = tensor("op_12190_cast")]; + tensor var_12194_begin_0 = const()[name = tensor("op_12194_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_12194_end_0 = const()[name = tensor("op_12194_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_12194_end_mask_0 = const()[name = tensor("op_12194_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12194_cast = slice_by_index(begin = var_12194_begin_0, end = var_12194_end_0, end_mask = var_12194_end_mask_0, x = v_49_cast)[name = tensor("op_12194_cast")]; + tensor var_12198_equation_0 = const()[name = tensor("op_12198_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12198_cast = einsum(equation = var_12198_equation_0, values = (var_12120_cast, var_12096_cast))[name = tensor("op_12198_cast")]; + tensor var_12199_to_fp16 = const()[name = tensor("op_12199_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_641_cast = mul(x = var_12198_cast, y = var_12199_to_fp16)[name = tensor("aw_chunk_641_cast")]; + tensor var_12202_equation_0 = const()[name = tensor("op_12202_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12202_cast = einsum(equation = var_12202_equation_0, values = (var_12120_cast, var_12097_cast))[name = tensor("op_12202_cast")]; + tensor var_12203_to_fp16 = const()[name = tensor("op_12203_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_643_cast = mul(x = var_12202_cast, y = var_12203_to_fp16)[name = tensor("aw_chunk_643_cast")]; + tensor var_12206_equation_0 = const()[name = tensor("op_12206_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12206_cast = einsum(equation = var_12206_equation_0, values = (var_12124_cast, var_12098_cast))[name = tensor("op_12206_cast")]; + tensor var_12207_to_fp16 = const()[name = tensor("op_12207_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_645_cast = mul(x = var_12206_cast, y = var_12207_to_fp16)[name = tensor("aw_chunk_645_cast")]; + tensor var_12210_equation_0 = const()[name = tensor("op_12210_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12210_cast = einsum(equation = var_12210_equation_0, values = (var_12124_cast, var_12099_cast))[name = tensor("op_12210_cast")]; + tensor var_12211_to_fp16 = const()[name = tensor("op_12211_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_647_cast = mul(x = var_12210_cast, y = var_12211_to_fp16)[name = tensor("aw_chunk_647_cast")]; + tensor var_12214_equation_0 = const()[name = tensor("op_12214_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12214_cast = einsum(equation = var_12214_equation_0, values = (var_12128_cast, var_12100_cast))[name = tensor("op_12214_cast")]; + tensor var_12215_to_fp16 = const()[name = tensor("op_12215_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_649_cast = mul(x = var_12214_cast, y = var_12215_to_fp16)[name = tensor("aw_chunk_649_cast")]; + tensor var_12218_equation_0 = const()[name = tensor("op_12218_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12218_cast = einsum(equation = var_12218_equation_0, values = (var_12128_cast, var_12101_cast))[name = tensor("op_12218_cast")]; + tensor var_12219_to_fp16 = const()[name = tensor("op_12219_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_651_cast = mul(x = var_12218_cast, y = var_12219_to_fp16)[name = tensor("aw_chunk_651_cast")]; + tensor var_12222_equation_0 = const()[name = tensor("op_12222_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12222_cast = einsum(equation = var_12222_equation_0, values = (var_12132_cast, var_12102_cast))[name = tensor("op_12222_cast")]; + tensor var_12223_to_fp16 = const()[name = tensor("op_12223_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_653_cast = mul(x = var_12222_cast, y = var_12223_to_fp16)[name = tensor("aw_chunk_653_cast")]; + tensor var_12226_equation_0 = const()[name = tensor("op_12226_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12226_cast = einsum(equation = var_12226_equation_0, values = (var_12132_cast, var_12103_cast))[name = tensor("op_12226_cast")]; + tensor var_12227_to_fp16 = const()[name = tensor("op_12227_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_655_cast = mul(x = var_12226_cast, y = var_12227_to_fp16)[name = tensor("aw_chunk_655_cast")]; + tensor var_12230_equation_0 = const()[name = tensor("op_12230_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12230_cast = einsum(equation = var_12230_equation_0, values = (var_12136_cast, var_12104_cast))[name = tensor("op_12230_cast")]; + tensor var_12231_to_fp16 = const()[name = tensor("op_12231_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_657_cast = mul(x = var_12230_cast, y = var_12231_to_fp16)[name = tensor("aw_chunk_657_cast")]; + tensor var_12234_equation_0 = const()[name = tensor("op_12234_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12234_cast = einsum(equation = var_12234_equation_0, values = (var_12136_cast, var_12105_cast))[name = tensor("op_12234_cast")]; + tensor var_12235_to_fp16 = const()[name = tensor("op_12235_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_659_cast = mul(x = var_12234_cast, y = var_12235_to_fp16)[name = tensor("aw_chunk_659_cast")]; + tensor var_12238_equation_0 = const()[name = tensor("op_12238_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12238_cast = einsum(equation = var_12238_equation_0, values = (var_12140_cast, var_12106_cast))[name = tensor("op_12238_cast")]; + tensor var_12239_to_fp16 = const()[name = tensor("op_12239_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_661_cast = mul(x = var_12238_cast, y = var_12239_to_fp16)[name = tensor("aw_chunk_661_cast")]; + tensor var_12242_equation_0 = const()[name = tensor("op_12242_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12242_cast = einsum(equation = var_12242_equation_0, values = (var_12140_cast, var_12107_cast))[name = tensor("op_12242_cast")]; + tensor var_12243_to_fp16 = const()[name = tensor("op_12243_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_663_cast = mul(x = var_12242_cast, y = var_12243_to_fp16)[name = tensor("aw_chunk_663_cast")]; + tensor var_12246_equation_0 = const()[name = tensor("op_12246_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12246_cast = einsum(equation = var_12246_equation_0, values = (var_12144_cast, var_12108_cast))[name = tensor("op_12246_cast")]; + tensor var_12247_to_fp16 = const()[name = tensor("op_12247_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_665_cast = mul(x = var_12246_cast, y = var_12247_to_fp16)[name = tensor("aw_chunk_665_cast")]; + tensor var_12250_equation_0 = const()[name = tensor("op_12250_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12250_cast = einsum(equation = var_12250_equation_0, values = (var_12144_cast, var_12109_cast))[name = tensor("op_12250_cast")]; + tensor var_12251_to_fp16 = const()[name = tensor("op_12251_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_667_cast = mul(x = var_12250_cast, y = var_12251_to_fp16)[name = tensor("aw_chunk_667_cast")]; + tensor var_12254_equation_0 = const()[name = tensor("op_12254_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12254_cast = einsum(equation = var_12254_equation_0, values = (var_12148_cast, var_12110_cast))[name = tensor("op_12254_cast")]; + tensor var_12255_to_fp16 = const()[name = tensor("op_12255_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_669_cast = mul(x = var_12254_cast, y = var_12255_to_fp16)[name = tensor("aw_chunk_669_cast")]; + tensor var_12258_equation_0 = const()[name = tensor("op_12258_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12258_cast = einsum(equation = var_12258_equation_0, values = (var_12148_cast, var_12111_cast))[name = tensor("op_12258_cast")]; + tensor var_12259_to_fp16 = const()[name = tensor("op_12259_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_671_cast = mul(x = var_12258_cast, y = var_12259_to_fp16)[name = tensor("aw_chunk_671_cast")]; + tensor var_12262_equation_0 = const()[name = tensor("op_12262_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12262_cast = einsum(equation = var_12262_equation_0, values = (var_12152_cast, var_12112_cast))[name = tensor("op_12262_cast")]; + tensor var_12263_to_fp16 = const()[name = tensor("op_12263_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_673_cast = mul(x = var_12262_cast, y = var_12263_to_fp16)[name = tensor("aw_chunk_673_cast")]; + tensor var_12266_equation_0 = const()[name = tensor("op_12266_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12266_cast = einsum(equation = var_12266_equation_0, values = (var_12152_cast, var_12113_cast))[name = tensor("op_12266_cast")]; + tensor var_12267_to_fp16 = const()[name = tensor("op_12267_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_675_cast = mul(x = var_12266_cast, y = var_12267_to_fp16)[name = tensor("aw_chunk_675_cast")]; + tensor var_12270_equation_0 = const()[name = tensor("op_12270_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12270_cast = einsum(equation = var_12270_equation_0, values = (var_12156_cast, var_12114_cast))[name = tensor("op_12270_cast")]; + tensor var_12271_to_fp16 = const()[name = tensor("op_12271_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_677_cast = mul(x = var_12270_cast, y = var_12271_to_fp16)[name = tensor("aw_chunk_677_cast")]; + tensor var_12274_equation_0 = const()[name = tensor("op_12274_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12274_cast = einsum(equation = var_12274_equation_0, values = (var_12156_cast, var_12115_cast))[name = tensor("op_12274_cast")]; + tensor var_12275_to_fp16 = const()[name = tensor("op_12275_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_679_cast = mul(x = var_12274_cast, y = var_12275_to_fp16)[name = tensor("aw_chunk_679_cast")]; + tensor var_12277_cast = softmax(axis = var_10257, x = aw_chunk_641_cast)[name = tensor("op_12277_cast")]; + tensor var_12278_cast = softmax(axis = var_10257, x = aw_chunk_643_cast)[name = tensor("op_12278_cast")]; + tensor var_12279_cast = softmax(axis = var_10257, x = aw_chunk_645_cast)[name = tensor("op_12279_cast")]; + tensor var_12280_cast = softmax(axis = var_10257, x = aw_chunk_647_cast)[name = tensor("op_12280_cast")]; + tensor var_12281_cast = softmax(axis = var_10257, x = aw_chunk_649_cast)[name = tensor("op_12281_cast")]; + tensor var_12282_cast = softmax(axis = var_10257, x = aw_chunk_651_cast)[name = tensor("op_12282_cast")]; + tensor var_12283_cast = softmax(axis = var_10257, x = aw_chunk_653_cast)[name = tensor("op_12283_cast")]; + tensor var_12284_cast = softmax(axis = var_10257, x = aw_chunk_655_cast)[name = tensor("op_12284_cast")]; + tensor var_12285_cast = softmax(axis = var_10257, x = aw_chunk_657_cast)[name = tensor("op_12285_cast")]; + tensor var_12286_cast = softmax(axis = var_10257, x = aw_chunk_659_cast)[name = tensor("op_12286_cast")]; + tensor var_12287_cast = softmax(axis = var_10257, x = aw_chunk_661_cast)[name = tensor("op_12287_cast")]; + tensor var_12288_cast = softmax(axis = var_10257, x = aw_chunk_663_cast)[name = tensor("op_12288_cast")]; + tensor var_12289_cast = softmax(axis = var_10257, x = aw_chunk_665_cast)[name = tensor("op_12289_cast")]; + tensor var_12290_cast = softmax(axis = var_10257, x = aw_chunk_667_cast)[name = tensor("op_12290_cast")]; + tensor var_12291_cast = softmax(axis = var_10257, x = aw_chunk_669_cast)[name = tensor("op_12291_cast")]; + tensor var_12292_cast = softmax(axis = var_10257, x = aw_chunk_671_cast)[name = tensor("op_12292_cast")]; + tensor var_12293_cast = softmax(axis = var_10257, x = aw_chunk_673_cast)[name = tensor("op_12293_cast")]; + tensor var_12294_cast = softmax(axis = var_10257, x = aw_chunk_675_cast)[name = tensor("op_12294_cast")]; + tensor var_12295_cast = softmax(axis = var_10257, x = aw_chunk_677_cast)[name = tensor("op_12295_cast")]; + tensor var_12296_cast = softmax(axis = var_10257, x = aw_chunk_679_cast)[name = tensor("op_12296_cast")]; + tensor var_12298_equation_0 = const()[name = tensor("op_12298_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_12298_cast = einsum(equation = var_12298_equation_0, values = (var_12158_cast, var_12277_cast))[name = tensor("op_12298_cast")]; + tensor var_12300_equation_0 = const()[name = tensor("op_12300_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_12300_cast = einsum(equation = var_12300_equation_0, values = (var_12158_cast, var_12278_cast))[name = tensor("op_12300_cast")]; + tensor var_12302_equation_0 = const()[name = tensor("op_12302_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_12302_cast = einsum(equation = var_12302_equation_0, values = (var_12162_cast, var_12279_cast))[name = tensor("op_12302_cast")]; + tensor var_12304_equation_0 = const()[name = tensor("op_12304_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_12304_cast = einsum(equation = var_12304_equation_0, values = (var_12162_cast, var_12280_cast))[name = tensor("op_12304_cast")]; + tensor var_12306_equation_0 = const()[name = tensor("op_12306_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_12306_cast = einsum(equation = var_12306_equation_0, values = (var_12166_cast, var_12281_cast))[name = tensor("op_12306_cast")]; + tensor var_12308_equation_0 = const()[name = tensor("op_12308_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_12308_cast = einsum(equation = var_12308_equation_0, values = (var_12166_cast, var_12282_cast))[name = tensor("op_12308_cast")]; + tensor var_12310_equation_0 = const()[name = tensor("op_12310_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_12310_cast = einsum(equation = var_12310_equation_0, values = (var_12170_cast, var_12283_cast))[name = tensor("op_12310_cast")]; + tensor var_12312_equation_0 = const()[name = tensor("op_12312_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_12312_cast = einsum(equation = var_12312_equation_0, values = (var_12170_cast, var_12284_cast))[name = tensor("op_12312_cast")]; + tensor var_12314_equation_0 = const()[name = tensor("op_12314_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_12314_cast = einsum(equation = var_12314_equation_0, values = (var_12174_cast, var_12285_cast))[name = tensor("op_12314_cast")]; + tensor var_12316_equation_0 = const()[name = tensor("op_12316_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_12316_cast = einsum(equation = var_12316_equation_0, values = (var_12174_cast, var_12286_cast))[name = tensor("op_12316_cast")]; + tensor var_12318_equation_0 = const()[name = tensor("op_12318_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_12318_cast = einsum(equation = var_12318_equation_0, values = (var_12178_cast, var_12287_cast))[name = tensor("op_12318_cast")]; + tensor var_12320_equation_0 = const()[name = tensor("op_12320_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_12320_cast = einsum(equation = var_12320_equation_0, values = (var_12178_cast, var_12288_cast))[name = tensor("op_12320_cast")]; + tensor var_12322_equation_0 = const()[name = tensor("op_12322_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_12322_cast = einsum(equation = var_12322_equation_0, values = (var_12182_cast, var_12289_cast))[name = tensor("op_12322_cast")]; + tensor var_12324_equation_0 = const()[name = tensor("op_12324_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_12324_cast = einsum(equation = var_12324_equation_0, values = (var_12182_cast, var_12290_cast))[name = tensor("op_12324_cast")]; + tensor var_12326_equation_0 = const()[name = tensor("op_12326_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_12326_cast = einsum(equation = var_12326_equation_0, values = (var_12186_cast, var_12291_cast))[name = tensor("op_12326_cast")]; + tensor var_12328_equation_0 = const()[name = tensor("op_12328_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_12328_cast = einsum(equation = var_12328_equation_0, values = (var_12186_cast, var_12292_cast))[name = tensor("op_12328_cast")]; + tensor var_12330_equation_0 = const()[name = tensor("op_12330_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_12330_cast = einsum(equation = var_12330_equation_0, values = (var_12190_cast, var_12293_cast))[name = tensor("op_12330_cast")]; + tensor var_12332_equation_0 = const()[name = tensor("op_12332_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_12332_cast = einsum(equation = var_12332_equation_0, values = (var_12190_cast, var_12294_cast))[name = tensor("op_12332_cast")]; + tensor var_12334_equation_0 = const()[name = tensor("op_12334_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_12334_cast = einsum(equation = var_12334_equation_0, values = (var_12194_cast, var_12295_cast))[name = tensor("op_12334_cast")]; + tensor var_12336_equation_0 = const()[name = tensor("op_12336_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_12336_cast = einsum(equation = var_12336_equation_0, values = (var_12194_cast, var_12296_cast))[name = tensor("op_12336_cast")]; + tensor var_12338_interleave_0 = const()[name = tensor("op_12338_interleave_0"), val = tensor(false)]; + tensor var_12338_cast = concat(axis = var_10234, interleave = var_12338_interleave_0, values = (var_12298_cast, var_12300_cast))[name = tensor("op_12338_cast")]; + tensor var_12340_interleave_0 = const()[name = tensor("op_12340_interleave_0"), val = tensor(false)]; + tensor var_12340_cast = concat(axis = var_10234, interleave = var_12340_interleave_0, values = (var_12302_cast, var_12304_cast))[name = tensor("op_12340_cast")]; + tensor var_12342_interleave_0 = const()[name = tensor("op_12342_interleave_0"), val = tensor(false)]; + tensor var_12342_cast = concat(axis = var_10234, interleave = var_12342_interleave_0, values = (var_12306_cast, var_12308_cast))[name = tensor("op_12342_cast")]; + tensor var_12344_interleave_0 = const()[name = tensor("op_12344_interleave_0"), val = tensor(false)]; + tensor var_12344_cast = concat(axis = var_10234, interleave = var_12344_interleave_0, values = (var_12310_cast, var_12312_cast))[name = tensor("op_12344_cast")]; + tensor var_12346_interleave_0 = const()[name = tensor("op_12346_interleave_0"), val = tensor(false)]; + tensor var_12346_cast = concat(axis = var_10234, interleave = var_12346_interleave_0, values = (var_12314_cast, var_12316_cast))[name = tensor("op_12346_cast")]; + tensor var_12348_interleave_0 = const()[name = tensor("op_12348_interleave_0"), val = tensor(false)]; + tensor var_12348_cast = concat(axis = var_10234, interleave = var_12348_interleave_0, values = (var_12318_cast, var_12320_cast))[name = tensor("op_12348_cast")]; + tensor var_12350_interleave_0 = const()[name = tensor("op_12350_interleave_0"), val = tensor(false)]; + tensor var_12350_cast = concat(axis = var_10234, interleave = var_12350_interleave_0, values = (var_12322_cast, var_12324_cast))[name = tensor("op_12350_cast")]; + tensor var_12352_interleave_0 = const()[name = tensor("op_12352_interleave_0"), val = tensor(false)]; + tensor var_12352_cast = concat(axis = var_10234, interleave = var_12352_interleave_0, values = (var_12326_cast, var_12328_cast))[name = tensor("op_12352_cast")]; + tensor var_12354_interleave_0 = const()[name = tensor("op_12354_interleave_0"), val = tensor(false)]; + tensor var_12354_cast = concat(axis = var_10234, interleave = var_12354_interleave_0, values = (var_12330_cast, var_12332_cast))[name = tensor("op_12354_cast")]; + tensor var_12356_interleave_0 = const()[name = tensor("op_12356_interleave_0"), val = tensor(false)]; + tensor var_12356_cast = concat(axis = var_10234, interleave = var_12356_interleave_0, values = (var_12334_cast, var_12336_cast))[name = tensor("op_12356_cast")]; + tensor input_435_interleave_0 = const()[name = tensor("input_435_interleave_0"), val = tensor(false)]; + tensor input_435_cast = concat(axis = var_10257, interleave = input_435_interleave_0, values = (var_12338_cast, var_12340_cast, var_12342_cast, var_12344_cast, var_12346_cast, var_12348_cast, var_12350_cast, var_12352_cast, var_12354_cast, var_12356_cast))[name = tensor("input_435_cast")]; + tensor var_12362 = const()[name = tensor("op_12362"), val = tensor([1, 1])]; + tensor var_12364 = const()[name = tensor("op_12364"), val = tensor([1, 1])]; + tensor var_12366_pad_type_0 = const()[name = tensor("op_12366_pad_type_0"), val = tensor("custom")]; + tensor var_12366_pad_0 = const()[name = tensor("op_12366_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_2_attentions_2_transformer_blocks_0_attn1_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(626746048))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(627053312))), name = tensor("up_blocks_2_attentions_2_transformer_blocks_0_attn1_to_out_0_weight_to_fp16_palettized"), shape = tensor([640, 640, 1, 1])]; + tensor up_blocks_2_attentions_2_transformer_blocks_0_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("up_blocks_2_attentions_2_transformer_blocks_0_attn1_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(627053504)))]; + tensor var_12366_cast = conv(bias = up_blocks_2_attentions_2_transformer_blocks_0_attn1_to_out_0_bias_to_fp16, dilations = var_12364, groups = var_10257, pad = var_12366_pad_0, pad_type = var_12366_pad_type_0, strides = var_12362, weight = up_blocks_2_attentions_2_transformer_blocks_0_attn1_to_out_0_weight_to_fp16_palettized, x = input_435_cast)[name = tensor("op_12366_cast")]; + tensor inputs_75_cast = add(x = var_12366_cast, y = inputs_73_cast)[name = tensor("inputs_75_cast")]; + tensor var_12370 = const()[name = tensor("op_12370"), val = tensor([1])]; + tensor channels_mean_75_cast = reduce_mean(axes = var_12370, keep_dims = var_10252, x = inputs_75_cast)[name = tensor("channels_mean_75_cast")]; + tensor zero_mean_75_cast = sub(x = inputs_75_cast, y = channels_mean_75_cast)[name = tensor("zero_mean_75_cast")]; + tensor zero_mean_sq_75_cast = mul(x = zero_mean_75_cast, y = zero_mean_75_cast)[name = tensor("zero_mean_sq_75_cast")]; + tensor var_12374 = const()[name = tensor("op_12374"), val = tensor([1])]; + tensor var_12375_cast = reduce_mean(axes = var_12374, keep_dims = var_10252, x = zero_mean_sq_75_cast)[name = tensor("op_12375_cast")]; + tensor var_12376_to_fp16 = const()[name = tensor("op_12376_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_12377_cast = add(x = var_12375_cast, y = var_12376_to_fp16)[name = tensor("op_12377_cast")]; + tensor denom_75_epsilon_0_to_fp16 = const()[name = tensor("denom_75_epsilon_0_to_fp16"), val = tensor(0x1p-24)]; + tensor denom_75_cast = rsqrt(epsilon = denom_75_epsilon_0_to_fp16, x = var_12377_cast)[name = tensor("denom_75_cast")]; + tensor out_75_cast = mul(x = zero_mean_75_cast, y = denom_75_cast)[name = tensor("out_75_cast")]; + tensor var_12381_to_fp16 = const()[name = tensor("op_12381_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(627054848)))]; + tensor var_12382_cast = add(x = out_75_cast, y = var_12381_to_fp16)[name = tensor("op_12382_cast")]; + tensor var_12384_to_fp16 = const()[name = tensor("op_12384_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(627056192)))]; + tensor hidden_states_269_cast = mul(x = var_12382_cast, y = var_12384_to_fp16)[name = tensor("hidden_states_269_cast")]; + tensor var_12391 = const()[name = tensor("op_12391"), val = tensor([1, 1])]; + tensor var_12393 = const()[name = tensor("op_12393"), val = tensor([1, 1])]; + tensor q_51_pad_type_0 = const()[name = tensor("q_51_pad_type_0"), val = tensor("custom")]; + tensor q_51_pad_0 = const()[name = tensor("q_51_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_2_attentions_2_transformer_blocks_0_attn2_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(627057536))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(627364800))), name = tensor("up_blocks_2_attentions_2_transformer_blocks_0_attn2_to_q_weight_to_fp16_palettized"), shape = tensor([640, 640, 1, 1])]; + tensor q_51_cast = conv(dilations = var_12393, groups = var_10257, pad = q_51_pad_0, pad_type = q_51_pad_type_0, strides = var_12391, weight = up_blocks_2_attentions_2_transformer_blocks_0_attn2_to_q_weight_to_fp16_palettized, x = hidden_states_269_cast)[name = tensor("q_51_cast")]; + tensor var_12397 = const()[name = tensor("op_12397"), val = tensor([1, 1])]; + tensor var_12399 = const()[name = tensor("op_12399"), val = tensor([1, 1])]; + tensor k_101_pad_type_0 = const()[name = tensor("k_101_pad_type_0"), val = tensor("custom")]; + tensor k_101_pad_0 = const()[name = tensor("k_101_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_2_attentions_2_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(627364992))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(627856576))), name = tensor("up_blocks_2_attentions_2_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized"), shape = tensor([640, 1024, 1, 1])]; + tensor k_101_cast = conv(dilations = var_12399, groups = var_10257, pad = k_101_pad_0, pad_type = k_101_pad_type_0, strides = var_12397, weight = up_blocks_2_attentions_2_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor("k_101_cast")]; + tensor var_12403 = const()[name = tensor("op_12403"), val = tensor([1, 1])]; + tensor var_12405 = const()[name = tensor("op_12405"), val = tensor([1, 1])]; + tensor v_51_pad_type_0 = const()[name = tensor("v_51_pad_type_0"), val = tensor("custom")]; + tensor v_51_pad_0 = const()[name = tensor("v_51_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_2_attentions_2_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(627856768))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(628348352))), name = tensor("up_blocks_2_attentions_2_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized"), shape = tensor([640, 1024, 1, 1])]; + tensor v_51_cast = conv(dilations = var_12405, groups = var_10257, pad = v_51_pad_0, pad_type = v_51_pad_type_0, strides = var_12403, weight = up_blocks_2_attentions_2_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor("v_51_cast")]; + tensor var_12409_begin_0 = const()[name = tensor("op_12409_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_12409_end_0 = const()[name = tensor("op_12409_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_12409_end_mask_0 = const()[name = tensor("op_12409_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12409_cast = slice_by_index(begin = var_12409_begin_0, end = var_12409_end_0, end_mask = var_12409_end_mask_0, x = q_51_cast)[name = tensor("op_12409_cast")]; + tensor var_12413_begin_0 = const()[name = tensor("op_12413_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_12413_end_0 = const()[name = tensor("op_12413_end_0"), val = tensor([2, 128, 1, 1024])]; + tensor var_12413_end_mask_0 = const()[name = tensor("op_12413_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12413_cast = slice_by_index(begin = var_12413_begin_0, end = var_12413_end_0, end_mask = var_12413_end_mask_0, x = q_51_cast)[name = tensor("op_12413_cast")]; + tensor var_12417_begin_0 = const()[name = tensor("op_12417_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_12417_end_0 = const()[name = tensor("op_12417_end_0"), val = tensor([2, 192, 1, 1024])]; + tensor var_12417_end_mask_0 = const()[name = tensor("op_12417_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12417_cast = slice_by_index(begin = var_12417_begin_0, end = var_12417_end_0, end_mask = var_12417_end_mask_0, x = q_51_cast)[name = tensor("op_12417_cast")]; + tensor var_12421_begin_0 = const()[name = tensor("op_12421_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_12421_end_0 = const()[name = tensor("op_12421_end_0"), val = tensor([2, 256, 1, 1024])]; + tensor var_12421_end_mask_0 = const()[name = tensor("op_12421_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12421_cast = slice_by_index(begin = var_12421_begin_0, end = var_12421_end_0, end_mask = var_12421_end_mask_0, x = q_51_cast)[name = tensor("op_12421_cast")]; + tensor var_12425_begin_0 = const()[name = tensor("op_12425_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_12425_end_0 = const()[name = tensor("op_12425_end_0"), val = tensor([2, 320, 1, 1024])]; + tensor var_12425_end_mask_0 = const()[name = tensor("op_12425_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12425_cast = slice_by_index(begin = var_12425_begin_0, end = var_12425_end_0, end_mask = var_12425_end_mask_0, x = q_51_cast)[name = tensor("op_12425_cast")]; + tensor var_12429_begin_0 = const()[name = tensor("op_12429_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_12429_end_0 = const()[name = tensor("op_12429_end_0"), val = tensor([2, 384, 1, 1024])]; + tensor var_12429_end_mask_0 = const()[name = tensor("op_12429_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12429_cast = slice_by_index(begin = var_12429_begin_0, end = var_12429_end_0, end_mask = var_12429_end_mask_0, x = q_51_cast)[name = tensor("op_12429_cast")]; + tensor var_12433_begin_0 = const()[name = tensor("op_12433_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_12433_end_0 = const()[name = tensor("op_12433_end_0"), val = tensor([2, 448, 1, 1024])]; + tensor var_12433_end_mask_0 = const()[name = tensor("op_12433_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12433_cast = slice_by_index(begin = var_12433_begin_0, end = var_12433_end_0, end_mask = var_12433_end_mask_0, x = q_51_cast)[name = tensor("op_12433_cast")]; + tensor var_12437_begin_0 = const()[name = tensor("op_12437_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_12437_end_0 = const()[name = tensor("op_12437_end_0"), val = tensor([2, 512, 1, 1024])]; + tensor var_12437_end_mask_0 = const()[name = tensor("op_12437_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12437_cast = slice_by_index(begin = var_12437_begin_0, end = var_12437_end_0, end_mask = var_12437_end_mask_0, x = q_51_cast)[name = tensor("op_12437_cast")]; + tensor var_12441_begin_0 = const()[name = tensor("op_12441_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_12441_end_0 = const()[name = tensor("op_12441_end_0"), val = tensor([2, 576, 1, 1024])]; + tensor var_12441_end_mask_0 = const()[name = tensor("op_12441_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12441_cast = slice_by_index(begin = var_12441_begin_0, end = var_12441_end_0, end_mask = var_12441_end_mask_0, x = q_51_cast)[name = tensor("op_12441_cast")]; + tensor var_12445_begin_0 = const()[name = tensor("op_12445_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_12445_end_0 = const()[name = tensor("op_12445_end_0"), val = tensor([2, 640, 1, 1024])]; + tensor var_12445_end_mask_0 = const()[name = tensor("op_12445_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12445_cast = slice_by_index(begin = var_12445_begin_0, end = var_12445_end_0, end_mask = var_12445_end_mask_0, x = q_51_cast)[name = tensor("op_12445_cast")]; + tensor var_12448_begin_0 = const()[name = tensor("op_12448_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_12448_end_0 = const()[name = tensor("op_12448_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_12448_end_mask_0 = const()[name = tensor("op_12448_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12448_cast = slice_by_index(begin = var_12448_begin_0, end = var_12448_end_0, end_mask = var_12448_end_mask_0, x = var_12409_cast)[name = tensor("op_12448_cast")]; + tensor var_12449_begin_0 = const()[name = tensor("op_12449_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_12449_end_0 = const()[name = tensor("op_12449_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_12449_end_mask_0 = const()[name = tensor("op_12449_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12449_cast = slice_by_index(begin = var_12449_begin_0, end = var_12449_end_0, end_mask = var_12449_end_mask_0, x = var_12409_cast)[name = tensor("op_12449_cast")]; + tensor var_12450_begin_0 = const()[name = tensor("op_12450_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_12450_end_0 = const()[name = tensor("op_12450_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_12450_end_mask_0 = const()[name = tensor("op_12450_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12450_cast = slice_by_index(begin = var_12450_begin_0, end = var_12450_end_0, end_mask = var_12450_end_mask_0, x = var_12413_cast)[name = tensor("op_12450_cast")]; + tensor var_12451_begin_0 = const()[name = tensor("op_12451_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_12451_end_0 = const()[name = tensor("op_12451_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_12451_end_mask_0 = const()[name = tensor("op_12451_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12451_cast = slice_by_index(begin = var_12451_begin_0, end = var_12451_end_0, end_mask = var_12451_end_mask_0, x = var_12413_cast)[name = tensor("op_12451_cast")]; + tensor var_12452_begin_0 = const()[name = tensor("op_12452_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_12452_end_0 = const()[name = tensor("op_12452_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_12452_end_mask_0 = const()[name = tensor("op_12452_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12452_cast = slice_by_index(begin = var_12452_begin_0, end = var_12452_end_0, end_mask = var_12452_end_mask_0, x = var_12417_cast)[name = tensor("op_12452_cast")]; + tensor var_12453_begin_0 = const()[name = tensor("op_12453_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_12453_end_0 = const()[name = tensor("op_12453_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_12453_end_mask_0 = const()[name = tensor("op_12453_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12453_cast = slice_by_index(begin = var_12453_begin_0, end = var_12453_end_0, end_mask = var_12453_end_mask_0, x = var_12417_cast)[name = tensor("op_12453_cast")]; + tensor var_12454_begin_0 = const()[name = tensor("op_12454_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_12454_end_0 = const()[name = tensor("op_12454_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_12454_end_mask_0 = const()[name = tensor("op_12454_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12454_cast = slice_by_index(begin = var_12454_begin_0, end = var_12454_end_0, end_mask = var_12454_end_mask_0, x = var_12421_cast)[name = tensor("op_12454_cast")]; + tensor var_12455_begin_0 = const()[name = tensor("op_12455_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_12455_end_0 = const()[name = tensor("op_12455_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_12455_end_mask_0 = const()[name = tensor("op_12455_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12455_cast = slice_by_index(begin = var_12455_begin_0, end = var_12455_end_0, end_mask = var_12455_end_mask_0, x = var_12421_cast)[name = tensor("op_12455_cast")]; + tensor var_12456_begin_0 = const()[name = tensor("op_12456_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_12456_end_0 = const()[name = tensor("op_12456_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_12456_end_mask_0 = const()[name = tensor("op_12456_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12456_cast = slice_by_index(begin = var_12456_begin_0, end = var_12456_end_0, end_mask = var_12456_end_mask_0, x = var_12425_cast)[name = tensor("op_12456_cast")]; + tensor var_12457_begin_0 = const()[name = tensor("op_12457_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_12457_end_0 = const()[name = tensor("op_12457_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_12457_end_mask_0 = const()[name = tensor("op_12457_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12457_cast = slice_by_index(begin = var_12457_begin_0, end = var_12457_end_0, end_mask = var_12457_end_mask_0, x = var_12425_cast)[name = tensor("op_12457_cast")]; + tensor var_12458_begin_0 = const()[name = tensor("op_12458_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_12458_end_0 = const()[name = tensor("op_12458_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_12458_end_mask_0 = const()[name = tensor("op_12458_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12458_cast = slice_by_index(begin = var_12458_begin_0, end = var_12458_end_0, end_mask = var_12458_end_mask_0, x = var_12429_cast)[name = tensor("op_12458_cast")]; + tensor var_12459_begin_0 = const()[name = tensor("op_12459_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_12459_end_0 = const()[name = tensor("op_12459_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_12459_end_mask_0 = const()[name = tensor("op_12459_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12459_cast = slice_by_index(begin = var_12459_begin_0, end = var_12459_end_0, end_mask = var_12459_end_mask_0, x = var_12429_cast)[name = tensor("op_12459_cast")]; + tensor var_12460_begin_0 = const()[name = tensor("op_12460_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_12460_end_0 = const()[name = tensor("op_12460_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_12460_end_mask_0 = const()[name = tensor("op_12460_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12460_cast = slice_by_index(begin = var_12460_begin_0, end = var_12460_end_0, end_mask = var_12460_end_mask_0, x = var_12433_cast)[name = tensor("op_12460_cast")]; + tensor var_12461_begin_0 = const()[name = tensor("op_12461_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_12461_end_0 = const()[name = tensor("op_12461_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_12461_end_mask_0 = const()[name = tensor("op_12461_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12461_cast = slice_by_index(begin = var_12461_begin_0, end = var_12461_end_0, end_mask = var_12461_end_mask_0, x = var_12433_cast)[name = tensor("op_12461_cast")]; + tensor var_12462_begin_0 = const()[name = tensor("op_12462_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_12462_end_0 = const()[name = tensor("op_12462_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_12462_end_mask_0 = const()[name = tensor("op_12462_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12462_cast = slice_by_index(begin = var_12462_begin_0, end = var_12462_end_0, end_mask = var_12462_end_mask_0, x = var_12437_cast)[name = tensor("op_12462_cast")]; + tensor var_12463_begin_0 = const()[name = tensor("op_12463_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_12463_end_0 = const()[name = tensor("op_12463_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_12463_end_mask_0 = const()[name = tensor("op_12463_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12463_cast = slice_by_index(begin = var_12463_begin_0, end = var_12463_end_0, end_mask = var_12463_end_mask_0, x = var_12437_cast)[name = tensor("op_12463_cast")]; + tensor var_12464_begin_0 = const()[name = tensor("op_12464_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_12464_end_0 = const()[name = tensor("op_12464_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_12464_end_mask_0 = const()[name = tensor("op_12464_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12464_cast = slice_by_index(begin = var_12464_begin_0, end = var_12464_end_0, end_mask = var_12464_end_mask_0, x = var_12441_cast)[name = tensor("op_12464_cast")]; + tensor var_12465_begin_0 = const()[name = tensor("op_12465_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_12465_end_0 = const()[name = tensor("op_12465_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_12465_end_mask_0 = const()[name = tensor("op_12465_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12465_cast = slice_by_index(begin = var_12465_begin_0, end = var_12465_end_0, end_mask = var_12465_end_mask_0, x = var_12441_cast)[name = tensor("op_12465_cast")]; + tensor var_12466_begin_0 = const()[name = tensor("op_12466_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_12466_end_0 = const()[name = tensor("op_12466_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_12466_end_mask_0 = const()[name = tensor("op_12466_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12466_cast = slice_by_index(begin = var_12466_begin_0, end = var_12466_end_0, end_mask = var_12466_end_mask_0, x = var_12445_cast)[name = tensor("op_12466_cast")]; + tensor var_12467_begin_0 = const()[name = tensor("op_12467_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_12467_end_0 = const()[name = tensor("op_12467_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_12467_end_mask_0 = const()[name = tensor("op_12467_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12467_cast = slice_by_index(begin = var_12467_begin_0, end = var_12467_end_0, end_mask = var_12467_end_mask_0, x = var_12445_cast)[name = tensor("op_12467_cast")]; + tensor k_103_perm_0 = const()[name = tensor("k_103_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_12472_begin_0 = const()[name = tensor("op_12472_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_12472_end_0 = const()[name = tensor("op_12472_end_0"), val = tensor([2, 77, 1, 64])]; + tensor var_12472_end_mask_0 = const()[name = tensor("op_12472_end_mask_0"), val = tensor([true, true, true, false])]; + tensor transpose_6 = transpose(perm = k_103_perm_0, x = k_101_cast)[name = tensor("transpose_6")]; + tensor var_12472_cast = slice_by_index(begin = var_12472_begin_0, end = var_12472_end_0, end_mask = var_12472_end_mask_0, x = transpose_6)[name = tensor("op_12472_cast")]; + tensor var_12476_begin_0 = const()[name = tensor("op_12476_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_12476_end_0 = const()[name = tensor("op_12476_end_0"), val = tensor([2, 77, 1, 128])]; + tensor var_12476_end_mask_0 = const()[name = tensor("op_12476_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12476_cast = slice_by_index(begin = var_12476_begin_0, end = var_12476_end_0, end_mask = var_12476_end_mask_0, x = transpose_6)[name = tensor("op_12476_cast")]; + tensor var_12480_begin_0 = const()[name = tensor("op_12480_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_12480_end_0 = const()[name = tensor("op_12480_end_0"), val = tensor([2, 77, 1, 192])]; + tensor var_12480_end_mask_0 = const()[name = tensor("op_12480_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12480_cast = slice_by_index(begin = var_12480_begin_0, end = var_12480_end_0, end_mask = var_12480_end_mask_0, x = transpose_6)[name = tensor("op_12480_cast")]; + tensor var_12484_begin_0 = const()[name = tensor("op_12484_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_12484_end_0 = const()[name = tensor("op_12484_end_0"), val = tensor([2, 77, 1, 256])]; + tensor var_12484_end_mask_0 = const()[name = tensor("op_12484_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12484_cast = slice_by_index(begin = var_12484_begin_0, end = var_12484_end_0, end_mask = var_12484_end_mask_0, x = transpose_6)[name = tensor("op_12484_cast")]; + tensor var_12488_begin_0 = const()[name = tensor("op_12488_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_12488_end_0 = const()[name = tensor("op_12488_end_0"), val = tensor([2, 77, 1, 320])]; + tensor var_12488_end_mask_0 = const()[name = tensor("op_12488_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12488_cast = slice_by_index(begin = var_12488_begin_0, end = var_12488_end_0, end_mask = var_12488_end_mask_0, x = transpose_6)[name = tensor("op_12488_cast")]; + tensor var_12492_begin_0 = const()[name = tensor("op_12492_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_12492_end_0 = const()[name = tensor("op_12492_end_0"), val = tensor([2, 77, 1, 384])]; + tensor var_12492_end_mask_0 = const()[name = tensor("op_12492_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12492_cast = slice_by_index(begin = var_12492_begin_0, end = var_12492_end_0, end_mask = var_12492_end_mask_0, x = transpose_6)[name = tensor("op_12492_cast")]; + tensor var_12496_begin_0 = const()[name = tensor("op_12496_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_12496_end_0 = const()[name = tensor("op_12496_end_0"), val = tensor([2, 77, 1, 448])]; + tensor var_12496_end_mask_0 = const()[name = tensor("op_12496_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12496_cast = slice_by_index(begin = var_12496_begin_0, end = var_12496_end_0, end_mask = var_12496_end_mask_0, x = transpose_6)[name = tensor("op_12496_cast")]; + tensor var_12500_begin_0 = const()[name = tensor("op_12500_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_12500_end_0 = const()[name = tensor("op_12500_end_0"), val = tensor([2, 77, 1, 512])]; + tensor var_12500_end_mask_0 = const()[name = tensor("op_12500_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12500_cast = slice_by_index(begin = var_12500_begin_0, end = var_12500_end_0, end_mask = var_12500_end_mask_0, x = transpose_6)[name = tensor("op_12500_cast")]; + tensor var_12504_begin_0 = const()[name = tensor("op_12504_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_12504_end_0 = const()[name = tensor("op_12504_end_0"), val = tensor([2, 77, 1, 576])]; + tensor var_12504_end_mask_0 = const()[name = tensor("op_12504_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12504_cast = slice_by_index(begin = var_12504_begin_0, end = var_12504_end_0, end_mask = var_12504_end_mask_0, x = transpose_6)[name = tensor("op_12504_cast")]; + tensor var_12508_begin_0 = const()[name = tensor("op_12508_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_12508_end_0 = const()[name = tensor("op_12508_end_0"), val = tensor([2, 77, 1, 640])]; + tensor var_12508_end_mask_0 = const()[name = tensor("op_12508_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12508_cast = slice_by_index(begin = var_12508_begin_0, end = var_12508_end_0, end_mask = var_12508_end_mask_0, x = transpose_6)[name = tensor("op_12508_cast")]; + tensor var_12510_begin_0 = const()[name = tensor("op_12510_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_12510_end_0 = const()[name = tensor("op_12510_end_0"), val = tensor([2, 64, 1, 77])]; + tensor var_12510_end_mask_0 = const()[name = tensor("op_12510_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12510_cast = slice_by_index(begin = var_12510_begin_0, end = var_12510_end_0, end_mask = var_12510_end_mask_0, x = v_51_cast)[name = tensor("op_12510_cast")]; + tensor var_12514_begin_0 = const()[name = tensor("op_12514_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_12514_end_0 = const()[name = tensor("op_12514_end_0"), val = tensor([2, 128, 1, 77])]; + tensor var_12514_end_mask_0 = const()[name = tensor("op_12514_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12514_cast = slice_by_index(begin = var_12514_begin_0, end = var_12514_end_0, end_mask = var_12514_end_mask_0, x = v_51_cast)[name = tensor("op_12514_cast")]; + tensor var_12518_begin_0 = const()[name = tensor("op_12518_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_12518_end_0 = const()[name = tensor("op_12518_end_0"), val = tensor([2, 192, 1, 77])]; + tensor var_12518_end_mask_0 = const()[name = tensor("op_12518_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12518_cast = slice_by_index(begin = var_12518_begin_0, end = var_12518_end_0, end_mask = var_12518_end_mask_0, x = v_51_cast)[name = tensor("op_12518_cast")]; + tensor var_12522_begin_0 = const()[name = tensor("op_12522_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_12522_end_0 = const()[name = tensor("op_12522_end_0"), val = tensor([2, 256, 1, 77])]; + tensor var_12522_end_mask_0 = const()[name = tensor("op_12522_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12522_cast = slice_by_index(begin = var_12522_begin_0, end = var_12522_end_0, end_mask = var_12522_end_mask_0, x = v_51_cast)[name = tensor("op_12522_cast")]; + tensor var_12526_begin_0 = const()[name = tensor("op_12526_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_12526_end_0 = const()[name = tensor("op_12526_end_0"), val = tensor([2, 320, 1, 77])]; + tensor var_12526_end_mask_0 = const()[name = tensor("op_12526_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12526_cast = slice_by_index(begin = var_12526_begin_0, end = var_12526_end_0, end_mask = var_12526_end_mask_0, x = v_51_cast)[name = tensor("op_12526_cast")]; + tensor var_12530_begin_0 = const()[name = tensor("op_12530_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_12530_end_0 = const()[name = tensor("op_12530_end_0"), val = tensor([2, 384, 1, 77])]; + tensor var_12530_end_mask_0 = const()[name = tensor("op_12530_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12530_cast = slice_by_index(begin = var_12530_begin_0, end = var_12530_end_0, end_mask = var_12530_end_mask_0, x = v_51_cast)[name = tensor("op_12530_cast")]; + tensor var_12534_begin_0 = const()[name = tensor("op_12534_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_12534_end_0 = const()[name = tensor("op_12534_end_0"), val = tensor([2, 448, 1, 77])]; + tensor var_12534_end_mask_0 = const()[name = tensor("op_12534_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12534_cast = slice_by_index(begin = var_12534_begin_0, end = var_12534_end_0, end_mask = var_12534_end_mask_0, x = v_51_cast)[name = tensor("op_12534_cast")]; + tensor var_12538_begin_0 = const()[name = tensor("op_12538_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_12538_end_0 = const()[name = tensor("op_12538_end_0"), val = tensor([2, 512, 1, 77])]; + tensor var_12538_end_mask_0 = const()[name = tensor("op_12538_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12538_cast = slice_by_index(begin = var_12538_begin_0, end = var_12538_end_0, end_mask = var_12538_end_mask_0, x = v_51_cast)[name = tensor("op_12538_cast")]; + tensor var_12542_begin_0 = const()[name = tensor("op_12542_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_12542_end_0 = const()[name = tensor("op_12542_end_0"), val = tensor([2, 576, 1, 77])]; + tensor var_12542_end_mask_0 = const()[name = tensor("op_12542_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12542_cast = slice_by_index(begin = var_12542_begin_0, end = var_12542_end_0, end_mask = var_12542_end_mask_0, x = v_51_cast)[name = tensor("op_12542_cast")]; + tensor var_12546_begin_0 = const()[name = tensor("op_12546_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_12546_end_0 = const()[name = tensor("op_12546_end_0"), val = tensor([2, 640, 1, 77])]; + tensor var_12546_end_mask_0 = const()[name = tensor("op_12546_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12546_cast = slice_by_index(begin = var_12546_begin_0, end = var_12546_end_0, end_mask = var_12546_end_mask_0, x = v_51_cast)[name = tensor("op_12546_cast")]; + tensor var_12550_equation_0 = const()[name = tensor("op_12550_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12550_cast = einsum(equation = var_12550_equation_0, values = (var_12472_cast, var_12448_cast))[name = tensor("op_12550_cast")]; + tensor var_12551_to_fp16 = const()[name = tensor("op_12551_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_681_cast = mul(x = var_12550_cast, y = var_12551_to_fp16)[name = tensor("aw_chunk_681_cast")]; + tensor var_12554_equation_0 = const()[name = tensor("op_12554_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12554_cast = einsum(equation = var_12554_equation_0, values = (var_12472_cast, var_12449_cast))[name = tensor("op_12554_cast")]; + tensor var_12555_to_fp16 = const()[name = tensor("op_12555_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_683_cast = mul(x = var_12554_cast, y = var_12555_to_fp16)[name = tensor("aw_chunk_683_cast")]; + tensor var_12558_equation_0 = const()[name = tensor("op_12558_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12558_cast = einsum(equation = var_12558_equation_0, values = (var_12476_cast, var_12450_cast))[name = tensor("op_12558_cast")]; + tensor var_12559_to_fp16 = const()[name = tensor("op_12559_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_685_cast = mul(x = var_12558_cast, y = var_12559_to_fp16)[name = tensor("aw_chunk_685_cast")]; + tensor var_12562_equation_0 = const()[name = tensor("op_12562_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12562_cast = einsum(equation = var_12562_equation_0, values = (var_12476_cast, var_12451_cast))[name = tensor("op_12562_cast")]; + tensor var_12563_to_fp16 = const()[name = tensor("op_12563_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_687_cast = mul(x = var_12562_cast, y = var_12563_to_fp16)[name = tensor("aw_chunk_687_cast")]; + tensor var_12566_equation_0 = const()[name = tensor("op_12566_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12566_cast = einsum(equation = var_12566_equation_0, values = (var_12480_cast, var_12452_cast))[name = tensor("op_12566_cast")]; + tensor var_12567_to_fp16 = const()[name = tensor("op_12567_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_689_cast = mul(x = var_12566_cast, y = var_12567_to_fp16)[name = tensor("aw_chunk_689_cast")]; + tensor var_12570_equation_0 = const()[name = tensor("op_12570_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12570_cast = einsum(equation = var_12570_equation_0, values = (var_12480_cast, var_12453_cast))[name = tensor("op_12570_cast")]; + tensor var_12571_to_fp16 = const()[name = tensor("op_12571_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_691_cast = mul(x = var_12570_cast, y = var_12571_to_fp16)[name = tensor("aw_chunk_691_cast")]; + tensor var_12574_equation_0 = const()[name = tensor("op_12574_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12574_cast = einsum(equation = var_12574_equation_0, values = (var_12484_cast, var_12454_cast))[name = tensor("op_12574_cast")]; + tensor var_12575_to_fp16 = const()[name = tensor("op_12575_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_693_cast = mul(x = var_12574_cast, y = var_12575_to_fp16)[name = tensor("aw_chunk_693_cast")]; + tensor var_12578_equation_0 = const()[name = tensor("op_12578_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12578_cast = einsum(equation = var_12578_equation_0, values = (var_12484_cast, var_12455_cast))[name = tensor("op_12578_cast")]; + tensor var_12579_to_fp16 = const()[name = tensor("op_12579_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_695_cast = mul(x = var_12578_cast, y = var_12579_to_fp16)[name = tensor("aw_chunk_695_cast")]; + tensor var_12582_equation_0 = const()[name = tensor("op_12582_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12582_cast = einsum(equation = var_12582_equation_0, values = (var_12488_cast, var_12456_cast))[name = tensor("op_12582_cast")]; + tensor var_12583_to_fp16 = const()[name = tensor("op_12583_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_697_cast = mul(x = var_12582_cast, y = var_12583_to_fp16)[name = tensor("aw_chunk_697_cast")]; + tensor var_12586_equation_0 = const()[name = tensor("op_12586_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12586_cast = einsum(equation = var_12586_equation_0, values = (var_12488_cast, var_12457_cast))[name = tensor("op_12586_cast")]; + tensor var_12587_to_fp16 = const()[name = tensor("op_12587_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_699_cast = mul(x = var_12586_cast, y = var_12587_to_fp16)[name = tensor("aw_chunk_699_cast")]; + tensor var_12590_equation_0 = const()[name = tensor("op_12590_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12590_cast = einsum(equation = var_12590_equation_0, values = (var_12492_cast, var_12458_cast))[name = tensor("op_12590_cast")]; + tensor var_12591_to_fp16 = const()[name = tensor("op_12591_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_701_cast = mul(x = var_12590_cast, y = var_12591_to_fp16)[name = tensor("aw_chunk_701_cast")]; + tensor var_12594_equation_0 = const()[name = tensor("op_12594_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12594_cast = einsum(equation = var_12594_equation_0, values = (var_12492_cast, var_12459_cast))[name = tensor("op_12594_cast")]; + tensor var_12595_to_fp16 = const()[name = tensor("op_12595_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_703_cast = mul(x = var_12594_cast, y = var_12595_to_fp16)[name = tensor("aw_chunk_703_cast")]; + tensor var_12598_equation_0 = const()[name = tensor("op_12598_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12598_cast = einsum(equation = var_12598_equation_0, values = (var_12496_cast, var_12460_cast))[name = tensor("op_12598_cast")]; + tensor var_12599_to_fp16 = const()[name = tensor("op_12599_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_705_cast = mul(x = var_12598_cast, y = var_12599_to_fp16)[name = tensor("aw_chunk_705_cast")]; + tensor var_12602_equation_0 = const()[name = tensor("op_12602_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12602_cast = einsum(equation = var_12602_equation_0, values = (var_12496_cast, var_12461_cast))[name = tensor("op_12602_cast")]; + tensor var_12603_to_fp16 = const()[name = tensor("op_12603_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_707_cast = mul(x = var_12602_cast, y = var_12603_to_fp16)[name = tensor("aw_chunk_707_cast")]; + tensor var_12606_equation_0 = const()[name = tensor("op_12606_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12606_cast = einsum(equation = var_12606_equation_0, values = (var_12500_cast, var_12462_cast))[name = tensor("op_12606_cast")]; + tensor var_12607_to_fp16 = const()[name = tensor("op_12607_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_709_cast = mul(x = var_12606_cast, y = var_12607_to_fp16)[name = tensor("aw_chunk_709_cast")]; + tensor var_12610_equation_0 = const()[name = tensor("op_12610_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12610_cast = einsum(equation = var_12610_equation_0, values = (var_12500_cast, var_12463_cast))[name = tensor("op_12610_cast")]; + tensor var_12611_to_fp16 = const()[name = tensor("op_12611_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_711_cast = mul(x = var_12610_cast, y = var_12611_to_fp16)[name = tensor("aw_chunk_711_cast")]; + tensor var_12614_equation_0 = const()[name = tensor("op_12614_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12614_cast = einsum(equation = var_12614_equation_0, values = (var_12504_cast, var_12464_cast))[name = tensor("op_12614_cast")]; + tensor var_12615_to_fp16 = const()[name = tensor("op_12615_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_713_cast = mul(x = var_12614_cast, y = var_12615_to_fp16)[name = tensor("aw_chunk_713_cast")]; + tensor var_12618_equation_0 = const()[name = tensor("op_12618_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12618_cast = einsum(equation = var_12618_equation_0, values = (var_12504_cast, var_12465_cast))[name = tensor("op_12618_cast")]; + tensor var_12619_to_fp16 = const()[name = tensor("op_12619_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_715_cast = mul(x = var_12618_cast, y = var_12619_to_fp16)[name = tensor("aw_chunk_715_cast")]; + tensor var_12622_equation_0 = const()[name = tensor("op_12622_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12622_cast = einsum(equation = var_12622_equation_0, values = (var_12508_cast, var_12466_cast))[name = tensor("op_12622_cast")]; + tensor var_12623_to_fp16 = const()[name = tensor("op_12623_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_717_cast = mul(x = var_12622_cast, y = var_12623_to_fp16)[name = tensor("aw_chunk_717_cast")]; + tensor var_12626_equation_0 = const()[name = tensor("op_12626_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12626_cast = einsum(equation = var_12626_equation_0, values = (var_12508_cast, var_12467_cast))[name = tensor("op_12626_cast")]; + tensor var_12627_to_fp16 = const()[name = tensor("op_12627_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_719_cast = mul(x = var_12626_cast, y = var_12627_to_fp16)[name = tensor("aw_chunk_719_cast")]; + tensor var_12629_cast = softmax(axis = var_10257, x = aw_chunk_681_cast)[name = tensor("op_12629_cast")]; + tensor var_12630_cast = softmax(axis = var_10257, x = aw_chunk_683_cast)[name = tensor("op_12630_cast")]; + tensor var_12631_cast = softmax(axis = var_10257, x = aw_chunk_685_cast)[name = tensor("op_12631_cast")]; + tensor var_12632_cast = softmax(axis = var_10257, x = aw_chunk_687_cast)[name = tensor("op_12632_cast")]; + tensor var_12633_cast = softmax(axis = var_10257, x = aw_chunk_689_cast)[name = tensor("op_12633_cast")]; + tensor var_12634_cast = softmax(axis = var_10257, x = aw_chunk_691_cast)[name = tensor("op_12634_cast")]; + tensor var_12635_cast = softmax(axis = var_10257, x = aw_chunk_693_cast)[name = tensor("op_12635_cast")]; + tensor var_12636_cast = softmax(axis = var_10257, x = aw_chunk_695_cast)[name = tensor("op_12636_cast")]; + tensor var_12637_cast = softmax(axis = var_10257, x = aw_chunk_697_cast)[name = tensor("op_12637_cast")]; + tensor var_12638_cast = softmax(axis = var_10257, x = aw_chunk_699_cast)[name = tensor("op_12638_cast")]; + tensor var_12639_cast = softmax(axis = var_10257, x = aw_chunk_701_cast)[name = tensor("op_12639_cast")]; + tensor var_12640_cast = softmax(axis = var_10257, x = aw_chunk_703_cast)[name = tensor("op_12640_cast")]; + tensor var_12641_cast = softmax(axis = var_10257, x = aw_chunk_705_cast)[name = tensor("op_12641_cast")]; + tensor var_12642_cast = softmax(axis = var_10257, x = aw_chunk_707_cast)[name = tensor("op_12642_cast")]; + tensor var_12643_cast = softmax(axis = var_10257, x = aw_chunk_709_cast)[name = tensor("op_12643_cast")]; + tensor var_12644_cast = softmax(axis = var_10257, x = aw_chunk_711_cast)[name = tensor("op_12644_cast")]; + tensor var_12645_cast = softmax(axis = var_10257, x = aw_chunk_713_cast)[name = tensor("op_12645_cast")]; + tensor var_12646_cast = softmax(axis = var_10257, x = aw_chunk_715_cast)[name = tensor("op_12646_cast")]; + tensor var_12647_cast = softmax(axis = var_10257, x = aw_chunk_717_cast)[name = tensor("op_12647_cast")]; + tensor var_12648_cast = softmax(axis = var_10257, x = aw_chunk_719_cast)[name = tensor("op_12648_cast")]; + tensor var_12650_equation_0 = const()[name = tensor("op_12650_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_12650_cast = einsum(equation = var_12650_equation_0, values = (var_12510_cast, var_12629_cast))[name = tensor("op_12650_cast")]; + tensor var_12652_equation_0 = const()[name = tensor("op_12652_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_12652_cast = einsum(equation = var_12652_equation_0, values = (var_12510_cast, var_12630_cast))[name = tensor("op_12652_cast")]; + tensor var_12654_equation_0 = const()[name = tensor("op_12654_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_12654_cast = einsum(equation = var_12654_equation_0, values = (var_12514_cast, var_12631_cast))[name = tensor("op_12654_cast")]; + tensor var_12656_equation_0 = const()[name = tensor("op_12656_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_12656_cast = einsum(equation = var_12656_equation_0, values = (var_12514_cast, var_12632_cast))[name = tensor("op_12656_cast")]; + tensor var_12658_equation_0 = const()[name = tensor("op_12658_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_12658_cast = einsum(equation = var_12658_equation_0, values = (var_12518_cast, var_12633_cast))[name = tensor("op_12658_cast")]; + tensor var_12660_equation_0 = const()[name = tensor("op_12660_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_12660_cast = einsum(equation = var_12660_equation_0, values = (var_12518_cast, var_12634_cast))[name = tensor("op_12660_cast")]; + tensor var_12662_equation_0 = const()[name = tensor("op_12662_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_12662_cast = einsum(equation = var_12662_equation_0, values = (var_12522_cast, var_12635_cast))[name = tensor("op_12662_cast")]; + tensor var_12664_equation_0 = const()[name = tensor("op_12664_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_12664_cast = einsum(equation = var_12664_equation_0, values = (var_12522_cast, var_12636_cast))[name = tensor("op_12664_cast")]; + tensor var_12666_equation_0 = const()[name = tensor("op_12666_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_12666_cast = einsum(equation = var_12666_equation_0, values = (var_12526_cast, var_12637_cast))[name = tensor("op_12666_cast")]; + tensor var_12668_equation_0 = const()[name = tensor("op_12668_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_12668_cast = einsum(equation = var_12668_equation_0, values = (var_12526_cast, var_12638_cast))[name = tensor("op_12668_cast")]; + tensor var_12670_equation_0 = const()[name = tensor("op_12670_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_12670_cast = einsum(equation = var_12670_equation_0, values = (var_12530_cast, var_12639_cast))[name = tensor("op_12670_cast")]; + tensor var_12672_equation_0 = const()[name = tensor("op_12672_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_12672_cast = einsum(equation = var_12672_equation_0, values = (var_12530_cast, var_12640_cast))[name = tensor("op_12672_cast")]; + tensor var_12674_equation_0 = const()[name = tensor("op_12674_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_12674_cast = einsum(equation = var_12674_equation_0, values = (var_12534_cast, var_12641_cast))[name = tensor("op_12674_cast")]; + tensor var_12676_equation_0 = const()[name = tensor("op_12676_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_12676_cast = einsum(equation = var_12676_equation_0, values = (var_12534_cast, var_12642_cast))[name = tensor("op_12676_cast")]; + tensor var_12678_equation_0 = const()[name = tensor("op_12678_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_12678_cast = einsum(equation = var_12678_equation_0, values = (var_12538_cast, var_12643_cast))[name = tensor("op_12678_cast")]; + tensor var_12680_equation_0 = const()[name = tensor("op_12680_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_12680_cast = einsum(equation = var_12680_equation_0, values = (var_12538_cast, var_12644_cast))[name = tensor("op_12680_cast")]; + tensor var_12682_equation_0 = const()[name = tensor("op_12682_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_12682_cast = einsum(equation = var_12682_equation_0, values = (var_12542_cast, var_12645_cast))[name = tensor("op_12682_cast")]; + tensor var_12684_equation_0 = const()[name = tensor("op_12684_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_12684_cast = einsum(equation = var_12684_equation_0, values = (var_12542_cast, var_12646_cast))[name = tensor("op_12684_cast")]; + tensor var_12686_equation_0 = const()[name = tensor("op_12686_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_12686_cast = einsum(equation = var_12686_equation_0, values = (var_12546_cast, var_12647_cast))[name = tensor("op_12686_cast")]; + tensor var_12688_equation_0 = const()[name = tensor("op_12688_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_12688_cast = einsum(equation = var_12688_equation_0, values = (var_12546_cast, var_12648_cast))[name = tensor("op_12688_cast")]; + tensor var_12690_interleave_0 = const()[name = tensor("op_12690_interleave_0"), val = tensor(false)]; + tensor var_12690_cast = concat(axis = var_10234, interleave = var_12690_interleave_0, values = (var_12650_cast, var_12652_cast))[name = tensor("op_12690_cast")]; + tensor var_12692_interleave_0 = const()[name = tensor("op_12692_interleave_0"), val = tensor(false)]; + tensor var_12692_cast = concat(axis = var_10234, interleave = var_12692_interleave_0, values = (var_12654_cast, var_12656_cast))[name = tensor("op_12692_cast")]; + tensor var_12694_interleave_0 = const()[name = tensor("op_12694_interleave_0"), val = tensor(false)]; + tensor var_12694_cast = concat(axis = var_10234, interleave = var_12694_interleave_0, values = (var_12658_cast, var_12660_cast))[name = tensor("op_12694_cast")]; + tensor var_12696_interleave_0 = const()[name = tensor("op_12696_interleave_0"), val = tensor(false)]; + tensor var_12696_cast = concat(axis = var_10234, interleave = var_12696_interleave_0, values = (var_12662_cast, var_12664_cast))[name = tensor("op_12696_cast")]; + tensor var_12698_interleave_0 = const()[name = tensor("op_12698_interleave_0"), val = tensor(false)]; + tensor var_12698_cast = concat(axis = var_10234, interleave = var_12698_interleave_0, values = (var_12666_cast, var_12668_cast))[name = tensor("op_12698_cast")]; + tensor var_12700_interleave_0 = const()[name = tensor("op_12700_interleave_0"), val = tensor(false)]; + tensor var_12700_cast = concat(axis = var_10234, interleave = var_12700_interleave_0, values = (var_12670_cast, var_12672_cast))[name = tensor("op_12700_cast")]; + tensor var_12702_interleave_0 = const()[name = tensor("op_12702_interleave_0"), val = tensor(false)]; + tensor var_12702_cast = concat(axis = var_10234, interleave = var_12702_interleave_0, values = (var_12674_cast, var_12676_cast))[name = tensor("op_12702_cast")]; + tensor var_12704_interleave_0 = const()[name = tensor("op_12704_interleave_0"), val = tensor(false)]; + tensor var_12704_cast = concat(axis = var_10234, interleave = var_12704_interleave_0, values = (var_12678_cast, var_12680_cast))[name = tensor("op_12704_cast")]; + tensor var_12706_interleave_0 = const()[name = tensor("op_12706_interleave_0"), val = tensor(false)]; + tensor var_12706_cast = concat(axis = var_10234, interleave = var_12706_interleave_0, values = (var_12682_cast, var_12684_cast))[name = tensor("op_12706_cast")]; + tensor var_12708_interleave_0 = const()[name = tensor("op_12708_interleave_0"), val = tensor(false)]; + tensor var_12708_cast = concat(axis = var_10234, interleave = var_12708_interleave_0, values = (var_12686_cast, var_12688_cast))[name = tensor("op_12708_cast")]; + tensor input_437_interleave_0 = const()[name = tensor("input_437_interleave_0"), val = tensor(false)]; + tensor input_437_cast = concat(axis = var_10257, interleave = input_437_interleave_0, values = (var_12690_cast, var_12692_cast, var_12694_cast, var_12696_cast, var_12698_cast, var_12700_cast, var_12702_cast, var_12704_cast, var_12706_cast, var_12708_cast))[name = tensor("input_437_cast")]; + tensor var_12714 = const()[name = tensor("op_12714"), val = tensor([1, 1])]; + tensor var_12716 = const()[name = tensor("op_12716"), val = tensor([1, 1])]; + tensor var_12718_pad_type_0 = const()[name = tensor("op_12718_pad_type_0"), val = tensor("custom")]; + tensor var_12718_pad_0 = const()[name = tensor("op_12718_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_2_attentions_2_transformer_blocks_0_attn2_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(628348544))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(628655808))), name = tensor("up_blocks_2_attentions_2_transformer_blocks_0_attn2_to_out_0_weight_to_fp16_palettized"), shape = tensor([640, 640, 1, 1])]; + tensor up_blocks_2_attentions_2_transformer_blocks_0_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("up_blocks_2_attentions_2_transformer_blocks_0_attn2_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(628656000)))]; + tensor var_12718_cast = conv(bias = up_blocks_2_attentions_2_transformer_blocks_0_attn2_to_out_0_bias_to_fp16, dilations = var_12716, groups = var_10257, pad = var_12718_pad_0, pad_type = var_12718_pad_type_0, strides = var_12714, weight = up_blocks_2_attentions_2_transformer_blocks_0_attn2_to_out_0_weight_to_fp16_palettized, x = input_437_cast)[name = tensor("op_12718_cast")]; + tensor inputs_77_cast = add(x = var_12718_cast, y = inputs_75_cast)[name = tensor("inputs_77_cast")]; + tensor var_12722 = const()[name = tensor("op_12722"), val = tensor([1])]; + tensor channels_mean_77_cast = reduce_mean(axes = var_12722, keep_dims = var_10252, x = inputs_77_cast)[name = tensor("channels_mean_77_cast")]; + tensor zero_mean_77_cast = sub(x = inputs_77_cast, y = channels_mean_77_cast)[name = tensor("zero_mean_77_cast")]; + tensor zero_mean_sq_77_cast = mul(x = zero_mean_77_cast, y = zero_mean_77_cast)[name = tensor("zero_mean_sq_77_cast")]; + tensor var_12726 = const()[name = tensor("op_12726"), val = tensor([1])]; + tensor var_12727_cast = reduce_mean(axes = var_12726, keep_dims = var_10252, x = zero_mean_sq_77_cast)[name = tensor("op_12727_cast")]; + tensor var_12728_to_fp16 = const()[name = tensor("op_12728_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_12729_cast = add(x = var_12727_cast, y = var_12728_to_fp16)[name = tensor("op_12729_cast")]; + tensor denom_77_epsilon_0_to_fp16 = const()[name = tensor("denom_77_epsilon_0_to_fp16"), val = tensor(0x1p-24)]; + tensor denom_77_cast = rsqrt(epsilon = denom_77_epsilon_0_to_fp16, x = var_12729_cast)[name = tensor("denom_77_cast")]; + tensor out_77_cast = mul(x = zero_mean_77_cast, y = denom_77_cast)[name = tensor("out_77_cast")]; + tensor var_12733_to_fp16 = const()[name = tensor("op_12733_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(628657344)))]; + tensor var_12734_cast = add(x = out_77_cast, y = var_12733_to_fp16)[name = tensor("op_12734_cast")]; + tensor var_12736_to_fp16 = const()[name = tensor("op_12736_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(628658688)))]; + tensor input_439_cast = mul(x = var_12734_cast, y = var_12736_to_fp16)[name = tensor("input_439_cast")]; + tensor var_12744 = const()[name = tensor("op_12744"), val = tensor([1, 1])]; + tensor var_12746 = const()[name = tensor("op_12746"), val = tensor([1, 1])]; + tensor var_12748_pad_type_0 = const()[name = tensor("op_12748_pad_type_0"), val = tensor("custom")]; + tensor var_12748_pad_0 = const()[name = tensor("op_12748_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_2_attentions_2_transformer_blocks_0_ff_net_0_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(628660032))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(631117696))), name = tensor("up_blocks_2_attentions_2_transformer_blocks_0_ff_net_0_proj_weight_to_fp16_palettized"), shape = tensor([5120, 640, 1, 1])]; + tensor up_blocks_2_attentions_2_transformer_blocks_0_ff_net_0_proj_bias_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(631117888))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(631121792))), name = tensor("up_blocks_2_attentions_2_transformer_blocks_0_ff_net_0_proj_bias_to_fp16_palettized"), shape = tensor([5120])]; + tensor var_12748_cast = conv(bias = up_blocks_2_attentions_2_transformer_blocks_0_ff_net_0_proj_bias_to_fp16_palettized, dilations = var_12746, groups = var_10257, pad = var_12748_pad_0, pad_type = var_12748_pad_type_0, strides = var_12744, weight = up_blocks_2_attentions_2_transformer_blocks_0_ff_net_0_proj_weight_to_fp16_palettized, x = input_439_cast)[name = tensor("op_12748_cast")]; + tensor var_12749_split_sizes_0 = const()[name = tensor("op_12749_split_sizes_0"), val = tensor([2560, 2560])]; + tensor var_12749_axis_0 = const()[name = tensor("op_12749_axis_0"), val = tensor(1)]; + tensor var_12749_cast_0, tensor var_12749_cast_1 = split(axis = var_12749_axis_0, split_sizes = var_12749_split_sizes_0, x = var_12748_cast)[name = tensor("op_12749_cast")]; + tensor var_12751_mode_0 = const()[name = tensor("op_12751_mode_0"), val = tensor("EXACT")]; + tensor var_12751_cast = gelu(mode = var_12751_mode_0, x = var_12749_cast_1)[name = tensor("op_12751_cast")]; + tensor input_441_cast = mul(x = var_12749_cast_0, y = var_12751_cast)[name = tensor("input_441_cast")]; + tensor var_12755 = const()[name = tensor("op_12755"), val = tensor([1, 1])]; + tensor var_12757 = const()[name = tensor("op_12757"), val = tensor([1, 1])]; + tensor var_12759_pad_type_0 = const()[name = tensor("op_12759_pad_type_0"), val = tensor("custom")]; + tensor var_12759_pad_0 = const()[name = tensor("op_12759_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_2_attentions_2_transformer_blocks_0_ff_net_2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(631121984))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(632350848))), name = tensor("up_blocks_2_attentions_2_transformer_blocks_0_ff_net_2_weight_to_fp16_palettized"), shape = tensor([640, 2560, 1, 1])]; + tensor up_blocks_2_attentions_2_transformer_blocks_0_ff_net_2_bias_to_fp16 = const()[name = tensor("up_blocks_2_attentions_2_transformer_blocks_0_ff_net_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(632351040)))]; + tensor var_12759_cast = conv(bias = up_blocks_2_attentions_2_transformer_blocks_0_ff_net_2_bias_to_fp16, dilations = var_12757, groups = var_10257, pad = var_12759_pad_0, pad_type = var_12759_pad_type_0, strides = var_12755, weight = up_blocks_2_attentions_2_transformer_blocks_0_ff_net_2_weight_to_fp16_palettized, x = input_441_cast)[name = tensor("op_12759_cast")]; + tensor hidden_states_273_cast = add(x = var_12759_cast, y = inputs_77_cast)[name = tensor("hidden_states_273_cast")]; + tensor var_12761 = const()[name = tensor("op_12761"), val = tensor([2, 640, 32, 32])]; + tensor input_443_cast = reshape(shape = var_12761, x = hidden_states_273_cast)[name = tensor("input_443_cast")]; + tensor var_12765 = const()[name = tensor("op_12765"), val = tensor([1, 1])]; + tensor var_12767 = const()[name = tensor("op_12767"), val = tensor([1, 1])]; + tensor hidden_states_275_pad_type_0 = const()[name = tensor("hidden_states_275_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_275_pad_0 = const()[name = tensor("hidden_states_275_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_2_attentions_2_proj_out_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(632352384))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(632659648))), name = tensor("up_blocks_2_attentions_2_proj_out_weight_to_fp16_palettized"), shape = tensor([640, 640, 1, 1])]; + tensor up_blocks_2_attentions_2_proj_out_bias_to_fp16 = const()[name = tensor("up_blocks_2_attentions_2_proj_out_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(632659840)))]; + tensor hidden_states_275_cast = conv(bias = up_blocks_2_attentions_2_proj_out_bias_to_fp16, dilations = var_12767, groups = var_10257, pad = hidden_states_275_pad_0, pad_type = hidden_states_275_pad_type_0, strides = var_12765, weight = up_blocks_2_attentions_2_proj_out_weight_to_fp16_palettized, x = input_443_cast)[name = tensor("hidden_states_275_cast")]; + tensor input_445_cast = add(x = hidden_states_275_cast, y = hidden_states_263_cast)[name = tensor("input_445_cast")]; + tensor input_447_scale_factor_height_0 = const()[name = tensor("input_447_scale_factor_height_0"), val = tensor(0x1p+1)]; + tensor input_447_scale_factor_width_0 = const()[name = tensor("input_447_scale_factor_width_0"), val = tensor(0x1p+1)]; + tensor input_447_cast = upsample_nearest_neighbor(scale_factor_height = input_447_scale_factor_height_0, scale_factor_width = input_447_scale_factor_width_0, x = input_445_cast)[name = tensor("input_447_cast")]; + tensor var_12776 = const()[name = tensor("op_12776"), val = tensor([1, 1])]; + tensor var_12778 = const()[name = tensor("op_12778"), val = tensor([1, 1])]; + tensor hidden_states_277_pad_type_0 = const()[name = tensor("hidden_states_277_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_277_pad_0 = const()[name = tensor("hidden_states_277_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor up_blocks_2_upsamplers_0_conv_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(632661184))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(635426048))), name = tensor("up_blocks_2_upsamplers_0_conv_weight_to_fp16_palettized"), shape = tensor([640, 640, 3, 3])]; + tensor up_blocks_2_upsamplers_0_conv_bias_to_fp16 = const()[name = tensor("up_blocks_2_upsamplers_0_conv_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(635426240)))]; + tensor hidden_states_277_cast = conv(bias = up_blocks_2_upsamplers_0_conv_bias_to_fp16, dilations = var_12778, groups = var_10257, pad = hidden_states_277_pad_0, pad_type = hidden_states_277_pad_type_0, strides = var_12776, weight = up_blocks_2_upsamplers_0_conv_weight_to_fp16_palettized, x = input_447_cast)[name = tensor("hidden_states_277_cast")]; + tensor var_12782 = const()[name = tensor("op_12782"), val = tensor(3)]; + tensor var_12802 = const()[name = tensor("op_12802"), val = tensor(true)]; + tensor var_12807 = const()[name = tensor("op_12807"), val = tensor(1)]; + tensor input_449_interleave_0 = const()[name = tensor("input_449_interleave_0"), val = tensor(false)]; + tensor input_449_cast = concat(axis = var_12807, interleave = input_449_interleave_0, values = (hidden_states_277_cast, input_61_cast))[name = tensor("input_449_cast")]; + tensor reshape_204_shape_0 = const()[name = tensor("reshape_204_shape_0"), val = tensor([2, 32, 30, 64, 64])]; + tensor reshape_204_cast = reshape(shape = reshape_204_shape_0, x = input_449_cast)[name = tensor("reshape_204_cast")]; + tensor reduce_mean_153_axes_0 = const()[name = tensor("reduce_mean_153_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_153_keep_dims_0 = const()[name = tensor("reduce_mean_153_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_153_cast = reduce_mean(axes = reduce_mean_153_axes_0, keep_dims = reduce_mean_153_keep_dims_0, x = reshape_204_cast)[name = tensor("reduce_mean_153_cast")]; + tensor sub_102_cast = sub(x = reshape_204_cast, y = reduce_mean_153_cast)[name = tensor("sub_102_cast")]; + tensor square_51_cast = square(x = sub_102_cast)[name = tensor("square_51_cast")]; + tensor reduce_mean_155_axes_0 = const()[name = tensor("reduce_mean_155_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_155_keep_dims_0 = const()[name = tensor("reduce_mean_155_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_155_cast = reduce_mean(axes = reduce_mean_155_axes_0, keep_dims = reduce_mean_155_keep_dims_0, x = square_51_cast)[name = tensor("reduce_mean_155_cast")]; + tensor add_102_y_0_to_fp16 = const()[name = tensor("add_102_y_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_102_cast = add(x = reduce_mean_155_cast, y = add_102_y_0_to_fp16)[name = tensor("add_102_cast")]; + tensor sqrt_51_cast = sqrt(x = add_102_cast)[name = tensor("sqrt_51_cast")]; + tensor real_div_51_cast = real_div(x = sub_102_cast, y = sqrt_51_cast)[name = tensor("real_div_51_cast")]; + tensor reshape_205_shape_0 = const()[name = tensor("reshape_205_shape_0"), val = tensor([2, 960, 64, 64])]; + tensor reshape_205_cast = reshape(shape = reshape_205_shape_0, x = real_div_51_cast)[name = tensor("reshape_205_cast")]; + tensor add_103_gamma_0_to_fp16 = const()[name = tensor("add_103_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(635427584)))]; + tensor add_103_beta_0_to_fp16 = const()[name = tensor("add_103_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(635429568)))]; + tensor add_103_epsilon_0_to_fp16 = const()[name = tensor("add_103_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_103_cast = batch_norm(beta = add_103_beta_0_to_fp16, epsilon = add_103_epsilon_0_to_fp16, gamma = add_103_gamma_0_to_fp16, mean = add_97_mean_0_to_fp16, variance = add_97_variance_0_to_fp16, x = reshape_205_cast)[name = tensor("add_103_cast")]; + tensor input_453_cast = silu(x = add_103_cast)[name = tensor("input_453_cast")]; + tensor var_12834 = const()[name = tensor("op_12834"), val = tensor([1, 1])]; + tensor var_12836 = const()[name = tensor("op_12836"), val = tensor([1, 1])]; + tensor hidden_states_279_pad_type_0 = const()[name = tensor("hidden_states_279_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_279_pad_0 = const()[name = tensor("hidden_states_279_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor up_blocks_3_resnets_0_conv1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(635431552))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(637505216))), name = tensor("up_blocks_3_resnets_0_conv1_weight_to_fp16_palettized"), shape = tensor([320, 960, 3, 3])]; + tensor up_blocks_3_resnets_0_conv1_bias_to_fp16 = const()[name = tensor("up_blocks_3_resnets_0_conv1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(637505408)))]; + tensor hidden_states_279_cast = conv(bias = up_blocks_3_resnets_0_conv1_bias_to_fp16, dilations = var_12836, groups = var_12807, pad = hidden_states_279_pad_0, pad_type = hidden_states_279_pad_type_0, strides = var_12834, weight = up_blocks_3_resnets_0_conv1_weight_to_fp16_palettized, x = input_453_cast)[name = tensor("hidden_states_279_cast")]; + tensor var_12842 = const()[name = tensor("op_12842"), val = tensor([1, 1])]; + tensor var_12844 = const()[name = tensor("op_12844"), val = tensor([1, 1])]; + tensor temb_39_pad_type_0 = const()[name = tensor("temb_39_pad_type_0"), val = tensor("custom")]; + tensor temb_39_pad_0 = const()[name = tensor("temb_39_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_3_resnets_0_time_emb_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(637506112))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(637813376))), name = tensor("up_blocks_3_resnets_0_time_emb_proj_weight_to_fp16_palettized"), shape = tensor([320, 1280, 1, 1])]; + tensor up_blocks_3_resnets_0_time_emb_proj_bias_to_fp16 = const()[name = tensor("up_blocks_3_resnets_0_time_emb_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(637813568)))]; + tensor temb_39_cast = conv(bias = up_blocks_3_resnets_0_time_emb_proj_bias_to_fp16, dilations = var_12844, groups = var_12807, pad = temb_39_pad_0, pad_type = temb_39_pad_type_0, strides = var_12842, weight = up_blocks_3_resnets_0_time_emb_proj_weight_to_fp16_palettized, x = input_15_cast)[name = tensor("temb_39_cast")]; + tensor input_457_cast = add(x = hidden_states_279_cast, y = temb_39_cast)[name = tensor("input_457_cast")]; + tensor reshape_208_shape_0 = const()[name = tensor("reshape_208_shape_0"), val = tensor([2, 32, 10, 64, 64])]; + tensor reshape_208_cast = reshape(shape = reshape_208_shape_0, x = input_457_cast)[name = tensor("reshape_208_cast")]; + tensor reduce_mean_156_axes_0 = const()[name = tensor("reduce_mean_156_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_156_keep_dims_0 = const()[name = tensor("reduce_mean_156_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_156_cast = reduce_mean(axes = reduce_mean_156_axes_0, keep_dims = reduce_mean_156_keep_dims_0, x = reshape_208_cast)[name = tensor("reduce_mean_156_cast")]; + tensor sub_104_cast = sub(x = reshape_208_cast, y = reduce_mean_156_cast)[name = tensor("sub_104_cast")]; + tensor square_52_cast = square(x = sub_104_cast)[name = tensor("square_52_cast")]; + tensor reduce_mean_158_axes_0 = const()[name = tensor("reduce_mean_158_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_158_keep_dims_0 = const()[name = tensor("reduce_mean_158_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_158_cast = reduce_mean(axes = reduce_mean_158_axes_0, keep_dims = reduce_mean_158_keep_dims_0, x = square_52_cast)[name = tensor("reduce_mean_158_cast")]; + tensor add_104_y_0_to_fp16 = const()[name = tensor("add_104_y_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_104_cast = add(x = reduce_mean_158_cast, y = add_104_y_0_to_fp16)[name = tensor("add_104_cast")]; + tensor sqrt_52_cast = sqrt(x = add_104_cast)[name = tensor("sqrt_52_cast")]; + tensor real_div_52_cast = real_div(x = sub_104_cast, y = sqrt_52_cast)[name = tensor("real_div_52_cast")]; + tensor reshape_209_shape_0 = const()[name = tensor("reshape_209_shape_0"), val = tensor([2, 320, 64, 64])]; + tensor reshape_209_cast = reshape(shape = reshape_209_shape_0, x = real_div_52_cast)[name = tensor("reshape_209_cast")]; + tensor add_105_gamma_0_to_fp16 = const()[name = tensor("add_105_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(637814272)))]; + tensor add_105_beta_0_to_fp16 = const()[name = tensor("add_105_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(637814976)))]; + tensor add_105_epsilon_0_to_fp16 = const()[name = tensor("add_105_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_105_cast = batch_norm(beta = add_105_beta_0_to_fp16, epsilon = add_105_epsilon_0_to_fp16, gamma = add_105_gamma_0_to_fp16, mean = add_1_mean_0_to_fp16, variance = add_1_variance_0_to_fp16, x = reshape_209_cast)[name = tensor("add_105_cast")]; + tensor input_461_cast = silu(x = add_105_cast)[name = tensor("input_461_cast")]; + tensor var_12854 = const()[name = tensor("op_12854"), val = tensor([1, 1])]; + tensor var_12856 = const()[name = tensor("op_12856"), val = tensor([1, 1])]; + tensor hidden_states_281_pad_type_0 = const()[name = tensor("hidden_states_281_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_281_pad_0 = const()[name = tensor("hidden_states_281_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor up_blocks_3_resnets_0_conv2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(637815680))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(638506944))), name = tensor("up_blocks_3_resnets_0_conv2_weight_to_fp16_palettized"), shape = tensor([320, 320, 3, 3])]; + tensor up_blocks_3_resnets_0_conv2_bias_to_fp16 = const()[name = tensor("up_blocks_3_resnets_0_conv2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(638507136)))]; + tensor hidden_states_281_cast = conv(bias = up_blocks_3_resnets_0_conv2_bias_to_fp16, dilations = var_12856, groups = var_12807, pad = hidden_states_281_pad_0, pad_type = hidden_states_281_pad_type_0, strides = var_12854, weight = up_blocks_3_resnets_0_conv2_weight_to_fp16_palettized, x = input_461_cast)[name = tensor("hidden_states_281_cast")]; + tensor var_12861 = const()[name = tensor("op_12861"), val = tensor([1, 1])]; + tensor var_12863 = const()[name = tensor("op_12863"), val = tensor([1, 1])]; + tensor x_23_pad_type_0 = const()[name = tensor("x_23_pad_type_0"), val = tensor("custom")]; + tensor x_23_pad_0 = const()[name = tensor("x_23_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_3_resnets_0_conv_shortcut_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(638507840))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(638738304))), name = tensor("up_blocks_3_resnets_0_conv_shortcut_weight_to_fp16_palettized"), shape = tensor([320, 960, 1, 1])]; + tensor up_blocks_3_resnets_0_conv_shortcut_bias_to_fp16 = const()[name = tensor("up_blocks_3_resnets_0_conv_shortcut_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(638738496)))]; + tensor x_23_cast = conv(bias = up_blocks_3_resnets_0_conv_shortcut_bias_to_fp16, dilations = var_12863, groups = var_12807, pad = x_23_pad_0, pad_type = x_23_pad_type_0, strides = var_12861, weight = up_blocks_3_resnets_0_conv_shortcut_weight_to_fp16_palettized, x = input_449_cast)[name = tensor("x_23_cast")]; + tensor hidden_states_283_cast = add(x = x_23_cast, y = hidden_states_281_cast)[name = tensor("hidden_states_283_cast")]; + tensor reshape_212_shape_0 = const()[name = tensor("reshape_212_shape_0"), val = tensor([2, 32, 10, 64, 64])]; + tensor reshape_212_cast = reshape(shape = reshape_212_shape_0, x = hidden_states_283_cast)[name = tensor("reshape_212_cast")]; + tensor reduce_mean_159_axes_0 = const()[name = tensor("reduce_mean_159_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_159_keep_dims_0 = const()[name = tensor("reduce_mean_159_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_159_cast = reduce_mean(axes = reduce_mean_159_axes_0, keep_dims = reduce_mean_159_keep_dims_0, x = reshape_212_cast)[name = tensor("reduce_mean_159_cast")]; + tensor sub_106_cast = sub(x = reshape_212_cast, y = reduce_mean_159_cast)[name = tensor("sub_106_cast")]; + tensor square_53_cast = square(x = sub_106_cast)[name = tensor("square_53_cast")]; + tensor reduce_mean_161_axes_0 = const()[name = tensor("reduce_mean_161_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_161_keep_dims_0 = const()[name = tensor("reduce_mean_161_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_161_cast = reduce_mean(axes = reduce_mean_161_axes_0, keep_dims = reduce_mean_161_keep_dims_0, x = square_53_cast)[name = tensor("reduce_mean_161_cast")]; + tensor add_106_y_0_to_fp16 = const()[name = tensor("add_106_y_0_to_fp16"), val = tensor(0x1.1p-20)]; + tensor add_106_cast = add(x = reduce_mean_161_cast, y = add_106_y_0_to_fp16)[name = tensor("add_106_cast")]; + tensor sqrt_53_cast = sqrt(x = add_106_cast)[name = tensor("sqrt_53_cast")]; + tensor real_div_53_cast = real_div(x = sub_106_cast, y = sqrt_53_cast)[name = tensor("real_div_53_cast")]; + tensor reshape_213_shape_0 = const()[name = tensor("reshape_213_shape_0"), val = tensor([2, 320, 64, 64])]; + tensor reshape_213_cast = reshape(shape = reshape_213_shape_0, x = real_div_53_cast)[name = tensor("reshape_213_cast")]; + tensor add_107_gamma_0_to_fp16 = const()[name = tensor("add_107_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(638739200)))]; + tensor add_107_beta_0_to_fp16 = const()[name = tensor("add_107_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(638739904)))]; + tensor add_107_epsilon_0_to_fp16 = const()[name = tensor("add_107_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_107_cast = batch_norm(beta = add_107_beta_0_to_fp16, epsilon = add_107_epsilon_0_to_fp16, gamma = add_107_gamma_0_to_fp16, mean = add_1_mean_0_to_fp16, variance = add_1_variance_0_to_fp16, x = reshape_213_cast)[name = tensor("add_107_cast")]; + tensor var_12883 = const()[name = tensor("op_12883"), val = tensor([1, 1])]; + tensor var_12885 = const()[name = tensor("op_12885"), val = tensor([1, 1])]; + tensor hidden_states_285_pad_type_0 = const()[name = tensor("hidden_states_285_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_285_pad_0 = const()[name = tensor("hidden_states_285_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_3_attentions_0_proj_in_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(638740608))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(638817472))), name = tensor("up_blocks_3_attentions_0_proj_in_weight_to_fp16_palettized"), shape = tensor([320, 320, 1, 1])]; + tensor up_blocks_3_attentions_0_proj_in_bias_to_fp16 = const()[name = tensor("up_blocks_3_attentions_0_proj_in_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(638817664)))]; + tensor hidden_states_285_cast = conv(bias = up_blocks_3_attentions_0_proj_in_bias_to_fp16, dilations = var_12885, groups = var_12807, pad = hidden_states_285_pad_0, pad_type = hidden_states_285_pad_type_0, strides = var_12883, weight = up_blocks_3_attentions_0_proj_in_weight_to_fp16_palettized, x = add_107_cast)[name = tensor("hidden_states_285_cast")]; + tensor var_12890 = const()[name = tensor("op_12890"), val = tensor([2, 320, 1, 4096])]; + tensor inputs_79_cast = reshape(shape = var_12890, x = hidden_states_285_cast)[name = tensor("inputs_79_cast")]; + tensor var_12900 = const()[name = tensor("op_12900"), val = tensor([1])]; + tensor channels_mean_79_cast = reduce_mean(axes = var_12900, keep_dims = var_12802, x = inputs_79_cast)[name = tensor("channels_mean_79_cast")]; + tensor zero_mean_79_cast = sub(x = inputs_79_cast, y = channels_mean_79_cast)[name = tensor("zero_mean_79_cast")]; + tensor zero_mean_sq_79_cast = mul(x = zero_mean_79_cast, y = zero_mean_79_cast)[name = tensor("zero_mean_sq_79_cast")]; + tensor var_12904 = const()[name = tensor("op_12904"), val = tensor([1])]; + tensor var_12905_cast = reduce_mean(axes = var_12904, keep_dims = var_12802, x = zero_mean_sq_79_cast)[name = tensor("op_12905_cast")]; + tensor var_12906_to_fp16 = const()[name = tensor("op_12906_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_12907_cast = add(x = var_12905_cast, y = var_12906_to_fp16)[name = tensor("op_12907_cast")]; + tensor denom_79_epsilon_0_to_fp16 = const()[name = tensor("denom_79_epsilon_0_to_fp16"), val = tensor(0x1p-24)]; + tensor denom_79_cast = rsqrt(epsilon = denom_79_epsilon_0_to_fp16, x = var_12907_cast)[name = tensor("denom_79_cast")]; + tensor out_79_cast = mul(x = zero_mean_79_cast, y = denom_79_cast)[name = tensor("out_79_cast")]; + tensor var_12911_to_fp16 = const()[name = tensor("op_12911_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(638818368)))]; + tensor var_12912_cast = add(x = out_79_cast, y = var_12911_to_fp16)[name = tensor("op_12912_cast")]; + tensor var_12914_to_fp16 = const()[name = tensor("op_12914_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(638819072)))]; + tensor hidden_states_287_cast = mul(x = var_12912_cast, y = var_12914_to_fp16)[name = tensor("hidden_states_287_cast")]; + tensor var_12921 = const()[name = tensor("op_12921"), val = tensor([1, 1])]; + tensor var_12923 = const()[name = tensor("op_12923"), val = tensor([1, 1])]; + tensor q_53_pad_type_0 = const()[name = tensor("q_53_pad_type_0"), val = tensor("custom")]; + tensor q_53_pad_0 = const()[name = tensor("q_53_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_3_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(638819776))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(638896640))), name = tensor("up_blocks_3_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16_palettized"), shape = tensor([320, 320, 1, 1])]; + tensor q_53_cast = conv(dilations = var_12923, groups = var_12807, pad = q_53_pad_0, pad_type = q_53_pad_type_0, strides = var_12921, weight = up_blocks_3_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16_palettized, x = hidden_states_287_cast)[name = tensor("q_53_cast")]; + tensor var_12927 = const()[name = tensor("op_12927"), val = tensor([1, 1])]; + tensor var_12929 = const()[name = tensor("op_12929"), val = tensor([1, 1])]; + tensor k_105_pad_type_0 = const()[name = tensor("k_105_pad_type_0"), val = tensor("custom")]; + tensor k_105_pad_0 = const()[name = tensor("k_105_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_3_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(638896832))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(638973696))), name = tensor("up_blocks_3_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16_palettized"), shape = tensor([320, 320, 1, 1])]; + tensor k_105_cast = conv(dilations = var_12929, groups = var_12807, pad = k_105_pad_0, pad_type = k_105_pad_type_0, strides = var_12927, weight = up_blocks_3_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16_palettized, x = hidden_states_287_cast)[name = tensor("k_105_cast")]; + tensor var_12933 = const()[name = tensor("op_12933"), val = tensor([1, 1])]; + tensor var_12935 = const()[name = tensor("op_12935"), val = tensor([1, 1])]; + tensor v_53_pad_type_0 = const()[name = tensor("v_53_pad_type_0"), val = tensor("custom")]; + tensor v_53_pad_0 = const()[name = tensor("v_53_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_3_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(638973888))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(639050752))), name = tensor("up_blocks_3_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16_palettized"), shape = tensor([320, 320, 1, 1])]; + tensor v_53_cast = conv(dilations = var_12935, groups = var_12807, pad = v_53_pad_0, pad_type = v_53_pad_type_0, strides = var_12933, weight = up_blocks_3_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16_palettized, x = hidden_states_287_cast)[name = tensor("v_53_cast")]; + tensor var_12939_begin_0 = const()[name = tensor("op_12939_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_12939_end_0 = const()[name = tensor("op_12939_end_0"), val = tensor([2, 64, 1, 4096])]; + tensor var_12939_end_mask_0 = const()[name = tensor("op_12939_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12939_cast = slice_by_index(begin = var_12939_begin_0, end = var_12939_end_0, end_mask = var_12939_end_mask_0, x = q_53_cast)[name = tensor("op_12939_cast")]; + tensor var_12943_begin_0 = const()[name = tensor("op_12943_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_12943_end_0 = const()[name = tensor("op_12943_end_0"), val = tensor([2, 128, 1, 4096])]; + tensor var_12943_end_mask_0 = const()[name = tensor("op_12943_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12943_cast = slice_by_index(begin = var_12943_begin_0, end = var_12943_end_0, end_mask = var_12943_end_mask_0, x = q_53_cast)[name = tensor("op_12943_cast")]; + tensor var_12947_begin_0 = const()[name = tensor("op_12947_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_12947_end_0 = const()[name = tensor("op_12947_end_0"), val = tensor([2, 192, 1, 4096])]; + tensor var_12947_end_mask_0 = const()[name = tensor("op_12947_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12947_cast = slice_by_index(begin = var_12947_begin_0, end = var_12947_end_0, end_mask = var_12947_end_mask_0, x = q_53_cast)[name = tensor("op_12947_cast")]; + tensor var_12951_begin_0 = const()[name = tensor("op_12951_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_12951_end_0 = const()[name = tensor("op_12951_end_0"), val = tensor([2, 256, 1, 4096])]; + tensor var_12951_end_mask_0 = const()[name = tensor("op_12951_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12951_cast = slice_by_index(begin = var_12951_begin_0, end = var_12951_end_0, end_mask = var_12951_end_mask_0, x = q_53_cast)[name = tensor("op_12951_cast")]; + tensor var_12955_begin_0 = const()[name = tensor("op_12955_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_12955_end_0 = const()[name = tensor("op_12955_end_0"), val = tensor([2, 320, 1, 4096])]; + tensor var_12955_end_mask_0 = const()[name = tensor("op_12955_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12955_cast = slice_by_index(begin = var_12955_begin_0, end = var_12955_end_0, end_mask = var_12955_end_mask_0, x = q_53_cast)[name = tensor("op_12955_cast")]; + tensor var_12958_begin_0 = const()[name = tensor("op_12958_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_12958_end_0 = const()[name = tensor("op_12958_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_12958_end_mask_0 = const()[name = tensor("op_12958_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12958_cast = slice_by_index(begin = var_12958_begin_0, end = var_12958_end_0, end_mask = var_12958_end_mask_0, x = var_12939_cast)[name = tensor("op_12958_cast")]; + tensor var_12959_begin_0 = const()[name = tensor("op_12959_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_12959_end_0 = const()[name = tensor("op_12959_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_12959_end_mask_0 = const()[name = tensor("op_12959_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12959_cast = slice_by_index(begin = var_12959_begin_0, end = var_12959_end_0, end_mask = var_12959_end_mask_0, x = var_12939_cast)[name = tensor("op_12959_cast")]; + tensor var_12960_begin_0 = const()[name = tensor("op_12960_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_12960_end_0 = const()[name = tensor("op_12960_end_0"), val = tensor([2, 64, 1, 1536])]; + tensor var_12960_end_mask_0 = const()[name = tensor("op_12960_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12960_cast = slice_by_index(begin = var_12960_begin_0, end = var_12960_end_0, end_mask = var_12960_end_mask_0, x = var_12939_cast)[name = tensor("op_12960_cast")]; + tensor var_12961_begin_0 = const()[name = tensor("op_12961_begin_0"), val = tensor([0, 0, 0, 1536])]; + tensor var_12961_end_0 = const()[name = tensor("op_12961_end_0"), val = tensor([2, 64, 1, 2048])]; + tensor var_12961_end_mask_0 = const()[name = tensor("op_12961_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12961_cast = slice_by_index(begin = var_12961_begin_0, end = var_12961_end_0, end_mask = var_12961_end_mask_0, x = var_12939_cast)[name = tensor("op_12961_cast")]; + tensor var_12962_begin_0 = const()[name = tensor("op_12962_begin_0"), val = tensor([0, 0, 0, 2048])]; + tensor var_12962_end_0 = const()[name = tensor("op_12962_end_0"), val = tensor([2, 64, 1, 2560])]; + tensor var_12962_end_mask_0 = const()[name = tensor("op_12962_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12962_cast = slice_by_index(begin = var_12962_begin_0, end = var_12962_end_0, end_mask = var_12962_end_mask_0, x = var_12939_cast)[name = tensor("op_12962_cast")]; + tensor var_12963_begin_0 = const()[name = tensor("op_12963_begin_0"), val = tensor([0, 0, 0, 2560])]; + tensor var_12963_end_0 = const()[name = tensor("op_12963_end_0"), val = tensor([2, 64, 1, 3072])]; + tensor var_12963_end_mask_0 = const()[name = tensor("op_12963_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12963_cast = slice_by_index(begin = var_12963_begin_0, end = var_12963_end_0, end_mask = var_12963_end_mask_0, x = var_12939_cast)[name = tensor("op_12963_cast")]; + tensor var_12964_begin_0 = const()[name = tensor("op_12964_begin_0"), val = tensor([0, 0, 0, 3072])]; + tensor var_12964_end_0 = const()[name = tensor("op_12964_end_0"), val = tensor([2, 64, 1, 3584])]; + tensor var_12964_end_mask_0 = const()[name = tensor("op_12964_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12964_cast = slice_by_index(begin = var_12964_begin_0, end = var_12964_end_0, end_mask = var_12964_end_mask_0, x = var_12939_cast)[name = tensor("op_12964_cast")]; + tensor var_12965_begin_0 = const()[name = tensor("op_12965_begin_0"), val = tensor([0, 0, 0, 3584])]; + tensor var_12965_end_0 = const()[name = tensor("op_12965_end_0"), val = tensor([2, 64, 1, 4096])]; + tensor var_12965_end_mask_0 = const()[name = tensor("op_12965_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12965_cast = slice_by_index(begin = var_12965_begin_0, end = var_12965_end_0, end_mask = var_12965_end_mask_0, x = var_12939_cast)[name = tensor("op_12965_cast")]; + tensor var_12966_begin_0 = const()[name = tensor("op_12966_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_12966_end_0 = const()[name = tensor("op_12966_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_12966_end_mask_0 = const()[name = tensor("op_12966_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12966_cast = slice_by_index(begin = var_12966_begin_0, end = var_12966_end_0, end_mask = var_12966_end_mask_0, x = var_12943_cast)[name = tensor("op_12966_cast")]; + tensor var_12967_begin_0 = const()[name = tensor("op_12967_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_12967_end_0 = const()[name = tensor("op_12967_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_12967_end_mask_0 = const()[name = tensor("op_12967_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12967_cast = slice_by_index(begin = var_12967_begin_0, end = var_12967_end_0, end_mask = var_12967_end_mask_0, x = var_12943_cast)[name = tensor("op_12967_cast")]; + tensor var_12968_begin_0 = const()[name = tensor("op_12968_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_12968_end_0 = const()[name = tensor("op_12968_end_0"), val = tensor([2, 64, 1, 1536])]; + tensor var_12968_end_mask_0 = const()[name = tensor("op_12968_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12968_cast = slice_by_index(begin = var_12968_begin_0, end = var_12968_end_0, end_mask = var_12968_end_mask_0, x = var_12943_cast)[name = tensor("op_12968_cast")]; + tensor var_12969_begin_0 = const()[name = tensor("op_12969_begin_0"), val = tensor([0, 0, 0, 1536])]; + tensor var_12969_end_0 = const()[name = tensor("op_12969_end_0"), val = tensor([2, 64, 1, 2048])]; + tensor var_12969_end_mask_0 = const()[name = tensor("op_12969_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12969_cast = slice_by_index(begin = var_12969_begin_0, end = var_12969_end_0, end_mask = var_12969_end_mask_0, x = var_12943_cast)[name = tensor("op_12969_cast")]; + tensor var_12970_begin_0 = const()[name = tensor("op_12970_begin_0"), val = tensor([0, 0, 0, 2048])]; + tensor var_12970_end_0 = const()[name = tensor("op_12970_end_0"), val = tensor([2, 64, 1, 2560])]; + tensor var_12970_end_mask_0 = const()[name = tensor("op_12970_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12970_cast = slice_by_index(begin = var_12970_begin_0, end = var_12970_end_0, end_mask = var_12970_end_mask_0, x = var_12943_cast)[name = tensor("op_12970_cast")]; + tensor var_12971_begin_0 = const()[name = tensor("op_12971_begin_0"), val = tensor([0, 0, 0, 2560])]; + tensor var_12971_end_0 = const()[name = tensor("op_12971_end_0"), val = tensor([2, 64, 1, 3072])]; + tensor var_12971_end_mask_0 = const()[name = tensor("op_12971_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12971_cast = slice_by_index(begin = var_12971_begin_0, end = var_12971_end_0, end_mask = var_12971_end_mask_0, x = var_12943_cast)[name = tensor("op_12971_cast")]; + tensor var_12972_begin_0 = const()[name = tensor("op_12972_begin_0"), val = tensor([0, 0, 0, 3072])]; + tensor var_12972_end_0 = const()[name = tensor("op_12972_end_0"), val = tensor([2, 64, 1, 3584])]; + tensor var_12972_end_mask_0 = const()[name = tensor("op_12972_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12972_cast = slice_by_index(begin = var_12972_begin_0, end = var_12972_end_0, end_mask = var_12972_end_mask_0, x = var_12943_cast)[name = tensor("op_12972_cast")]; + tensor var_12973_begin_0 = const()[name = tensor("op_12973_begin_0"), val = tensor([0, 0, 0, 3584])]; + tensor var_12973_end_0 = const()[name = tensor("op_12973_end_0"), val = tensor([2, 64, 1, 4096])]; + tensor var_12973_end_mask_0 = const()[name = tensor("op_12973_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12973_cast = slice_by_index(begin = var_12973_begin_0, end = var_12973_end_0, end_mask = var_12973_end_mask_0, x = var_12943_cast)[name = tensor("op_12973_cast")]; + tensor var_12974_begin_0 = const()[name = tensor("op_12974_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_12974_end_0 = const()[name = tensor("op_12974_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_12974_end_mask_0 = const()[name = tensor("op_12974_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12974_cast = slice_by_index(begin = var_12974_begin_0, end = var_12974_end_0, end_mask = var_12974_end_mask_0, x = var_12947_cast)[name = tensor("op_12974_cast")]; + tensor var_12975_begin_0 = const()[name = tensor("op_12975_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_12975_end_0 = const()[name = tensor("op_12975_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_12975_end_mask_0 = const()[name = tensor("op_12975_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12975_cast = slice_by_index(begin = var_12975_begin_0, end = var_12975_end_0, end_mask = var_12975_end_mask_0, x = var_12947_cast)[name = tensor("op_12975_cast")]; + tensor var_12976_begin_0 = const()[name = tensor("op_12976_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_12976_end_0 = const()[name = tensor("op_12976_end_0"), val = tensor([2, 64, 1, 1536])]; + tensor var_12976_end_mask_0 = const()[name = tensor("op_12976_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12976_cast = slice_by_index(begin = var_12976_begin_0, end = var_12976_end_0, end_mask = var_12976_end_mask_0, x = var_12947_cast)[name = tensor("op_12976_cast")]; + tensor var_12977_begin_0 = const()[name = tensor("op_12977_begin_0"), val = tensor([0, 0, 0, 1536])]; + tensor var_12977_end_0 = const()[name = tensor("op_12977_end_0"), val = tensor([2, 64, 1, 2048])]; + tensor var_12977_end_mask_0 = const()[name = tensor("op_12977_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12977_cast = slice_by_index(begin = var_12977_begin_0, end = var_12977_end_0, end_mask = var_12977_end_mask_0, x = var_12947_cast)[name = tensor("op_12977_cast")]; + tensor var_12978_begin_0 = const()[name = tensor("op_12978_begin_0"), val = tensor([0, 0, 0, 2048])]; + tensor var_12978_end_0 = const()[name = tensor("op_12978_end_0"), val = tensor([2, 64, 1, 2560])]; + tensor var_12978_end_mask_0 = const()[name = tensor("op_12978_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12978_cast = slice_by_index(begin = var_12978_begin_0, end = var_12978_end_0, end_mask = var_12978_end_mask_0, x = var_12947_cast)[name = tensor("op_12978_cast")]; + tensor var_12979_begin_0 = const()[name = tensor("op_12979_begin_0"), val = tensor([0, 0, 0, 2560])]; + tensor var_12979_end_0 = const()[name = tensor("op_12979_end_0"), val = tensor([2, 64, 1, 3072])]; + tensor var_12979_end_mask_0 = const()[name = tensor("op_12979_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12979_cast = slice_by_index(begin = var_12979_begin_0, end = var_12979_end_0, end_mask = var_12979_end_mask_0, x = var_12947_cast)[name = tensor("op_12979_cast")]; + tensor var_12980_begin_0 = const()[name = tensor("op_12980_begin_0"), val = tensor([0, 0, 0, 3072])]; + tensor var_12980_end_0 = const()[name = tensor("op_12980_end_0"), val = tensor([2, 64, 1, 3584])]; + tensor var_12980_end_mask_0 = const()[name = tensor("op_12980_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12980_cast = slice_by_index(begin = var_12980_begin_0, end = var_12980_end_0, end_mask = var_12980_end_mask_0, x = var_12947_cast)[name = tensor("op_12980_cast")]; + tensor var_12981_begin_0 = const()[name = tensor("op_12981_begin_0"), val = tensor([0, 0, 0, 3584])]; + tensor var_12981_end_0 = const()[name = tensor("op_12981_end_0"), val = tensor([2, 64, 1, 4096])]; + tensor var_12981_end_mask_0 = const()[name = tensor("op_12981_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12981_cast = slice_by_index(begin = var_12981_begin_0, end = var_12981_end_0, end_mask = var_12981_end_mask_0, x = var_12947_cast)[name = tensor("op_12981_cast")]; + tensor var_12982_begin_0 = const()[name = tensor("op_12982_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_12982_end_0 = const()[name = tensor("op_12982_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_12982_end_mask_0 = const()[name = tensor("op_12982_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12982_cast = slice_by_index(begin = var_12982_begin_0, end = var_12982_end_0, end_mask = var_12982_end_mask_0, x = var_12951_cast)[name = tensor("op_12982_cast")]; + tensor var_12983_begin_0 = const()[name = tensor("op_12983_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_12983_end_0 = const()[name = tensor("op_12983_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_12983_end_mask_0 = const()[name = tensor("op_12983_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12983_cast = slice_by_index(begin = var_12983_begin_0, end = var_12983_end_0, end_mask = var_12983_end_mask_0, x = var_12951_cast)[name = tensor("op_12983_cast")]; + tensor var_12984_begin_0 = const()[name = tensor("op_12984_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_12984_end_0 = const()[name = tensor("op_12984_end_0"), val = tensor([2, 64, 1, 1536])]; + tensor var_12984_end_mask_0 = const()[name = tensor("op_12984_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12984_cast = slice_by_index(begin = var_12984_begin_0, end = var_12984_end_0, end_mask = var_12984_end_mask_0, x = var_12951_cast)[name = tensor("op_12984_cast")]; + tensor var_12985_begin_0 = const()[name = tensor("op_12985_begin_0"), val = tensor([0, 0, 0, 1536])]; + tensor var_12985_end_0 = const()[name = tensor("op_12985_end_0"), val = tensor([2, 64, 1, 2048])]; + tensor var_12985_end_mask_0 = const()[name = tensor("op_12985_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12985_cast = slice_by_index(begin = var_12985_begin_0, end = var_12985_end_0, end_mask = var_12985_end_mask_0, x = var_12951_cast)[name = tensor("op_12985_cast")]; + tensor var_12986_begin_0 = const()[name = tensor("op_12986_begin_0"), val = tensor([0, 0, 0, 2048])]; + tensor var_12986_end_0 = const()[name = tensor("op_12986_end_0"), val = tensor([2, 64, 1, 2560])]; + tensor var_12986_end_mask_0 = const()[name = tensor("op_12986_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12986_cast = slice_by_index(begin = var_12986_begin_0, end = var_12986_end_0, end_mask = var_12986_end_mask_0, x = var_12951_cast)[name = tensor("op_12986_cast")]; + tensor var_12987_begin_0 = const()[name = tensor("op_12987_begin_0"), val = tensor([0, 0, 0, 2560])]; + tensor var_12987_end_0 = const()[name = tensor("op_12987_end_0"), val = tensor([2, 64, 1, 3072])]; + tensor var_12987_end_mask_0 = const()[name = tensor("op_12987_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12987_cast = slice_by_index(begin = var_12987_begin_0, end = var_12987_end_0, end_mask = var_12987_end_mask_0, x = var_12951_cast)[name = tensor("op_12987_cast")]; + tensor var_12988_begin_0 = const()[name = tensor("op_12988_begin_0"), val = tensor([0, 0, 0, 3072])]; + tensor var_12988_end_0 = const()[name = tensor("op_12988_end_0"), val = tensor([2, 64, 1, 3584])]; + tensor var_12988_end_mask_0 = const()[name = tensor("op_12988_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12988_cast = slice_by_index(begin = var_12988_begin_0, end = var_12988_end_0, end_mask = var_12988_end_mask_0, x = var_12951_cast)[name = tensor("op_12988_cast")]; + tensor var_12989_begin_0 = const()[name = tensor("op_12989_begin_0"), val = tensor([0, 0, 0, 3584])]; + tensor var_12989_end_0 = const()[name = tensor("op_12989_end_0"), val = tensor([2, 64, 1, 4096])]; + tensor var_12989_end_mask_0 = const()[name = tensor("op_12989_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12989_cast = slice_by_index(begin = var_12989_begin_0, end = var_12989_end_0, end_mask = var_12989_end_mask_0, x = var_12951_cast)[name = tensor("op_12989_cast")]; + tensor var_12990_begin_0 = const()[name = tensor("op_12990_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_12990_end_0 = const()[name = tensor("op_12990_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_12990_end_mask_0 = const()[name = tensor("op_12990_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12990_cast = slice_by_index(begin = var_12990_begin_0, end = var_12990_end_0, end_mask = var_12990_end_mask_0, x = var_12955_cast)[name = tensor("op_12990_cast")]; + tensor var_12991_begin_0 = const()[name = tensor("op_12991_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_12991_end_0 = const()[name = tensor("op_12991_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_12991_end_mask_0 = const()[name = tensor("op_12991_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12991_cast = slice_by_index(begin = var_12991_begin_0, end = var_12991_end_0, end_mask = var_12991_end_mask_0, x = var_12955_cast)[name = tensor("op_12991_cast")]; + tensor var_12992_begin_0 = const()[name = tensor("op_12992_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_12992_end_0 = const()[name = tensor("op_12992_end_0"), val = tensor([2, 64, 1, 1536])]; + tensor var_12992_end_mask_0 = const()[name = tensor("op_12992_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12992_cast = slice_by_index(begin = var_12992_begin_0, end = var_12992_end_0, end_mask = var_12992_end_mask_0, x = var_12955_cast)[name = tensor("op_12992_cast")]; + tensor var_12993_begin_0 = const()[name = tensor("op_12993_begin_0"), val = tensor([0, 0, 0, 1536])]; + tensor var_12993_end_0 = const()[name = tensor("op_12993_end_0"), val = tensor([2, 64, 1, 2048])]; + tensor var_12993_end_mask_0 = const()[name = tensor("op_12993_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12993_cast = slice_by_index(begin = var_12993_begin_0, end = var_12993_end_0, end_mask = var_12993_end_mask_0, x = var_12955_cast)[name = tensor("op_12993_cast")]; + tensor var_12994_begin_0 = const()[name = tensor("op_12994_begin_0"), val = tensor([0, 0, 0, 2048])]; + tensor var_12994_end_0 = const()[name = tensor("op_12994_end_0"), val = tensor([2, 64, 1, 2560])]; + tensor var_12994_end_mask_0 = const()[name = tensor("op_12994_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12994_cast = slice_by_index(begin = var_12994_begin_0, end = var_12994_end_0, end_mask = var_12994_end_mask_0, x = var_12955_cast)[name = tensor("op_12994_cast")]; + tensor var_12995_begin_0 = const()[name = tensor("op_12995_begin_0"), val = tensor([0, 0, 0, 2560])]; + tensor var_12995_end_0 = const()[name = tensor("op_12995_end_0"), val = tensor([2, 64, 1, 3072])]; + tensor var_12995_end_mask_0 = const()[name = tensor("op_12995_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12995_cast = slice_by_index(begin = var_12995_begin_0, end = var_12995_end_0, end_mask = var_12995_end_mask_0, x = var_12955_cast)[name = tensor("op_12995_cast")]; + tensor var_12996_begin_0 = const()[name = tensor("op_12996_begin_0"), val = tensor([0, 0, 0, 3072])]; + tensor var_12996_end_0 = const()[name = tensor("op_12996_end_0"), val = tensor([2, 64, 1, 3584])]; + tensor var_12996_end_mask_0 = const()[name = tensor("op_12996_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12996_cast = slice_by_index(begin = var_12996_begin_0, end = var_12996_end_0, end_mask = var_12996_end_mask_0, x = var_12955_cast)[name = tensor("op_12996_cast")]; + tensor var_12997_begin_0 = const()[name = tensor("op_12997_begin_0"), val = tensor([0, 0, 0, 3584])]; + tensor var_12997_end_0 = const()[name = tensor("op_12997_end_0"), val = tensor([2, 64, 1, 4096])]; + tensor var_12997_end_mask_0 = const()[name = tensor("op_12997_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12997_cast = slice_by_index(begin = var_12997_begin_0, end = var_12997_end_0, end_mask = var_12997_end_mask_0, x = var_12955_cast)[name = tensor("op_12997_cast")]; + tensor k_107_perm_0 = const()[name = tensor("k_107_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_13002_begin_0 = const()[name = tensor("op_13002_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_13002_end_0 = const()[name = tensor("op_13002_end_0"), val = tensor([2, 4096, 1, 64])]; + tensor var_13002_end_mask_0 = const()[name = tensor("op_13002_end_mask_0"), val = tensor([true, true, true, false])]; + tensor transpose_5 = transpose(perm = k_107_perm_0, x = k_105_cast)[name = tensor("transpose_5")]; + tensor var_13002_cast = slice_by_index(begin = var_13002_begin_0, end = var_13002_end_0, end_mask = var_13002_end_mask_0, x = transpose_5)[name = tensor("op_13002_cast")]; + tensor var_13006_begin_0 = const()[name = tensor("op_13006_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_13006_end_0 = const()[name = tensor("op_13006_end_0"), val = tensor([2, 4096, 1, 128])]; + tensor var_13006_end_mask_0 = const()[name = tensor("op_13006_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13006_cast = slice_by_index(begin = var_13006_begin_0, end = var_13006_end_0, end_mask = var_13006_end_mask_0, x = transpose_5)[name = tensor("op_13006_cast")]; + tensor var_13010_begin_0 = const()[name = tensor("op_13010_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_13010_end_0 = const()[name = tensor("op_13010_end_0"), val = tensor([2, 4096, 1, 192])]; + tensor var_13010_end_mask_0 = const()[name = tensor("op_13010_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13010_cast = slice_by_index(begin = var_13010_begin_0, end = var_13010_end_0, end_mask = var_13010_end_mask_0, x = transpose_5)[name = tensor("op_13010_cast")]; + tensor var_13014_begin_0 = const()[name = tensor("op_13014_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_13014_end_0 = const()[name = tensor("op_13014_end_0"), val = tensor([2, 4096, 1, 256])]; + tensor var_13014_end_mask_0 = const()[name = tensor("op_13014_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13014_cast = slice_by_index(begin = var_13014_begin_0, end = var_13014_end_0, end_mask = var_13014_end_mask_0, x = transpose_5)[name = tensor("op_13014_cast")]; + tensor var_13018_begin_0 = const()[name = tensor("op_13018_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_13018_end_0 = const()[name = tensor("op_13018_end_0"), val = tensor([2, 4096, 1, 320])]; + tensor var_13018_end_mask_0 = const()[name = tensor("op_13018_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13018_cast = slice_by_index(begin = var_13018_begin_0, end = var_13018_end_0, end_mask = var_13018_end_mask_0, x = transpose_5)[name = tensor("op_13018_cast")]; + tensor var_13020_begin_0 = const()[name = tensor("op_13020_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_13020_end_0 = const()[name = tensor("op_13020_end_0"), val = tensor([2, 64, 1, 4096])]; + tensor var_13020_end_mask_0 = const()[name = tensor("op_13020_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13020_cast = slice_by_index(begin = var_13020_begin_0, end = var_13020_end_0, end_mask = var_13020_end_mask_0, x = v_53_cast)[name = tensor("op_13020_cast")]; + tensor var_13024_begin_0 = const()[name = tensor("op_13024_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_13024_end_0 = const()[name = tensor("op_13024_end_0"), val = tensor([2, 128, 1, 4096])]; + tensor var_13024_end_mask_0 = const()[name = tensor("op_13024_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13024_cast = slice_by_index(begin = var_13024_begin_0, end = var_13024_end_0, end_mask = var_13024_end_mask_0, x = v_53_cast)[name = tensor("op_13024_cast")]; + tensor var_13028_begin_0 = const()[name = tensor("op_13028_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_13028_end_0 = const()[name = tensor("op_13028_end_0"), val = tensor([2, 192, 1, 4096])]; + tensor var_13028_end_mask_0 = const()[name = tensor("op_13028_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13028_cast = slice_by_index(begin = var_13028_begin_0, end = var_13028_end_0, end_mask = var_13028_end_mask_0, x = v_53_cast)[name = tensor("op_13028_cast")]; + tensor var_13032_begin_0 = const()[name = tensor("op_13032_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_13032_end_0 = const()[name = tensor("op_13032_end_0"), val = tensor([2, 256, 1, 4096])]; + tensor var_13032_end_mask_0 = const()[name = tensor("op_13032_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13032_cast = slice_by_index(begin = var_13032_begin_0, end = var_13032_end_0, end_mask = var_13032_end_mask_0, x = v_53_cast)[name = tensor("op_13032_cast")]; + tensor var_13036_begin_0 = const()[name = tensor("op_13036_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_13036_end_0 = const()[name = tensor("op_13036_end_0"), val = tensor([2, 320, 1, 4096])]; + tensor var_13036_end_mask_0 = const()[name = tensor("op_13036_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13036_cast = slice_by_index(begin = var_13036_begin_0, end = var_13036_end_0, end_mask = var_13036_end_mask_0, x = v_53_cast)[name = tensor("op_13036_cast")]; + tensor var_13040_equation_0 = const()[name = tensor("op_13040_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13040_cast = einsum(equation = var_13040_equation_0, values = (var_13002_cast, var_12958_cast))[name = tensor("op_13040_cast")]; + tensor var_13041_to_fp16 = const()[name = tensor("op_13041_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_721_cast = mul(x = var_13040_cast, y = var_13041_to_fp16)[name = tensor("aw_chunk_721_cast")]; + tensor var_13044_equation_0 = const()[name = tensor("op_13044_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13044_cast = einsum(equation = var_13044_equation_0, values = (var_13002_cast, var_12959_cast))[name = tensor("op_13044_cast")]; + tensor var_13045_to_fp16 = const()[name = tensor("op_13045_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_723_cast = mul(x = var_13044_cast, y = var_13045_to_fp16)[name = tensor("aw_chunk_723_cast")]; + tensor var_13048_equation_0 = const()[name = tensor("op_13048_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13048_cast = einsum(equation = var_13048_equation_0, values = (var_13002_cast, var_12960_cast))[name = tensor("op_13048_cast")]; + tensor var_13049_to_fp16 = const()[name = tensor("op_13049_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_725_cast = mul(x = var_13048_cast, y = var_13049_to_fp16)[name = tensor("aw_chunk_725_cast")]; + tensor var_13052_equation_0 = const()[name = tensor("op_13052_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13052_cast = einsum(equation = var_13052_equation_0, values = (var_13002_cast, var_12961_cast))[name = tensor("op_13052_cast")]; + tensor var_13053_to_fp16 = const()[name = tensor("op_13053_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_727_cast = mul(x = var_13052_cast, y = var_13053_to_fp16)[name = tensor("aw_chunk_727_cast")]; + tensor var_13056_equation_0 = const()[name = tensor("op_13056_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13056_cast = einsum(equation = var_13056_equation_0, values = (var_13002_cast, var_12962_cast))[name = tensor("op_13056_cast")]; + tensor var_13057_to_fp16 = const()[name = tensor("op_13057_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_729_cast = mul(x = var_13056_cast, y = var_13057_to_fp16)[name = tensor("aw_chunk_729_cast")]; + tensor var_13060_equation_0 = const()[name = tensor("op_13060_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13060_cast = einsum(equation = var_13060_equation_0, values = (var_13002_cast, var_12963_cast))[name = tensor("op_13060_cast")]; + tensor var_13061_to_fp16 = const()[name = tensor("op_13061_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_731_cast = mul(x = var_13060_cast, y = var_13061_to_fp16)[name = tensor("aw_chunk_731_cast")]; + tensor var_13064_equation_0 = const()[name = tensor("op_13064_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13064_cast = einsum(equation = var_13064_equation_0, values = (var_13002_cast, var_12964_cast))[name = tensor("op_13064_cast")]; + tensor var_13065_to_fp16 = const()[name = tensor("op_13065_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_733_cast = mul(x = var_13064_cast, y = var_13065_to_fp16)[name = tensor("aw_chunk_733_cast")]; + tensor var_13068_equation_0 = const()[name = tensor("op_13068_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13068_cast = einsum(equation = var_13068_equation_0, values = (var_13002_cast, var_12965_cast))[name = tensor("op_13068_cast")]; + tensor var_13069_to_fp16 = const()[name = tensor("op_13069_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_735_cast = mul(x = var_13068_cast, y = var_13069_to_fp16)[name = tensor("aw_chunk_735_cast")]; + tensor var_13072_equation_0 = const()[name = tensor("op_13072_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13072_cast = einsum(equation = var_13072_equation_0, values = (var_13006_cast, var_12966_cast))[name = tensor("op_13072_cast")]; + tensor var_13073_to_fp16 = const()[name = tensor("op_13073_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_737_cast = mul(x = var_13072_cast, y = var_13073_to_fp16)[name = tensor("aw_chunk_737_cast")]; + tensor var_13076_equation_0 = const()[name = tensor("op_13076_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13076_cast = einsum(equation = var_13076_equation_0, values = (var_13006_cast, var_12967_cast))[name = tensor("op_13076_cast")]; + tensor var_13077_to_fp16 = const()[name = tensor("op_13077_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_739_cast = mul(x = var_13076_cast, y = var_13077_to_fp16)[name = tensor("aw_chunk_739_cast")]; + tensor var_13080_equation_0 = const()[name = tensor("op_13080_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13080_cast = einsum(equation = var_13080_equation_0, values = (var_13006_cast, var_12968_cast))[name = tensor("op_13080_cast")]; + tensor var_13081_to_fp16 = const()[name = tensor("op_13081_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_741_cast = mul(x = var_13080_cast, y = var_13081_to_fp16)[name = tensor("aw_chunk_741_cast")]; + tensor var_13084_equation_0 = const()[name = tensor("op_13084_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13084_cast = einsum(equation = var_13084_equation_0, values = (var_13006_cast, var_12969_cast))[name = tensor("op_13084_cast")]; + tensor var_13085_to_fp16 = const()[name = tensor("op_13085_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_743_cast = mul(x = var_13084_cast, y = var_13085_to_fp16)[name = tensor("aw_chunk_743_cast")]; + tensor var_13088_equation_0 = const()[name = tensor("op_13088_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13088_cast = einsum(equation = var_13088_equation_0, values = (var_13006_cast, var_12970_cast))[name = tensor("op_13088_cast")]; + tensor var_13089_to_fp16 = const()[name = tensor("op_13089_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_745_cast = mul(x = var_13088_cast, y = var_13089_to_fp16)[name = tensor("aw_chunk_745_cast")]; + tensor var_13092_equation_0 = const()[name = tensor("op_13092_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13092_cast = einsum(equation = var_13092_equation_0, values = (var_13006_cast, var_12971_cast))[name = tensor("op_13092_cast")]; + tensor var_13093_to_fp16 = const()[name = tensor("op_13093_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_747_cast = mul(x = var_13092_cast, y = var_13093_to_fp16)[name = tensor("aw_chunk_747_cast")]; + tensor var_13096_equation_0 = const()[name = tensor("op_13096_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13096_cast = einsum(equation = var_13096_equation_0, values = (var_13006_cast, var_12972_cast))[name = tensor("op_13096_cast")]; + tensor var_13097_to_fp16 = const()[name = tensor("op_13097_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_749_cast = mul(x = var_13096_cast, y = var_13097_to_fp16)[name = tensor("aw_chunk_749_cast")]; + tensor var_13100_equation_0 = const()[name = tensor("op_13100_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13100_cast = einsum(equation = var_13100_equation_0, values = (var_13006_cast, var_12973_cast))[name = tensor("op_13100_cast")]; + tensor var_13101_to_fp16 = const()[name = tensor("op_13101_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_751_cast = mul(x = var_13100_cast, y = var_13101_to_fp16)[name = tensor("aw_chunk_751_cast")]; + tensor var_13104_equation_0 = const()[name = tensor("op_13104_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13104_cast = einsum(equation = var_13104_equation_0, values = (var_13010_cast, var_12974_cast))[name = tensor("op_13104_cast")]; + tensor var_13105_to_fp16 = const()[name = tensor("op_13105_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_753_cast = mul(x = var_13104_cast, y = var_13105_to_fp16)[name = tensor("aw_chunk_753_cast")]; + tensor var_13108_equation_0 = const()[name = tensor("op_13108_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13108_cast = einsum(equation = var_13108_equation_0, values = (var_13010_cast, var_12975_cast))[name = tensor("op_13108_cast")]; + tensor var_13109_to_fp16 = const()[name = tensor("op_13109_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_755_cast = mul(x = var_13108_cast, y = var_13109_to_fp16)[name = tensor("aw_chunk_755_cast")]; + tensor var_13112_equation_0 = const()[name = tensor("op_13112_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13112_cast = einsum(equation = var_13112_equation_0, values = (var_13010_cast, var_12976_cast))[name = tensor("op_13112_cast")]; + tensor var_13113_to_fp16 = const()[name = tensor("op_13113_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_757_cast = mul(x = var_13112_cast, y = var_13113_to_fp16)[name = tensor("aw_chunk_757_cast")]; + tensor var_13116_equation_0 = const()[name = tensor("op_13116_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13116_cast = einsum(equation = var_13116_equation_0, values = (var_13010_cast, var_12977_cast))[name = tensor("op_13116_cast")]; + tensor var_13117_to_fp16 = const()[name = tensor("op_13117_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_759_cast = mul(x = var_13116_cast, y = var_13117_to_fp16)[name = tensor("aw_chunk_759_cast")]; + tensor var_13120_equation_0 = const()[name = tensor("op_13120_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13120_cast = einsum(equation = var_13120_equation_0, values = (var_13010_cast, var_12978_cast))[name = tensor("op_13120_cast")]; + tensor var_13121_to_fp16 = const()[name = tensor("op_13121_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_761_cast = mul(x = var_13120_cast, y = var_13121_to_fp16)[name = tensor("aw_chunk_761_cast")]; + tensor var_13124_equation_0 = const()[name = tensor("op_13124_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13124_cast = einsum(equation = var_13124_equation_0, values = (var_13010_cast, var_12979_cast))[name = tensor("op_13124_cast")]; + tensor var_13125_to_fp16 = const()[name = tensor("op_13125_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_763_cast = mul(x = var_13124_cast, y = var_13125_to_fp16)[name = tensor("aw_chunk_763_cast")]; + tensor var_13128_equation_0 = const()[name = tensor("op_13128_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13128_cast = einsum(equation = var_13128_equation_0, values = (var_13010_cast, var_12980_cast))[name = tensor("op_13128_cast")]; + tensor var_13129_to_fp16 = const()[name = tensor("op_13129_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_765_cast = mul(x = var_13128_cast, y = var_13129_to_fp16)[name = tensor("aw_chunk_765_cast")]; + tensor var_13132_equation_0 = const()[name = tensor("op_13132_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13132_cast = einsum(equation = var_13132_equation_0, values = (var_13010_cast, var_12981_cast))[name = tensor("op_13132_cast")]; + tensor var_13133_to_fp16 = const()[name = tensor("op_13133_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_767_cast = mul(x = var_13132_cast, y = var_13133_to_fp16)[name = tensor("aw_chunk_767_cast")]; + tensor var_13136_equation_0 = const()[name = tensor("op_13136_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13136_cast = einsum(equation = var_13136_equation_0, values = (var_13014_cast, var_12982_cast))[name = tensor("op_13136_cast")]; + tensor var_13137_to_fp16 = const()[name = tensor("op_13137_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_769_cast = mul(x = var_13136_cast, y = var_13137_to_fp16)[name = tensor("aw_chunk_769_cast")]; + tensor var_13140_equation_0 = const()[name = tensor("op_13140_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13140_cast = einsum(equation = var_13140_equation_0, values = (var_13014_cast, var_12983_cast))[name = tensor("op_13140_cast")]; + tensor var_13141_to_fp16 = const()[name = tensor("op_13141_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_771_cast = mul(x = var_13140_cast, y = var_13141_to_fp16)[name = tensor("aw_chunk_771_cast")]; + tensor var_13144_equation_0 = const()[name = tensor("op_13144_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13144_cast = einsum(equation = var_13144_equation_0, values = (var_13014_cast, var_12984_cast))[name = tensor("op_13144_cast")]; + tensor var_13145_to_fp16 = const()[name = tensor("op_13145_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_773_cast = mul(x = var_13144_cast, y = var_13145_to_fp16)[name = tensor("aw_chunk_773_cast")]; + tensor var_13148_equation_0 = const()[name = tensor("op_13148_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13148_cast = einsum(equation = var_13148_equation_0, values = (var_13014_cast, var_12985_cast))[name = tensor("op_13148_cast")]; + tensor var_13149_to_fp16 = const()[name = tensor("op_13149_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_775_cast = mul(x = var_13148_cast, y = var_13149_to_fp16)[name = tensor("aw_chunk_775_cast")]; + tensor var_13152_equation_0 = const()[name = tensor("op_13152_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13152_cast = einsum(equation = var_13152_equation_0, values = (var_13014_cast, var_12986_cast))[name = tensor("op_13152_cast")]; + tensor var_13153_to_fp16 = const()[name = tensor("op_13153_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_777_cast = mul(x = var_13152_cast, y = var_13153_to_fp16)[name = tensor("aw_chunk_777_cast")]; + tensor var_13156_equation_0 = const()[name = tensor("op_13156_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13156_cast = einsum(equation = var_13156_equation_0, values = (var_13014_cast, var_12987_cast))[name = tensor("op_13156_cast")]; + tensor var_13157_to_fp16 = const()[name = tensor("op_13157_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_779_cast = mul(x = var_13156_cast, y = var_13157_to_fp16)[name = tensor("aw_chunk_779_cast")]; + tensor var_13160_equation_0 = const()[name = tensor("op_13160_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13160_cast = einsum(equation = var_13160_equation_0, values = (var_13014_cast, var_12988_cast))[name = tensor("op_13160_cast")]; + tensor var_13161_to_fp16 = const()[name = tensor("op_13161_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_781_cast = mul(x = var_13160_cast, y = var_13161_to_fp16)[name = tensor("aw_chunk_781_cast")]; + tensor var_13164_equation_0 = const()[name = tensor("op_13164_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13164_cast = einsum(equation = var_13164_equation_0, values = (var_13014_cast, var_12989_cast))[name = tensor("op_13164_cast")]; + tensor var_13165_to_fp16 = const()[name = tensor("op_13165_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_783_cast = mul(x = var_13164_cast, y = var_13165_to_fp16)[name = tensor("aw_chunk_783_cast")]; + tensor var_13168_equation_0 = const()[name = tensor("op_13168_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13168_cast = einsum(equation = var_13168_equation_0, values = (var_13018_cast, var_12990_cast))[name = tensor("op_13168_cast")]; + tensor var_13169_to_fp16 = const()[name = tensor("op_13169_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_785_cast = mul(x = var_13168_cast, y = var_13169_to_fp16)[name = tensor("aw_chunk_785_cast")]; + tensor var_13172_equation_0 = const()[name = tensor("op_13172_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13172_cast = einsum(equation = var_13172_equation_0, values = (var_13018_cast, var_12991_cast))[name = tensor("op_13172_cast")]; + tensor var_13173_to_fp16 = const()[name = tensor("op_13173_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_787_cast = mul(x = var_13172_cast, y = var_13173_to_fp16)[name = tensor("aw_chunk_787_cast")]; + tensor var_13176_equation_0 = const()[name = tensor("op_13176_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13176_cast = einsum(equation = var_13176_equation_0, values = (var_13018_cast, var_12992_cast))[name = tensor("op_13176_cast")]; + tensor var_13177_to_fp16 = const()[name = tensor("op_13177_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_789_cast = mul(x = var_13176_cast, y = var_13177_to_fp16)[name = tensor("aw_chunk_789_cast")]; + tensor var_13180_equation_0 = const()[name = tensor("op_13180_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13180_cast = einsum(equation = var_13180_equation_0, values = (var_13018_cast, var_12993_cast))[name = tensor("op_13180_cast")]; + tensor var_13181_to_fp16 = const()[name = tensor("op_13181_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_791_cast = mul(x = var_13180_cast, y = var_13181_to_fp16)[name = tensor("aw_chunk_791_cast")]; + tensor var_13184_equation_0 = const()[name = tensor("op_13184_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13184_cast = einsum(equation = var_13184_equation_0, values = (var_13018_cast, var_12994_cast))[name = tensor("op_13184_cast")]; + tensor var_13185_to_fp16 = const()[name = tensor("op_13185_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_793_cast = mul(x = var_13184_cast, y = var_13185_to_fp16)[name = tensor("aw_chunk_793_cast")]; + tensor var_13188_equation_0 = const()[name = tensor("op_13188_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13188_cast = einsum(equation = var_13188_equation_0, values = (var_13018_cast, var_12995_cast))[name = tensor("op_13188_cast")]; + tensor var_13189_to_fp16 = const()[name = tensor("op_13189_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_795_cast = mul(x = var_13188_cast, y = var_13189_to_fp16)[name = tensor("aw_chunk_795_cast")]; + tensor var_13192_equation_0 = const()[name = tensor("op_13192_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13192_cast = einsum(equation = var_13192_equation_0, values = (var_13018_cast, var_12996_cast))[name = tensor("op_13192_cast")]; + tensor var_13193_to_fp16 = const()[name = tensor("op_13193_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_797_cast = mul(x = var_13192_cast, y = var_13193_to_fp16)[name = tensor("aw_chunk_797_cast")]; + tensor var_13196_equation_0 = const()[name = tensor("op_13196_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13196_cast = einsum(equation = var_13196_equation_0, values = (var_13018_cast, var_12997_cast))[name = tensor("op_13196_cast")]; + tensor var_13197_to_fp16 = const()[name = tensor("op_13197_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_799_cast = mul(x = var_13196_cast, y = var_13197_to_fp16)[name = tensor("aw_chunk_799_cast")]; + tensor var_13199_cast = softmax(axis = var_12807, x = aw_chunk_721_cast)[name = tensor("op_13199_cast")]; + tensor var_13200_cast = softmax(axis = var_12807, x = aw_chunk_723_cast)[name = tensor("op_13200_cast")]; + tensor var_13201_cast = softmax(axis = var_12807, x = aw_chunk_725_cast)[name = tensor("op_13201_cast")]; + tensor var_13202_cast = softmax(axis = var_12807, x = aw_chunk_727_cast)[name = tensor("op_13202_cast")]; + tensor var_13203_cast = softmax(axis = var_12807, x = aw_chunk_729_cast)[name = tensor("op_13203_cast")]; + tensor var_13204_cast = softmax(axis = var_12807, x = aw_chunk_731_cast)[name = tensor("op_13204_cast")]; + tensor var_13205_cast = softmax(axis = var_12807, x = aw_chunk_733_cast)[name = tensor("op_13205_cast")]; + tensor var_13206_cast = softmax(axis = var_12807, x = aw_chunk_735_cast)[name = tensor("op_13206_cast")]; + tensor var_13207_cast = softmax(axis = var_12807, x = aw_chunk_737_cast)[name = tensor("op_13207_cast")]; + tensor var_13208_cast = softmax(axis = var_12807, x = aw_chunk_739_cast)[name = tensor("op_13208_cast")]; + tensor var_13209_cast = softmax(axis = var_12807, x = aw_chunk_741_cast)[name = tensor("op_13209_cast")]; + tensor var_13210_cast = softmax(axis = var_12807, x = aw_chunk_743_cast)[name = tensor("op_13210_cast")]; + tensor var_13211_cast = softmax(axis = var_12807, x = aw_chunk_745_cast)[name = tensor("op_13211_cast")]; + tensor var_13212_cast = softmax(axis = var_12807, x = aw_chunk_747_cast)[name = tensor("op_13212_cast")]; + tensor var_13213_cast = softmax(axis = var_12807, x = aw_chunk_749_cast)[name = tensor("op_13213_cast")]; + tensor var_13214_cast = softmax(axis = var_12807, x = aw_chunk_751_cast)[name = tensor("op_13214_cast")]; + tensor var_13215_cast = softmax(axis = var_12807, x = aw_chunk_753_cast)[name = tensor("op_13215_cast")]; + tensor var_13216_cast = softmax(axis = var_12807, x = aw_chunk_755_cast)[name = tensor("op_13216_cast")]; + tensor var_13217_cast = softmax(axis = var_12807, x = aw_chunk_757_cast)[name = tensor("op_13217_cast")]; + tensor var_13218_cast = softmax(axis = var_12807, x = aw_chunk_759_cast)[name = tensor("op_13218_cast")]; + tensor var_13219_cast = softmax(axis = var_12807, x = aw_chunk_761_cast)[name = tensor("op_13219_cast")]; + tensor var_13220_cast = softmax(axis = var_12807, x = aw_chunk_763_cast)[name = tensor("op_13220_cast")]; + tensor var_13221_cast = softmax(axis = var_12807, x = aw_chunk_765_cast)[name = tensor("op_13221_cast")]; + tensor var_13222_cast = softmax(axis = var_12807, x = aw_chunk_767_cast)[name = tensor("op_13222_cast")]; + tensor var_13223_cast = softmax(axis = var_12807, x = aw_chunk_769_cast)[name = tensor("op_13223_cast")]; + tensor var_13224_cast = softmax(axis = var_12807, x = aw_chunk_771_cast)[name = tensor("op_13224_cast")]; + tensor var_13225_cast = softmax(axis = var_12807, x = aw_chunk_773_cast)[name = tensor("op_13225_cast")]; + tensor var_13226_cast = softmax(axis = var_12807, x = aw_chunk_775_cast)[name = tensor("op_13226_cast")]; + tensor var_13227_cast = softmax(axis = var_12807, x = aw_chunk_777_cast)[name = tensor("op_13227_cast")]; + tensor var_13228_cast = softmax(axis = var_12807, x = aw_chunk_779_cast)[name = tensor("op_13228_cast")]; + tensor var_13229_cast = softmax(axis = var_12807, x = aw_chunk_781_cast)[name = tensor("op_13229_cast")]; + tensor var_13230_cast = softmax(axis = var_12807, x = aw_chunk_783_cast)[name = tensor("op_13230_cast")]; + tensor var_13231_cast = softmax(axis = var_12807, x = aw_chunk_785_cast)[name = tensor("op_13231_cast")]; + tensor var_13232_cast = softmax(axis = var_12807, x = aw_chunk_787_cast)[name = tensor("op_13232_cast")]; + tensor var_13233_cast = softmax(axis = var_12807, x = aw_chunk_789_cast)[name = tensor("op_13233_cast")]; + tensor var_13234_cast = softmax(axis = var_12807, x = aw_chunk_791_cast)[name = tensor("op_13234_cast")]; + tensor var_13235_cast = softmax(axis = var_12807, x = aw_chunk_793_cast)[name = tensor("op_13235_cast")]; + tensor var_13236_cast = softmax(axis = var_12807, x = aw_chunk_795_cast)[name = tensor("op_13236_cast")]; + tensor var_13237_cast = softmax(axis = var_12807, x = aw_chunk_797_cast)[name = tensor("op_13237_cast")]; + tensor var_13238_cast = softmax(axis = var_12807, x = aw_chunk_799_cast)[name = tensor("op_13238_cast")]; + tensor var_13240_equation_0 = const()[name = tensor("op_13240_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13240_cast = einsum(equation = var_13240_equation_0, values = (var_13020_cast, var_13199_cast))[name = tensor("op_13240_cast")]; + tensor var_13242_equation_0 = const()[name = tensor("op_13242_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13242_cast = einsum(equation = var_13242_equation_0, values = (var_13020_cast, var_13200_cast))[name = tensor("op_13242_cast")]; + tensor var_13244_equation_0 = const()[name = tensor("op_13244_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13244_cast = einsum(equation = var_13244_equation_0, values = (var_13020_cast, var_13201_cast))[name = tensor("op_13244_cast")]; + tensor var_13246_equation_0 = const()[name = tensor("op_13246_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13246_cast = einsum(equation = var_13246_equation_0, values = (var_13020_cast, var_13202_cast))[name = tensor("op_13246_cast")]; + tensor var_13248_equation_0 = const()[name = tensor("op_13248_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13248_cast = einsum(equation = var_13248_equation_0, values = (var_13020_cast, var_13203_cast))[name = tensor("op_13248_cast")]; + tensor var_13250_equation_0 = const()[name = tensor("op_13250_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13250_cast = einsum(equation = var_13250_equation_0, values = (var_13020_cast, var_13204_cast))[name = tensor("op_13250_cast")]; + tensor var_13252_equation_0 = const()[name = tensor("op_13252_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13252_cast = einsum(equation = var_13252_equation_0, values = (var_13020_cast, var_13205_cast))[name = tensor("op_13252_cast")]; + tensor var_13254_equation_0 = const()[name = tensor("op_13254_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13254_cast = einsum(equation = var_13254_equation_0, values = (var_13020_cast, var_13206_cast))[name = tensor("op_13254_cast")]; + tensor var_13256_equation_0 = const()[name = tensor("op_13256_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13256_cast = einsum(equation = var_13256_equation_0, values = (var_13024_cast, var_13207_cast))[name = tensor("op_13256_cast")]; + tensor var_13258_equation_0 = const()[name = tensor("op_13258_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13258_cast = einsum(equation = var_13258_equation_0, values = (var_13024_cast, var_13208_cast))[name = tensor("op_13258_cast")]; + tensor var_13260_equation_0 = const()[name = tensor("op_13260_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13260_cast = einsum(equation = var_13260_equation_0, values = (var_13024_cast, var_13209_cast))[name = tensor("op_13260_cast")]; + tensor var_13262_equation_0 = const()[name = tensor("op_13262_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13262_cast = einsum(equation = var_13262_equation_0, values = (var_13024_cast, var_13210_cast))[name = tensor("op_13262_cast")]; + tensor var_13264_equation_0 = const()[name = tensor("op_13264_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13264_cast = einsum(equation = var_13264_equation_0, values = (var_13024_cast, var_13211_cast))[name = tensor("op_13264_cast")]; + tensor var_13266_equation_0 = const()[name = tensor("op_13266_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13266_cast = einsum(equation = var_13266_equation_0, values = (var_13024_cast, var_13212_cast))[name = tensor("op_13266_cast")]; + tensor var_13268_equation_0 = const()[name = tensor("op_13268_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13268_cast = einsum(equation = var_13268_equation_0, values = (var_13024_cast, var_13213_cast))[name = tensor("op_13268_cast")]; + tensor var_13270_equation_0 = const()[name = tensor("op_13270_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13270_cast = einsum(equation = var_13270_equation_0, values = (var_13024_cast, var_13214_cast))[name = tensor("op_13270_cast")]; + tensor var_13272_equation_0 = const()[name = tensor("op_13272_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13272_cast = einsum(equation = var_13272_equation_0, values = (var_13028_cast, var_13215_cast))[name = tensor("op_13272_cast")]; + tensor var_13274_equation_0 = const()[name = tensor("op_13274_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13274_cast = einsum(equation = var_13274_equation_0, values = (var_13028_cast, var_13216_cast))[name = tensor("op_13274_cast")]; + tensor var_13276_equation_0 = const()[name = tensor("op_13276_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13276_cast = einsum(equation = var_13276_equation_0, values = (var_13028_cast, var_13217_cast))[name = tensor("op_13276_cast")]; + tensor var_13278_equation_0 = const()[name = tensor("op_13278_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13278_cast = einsum(equation = var_13278_equation_0, values = (var_13028_cast, var_13218_cast))[name = tensor("op_13278_cast")]; + tensor var_13280_equation_0 = const()[name = tensor("op_13280_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13280_cast = einsum(equation = var_13280_equation_0, values = (var_13028_cast, var_13219_cast))[name = tensor("op_13280_cast")]; + tensor var_13282_equation_0 = const()[name = tensor("op_13282_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13282_cast = einsum(equation = var_13282_equation_0, values = (var_13028_cast, var_13220_cast))[name = tensor("op_13282_cast")]; + tensor var_13284_equation_0 = const()[name = tensor("op_13284_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13284_cast = einsum(equation = var_13284_equation_0, values = (var_13028_cast, var_13221_cast))[name = tensor("op_13284_cast")]; + tensor var_13286_equation_0 = const()[name = tensor("op_13286_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13286_cast = einsum(equation = var_13286_equation_0, values = (var_13028_cast, var_13222_cast))[name = tensor("op_13286_cast")]; + tensor var_13288_equation_0 = const()[name = tensor("op_13288_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13288_cast = einsum(equation = var_13288_equation_0, values = (var_13032_cast, var_13223_cast))[name = tensor("op_13288_cast")]; + tensor var_13290_equation_0 = const()[name = tensor("op_13290_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13290_cast = einsum(equation = var_13290_equation_0, values = (var_13032_cast, var_13224_cast))[name = tensor("op_13290_cast")]; + tensor var_13292_equation_0 = const()[name = tensor("op_13292_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13292_cast = einsum(equation = var_13292_equation_0, values = (var_13032_cast, var_13225_cast))[name = tensor("op_13292_cast")]; + tensor var_13294_equation_0 = const()[name = tensor("op_13294_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13294_cast = einsum(equation = var_13294_equation_0, values = (var_13032_cast, var_13226_cast))[name = tensor("op_13294_cast")]; + tensor var_13296_equation_0 = const()[name = tensor("op_13296_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13296_cast = einsum(equation = var_13296_equation_0, values = (var_13032_cast, var_13227_cast))[name = tensor("op_13296_cast")]; + tensor var_13298_equation_0 = const()[name = tensor("op_13298_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13298_cast = einsum(equation = var_13298_equation_0, values = (var_13032_cast, var_13228_cast))[name = tensor("op_13298_cast")]; + tensor var_13300_equation_0 = const()[name = tensor("op_13300_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13300_cast = einsum(equation = var_13300_equation_0, values = (var_13032_cast, var_13229_cast))[name = tensor("op_13300_cast")]; + tensor var_13302_equation_0 = const()[name = tensor("op_13302_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13302_cast = einsum(equation = var_13302_equation_0, values = (var_13032_cast, var_13230_cast))[name = tensor("op_13302_cast")]; + tensor var_13304_equation_0 = const()[name = tensor("op_13304_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13304_cast = einsum(equation = var_13304_equation_0, values = (var_13036_cast, var_13231_cast))[name = tensor("op_13304_cast")]; + tensor var_13306_equation_0 = const()[name = tensor("op_13306_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13306_cast = einsum(equation = var_13306_equation_0, values = (var_13036_cast, var_13232_cast))[name = tensor("op_13306_cast")]; + tensor var_13308_equation_0 = const()[name = tensor("op_13308_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13308_cast = einsum(equation = var_13308_equation_0, values = (var_13036_cast, var_13233_cast))[name = tensor("op_13308_cast")]; + tensor var_13310_equation_0 = const()[name = tensor("op_13310_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13310_cast = einsum(equation = var_13310_equation_0, values = (var_13036_cast, var_13234_cast))[name = tensor("op_13310_cast")]; + tensor var_13312_equation_0 = const()[name = tensor("op_13312_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13312_cast = einsum(equation = var_13312_equation_0, values = (var_13036_cast, var_13235_cast))[name = tensor("op_13312_cast")]; + tensor var_13314_equation_0 = const()[name = tensor("op_13314_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13314_cast = einsum(equation = var_13314_equation_0, values = (var_13036_cast, var_13236_cast))[name = tensor("op_13314_cast")]; + tensor var_13316_equation_0 = const()[name = tensor("op_13316_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13316_cast = einsum(equation = var_13316_equation_0, values = (var_13036_cast, var_13237_cast))[name = tensor("op_13316_cast")]; + tensor var_13318_equation_0 = const()[name = tensor("op_13318_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13318_cast = einsum(equation = var_13318_equation_0, values = (var_13036_cast, var_13238_cast))[name = tensor("op_13318_cast")]; + tensor var_13320_interleave_0 = const()[name = tensor("op_13320_interleave_0"), val = tensor(false)]; + tensor var_13320_cast = concat(axis = var_12782, interleave = var_13320_interleave_0, values = (var_13240_cast, var_13242_cast, var_13244_cast, var_13246_cast, var_13248_cast, var_13250_cast, var_13252_cast, var_13254_cast))[name = tensor("op_13320_cast")]; + tensor var_13322_interleave_0 = const()[name = tensor("op_13322_interleave_0"), val = tensor(false)]; + tensor var_13322_cast = concat(axis = var_12782, interleave = var_13322_interleave_0, values = (var_13256_cast, var_13258_cast, var_13260_cast, var_13262_cast, var_13264_cast, var_13266_cast, var_13268_cast, var_13270_cast))[name = tensor("op_13322_cast")]; + tensor var_13324_interleave_0 = const()[name = tensor("op_13324_interleave_0"), val = tensor(false)]; + tensor var_13324_cast = concat(axis = var_12782, interleave = var_13324_interleave_0, values = (var_13272_cast, var_13274_cast, var_13276_cast, var_13278_cast, var_13280_cast, var_13282_cast, var_13284_cast, var_13286_cast))[name = tensor("op_13324_cast")]; + tensor var_13326_interleave_0 = const()[name = tensor("op_13326_interleave_0"), val = tensor(false)]; + tensor var_13326_cast = concat(axis = var_12782, interleave = var_13326_interleave_0, values = (var_13288_cast, var_13290_cast, var_13292_cast, var_13294_cast, var_13296_cast, var_13298_cast, var_13300_cast, var_13302_cast))[name = tensor("op_13326_cast")]; + tensor var_13328_interleave_0 = const()[name = tensor("op_13328_interleave_0"), val = tensor(false)]; + tensor var_13328_cast = concat(axis = var_12782, interleave = var_13328_interleave_0, values = (var_13304_cast, var_13306_cast, var_13308_cast, var_13310_cast, var_13312_cast, var_13314_cast, var_13316_cast, var_13318_cast))[name = tensor("op_13328_cast")]; + tensor input_465_interleave_0 = const()[name = tensor("input_465_interleave_0"), val = tensor(false)]; + tensor input_465_cast = concat(axis = var_12807, interleave = input_465_interleave_0, values = (var_13320_cast, var_13322_cast, var_13324_cast, var_13326_cast, var_13328_cast))[name = tensor("input_465_cast")]; + tensor var_13334 = const()[name = tensor("op_13334"), val = tensor([1, 1])]; + tensor var_13336 = const()[name = tensor("op_13336"), val = tensor([1, 1])]; + tensor var_13338_pad_type_0 = const()[name = tensor("op_13338_pad_type_0"), val = tensor("custom")]; + tensor var_13338_pad_0 = const()[name = tensor("op_13338_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_3_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(639050944))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(639127808))), name = tensor("up_blocks_3_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16_palettized"), shape = tensor([320, 320, 1, 1])]; + tensor up_blocks_3_attentions_0_transformer_blocks_0_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("up_blocks_3_attentions_0_transformer_blocks_0_attn1_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(639128000)))]; + tensor var_13338_cast = conv(bias = up_blocks_3_attentions_0_transformer_blocks_0_attn1_to_out_0_bias_to_fp16, dilations = var_13336, groups = var_12807, pad = var_13338_pad_0, pad_type = var_13338_pad_type_0, strides = var_13334, weight = up_blocks_3_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16_palettized, x = input_465_cast)[name = tensor("op_13338_cast")]; + tensor inputs_81_cast = add(x = var_13338_cast, y = inputs_79_cast)[name = tensor("inputs_81_cast")]; + tensor var_13342 = const()[name = tensor("op_13342"), val = tensor([1])]; + tensor channels_mean_81_cast = reduce_mean(axes = var_13342, keep_dims = var_12802, x = inputs_81_cast)[name = tensor("channels_mean_81_cast")]; + tensor zero_mean_81_cast = sub(x = inputs_81_cast, y = channels_mean_81_cast)[name = tensor("zero_mean_81_cast")]; + tensor zero_mean_sq_81_cast = mul(x = zero_mean_81_cast, y = zero_mean_81_cast)[name = tensor("zero_mean_sq_81_cast")]; + tensor var_13346 = const()[name = tensor("op_13346"), val = tensor([1])]; + tensor var_13347_cast = reduce_mean(axes = var_13346, keep_dims = var_12802, x = zero_mean_sq_81_cast)[name = tensor("op_13347_cast")]; + tensor var_13348_to_fp16 = const()[name = tensor("op_13348_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_13349_cast = add(x = var_13347_cast, y = var_13348_to_fp16)[name = tensor("op_13349_cast")]; + tensor denom_81_epsilon_0_to_fp16 = const()[name = tensor("denom_81_epsilon_0_to_fp16"), val = tensor(0x1p-24)]; + tensor denom_81_cast = rsqrt(epsilon = denom_81_epsilon_0_to_fp16, x = var_13349_cast)[name = tensor("denom_81_cast")]; + tensor out_81_cast = mul(x = zero_mean_81_cast, y = denom_81_cast)[name = tensor("out_81_cast")]; + tensor var_13353_to_fp16 = const()[name = tensor("op_13353_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(639128704)))]; + tensor var_13354_cast = add(x = out_81_cast, y = var_13353_to_fp16)[name = tensor("op_13354_cast")]; + tensor var_13356_to_fp16 = const()[name = tensor("op_13356_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(639129408)))]; + tensor hidden_states_289_cast = mul(x = var_13354_cast, y = var_13356_to_fp16)[name = tensor("hidden_states_289_cast")]; + tensor var_13363 = const()[name = tensor("op_13363"), val = tensor([1, 1])]; + tensor var_13365 = const()[name = tensor("op_13365"), val = tensor([1, 1])]; + tensor q_55_pad_type_0 = const()[name = tensor("q_55_pad_type_0"), val = tensor("custom")]; + tensor q_55_pad_0 = const()[name = tensor("q_55_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_3_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(639130112))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(639206976))), name = tensor("up_blocks_3_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16_palettized"), shape = tensor([320, 320, 1, 1])]; + tensor q_55_cast = conv(dilations = var_13365, groups = var_12807, pad = q_55_pad_0, pad_type = q_55_pad_type_0, strides = var_13363, weight = up_blocks_3_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16_palettized, x = hidden_states_289_cast)[name = tensor("q_55_cast")]; + tensor var_13369 = const()[name = tensor("op_13369"), val = tensor([1, 1])]; + tensor var_13371 = const()[name = tensor("op_13371"), val = tensor([1, 1])]; + tensor k_109_pad_type_0 = const()[name = tensor("k_109_pad_type_0"), val = tensor("custom")]; + tensor k_109_pad_0 = const()[name = tensor("k_109_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_3_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(639207168))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(639452992))), name = tensor("up_blocks_3_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized"), shape = tensor([320, 1024, 1, 1])]; + tensor k_109_cast = conv(dilations = var_13371, groups = var_12807, pad = k_109_pad_0, pad_type = k_109_pad_type_0, strides = var_13369, weight = up_blocks_3_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor("k_109_cast")]; + tensor var_13375 = const()[name = tensor("op_13375"), val = tensor([1, 1])]; + tensor var_13377 = const()[name = tensor("op_13377"), val = tensor([1, 1])]; + tensor v_55_pad_type_0 = const()[name = tensor("v_55_pad_type_0"), val = tensor("custom")]; + tensor v_55_pad_0 = const()[name = tensor("v_55_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_3_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(639453184))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(639699008))), name = tensor("up_blocks_3_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized"), shape = tensor([320, 1024, 1, 1])]; + tensor v_55_cast = conv(dilations = var_13377, groups = var_12807, pad = v_55_pad_0, pad_type = v_55_pad_type_0, strides = var_13375, weight = up_blocks_3_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor("v_55_cast")]; + tensor var_13381_begin_0 = const()[name = tensor("op_13381_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_13381_end_0 = const()[name = tensor("op_13381_end_0"), val = tensor([2, 64, 1, 4096])]; + tensor var_13381_end_mask_0 = const()[name = tensor("op_13381_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13381_cast = slice_by_index(begin = var_13381_begin_0, end = var_13381_end_0, end_mask = var_13381_end_mask_0, x = q_55_cast)[name = tensor("op_13381_cast")]; + tensor var_13385_begin_0 = const()[name = tensor("op_13385_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_13385_end_0 = const()[name = tensor("op_13385_end_0"), val = tensor([2, 128, 1, 4096])]; + tensor var_13385_end_mask_0 = const()[name = tensor("op_13385_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13385_cast = slice_by_index(begin = var_13385_begin_0, end = var_13385_end_0, end_mask = var_13385_end_mask_0, x = q_55_cast)[name = tensor("op_13385_cast")]; + tensor var_13389_begin_0 = const()[name = tensor("op_13389_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_13389_end_0 = const()[name = tensor("op_13389_end_0"), val = tensor([2, 192, 1, 4096])]; + tensor var_13389_end_mask_0 = const()[name = tensor("op_13389_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13389_cast = slice_by_index(begin = var_13389_begin_0, end = var_13389_end_0, end_mask = var_13389_end_mask_0, x = q_55_cast)[name = tensor("op_13389_cast")]; + tensor var_13393_begin_0 = const()[name = tensor("op_13393_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_13393_end_0 = const()[name = tensor("op_13393_end_0"), val = tensor([2, 256, 1, 4096])]; + tensor var_13393_end_mask_0 = const()[name = tensor("op_13393_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13393_cast = slice_by_index(begin = var_13393_begin_0, end = var_13393_end_0, end_mask = var_13393_end_mask_0, x = q_55_cast)[name = tensor("op_13393_cast")]; + tensor var_13397_begin_0 = const()[name = tensor("op_13397_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_13397_end_0 = const()[name = tensor("op_13397_end_0"), val = tensor([2, 320, 1, 4096])]; + tensor var_13397_end_mask_0 = const()[name = tensor("op_13397_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13397_cast = slice_by_index(begin = var_13397_begin_0, end = var_13397_end_0, end_mask = var_13397_end_mask_0, x = q_55_cast)[name = tensor("op_13397_cast")]; + tensor var_13400_begin_0 = const()[name = tensor("op_13400_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_13400_end_0 = const()[name = tensor("op_13400_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_13400_end_mask_0 = const()[name = tensor("op_13400_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13400_cast = slice_by_index(begin = var_13400_begin_0, end = var_13400_end_0, end_mask = var_13400_end_mask_0, x = var_13381_cast)[name = tensor("op_13400_cast")]; + tensor var_13401_begin_0 = const()[name = tensor("op_13401_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_13401_end_0 = const()[name = tensor("op_13401_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_13401_end_mask_0 = const()[name = tensor("op_13401_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13401_cast = slice_by_index(begin = var_13401_begin_0, end = var_13401_end_0, end_mask = var_13401_end_mask_0, x = var_13381_cast)[name = tensor("op_13401_cast")]; + tensor var_13402_begin_0 = const()[name = tensor("op_13402_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_13402_end_0 = const()[name = tensor("op_13402_end_0"), val = tensor([2, 64, 1, 1536])]; + tensor var_13402_end_mask_0 = const()[name = tensor("op_13402_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13402_cast = slice_by_index(begin = var_13402_begin_0, end = var_13402_end_0, end_mask = var_13402_end_mask_0, x = var_13381_cast)[name = tensor("op_13402_cast")]; + tensor var_13403_begin_0 = const()[name = tensor("op_13403_begin_0"), val = tensor([0, 0, 0, 1536])]; + tensor var_13403_end_0 = const()[name = tensor("op_13403_end_0"), val = tensor([2, 64, 1, 2048])]; + tensor var_13403_end_mask_0 = const()[name = tensor("op_13403_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13403_cast = slice_by_index(begin = var_13403_begin_0, end = var_13403_end_0, end_mask = var_13403_end_mask_0, x = var_13381_cast)[name = tensor("op_13403_cast")]; + tensor var_13404_begin_0 = const()[name = tensor("op_13404_begin_0"), val = tensor([0, 0, 0, 2048])]; + tensor var_13404_end_0 = const()[name = tensor("op_13404_end_0"), val = tensor([2, 64, 1, 2560])]; + tensor var_13404_end_mask_0 = const()[name = tensor("op_13404_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13404_cast = slice_by_index(begin = var_13404_begin_0, end = var_13404_end_0, end_mask = var_13404_end_mask_0, x = var_13381_cast)[name = tensor("op_13404_cast")]; + tensor var_13405_begin_0 = const()[name = tensor("op_13405_begin_0"), val = tensor([0, 0, 0, 2560])]; + tensor var_13405_end_0 = const()[name = tensor("op_13405_end_0"), val = tensor([2, 64, 1, 3072])]; + tensor var_13405_end_mask_0 = const()[name = tensor("op_13405_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13405_cast = slice_by_index(begin = var_13405_begin_0, end = var_13405_end_0, end_mask = var_13405_end_mask_0, x = var_13381_cast)[name = tensor("op_13405_cast")]; + tensor var_13406_begin_0 = const()[name = tensor("op_13406_begin_0"), val = tensor([0, 0, 0, 3072])]; + tensor var_13406_end_0 = const()[name = tensor("op_13406_end_0"), val = tensor([2, 64, 1, 3584])]; + tensor var_13406_end_mask_0 = const()[name = tensor("op_13406_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13406_cast = slice_by_index(begin = var_13406_begin_0, end = var_13406_end_0, end_mask = var_13406_end_mask_0, x = var_13381_cast)[name = tensor("op_13406_cast")]; + tensor var_13407_begin_0 = const()[name = tensor("op_13407_begin_0"), val = tensor([0, 0, 0, 3584])]; + tensor var_13407_end_0 = const()[name = tensor("op_13407_end_0"), val = tensor([2, 64, 1, 4096])]; + tensor var_13407_end_mask_0 = const()[name = tensor("op_13407_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13407_cast = slice_by_index(begin = var_13407_begin_0, end = var_13407_end_0, end_mask = var_13407_end_mask_0, x = var_13381_cast)[name = tensor("op_13407_cast")]; + tensor var_13408_begin_0 = const()[name = tensor("op_13408_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_13408_end_0 = const()[name = tensor("op_13408_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_13408_end_mask_0 = const()[name = tensor("op_13408_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13408_cast = slice_by_index(begin = var_13408_begin_0, end = var_13408_end_0, end_mask = var_13408_end_mask_0, x = var_13385_cast)[name = tensor("op_13408_cast")]; + tensor var_13409_begin_0 = const()[name = tensor("op_13409_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_13409_end_0 = const()[name = tensor("op_13409_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_13409_end_mask_0 = const()[name = tensor("op_13409_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13409_cast = slice_by_index(begin = var_13409_begin_0, end = var_13409_end_0, end_mask = var_13409_end_mask_0, x = var_13385_cast)[name = tensor("op_13409_cast")]; + tensor var_13410_begin_0 = const()[name = tensor("op_13410_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_13410_end_0 = const()[name = tensor("op_13410_end_0"), val = tensor([2, 64, 1, 1536])]; + tensor var_13410_end_mask_0 = const()[name = tensor("op_13410_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13410_cast = slice_by_index(begin = var_13410_begin_0, end = var_13410_end_0, end_mask = var_13410_end_mask_0, x = var_13385_cast)[name = tensor("op_13410_cast")]; + tensor var_13411_begin_0 = const()[name = tensor("op_13411_begin_0"), val = tensor([0, 0, 0, 1536])]; + tensor var_13411_end_0 = const()[name = tensor("op_13411_end_0"), val = tensor([2, 64, 1, 2048])]; + tensor var_13411_end_mask_0 = const()[name = tensor("op_13411_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13411_cast = slice_by_index(begin = var_13411_begin_0, end = var_13411_end_0, end_mask = var_13411_end_mask_0, x = var_13385_cast)[name = tensor("op_13411_cast")]; + tensor var_13412_begin_0 = const()[name = tensor("op_13412_begin_0"), val = tensor([0, 0, 0, 2048])]; + tensor var_13412_end_0 = const()[name = tensor("op_13412_end_0"), val = tensor([2, 64, 1, 2560])]; + tensor var_13412_end_mask_0 = const()[name = tensor("op_13412_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13412_cast = slice_by_index(begin = var_13412_begin_0, end = var_13412_end_0, end_mask = var_13412_end_mask_0, x = var_13385_cast)[name = tensor("op_13412_cast")]; + tensor var_13413_begin_0 = const()[name = tensor("op_13413_begin_0"), val = tensor([0, 0, 0, 2560])]; + tensor var_13413_end_0 = const()[name = tensor("op_13413_end_0"), val = tensor([2, 64, 1, 3072])]; + tensor var_13413_end_mask_0 = const()[name = tensor("op_13413_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13413_cast = slice_by_index(begin = var_13413_begin_0, end = var_13413_end_0, end_mask = var_13413_end_mask_0, x = var_13385_cast)[name = tensor("op_13413_cast")]; + tensor var_13414_begin_0 = const()[name = tensor("op_13414_begin_0"), val = tensor([0, 0, 0, 3072])]; + tensor var_13414_end_0 = const()[name = tensor("op_13414_end_0"), val = tensor([2, 64, 1, 3584])]; + tensor var_13414_end_mask_0 = const()[name = tensor("op_13414_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13414_cast = slice_by_index(begin = var_13414_begin_0, end = var_13414_end_0, end_mask = var_13414_end_mask_0, x = var_13385_cast)[name = tensor("op_13414_cast")]; + tensor var_13415_begin_0 = const()[name = tensor("op_13415_begin_0"), val = tensor([0, 0, 0, 3584])]; + tensor var_13415_end_0 = const()[name = tensor("op_13415_end_0"), val = tensor([2, 64, 1, 4096])]; + tensor var_13415_end_mask_0 = const()[name = tensor("op_13415_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13415_cast = slice_by_index(begin = var_13415_begin_0, end = var_13415_end_0, end_mask = var_13415_end_mask_0, x = var_13385_cast)[name = tensor("op_13415_cast")]; + tensor var_13416_begin_0 = const()[name = tensor("op_13416_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_13416_end_0 = const()[name = tensor("op_13416_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_13416_end_mask_0 = const()[name = tensor("op_13416_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13416_cast = slice_by_index(begin = var_13416_begin_0, end = var_13416_end_0, end_mask = var_13416_end_mask_0, x = var_13389_cast)[name = tensor("op_13416_cast")]; + tensor var_13417_begin_0 = const()[name = tensor("op_13417_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_13417_end_0 = const()[name = tensor("op_13417_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_13417_end_mask_0 = const()[name = tensor("op_13417_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13417_cast = slice_by_index(begin = var_13417_begin_0, end = var_13417_end_0, end_mask = var_13417_end_mask_0, x = var_13389_cast)[name = tensor("op_13417_cast")]; + tensor var_13418_begin_0 = const()[name = tensor("op_13418_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_13418_end_0 = const()[name = tensor("op_13418_end_0"), val = tensor([2, 64, 1, 1536])]; + tensor var_13418_end_mask_0 = const()[name = tensor("op_13418_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13418_cast = slice_by_index(begin = var_13418_begin_0, end = var_13418_end_0, end_mask = var_13418_end_mask_0, x = var_13389_cast)[name = tensor("op_13418_cast")]; + tensor var_13419_begin_0 = const()[name = tensor("op_13419_begin_0"), val = tensor([0, 0, 0, 1536])]; + tensor var_13419_end_0 = const()[name = tensor("op_13419_end_0"), val = tensor([2, 64, 1, 2048])]; + tensor var_13419_end_mask_0 = const()[name = tensor("op_13419_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13419_cast = slice_by_index(begin = var_13419_begin_0, end = var_13419_end_0, end_mask = var_13419_end_mask_0, x = var_13389_cast)[name = tensor("op_13419_cast")]; + tensor var_13420_begin_0 = const()[name = tensor("op_13420_begin_0"), val = tensor([0, 0, 0, 2048])]; + tensor var_13420_end_0 = const()[name = tensor("op_13420_end_0"), val = tensor([2, 64, 1, 2560])]; + tensor var_13420_end_mask_0 = const()[name = tensor("op_13420_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13420_cast = slice_by_index(begin = var_13420_begin_0, end = var_13420_end_0, end_mask = var_13420_end_mask_0, x = var_13389_cast)[name = tensor("op_13420_cast")]; + tensor var_13421_begin_0 = const()[name = tensor("op_13421_begin_0"), val = tensor([0, 0, 0, 2560])]; + tensor var_13421_end_0 = const()[name = tensor("op_13421_end_0"), val = tensor([2, 64, 1, 3072])]; + tensor var_13421_end_mask_0 = const()[name = tensor("op_13421_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13421_cast = slice_by_index(begin = var_13421_begin_0, end = var_13421_end_0, end_mask = var_13421_end_mask_0, x = var_13389_cast)[name = tensor("op_13421_cast")]; + tensor var_13422_begin_0 = const()[name = tensor("op_13422_begin_0"), val = tensor([0, 0, 0, 3072])]; + tensor var_13422_end_0 = const()[name = tensor("op_13422_end_0"), val = tensor([2, 64, 1, 3584])]; + tensor var_13422_end_mask_0 = const()[name = tensor("op_13422_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13422_cast = slice_by_index(begin = var_13422_begin_0, end = var_13422_end_0, end_mask = var_13422_end_mask_0, x = var_13389_cast)[name = tensor("op_13422_cast")]; + tensor var_13423_begin_0 = const()[name = tensor("op_13423_begin_0"), val = tensor([0, 0, 0, 3584])]; + tensor var_13423_end_0 = const()[name = tensor("op_13423_end_0"), val = tensor([2, 64, 1, 4096])]; + tensor var_13423_end_mask_0 = const()[name = tensor("op_13423_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13423_cast = slice_by_index(begin = var_13423_begin_0, end = var_13423_end_0, end_mask = var_13423_end_mask_0, x = var_13389_cast)[name = tensor("op_13423_cast")]; + tensor var_13424_begin_0 = const()[name = tensor("op_13424_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_13424_end_0 = const()[name = tensor("op_13424_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_13424_end_mask_0 = const()[name = tensor("op_13424_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13424_cast = slice_by_index(begin = var_13424_begin_0, end = var_13424_end_0, end_mask = var_13424_end_mask_0, x = var_13393_cast)[name = tensor("op_13424_cast")]; + tensor var_13425_begin_0 = const()[name = tensor("op_13425_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_13425_end_0 = const()[name = tensor("op_13425_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_13425_end_mask_0 = const()[name = tensor("op_13425_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13425_cast = slice_by_index(begin = var_13425_begin_0, end = var_13425_end_0, end_mask = var_13425_end_mask_0, x = var_13393_cast)[name = tensor("op_13425_cast")]; + tensor var_13426_begin_0 = const()[name = tensor("op_13426_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_13426_end_0 = const()[name = tensor("op_13426_end_0"), val = tensor([2, 64, 1, 1536])]; + tensor var_13426_end_mask_0 = const()[name = tensor("op_13426_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13426_cast = slice_by_index(begin = var_13426_begin_0, end = var_13426_end_0, end_mask = var_13426_end_mask_0, x = var_13393_cast)[name = tensor("op_13426_cast")]; + tensor var_13427_begin_0 = const()[name = tensor("op_13427_begin_0"), val = tensor([0, 0, 0, 1536])]; + tensor var_13427_end_0 = const()[name = tensor("op_13427_end_0"), val = tensor([2, 64, 1, 2048])]; + tensor var_13427_end_mask_0 = const()[name = tensor("op_13427_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13427_cast = slice_by_index(begin = var_13427_begin_0, end = var_13427_end_0, end_mask = var_13427_end_mask_0, x = var_13393_cast)[name = tensor("op_13427_cast")]; + tensor var_13428_begin_0 = const()[name = tensor("op_13428_begin_0"), val = tensor([0, 0, 0, 2048])]; + tensor var_13428_end_0 = const()[name = tensor("op_13428_end_0"), val = tensor([2, 64, 1, 2560])]; + tensor var_13428_end_mask_0 = const()[name = tensor("op_13428_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13428_cast = slice_by_index(begin = var_13428_begin_0, end = var_13428_end_0, end_mask = var_13428_end_mask_0, x = var_13393_cast)[name = tensor("op_13428_cast")]; + tensor var_13429_begin_0 = const()[name = tensor("op_13429_begin_0"), val = tensor([0, 0, 0, 2560])]; + tensor var_13429_end_0 = const()[name = tensor("op_13429_end_0"), val = tensor([2, 64, 1, 3072])]; + tensor var_13429_end_mask_0 = const()[name = tensor("op_13429_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13429_cast = slice_by_index(begin = var_13429_begin_0, end = var_13429_end_0, end_mask = var_13429_end_mask_0, x = var_13393_cast)[name = tensor("op_13429_cast")]; + tensor var_13430_begin_0 = const()[name = tensor("op_13430_begin_0"), val = tensor([0, 0, 0, 3072])]; + tensor var_13430_end_0 = const()[name = tensor("op_13430_end_0"), val = tensor([2, 64, 1, 3584])]; + tensor var_13430_end_mask_0 = const()[name = tensor("op_13430_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13430_cast = slice_by_index(begin = var_13430_begin_0, end = var_13430_end_0, end_mask = var_13430_end_mask_0, x = var_13393_cast)[name = tensor("op_13430_cast")]; + tensor var_13431_begin_0 = const()[name = tensor("op_13431_begin_0"), val = tensor([0, 0, 0, 3584])]; + tensor var_13431_end_0 = const()[name = tensor("op_13431_end_0"), val = tensor([2, 64, 1, 4096])]; + tensor var_13431_end_mask_0 = const()[name = tensor("op_13431_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13431_cast = slice_by_index(begin = var_13431_begin_0, end = var_13431_end_0, end_mask = var_13431_end_mask_0, x = var_13393_cast)[name = tensor("op_13431_cast")]; + tensor var_13432_begin_0 = const()[name = tensor("op_13432_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_13432_end_0 = const()[name = tensor("op_13432_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_13432_end_mask_0 = const()[name = tensor("op_13432_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13432_cast = slice_by_index(begin = var_13432_begin_0, end = var_13432_end_0, end_mask = var_13432_end_mask_0, x = var_13397_cast)[name = tensor("op_13432_cast")]; + tensor var_13433_begin_0 = const()[name = tensor("op_13433_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_13433_end_0 = const()[name = tensor("op_13433_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_13433_end_mask_0 = const()[name = tensor("op_13433_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13433_cast = slice_by_index(begin = var_13433_begin_0, end = var_13433_end_0, end_mask = var_13433_end_mask_0, x = var_13397_cast)[name = tensor("op_13433_cast")]; + tensor var_13434_begin_0 = const()[name = tensor("op_13434_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_13434_end_0 = const()[name = tensor("op_13434_end_0"), val = tensor([2, 64, 1, 1536])]; + tensor var_13434_end_mask_0 = const()[name = tensor("op_13434_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13434_cast = slice_by_index(begin = var_13434_begin_0, end = var_13434_end_0, end_mask = var_13434_end_mask_0, x = var_13397_cast)[name = tensor("op_13434_cast")]; + tensor var_13435_begin_0 = const()[name = tensor("op_13435_begin_0"), val = tensor([0, 0, 0, 1536])]; + tensor var_13435_end_0 = const()[name = tensor("op_13435_end_0"), val = tensor([2, 64, 1, 2048])]; + tensor var_13435_end_mask_0 = const()[name = tensor("op_13435_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13435_cast = slice_by_index(begin = var_13435_begin_0, end = var_13435_end_0, end_mask = var_13435_end_mask_0, x = var_13397_cast)[name = tensor("op_13435_cast")]; + tensor var_13436_begin_0 = const()[name = tensor("op_13436_begin_0"), val = tensor([0, 0, 0, 2048])]; + tensor var_13436_end_0 = const()[name = tensor("op_13436_end_0"), val = tensor([2, 64, 1, 2560])]; + tensor var_13436_end_mask_0 = const()[name = tensor("op_13436_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13436_cast = slice_by_index(begin = var_13436_begin_0, end = var_13436_end_0, end_mask = var_13436_end_mask_0, x = var_13397_cast)[name = tensor("op_13436_cast")]; + tensor var_13437_begin_0 = const()[name = tensor("op_13437_begin_0"), val = tensor([0, 0, 0, 2560])]; + tensor var_13437_end_0 = const()[name = tensor("op_13437_end_0"), val = tensor([2, 64, 1, 3072])]; + tensor var_13437_end_mask_0 = const()[name = tensor("op_13437_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13437_cast = slice_by_index(begin = var_13437_begin_0, end = var_13437_end_0, end_mask = var_13437_end_mask_0, x = var_13397_cast)[name = tensor("op_13437_cast")]; + tensor var_13438_begin_0 = const()[name = tensor("op_13438_begin_0"), val = tensor([0, 0, 0, 3072])]; + tensor var_13438_end_0 = const()[name = tensor("op_13438_end_0"), val = tensor([2, 64, 1, 3584])]; + tensor var_13438_end_mask_0 = const()[name = tensor("op_13438_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13438_cast = slice_by_index(begin = var_13438_begin_0, end = var_13438_end_0, end_mask = var_13438_end_mask_0, x = var_13397_cast)[name = tensor("op_13438_cast")]; + tensor var_13439_begin_0 = const()[name = tensor("op_13439_begin_0"), val = tensor([0, 0, 0, 3584])]; + tensor var_13439_end_0 = const()[name = tensor("op_13439_end_0"), val = tensor([2, 64, 1, 4096])]; + tensor var_13439_end_mask_0 = const()[name = tensor("op_13439_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13439_cast = slice_by_index(begin = var_13439_begin_0, end = var_13439_end_0, end_mask = var_13439_end_mask_0, x = var_13397_cast)[name = tensor("op_13439_cast")]; + tensor k_111_perm_0 = const()[name = tensor("k_111_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_13444_begin_0 = const()[name = tensor("op_13444_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_13444_end_0 = const()[name = tensor("op_13444_end_0"), val = tensor([2, 77, 1, 64])]; + tensor var_13444_end_mask_0 = const()[name = tensor("op_13444_end_mask_0"), val = tensor([true, true, true, false])]; + tensor transpose_4 = transpose(perm = k_111_perm_0, x = k_109_cast)[name = tensor("transpose_4")]; + tensor var_13444_cast = slice_by_index(begin = var_13444_begin_0, end = var_13444_end_0, end_mask = var_13444_end_mask_0, x = transpose_4)[name = tensor("op_13444_cast")]; + tensor var_13448_begin_0 = const()[name = tensor("op_13448_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_13448_end_0 = const()[name = tensor("op_13448_end_0"), val = tensor([2, 77, 1, 128])]; + tensor var_13448_end_mask_0 = const()[name = tensor("op_13448_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13448_cast = slice_by_index(begin = var_13448_begin_0, end = var_13448_end_0, end_mask = var_13448_end_mask_0, x = transpose_4)[name = tensor("op_13448_cast")]; + tensor var_13452_begin_0 = const()[name = tensor("op_13452_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_13452_end_0 = const()[name = tensor("op_13452_end_0"), val = tensor([2, 77, 1, 192])]; + tensor var_13452_end_mask_0 = const()[name = tensor("op_13452_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13452_cast = slice_by_index(begin = var_13452_begin_0, end = var_13452_end_0, end_mask = var_13452_end_mask_0, x = transpose_4)[name = tensor("op_13452_cast")]; + tensor var_13456_begin_0 = const()[name = tensor("op_13456_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_13456_end_0 = const()[name = tensor("op_13456_end_0"), val = tensor([2, 77, 1, 256])]; + tensor var_13456_end_mask_0 = const()[name = tensor("op_13456_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13456_cast = slice_by_index(begin = var_13456_begin_0, end = var_13456_end_0, end_mask = var_13456_end_mask_0, x = transpose_4)[name = tensor("op_13456_cast")]; + tensor var_13460_begin_0 = const()[name = tensor("op_13460_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_13460_end_0 = const()[name = tensor("op_13460_end_0"), val = tensor([2, 77, 1, 320])]; + tensor var_13460_end_mask_0 = const()[name = tensor("op_13460_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13460_cast = slice_by_index(begin = var_13460_begin_0, end = var_13460_end_0, end_mask = var_13460_end_mask_0, x = transpose_4)[name = tensor("op_13460_cast")]; + tensor var_13462_begin_0 = const()[name = tensor("op_13462_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_13462_end_0 = const()[name = tensor("op_13462_end_0"), val = tensor([2, 64, 1, 77])]; + tensor var_13462_end_mask_0 = const()[name = tensor("op_13462_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13462_cast = slice_by_index(begin = var_13462_begin_0, end = var_13462_end_0, end_mask = var_13462_end_mask_0, x = v_55_cast)[name = tensor("op_13462_cast")]; + tensor var_13466_begin_0 = const()[name = tensor("op_13466_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_13466_end_0 = const()[name = tensor("op_13466_end_0"), val = tensor([2, 128, 1, 77])]; + tensor var_13466_end_mask_0 = const()[name = tensor("op_13466_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13466_cast = slice_by_index(begin = var_13466_begin_0, end = var_13466_end_0, end_mask = var_13466_end_mask_0, x = v_55_cast)[name = tensor("op_13466_cast")]; + tensor var_13470_begin_0 = const()[name = tensor("op_13470_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_13470_end_0 = const()[name = tensor("op_13470_end_0"), val = tensor([2, 192, 1, 77])]; + tensor var_13470_end_mask_0 = const()[name = tensor("op_13470_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13470_cast = slice_by_index(begin = var_13470_begin_0, end = var_13470_end_0, end_mask = var_13470_end_mask_0, x = v_55_cast)[name = tensor("op_13470_cast")]; + tensor var_13474_begin_0 = const()[name = tensor("op_13474_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_13474_end_0 = const()[name = tensor("op_13474_end_0"), val = tensor([2, 256, 1, 77])]; + tensor var_13474_end_mask_0 = const()[name = tensor("op_13474_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13474_cast = slice_by_index(begin = var_13474_begin_0, end = var_13474_end_0, end_mask = var_13474_end_mask_0, x = v_55_cast)[name = tensor("op_13474_cast")]; + tensor var_13478_begin_0 = const()[name = tensor("op_13478_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_13478_end_0 = const()[name = tensor("op_13478_end_0"), val = tensor([2, 320, 1, 77])]; + tensor var_13478_end_mask_0 = const()[name = tensor("op_13478_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13478_cast = slice_by_index(begin = var_13478_begin_0, end = var_13478_end_0, end_mask = var_13478_end_mask_0, x = v_55_cast)[name = tensor("op_13478_cast")]; + tensor var_13482_equation_0 = const()[name = tensor("op_13482_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13482_cast = einsum(equation = var_13482_equation_0, values = (var_13444_cast, var_13400_cast))[name = tensor("op_13482_cast")]; + tensor var_13483_to_fp16 = const()[name = tensor("op_13483_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_801_cast = mul(x = var_13482_cast, y = var_13483_to_fp16)[name = tensor("aw_chunk_801_cast")]; + tensor var_13486_equation_0 = const()[name = tensor("op_13486_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13486_cast = einsum(equation = var_13486_equation_0, values = (var_13444_cast, var_13401_cast))[name = tensor("op_13486_cast")]; + tensor var_13487_to_fp16 = const()[name = tensor("op_13487_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_803_cast = mul(x = var_13486_cast, y = var_13487_to_fp16)[name = tensor("aw_chunk_803_cast")]; + tensor var_13490_equation_0 = const()[name = tensor("op_13490_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13490_cast = einsum(equation = var_13490_equation_0, values = (var_13444_cast, var_13402_cast))[name = tensor("op_13490_cast")]; + tensor var_13491_to_fp16 = const()[name = tensor("op_13491_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_805_cast = mul(x = var_13490_cast, y = var_13491_to_fp16)[name = tensor("aw_chunk_805_cast")]; + tensor var_13494_equation_0 = const()[name = tensor("op_13494_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13494_cast = einsum(equation = var_13494_equation_0, values = (var_13444_cast, var_13403_cast))[name = tensor("op_13494_cast")]; + tensor var_13495_to_fp16 = const()[name = tensor("op_13495_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_807_cast = mul(x = var_13494_cast, y = var_13495_to_fp16)[name = tensor("aw_chunk_807_cast")]; + tensor var_13498_equation_0 = const()[name = tensor("op_13498_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13498_cast = einsum(equation = var_13498_equation_0, values = (var_13444_cast, var_13404_cast))[name = tensor("op_13498_cast")]; + tensor var_13499_to_fp16 = const()[name = tensor("op_13499_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_809_cast = mul(x = var_13498_cast, y = var_13499_to_fp16)[name = tensor("aw_chunk_809_cast")]; + tensor var_13502_equation_0 = const()[name = tensor("op_13502_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13502_cast = einsum(equation = var_13502_equation_0, values = (var_13444_cast, var_13405_cast))[name = tensor("op_13502_cast")]; + tensor var_13503_to_fp16 = const()[name = tensor("op_13503_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_811_cast = mul(x = var_13502_cast, y = var_13503_to_fp16)[name = tensor("aw_chunk_811_cast")]; + tensor var_13506_equation_0 = const()[name = tensor("op_13506_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13506_cast = einsum(equation = var_13506_equation_0, values = (var_13444_cast, var_13406_cast))[name = tensor("op_13506_cast")]; + tensor var_13507_to_fp16 = const()[name = tensor("op_13507_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_813_cast = mul(x = var_13506_cast, y = var_13507_to_fp16)[name = tensor("aw_chunk_813_cast")]; + tensor var_13510_equation_0 = const()[name = tensor("op_13510_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13510_cast = einsum(equation = var_13510_equation_0, values = (var_13444_cast, var_13407_cast))[name = tensor("op_13510_cast")]; + tensor var_13511_to_fp16 = const()[name = tensor("op_13511_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_815_cast = mul(x = var_13510_cast, y = var_13511_to_fp16)[name = tensor("aw_chunk_815_cast")]; + tensor var_13514_equation_0 = const()[name = tensor("op_13514_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13514_cast = einsum(equation = var_13514_equation_0, values = (var_13448_cast, var_13408_cast))[name = tensor("op_13514_cast")]; + tensor var_13515_to_fp16 = const()[name = tensor("op_13515_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_817_cast = mul(x = var_13514_cast, y = var_13515_to_fp16)[name = tensor("aw_chunk_817_cast")]; + tensor var_13518_equation_0 = const()[name = tensor("op_13518_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13518_cast = einsum(equation = var_13518_equation_0, values = (var_13448_cast, var_13409_cast))[name = tensor("op_13518_cast")]; + tensor var_13519_to_fp16 = const()[name = tensor("op_13519_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_819_cast = mul(x = var_13518_cast, y = var_13519_to_fp16)[name = tensor("aw_chunk_819_cast")]; + tensor var_13522_equation_0 = const()[name = tensor("op_13522_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13522_cast = einsum(equation = var_13522_equation_0, values = (var_13448_cast, var_13410_cast))[name = tensor("op_13522_cast")]; + tensor var_13523_to_fp16 = const()[name = tensor("op_13523_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_821_cast = mul(x = var_13522_cast, y = var_13523_to_fp16)[name = tensor("aw_chunk_821_cast")]; + tensor var_13526_equation_0 = const()[name = tensor("op_13526_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13526_cast = einsum(equation = var_13526_equation_0, values = (var_13448_cast, var_13411_cast))[name = tensor("op_13526_cast")]; + tensor var_13527_to_fp16 = const()[name = tensor("op_13527_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_823_cast = mul(x = var_13526_cast, y = var_13527_to_fp16)[name = tensor("aw_chunk_823_cast")]; + tensor var_13530_equation_0 = const()[name = tensor("op_13530_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13530_cast = einsum(equation = var_13530_equation_0, values = (var_13448_cast, var_13412_cast))[name = tensor("op_13530_cast")]; + tensor var_13531_to_fp16 = const()[name = tensor("op_13531_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_825_cast = mul(x = var_13530_cast, y = var_13531_to_fp16)[name = tensor("aw_chunk_825_cast")]; + tensor var_13534_equation_0 = const()[name = tensor("op_13534_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13534_cast = einsum(equation = var_13534_equation_0, values = (var_13448_cast, var_13413_cast))[name = tensor("op_13534_cast")]; + tensor var_13535_to_fp16 = const()[name = tensor("op_13535_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_827_cast = mul(x = var_13534_cast, y = var_13535_to_fp16)[name = tensor("aw_chunk_827_cast")]; + tensor var_13538_equation_0 = const()[name = tensor("op_13538_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13538_cast = einsum(equation = var_13538_equation_0, values = (var_13448_cast, var_13414_cast))[name = tensor("op_13538_cast")]; + tensor var_13539_to_fp16 = const()[name = tensor("op_13539_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_829_cast = mul(x = var_13538_cast, y = var_13539_to_fp16)[name = tensor("aw_chunk_829_cast")]; + tensor var_13542_equation_0 = const()[name = tensor("op_13542_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13542_cast = einsum(equation = var_13542_equation_0, values = (var_13448_cast, var_13415_cast))[name = tensor("op_13542_cast")]; + tensor var_13543_to_fp16 = const()[name = tensor("op_13543_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_831_cast = mul(x = var_13542_cast, y = var_13543_to_fp16)[name = tensor("aw_chunk_831_cast")]; + tensor var_13546_equation_0 = const()[name = tensor("op_13546_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13546_cast = einsum(equation = var_13546_equation_0, values = (var_13452_cast, var_13416_cast))[name = tensor("op_13546_cast")]; + tensor var_13547_to_fp16 = const()[name = tensor("op_13547_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_833_cast = mul(x = var_13546_cast, y = var_13547_to_fp16)[name = tensor("aw_chunk_833_cast")]; + tensor var_13550_equation_0 = const()[name = tensor("op_13550_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13550_cast = einsum(equation = var_13550_equation_0, values = (var_13452_cast, var_13417_cast))[name = tensor("op_13550_cast")]; + tensor var_13551_to_fp16 = const()[name = tensor("op_13551_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_835_cast = mul(x = var_13550_cast, y = var_13551_to_fp16)[name = tensor("aw_chunk_835_cast")]; + tensor var_13554_equation_0 = const()[name = tensor("op_13554_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13554_cast = einsum(equation = var_13554_equation_0, values = (var_13452_cast, var_13418_cast))[name = tensor("op_13554_cast")]; + tensor var_13555_to_fp16 = const()[name = tensor("op_13555_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_837_cast = mul(x = var_13554_cast, y = var_13555_to_fp16)[name = tensor("aw_chunk_837_cast")]; + tensor var_13558_equation_0 = const()[name = tensor("op_13558_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13558_cast = einsum(equation = var_13558_equation_0, values = (var_13452_cast, var_13419_cast))[name = tensor("op_13558_cast")]; + tensor var_13559_to_fp16 = const()[name = tensor("op_13559_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_839_cast = mul(x = var_13558_cast, y = var_13559_to_fp16)[name = tensor("aw_chunk_839_cast")]; + tensor var_13562_equation_0 = const()[name = tensor("op_13562_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13562_cast = einsum(equation = var_13562_equation_0, values = (var_13452_cast, var_13420_cast))[name = tensor("op_13562_cast")]; + tensor var_13563_to_fp16 = const()[name = tensor("op_13563_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_841_cast = mul(x = var_13562_cast, y = var_13563_to_fp16)[name = tensor("aw_chunk_841_cast")]; + tensor var_13566_equation_0 = const()[name = tensor("op_13566_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13566_cast = einsum(equation = var_13566_equation_0, values = (var_13452_cast, var_13421_cast))[name = tensor("op_13566_cast")]; + tensor var_13567_to_fp16 = const()[name = tensor("op_13567_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_843_cast = mul(x = var_13566_cast, y = var_13567_to_fp16)[name = tensor("aw_chunk_843_cast")]; + tensor var_13570_equation_0 = const()[name = tensor("op_13570_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13570_cast = einsum(equation = var_13570_equation_0, values = (var_13452_cast, var_13422_cast))[name = tensor("op_13570_cast")]; + tensor var_13571_to_fp16 = const()[name = tensor("op_13571_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_845_cast = mul(x = var_13570_cast, y = var_13571_to_fp16)[name = tensor("aw_chunk_845_cast")]; + tensor var_13574_equation_0 = const()[name = tensor("op_13574_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13574_cast = einsum(equation = var_13574_equation_0, values = (var_13452_cast, var_13423_cast))[name = tensor("op_13574_cast")]; + tensor var_13575_to_fp16 = const()[name = tensor("op_13575_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_847_cast = mul(x = var_13574_cast, y = var_13575_to_fp16)[name = tensor("aw_chunk_847_cast")]; + tensor var_13578_equation_0 = const()[name = tensor("op_13578_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13578_cast = einsum(equation = var_13578_equation_0, values = (var_13456_cast, var_13424_cast))[name = tensor("op_13578_cast")]; + tensor var_13579_to_fp16 = const()[name = tensor("op_13579_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_849_cast = mul(x = var_13578_cast, y = var_13579_to_fp16)[name = tensor("aw_chunk_849_cast")]; + tensor var_13582_equation_0 = const()[name = tensor("op_13582_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13582_cast = einsum(equation = var_13582_equation_0, values = (var_13456_cast, var_13425_cast))[name = tensor("op_13582_cast")]; + tensor var_13583_to_fp16 = const()[name = tensor("op_13583_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_851_cast = mul(x = var_13582_cast, y = var_13583_to_fp16)[name = tensor("aw_chunk_851_cast")]; + tensor var_13586_equation_0 = const()[name = tensor("op_13586_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13586_cast = einsum(equation = var_13586_equation_0, values = (var_13456_cast, var_13426_cast))[name = tensor("op_13586_cast")]; + tensor var_13587_to_fp16 = const()[name = tensor("op_13587_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_853_cast = mul(x = var_13586_cast, y = var_13587_to_fp16)[name = tensor("aw_chunk_853_cast")]; + tensor var_13590_equation_0 = const()[name = tensor("op_13590_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13590_cast = einsum(equation = var_13590_equation_0, values = (var_13456_cast, var_13427_cast))[name = tensor("op_13590_cast")]; + tensor var_13591_to_fp16 = const()[name = tensor("op_13591_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_855_cast = mul(x = var_13590_cast, y = var_13591_to_fp16)[name = tensor("aw_chunk_855_cast")]; + tensor var_13594_equation_0 = const()[name = tensor("op_13594_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13594_cast = einsum(equation = var_13594_equation_0, values = (var_13456_cast, var_13428_cast))[name = tensor("op_13594_cast")]; + tensor var_13595_to_fp16 = const()[name = tensor("op_13595_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_857_cast = mul(x = var_13594_cast, y = var_13595_to_fp16)[name = tensor("aw_chunk_857_cast")]; + tensor var_13598_equation_0 = const()[name = tensor("op_13598_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13598_cast = einsum(equation = var_13598_equation_0, values = (var_13456_cast, var_13429_cast))[name = tensor("op_13598_cast")]; + tensor var_13599_to_fp16 = const()[name = tensor("op_13599_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_859_cast = mul(x = var_13598_cast, y = var_13599_to_fp16)[name = tensor("aw_chunk_859_cast")]; + tensor var_13602_equation_0 = const()[name = tensor("op_13602_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13602_cast = einsum(equation = var_13602_equation_0, values = (var_13456_cast, var_13430_cast))[name = tensor("op_13602_cast")]; + tensor var_13603_to_fp16 = const()[name = tensor("op_13603_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_861_cast = mul(x = var_13602_cast, y = var_13603_to_fp16)[name = tensor("aw_chunk_861_cast")]; + tensor var_13606_equation_0 = const()[name = tensor("op_13606_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13606_cast = einsum(equation = var_13606_equation_0, values = (var_13456_cast, var_13431_cast))[name = tensor("op_13606_cast")]; + tensor var_13607_to_fp16 = const()[name = tensor("op_13607_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_863_cast = mul(x = var_13606_cast, y = var_13607_to_fp16)[name = tensor("aw_chunk_863_cast")]; + tensor var_13610_equation_0 = const()[name = tensor("op_13610_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13610_cast = einsum(equation = var_13610_equation_0, values = (var_13460_cast, var_13432_cast))[name = tensor("op_13610_cast")]; + tensor var_13611_to_fp16 = const()[name = tensor("op_13611_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_865_cast = mul(x = var_13610_cast, y = var_13611_to_fp16)[name = tensor("aw_chunk_865_cast")]; + tensor var_13614_equation_0 = const()[name = tensor("op_13614_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13614_cast = einsum(equation = var_13614_equation_0, values = (var_13460_cast, var_13433_cast))[name = tensor("op_13614_cast")]; + tensor var_13615_to_fp16 = const()[name = tensor("op_13615_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_867_cast = mul(x = var_13614_cast, y = var_13615_to_fp16)[name = tensor("aw_chunk_867_cast")]; + tensor var_13618_equation_0 = const()[name = tensor("op_13618_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13618_cast = einsum(equation = var_13618_equation_0, values = (var_13460_cast, var_13434_cast))[name = tensor("op_13618_cast")]; + tensor var_13619_to_fp16 = const()[name = tensor("op_13619_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_869_cast = mul(x = var_13618_cast, y = var_13619_to_fp16)[name = tensor("aw_chunk_869_cast")]; + tensor var_13622_equation_0 = const()[name = tensor("op_13622_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13622_cast = einsum(equation = var_13622_equation_0, values = (var_13460_cast, var_13435_cast))[name = tensor("op_13622_cast")]; + tensor var_13623_to_fp16 = const()[name = tensor("op_13623_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_871_cast = mul(x = var_13622_cast, y = var_13623_to_fp16)[name = tensor("aw_chunk_871_cast")]; + tensor var_13626_equation_0 = const()[name = tensor("op_13626_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13626_cast = einsum(equation = var_13626_equation_0, values = (var_13460_cast, var_13436_cast))[name = tensor("op_13626_cast")]; + tensor var_13627_to_fp16 = const()[name = tensor("op_13627_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_873_cast = mul(x = var_13626_cast, y = var_13627_to_fp16)[name = tensor("aw_chunk_873_cast")]; + tensor var_13630_equation_0 = const()[name = tensor("op_13630_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13630_cast = einsum(equation = var_13630_equation_0, values = (var_13460_cast, var_13437_cast))[name = tensor("op_13630_cast")]; + tensor var_13631_to_fp16 = const()[name = tensor("op_13631_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_875_cast = mul(x = var_13630_cast, y = var_13631_to_fp16)[name = tensor("aw_chunk_875_cast")]; + tensor var_13634_equation_0 = const()[name = tensor("op_13634_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13634_cast = einsum(equation = var_13634_equation_0, values = (var_13460_cast, var_13438_cast))[name = tensor("op_13634_cast")]; + tensor var_13635_to_fp16 = const()[name = tensor("op_13635_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_877_cast = mul(x = var_13634_cast, y = var_13635_to_fp16)[name = tensor("aw_chunk_877_cast")]; + tensor var_13638_equation_0 = const()[name = tensor("op_13638_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_13638_cast = einsum(equation = var_13638_equation_0, values = (var_13460_cast, var_13439_cast))[name = tensor("op_13638_cast")]; + tensor var_13639_to_fp16 = const()[name = tensor("op_13639_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_879_cast = mul(x = var_13638_cast, y = var_13639_to_fp16)[name = tensor("aw_chunk_879_cast")]; + tensor var_13641_cast = softmax(axis = var_12807, x = aw_chunk_801_cast)[name = tensor("op_13641_cast")]; + tensor var_13642_cast = softmax(axis = var_12807, x = aw_chunk_803_cast)[name = tensor("op_13642_cast")]; + tensor var_13643_cast = softmax(axis = var_12807, x = aw_chunk_805_cast)[name = tensor("op_13643_cast")]; + tensor var_13644_cast = softmax(axis = var_12807, x = aw_chunk_807_cast)[name = tensor("op_13644_cast")]; + tensor var_13645_cast = softmax(axis = var_12807, x = aw_chunk_809_cast)[name = tensor("op_13645_cast")]; + tensor var_13646_cast = softmax(axis = var_12807, x = aw_chunk_811_cast)[name = tensor("op_13646_cast")]; + tensor var_13647_cast = softmax(axis = var_12807, x = aw_chunk_813_cast)[name = tensor("op_13647_cast")]; + tensor var_13648_cast = softmax(axis = var_12807, x = aw_chunk_815_cast)[name = tensor("op_13648_cast")]; + tensor var_13649_cast = softmax(axis = var_12807, x = aw_chunk_817_cast)[name = tensor("op_13649_cast")]; + tensor var_13650_cast = softmax(axis = var_12807, x = aw_chunk_819_cast)[name = tensor("op_13650_cast")]; + tensor var_13651_cast = softmax(axis = var_12807, x = aw_chunk_821_cast)[name = tensor("op_13651_cast")]; + tensor var_13652_cast = softmax(axis = var_12807, x = aw_chunk_823_cast)[name = tensor("op_13652_cast")]; + tensor var_13653_cast = softmax(axis = var_12807, x = aw_chunk_825_cast)[name = tensor("op_13653_cast")]; + tensor var_13654_cast = softmax(axis = var_12807, x = aw_chunk_827_cast)[name = tensor("op_13654_cast")]; + tensor var_13655_cast = softmax(axis = var_12807, x = aw_chunk_829_cast)[name = tensor("op_13655_cast")]; + tensor var_13656_cast = softmax(axis = var_12807, x = aw_chunk_831_cast)[name = tensor("op_13656_cast")]; + tensor var_13657_cast = softmax(axis = var_12807, x = aw_chunk_833_cast)[name = tensor("op_13657_cast")]; + tensor var_13658_cast = softmax(axis = var_12807, x = aw_chunk_835_cast)[name = tensor("op_13658_cast")]; + tensor var_13659_cast = softmax(axis = var_12807, x = aw_chunk_837_cast)[name = tensor("op_13659_cast")]; + tensor var_13660_cast = softmax(axis = var_12807, x = aw_chunk_839_cast)[name = tensor("op_13660_cast")]; + tensor var_13661_cast = softmax(axis = var_12807, x = aw_chunk_841_cast)[name = tensor("op_13661_cast")]; + tensor var_13662_cast = softmax(axis = var_12807, x = aw_chunk_843_cast)[name = tensor("op_13662_cast")]; + tensor var_13663_cast = softmax(axis = var_12807, x = aw_chunk_845_cast)[name = tensor("op_13663_cast")]; + tensor var_13664_cast = softmax(axis = var_12807, x = aw_chunk_847_cast)[name = tensor("op_13664_cast")]; + tensor var_13665_cast = softmax(axis = var_12807, x = aw_chunk_849_cast)[name = tensor("op_13665_cast")]; + tensor var_13666_cast = softmax(axis = var_12807, x = aw_chunk_851_cast)[name = tensor("op_13666_cast")]; + tensor var_13667_cast = softmax(axis = var_12807, x = aw_chunk_853_cast)[name = tensor("op_13667_cast")]; + tensor var_13668_cast = softmax(axis = var_12807, x = aw_chunk_855_cast)[name = tensor("op_13668_cast")]; + tensor var_13669_cast = softmax(axis = var_12807, x = aw_chunk_857_cast)[name = tensor("op_13669_cast")]; + tensor var_13670_cast = softmax(axis = var_12807, x = aw_chunk_859_cast)[name = tensor("op_13670_cast")]; + tensor var_13671_cast = softmax(axis = var_12807, x = aw_chunk_861_cast)[name = tensor("op_13671_cast")]; + tensor var_13672_cast = softmax(axis = var_12807, x = aw_chunk_863_cast)[name = tensor("op_13672_cast")]; + tensor var_13673_cast = softmax(axis = var_12807, x = aw_chunk_865_cast)[name = tensor("op_13673_cast")]; + tensor var_13674_cast = softmax(axis = var_12807, x = aw_chunk_867_cast)[name = tensor("op_13674_cast")]; + tensor var_13675_cast = softmax(axis = var_12807, x = aw_chunk_869_cast)[name = tensor("op_13675_cast")]; + tensor var_13676_cast = softmax(axis = var_12807, x = aw_chunk_871_cast)[name = tensor("op_13676_cast")]; + tensor var_13677_cast = softmax(axis = var_12807, x = aw_chunk_873_cast)[name = tensor("op_13677_cast")]; + tensor var_13678_cast = softmax(axis = var_12807, x = aw_chunk_875_cast)[name = tensor("op_13678_cast")]; + tensor var_13679_cast = softmax(axis = var_12807, x = aw_chunk_877_cast)[name = tensor("op_13679_cast")]; + tensor var_13680_cast = softmax(axis = var_12807, x = aw_chunk_879_cast)[name = tensor("op_13680_cast")]; + tensor var_13682_equation_0 = const()[name = tensor("op_13682_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13682_cast = einsum(equation = var_13682_equation_0, values = (var_13462_cast, var_13641_cast))[name = tensor("op_13682_cast")]; + tensor var_13684_equation_0 = const()[name = tensor("op_13684_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13684_cast = einsum(equation = var_13684_equation_0, values = (var_13462_cast, var_13642_cast))[name = tensor("op_13684_cast")]; + tensor var_13686_equation_0 = const()[name = tensor("op_13686_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13686_cast = einsum(equation = var_13686_equation_0, values = (var_13462_cast, var_13643_cast))[name = tensor("op_13686_cast")]; + tensor var_13688_equation_0 = const()[name = tensor("op_13688_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13688_cast = einsum(equation = var_13688_equation_0, values = (var_13462_cast, var_13644_cast))[name = tensor("op_13688_cast")]; + tensor var_13690_equation_0 = const()[name = tensor("op_13690_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13690_cast = einsum(equation = var_13690_equation_0, values = (var_13462_cast, var_13645_cast))[name = tensor("op_13690_cast")]; + tensor var_13692_equation_0 = const()[name = tensor("op_13692_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13692_cast = einsum(equation = var_13692_equation_0, values = (var_13462_cast, var_13646_cast))[name = tensor("op_13692_cast")]; + tensor var_13694_equation_0 = const()[name = tensor("op_13694_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13694_cast = einsum(equation = var_13694_equation_0, values = (var_13462_cast, var_13647_cast))[name = tensor("op_13694_cast")]; + tensor var_13696_equation_0 = const()[name = tensor("op_13696_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13696_cast = einsum(equation = var_13696_equation_0, values = (var_13462_cast, var_13648_cast))[name = tensor("op_13696_cast")]; + tensor var_13698_equation_0 = const()[name = tensor("op_13698_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13698_cast = einsum(equation = var_13698_equation_0, values = (var_13466_cast, var_13649_cast))[name = tensor("op_13698_cast")]; + tensor var_13700_equation_0 = const()[name = tensor("op_13700_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13700_cast = einsum(equation = var_13700_equation_0, values = (var_13466_cast, var_13650_cast))[name = tensor("op_13700_cast")]; + tensor var_13702_equation_0 = const()[name = tensor("op_13702_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13702_cast = einsum(equation = var_13702_equation_0, values = (var_13466_cast, var_13651_cast))[name = tensor("op_13702_cast")]; + tensor var_13704_equation_0 = const()[name = tensor("op_13704_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13704_cast = einsum(equation = var_13704_equation_0, values = (var_13466_cast, var_13652_cast))[name = tensor("op_13704_cast")]; + tensor var_13706_equation_0 = const()[name = tensor("op_13706_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13706_cast = einsum(equation = var_13706_equation_0, values = (var_13466_cast, var_13653_cast))[name = tensor("op_13706_cast")]; + tensor var_13708_equation_0 = const()[name = tensor("op_13708_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13708_cast = einsum(equation = var_13708_equation_0, values = (var_13466_cast, var_13654_cast))[name = tensor("op_13708_cast")]; + tensor var_13710_equation_0 = const()[name = tensor("op_13710_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13710_cast = einsum(equation = var_13710_equation_0, values = (var_13466_cast, var_13655_cast))[name = tensor("op_13710_cast")]; + tensor var_13712_equation_0 = const()[name = tensor("op_13712_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13712_cast = einsum(equation = var_13712_equation_0, values = (var_13466_cast, var_13656_cast))[name = tensor("op_13712_cast")]; + tensor var_13714_equation_0 = const()[name = tensor("op_13714_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13714_cast = einsum(equation = var_13714_equation_0, values = (var_13470_cast, var_13657_cast))[name = tensor("op_13714_cast")]; + tensor var_13716_equation_0 = const()[name = tensor("op_13716_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13716_cast = einsum(equation = var_13716_equation_0, values = (var_13470_cast, var_13658_cast))[name = tensor("op_13716_cast")]; + tensor var_13718_equation_0 = const()[name = tensor("op_13718_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13718_cast = einsum(equation = var_13718_equation_0, values = (var_13470_cast, var_13659_cast))[name = tensor("op_13718_cast")]; + tensor var_13720_equation_0 = const()[name = tensor("op_13720_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13720_cast = einsum(equation = var_13720_equation_0, values = (var_13470_cast, var_13660_cast))[name = tensor("op_13720_cast")]; + tensor var_13722_equation_0 = const()[name = tensor("op_13722_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13722_cast = einsum(equation = var_13722_equation_0, values = (var_13470_cast, var_13661_cast))[name = tensor("op_13722_cast")]; + tensor var_13724_equation_0 = const()[name = tensor("op_13724_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13724_cast = einsum(equation = var_13724_equation_0, values = (var_13470_cast, var_13662_cast))[name = tensor("op_13724_cast")]; + tensor var_13726_equation_0 = const()[name = tensor("op_13726_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13726_cast = einsum(equation = var_13726_equation_0, values = (var_13470_cast, var_13663_cast))[name = tensor("op_13726_cast")]; + tensor var_13728_equation_0 = const()[name = tensor("op_13728_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13728_cast = einsum(equation = var_13728_equation_0, values = (var_13470_cast, var_13664_cast))[name = tensor("op_13728_cast")]; + tensor var_13730_equation_0 = const()[name = tensor("op_13730_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13730_cast = einsum(equation = var_13730_equation_0, values = (var_13474_cast, var_13665_cast))[name = tensor("op_13730_cast")]; + tensor var_13732_equation_0 = const()[name = tensor("op_13732_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13732_cast = einsum(equation = var_13732_equation_0, values = (var_13474_cast, var_13666_cast))[name = tensor("op_13732_cast")]; + tensor var_13734_equation_0 = const()[name = tensor("op_13734_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13734_cast = einsum(equation = var_13734_equation_0, values = (var_13474_cast, var_13667_cast))[name = tensor("op_13734_cast")]; + tensor var_13736_equation_0 = const()[name = tensor("op_13736_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13736_cast = einsum(equation = var_13736_equation_0, values = (var_13474_cast, var_13668_cast))[name = tensor("op_13736_cast")]; + tensor var_13738_equation_0 = const()[name = tensor("op_13738_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13738_cast = einsum(equation = var_13738_equation_0, values = (var_13474_cast, var_13669_cast))[name = tensor("op_13738_cast")]; + tensor var_13740_equation_0 = const()[name = tensor("op_13740_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13740_cast = einsum(equation = var_13740_equation_0, values = (var_13474_cast, var_13670_cast))[name = tensor("op_13740_cast")]; + tensor var_13742_equation_0 = const()[name = tensor("op_13742_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13742_cast = einsum(equation = var_13742_equation_0, values = (var_13474_cast, var_13671_cast))[name = tensor("op_13742_cast")]; + tensor var_13744_equation_0 = const()[name = tensor("op_13744_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13744_cast = einsum(equation = var_13744_equation_0, values = (var_13474_cast, var_13672_cast))[name = tensor("op_13744_cast")]; + tensor var_13746_equation_0 = const()[name = tensor("op_13746_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13746_cast = einsum(equation = var_13746_equation_0, values = (var_13478_cast, var_13673_cast))[name = tensor("op_13746_cast")]; + tensor var_13748_equation_0 = const()[name = tensor("op_13748_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13748_cast = einsum(equation = var_13748_equation_0, values = (var_13478_cast, var_13674_cast))[name = tensor("op_13748_cast")]; + tensor var_13750_equation_0 = const()[name = tensor("op_13750_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13750_cast = einsum(equation = var_13750_equation_0, values = (var_13478_cast, var_13675_cast))[name = tensor("op_13750_cast")]; + tensor var_13752_equation_0 = const()[name = tensor("op_13752_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13752_cast = einsum(equation = var_13752_equation_0, values = (var_13478_cast, var_13676_cast))[name = tensor("op_13752_cast")]; + tensor var_13754_equation_0 = const()[name = tensor("op_13754_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13754_cast = einsum(equation = var_13754_equation_0, values = (var_13478_cast, var_13677_cast))[name = tensor("op_13754_cast")]; + tensor var_13756_equation_0 = const()[name = tensor("op_13756_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13756_cast = einsum(equation = var_13756_equation_0, values = (var_13478_cast, var_13678_cast))[name = tensor("op_13756_cast")]; + tensor var_13758_equation_0 = const()[name = tensor("op_13758_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13758_cast = einsum(equation = var_13758_equation_0, values = (var_13478_cast, var_13679_cast))[name = tensor("op_13758_cast")]; + tensor var_13760_equation_0 = const()[name = tensor("op_13760_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13760_cast = einsum(equation = var_13760_equation_0, values = (var_13478_cast, var_13680_cast))[name = tensor("op_13760_cast")]; + tensor var_13762_interleave_0 = const()[name = tensor("op_13762_interleave_0"), val = tensor(false)]; + tensor var_13762_cast = concat(axis = var_12782, interleave = var_13762_interleave_0, values = (var_13682_cast, var_13684_cast, var_13686_cast, var_13688_cast, var_13690_cast, var_13692_cast, var_13694_cast, var_13696_cast))[name = tensor("op_13762_cast")]; + tensor var_13764_interleave_0 = const()[name = tensor("op_13764_interleave_0"), val = tensor(false)]; + tensor var_13764_cast = concat(axis = var_12782, interleave = var_13764_interleave_0, values = (var_13698_cast, var_13700_cast, var_13702_cast, var_13704_cast, var_13706_cast, var_13708_cast, var_13710_cast, var_13712_cast))[name = tensor("op_13764_cast")]; + tensor var_13766_interleave_0 = const()[name = tensor("op_13766_interleave_0"), val = tensor(false)]; + tensor var_13766_cast = concat(axis = var_12782, interleave = var_13766_interleave_0, values = (var_13714_cast, var_13716_cast, var_13718_cast, var_13720_cast, var_13722_cast, var_13724_cast, var_13726_cast, var_13728_cast))[name = tensor("op_13766_cast")]; + tensor var_13768_interleave_0 = const()[name = tensor("op_13768_interleave_0"), val = tensor(false)]; + tensor var_13768_cast = concat(axis = var_12782, interleave = var_13768_interleave_0, values = (var_13730_cast, var_13732_cast, var_13734_cast, var_13736_cast, var_13738_cast, var_13740_cast, var_13742_cast, var_13744_cast))[name = tensor("op_13768_cast")]; + tensor var_13770_interleave_0 = const()[name = tensor("op_13770_interleave_0"), val = tensor(false)]; + tensor var_13770_cast = concat(axis = var_12782, interleave = var_13770_interleave_0, values = (var_13746_cast, var_13748_cast, var_13750_cast, var_13752_cast, var_13754_cast, var_13756_cast, var_13758_cast, var_13760_cast))[name = tensor("op_13770_cast")]; + tensor input_467_interleave_0 = const()[name = tensor("input_467_interleave_0"), val = tensor(false)]; + tensor input_467_cast = concat(axis = var_12807, interleave = input_467_interleave_0, values = (var_13762_cast, var_13764_cast, var_13766_cast, var_13768_cast, var_13770_cast))[name = tensor("input_467_cast")]; + tensor var_13776 = const()[name = tensor("op_13776"), val = tensor([1, 1])]; + tensor var_13778 = const()[name = tensor("op_13778"), val = tensor([1, 1])]; + tensor var_13780_pad_type_0 = const()[name = tensor("op_13780_pad_type_0"), val = tensor("custom")]; + tensor var_13780_pad_0 = const()[name = tensor("op_13780_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_3_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(639699200))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(639776064))), name = tensor("up_blocks_3_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16_palettized"), shape = tensor([320, 320, 1, 1])]; + tensor up_blocks_3_attentions_0_transformer_blocks_0_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("up_blocks_3_attentions_0_transformer_blocks_0_attn2_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(639776256)))]; + tensor var_13780_cast = conv(bias = up_blocks_3_attentions_0_transformer_blocks_0_attn2_to_out_0_bias_to_fp16, dilations = var_13778, groups = var_12807, pad = var_13780_pad_0, pad_type = var_13780_pad_type_0, strides = var_13776, weight = up_blocks_3_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16_palettized, x = input_467_cast)[name = tensor("op_13780_cast")]; + tensor inputs_83_cast = add(x = var_13780_cast, y = inputs_81_cast)[name = tensor("inputs_83_cast")]; + tensor var_13784 = const()[name = tensor("op_13784"), val = tensor([1])]; + tensor channels_mean_83_cast = reduce_mean(axes = var_13784, keep_dims = var_12802, x = inputs_83_cast)[name = tensor("channels_mean_83_cast")]; + tensor zero_mean_83_cast = sub(x = inputs_83_cast, y = channels_mean_83_cast)[name = tensor("zero_mean_83_cast")]; + tensor zero_mean_sq_83_cast = mul(x = zero_mean_83_cast, y = zero_mean_83_cast)[name = tensor("zero_mean_sq_83_cast")]; + tensor var_13788 = const()[name = tensor("op_13788"), val = tensor([1])]; + tensor var_13789_cast = reduce_mean(axes = var_13788, keep_dims = var_12802, x = zero_mean_sq_83_cast)[name = tensor("op_13789_cast")]; + tensor var_13790_to_fp16 = const()[name = tensor("op_13790_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_13791_cast = add(x = var_13789_cast, y = var_13790_to_fp16)[name = tensor("op_13791_cast")]; + tensor denom_83_epsilon_0_to_fp16 = const()[name = tensor("denom_83_epsilon_0_to_fp16"), val = tensor(0x1p-24)]; + tensor denom_83_cast = rsqrt(epsilon = denom_83_epsilon_0_to_fp16, x = var_13791_cast)[name = tensor("denom_83_cast")]; + tensor out_83_cast = mul(x = zero_mean_83_cast, y = denom_83_cast)[name = tensor("out_83_cast")]; + tensor var_13795_to_fp16 = const()[name = tensor("op_13795_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(639776960)))]; + tensor var_13796_cast = add(x = out_83_cast, y = var_13795_to_fp16)[name = tensor("op_13796_cast")]; + tensor var_13798_to_fp16 = const()[name = tensor("op_13798_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(639777664)))]; + tensor input_469_cast = mul(x = var_13796_cast, y = var_13798_to_fp16)[name = tensor("input_469_cast")]; + tensor var_13806 = const()[name = tensor("op_13806"), val = tensor([1, 1])]; + tensor var_13808 = const()[name = tensor("op_13808"), val = tensor([1, 1])]; + tensor var_13810_pad_type_0 = const()[name = tensor("op_13810_pad_type_0"), val = tensor("custom")]; + tensor var_13810_pad_0 = const()[name = tensor("op_13810_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_3_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(639778368))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(640392832))), name = tensor("up_blocks_3_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16_palettized"), shape = tensor([2560, 320, 1, 1])]; + tensor up_blocks_3_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(640393024))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(640395008))), name = tensor("up_blocks_3_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16_palettized"), shape = tensor([2560])]; + tensor var_13810_cast = conv(bias = up_blocks_3_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16_palettized, dilations = var_13808, groups = var_12807, pad = var_13810_pad_0, pad_type = var_13810_pad_type_0, strides = var_13806, weight = up_blocks_3_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16_palettized, x = input_469_cast)[name = tensor("op_13810_cast")]; + tensor var_13811_split_sizes_0 = const()[name = tensor("op_13811_split_sizes_0"), val = tensor([1280, 1280])]; + tensor var_13811_axis_0 = const()[name = tensor("op_13811_axis_0"), val = tensor(1)]; + tensor var_13811_cast_0, tensor var_13811_cast_1 = split(axis = var_13811_axis_0, split_sizes = var_13811_split_sizes_0, x = var_13810_cast)[name = tensor("op_13811_cast")]; + tensor var_13813_mode_0 = const()[name = tensor("op_13813_mode_0"), val = tensor("EXACT")]; + tensor var_13813_cast = gelu(mode = var_13813_mode_0, x = var_13811_cast_1)[name = tensor("op_13813_cast")]; + tensor input_471_cast = mul(x = var_13811_cast_0, y = var_13813_cast)[name = tensor("input_471_cast")]; + tensor var_13817 = const()[name = tensor("op_13817"), val = tensor([1, 1])]; + tensor var_13819 = const()[name = tensor("op_13819"), val = tensor([1, 1])]; + tensor var_13821_pad_type_0 = const()[name = tensor("op_13821_pad_type_0"), val = tensor("custom")]; + tensor var_13821_pad_0 = const()[name = tensor("op_13821_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_3_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(640395200))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(640702464))), name = tensor("up_blocks_3_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16_palettized"), shape = tensor([320, 1280, 1, 1])]; + tensor up_blocks_3_attentions_0_transformer_blocks_0_ff_net_2_bias_to_fp16 = const()[name = tensor("up_blocks_3_attentions_0_transformer_blocks_0_ff_net_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(640702656)))]; + tensor var_13821_cast = conv(bias = up_blocks_3_attentions_0_transformer_blocks_0_ff_net_2_bias_to_fp16, dilations = var_13819, groups = var_12807, pad = var_13821_pad_0, pad_type = var_13821_pad_type_0, strides = var_13817, weight = up_blocks_3_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16_palettized, x = input_471_cast)[name = tensor("op_13821_cast")]; + tensor hidden_states_293_cast = add(x = var_13821_cast, y = inputs_83_cast)[name = tensor("hidden_states_293_cast")]; + tensor var_13823 = const()[name = tensor("op_13823"), val = tensor([2, 320, 64, 64])]; + tensor input_473_cast = reshape(shape = var_13823, x = hidden_states_293_cast)[name = tensor("input_473_cast")]; + tensor var_13827 = const()[name = tensor("op_13827"), val = tensor([1, 1])]; + tensor var_13829 = const()[name = tensor("op_13829"), val = tensor([1, 1])]; + tensor hidden_states_295_pad_type_0 = const()[name = tensor("hidden_states_295_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_295_pad_0 = const()[name = tensor("hidden_states_295_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_3_attentions_0_proj_out_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(640703360))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(640780224))), name = tensor("up_blocks_3_attentions_0_proj_out_weight_to_fp16_palettized"), shape = tensor([320, 320, 1, 1])]; + tensor up_blocks_3_attentions_0_proj_out_bias_to_fp16 = const()[name = tensor("up_blocks_3_attentions_0_proj_out_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(640780416)))]; + tensor hidden_states_295_cast = conv(bias = up_blocks_3_attentions_0_proj_out_bias_to_fp16, dilations = var_13829, groups = var_12807, pad = hidden_states_295_pad_0, pad_type = hidden_states_295_pad_type_0, strides = var_13827, weight = up_blocks_3_attentions_0_proj_out_weight_to_fp16_palettized, x = input_473_cast)[name = tensor("hidden_states_295_cast")]; + tensor hidden_states_297_cast = add(x = hidden_states_295_cast, y = hidden_states_283_cast)[name = tensor("hidden_states_297_cast")]; + tensor input_475_interleave_0 = const()[name = tensor("input_475_interleave_0"), val = tensor(false)]; + tensor input_475_cast = concat(axis = var_12807, interleave = input_475_interleave_0, values = (hidden_states_297_cast, input_35_cast))[name = tensor("input_475_cast")]; + tensor reshape_216_shape_0 = const()[name = tensor("reshape_216_shape_0"), val = tensor([2, 32, 20, 64, 64])]; + tensor reshape_216_cast = reshape(shape = reshape_216_shape_0, x = input_475_cast)[name = tensor("reshape_216_cast")]; + tensor reduce_mean_162_axes_0 = const()[name = tensor("reduce_mean_162_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_162_keep_dims_0 = const()[name = tensor("reduce_mean_162_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_162_cast = reduce_mean(axes = reduce_mean_162_axes_0, keep_dims = reduce_mean_162_keep_dims_0, x = reshape_216_cast)[name = tensor("reduce_mean_162_cast")]; + tensor sub_108_cast = sub(x = reshape_216_cast, y = reduce_mean_162_cast)[name = tensor("sub_108_cast")]; + tensor square_54_cast = square(x = sub_108_cast)[name = tensor("square_54_cast")]; + tensor reduce_mean_164_axes_0 = const()[name = tensor("reduce_mean_164_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_164_keep_dims_0 = const()[name = tensor("reduce_mean_164_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_164_cast = reduce_mean(axes = reduce_mean_164_axes_0, keep_dims = reduce_mean_164_keep_dims_0, x = square_54_cast)[name = tensor("reduce_mean_164_cast")]; + tensor add_108_y_0_to_fp16 = const()[name = tensor("add_108_y_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_108_cast = add(x = reduce_mean_164_cast, y = add_108_y_0_to_fp16)[name = tensor("add_108_cast")]; + tensor sqrt_54_cast = sqrt(x = add_108_cast)[name = tensor("sqrt_54_cast")]; + tensor real_div_54_cast = real_div(x = sub_108_cast, y = sqrt_54_cast)[name = tensor("real_div_54_cast")]; + tensor reshape_217_shape_0 = const()[name = tensor("reshape_217_shape_0"), val = tensor([2, 640, 64, 64])]; + tensor reshape_217_cast = reshape(shape = reshape_217_shape_0, x = real_div_54_cast)[name = tensor("reshape_217_cast")]; + tensor add_109_gamma_0_to_fp16 = const()[name = tensor("add_109_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(640781120)))]; + tensor add_109_beta_0_to_fp16 = const()[name = tensor("add_109_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(640782464)))]; + tensor add_109_epsilon_0_to_fp16 = const()[name = tensor("add_109_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_109_cast = batch_norm(beta = add_109_beta_0_to_fp16, epsilon = add_109_epsilon_0_to_fp16, gamma = add_109_gamma_0_to_fp16, mean = add_15_mean_0_to_fp16, variance = add_15_variance_0_to_fp16, x = reshape_217_cast)[name = tensor("add_109_cast")]; + tensor input_479_cast = silu(x = add_109_cast)[name = tensor("input_479_cast")]; + tensor var_13847 = const()[name = tensor("op_13847"), val = tensor([1, 1])]; + tensor var_13849 = const()[name = tensor("op_13849"), val = tensor([1, 1])]; + tensor hidden_states_299_pad_type_0 = const()[name = tensor("hidden_states_299_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_299_pad_0 = const()[name = tensor("hidden_states_299_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor up_blocks_3_resnets_1_conv1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(640783808))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(642166272))), name = tensor("up_blocks_3_resnets_1_conv1_weight_to_fp16_palettized"), shape = tensor([320, 640, 3, 3])]; + tensor up_blocks_3_resnets_1_conv1_bias_to_fp16 = const()[name = tensor("up_blocks_3_resnets_1_conv1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(642166464)))]; + tensor hidden_states_299_cast = conv(bias = up_blocks_3_resnets_1_conv1_bias_to_fp16, dilations = var_13849, groups = var_12807, pad = hidden_states_299_pad_0, pad_type = hidden_states_299_pad_type_0, strides = var_13847, weight = up_blocks_3_resnets_1_conv1_weight_to_fp16_palettized, x = input_479_cast)[name = tensor("hidden_states_299_cast")]; + tensor var_13855 = const()[name = tensor("op_13855"), val = tensor([1, 1])]; + tensor var_13857 = const()[name = tensor("op_13857"), val = tensor([1, 1])]; + tensor temb_41_pad_type_0 = const()[name = tensor("temb_41_pad_type_0"), val = tensor("custom")]; + tensor temb_41_pad_0 = const()[name = tensor("temb_41_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_3_resnets_1_time_emb_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(642167168))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(642474432))), name = tensor("up_blocks_3_resnets_1_time_emb_proj_weight_to_fp16_palettized"), shape = tensor([320, 1280, 1, 1])]; + tensor up_blocks_3_resnets_1_time_emb_proj_bias_to_fp16 = const()[name = tensor("up_blocks_3_resnets_1_time_emb_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(642474624)))]; + tensor temb_41_cast = conv(bias = up_blocks_3_resnets_1_time_emb_proj_bias_to_fp16, dilations = var_13857, groups = var_12807, pad = temb_41_pad_0, pad_type = temb_41_pad_type_0, strides = var_13855, weight = up_blocks_3_resnets_1_time_emb_proj_weight_to_fp16_palettized, x = input_15_cast)[name = tensor("temb_41_cast")]; + tensor input_483_cast = add(x = hidden_states_299_cast, y = temb_41_cast)[name = tensor("input_483_cast")]; + tensor reshape_220_shape_0 = const()[name = tensor("reshape_220_shape_0"), val = tensor([2, 32, 10, 64, 64])]; + tensor reshape_220_cast = reshape(shape = reshape_220_shape_0, x = input_483_cast)[name = tensor("reshape_220_cast")]; + tensor reduce_mean_165_axes_0 = const()[name = tensor("reduce_mean_165_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_165_keep_dims_0 = const()[name = tensor("reduce_mean_165_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_165_cast = reduce_mean(axes = reduce_mean_165_axes_0, keep_dims = reduce_mean_165_keep_dims_0, x = reshape_220_cast)[name = tensor("reduce_mean_165_cast")]; + tensor sub_110_cast = sub(x = reshape_220_cast, y = reduce_mean_165_cast)[name = tensor("sub_110_cast")]; + tensor square_55_cast = square(x = sub_110_cast)[name = tensor("square_55_cast")]; + tensor reduce_mean_167_axes_0 = const()[name = tensor("reduce_mean_167_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_167_keep_dims_0 = const()[name = tensor("reduce_mean_167_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_167_cast = reduce_mean(axes = reduce_mean_167_axes_0, keep_dims = reduce_mean_167_keep_dims_0, x = square_55_cast)[name = tensor("reduce_mean_167_cast")]; + tensor add_110_y_0_to_fp16 = const()[name = tensor("add_110_y_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_110_cast = add(x = reduce_mean_167_cast, y = add_110_y_0_to_fp16)[name = tensor("add_110_cast")]; + tensor sqrt_55_cast = sqrt(x = add_110_cast)[name = tensor("sqrt_55_cast")]; + tensor real_div_55_cast = real_div(x = sub_110_cast, y = sqrt_55_cast)[name = tensor("real_div_55_cast")]; + tensor reshape_221_shape_0 = const()[name = tensor("reshape_221_shape_0"), val = tensor([2, 320, 64, 64])]; + tensor reshape_221_cast = reshape(shape = reshape_221_shape_0, x = real_div_55_cast)[name = tensor("reshape_221_cast")]; + tensor add_111_gamma_0_to_fp16 = const()[name = tensor("add_111_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(642475328)))]; + tensor add_111_beta_0_to_fp16 = const()[name = tensor("add_111_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(642476032)))]; + tensor add_111_epsilon_0_to_fp16 = const()[name = tensor("add_111_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_111_cast = batch_norm(beta = add_111_beta_0_to_fp16, epsilon = add_111_epsilon_0_to_fp16, gamma = add_111_gamma_0_to_fp16, mean = add_1_mean_0_to_fp16, variance = add_1_variance_0_to_fp16, x = reshape_221_cast)[name = tensor("add_111_cast")]; + tensor input_487_cast = silu(x = add_111_cast)[name = tensor("input_487_cast")]; + tensor var_13867 = const()[name = tensor("op_13867"), val = tensor([1, 1])]; + tensor var_13869 = const()[name = tensor("op_13869"), val = tensor([1, 1])]; + tensor hidden_states_301_pad_type_0 = const()[name = tensor("hidden_states_301_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_301_pad_0 = const()[name = tensor("hidden_states_301_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor up_blocks_3_resnets_1_conv2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(642476736))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(643168000))), name = tensor("up_blocks_3_resnets_1_conv2_weight_to_fp16_palettized"), shape = tensor([320, 320, 3, 3])]; + tensor up_blocks_3_resnets_1_conv2_bias_to_fp16 = const()[name = tensor("up_blocks_3_resnets_1_conv2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(643168192)))]; + tensor hidden_states_301_cast = conv(bias = up_blocks_3_resnets_1_conv2_bias_to_fp16, dilations = var_13869, groups = var_12807, pad = hidden_states_301_pad_0, pad_type = hidden_states_301_pad_type_0, strides = var_13867, weight = up_blocks_3_resnets_1_conv2_weight_to_fp16_palettized, x = input_487_cast)[name = tensor("hidden_states_301_cast")]; + tensor var_13874 = const()[name = tensor("op_13874"), val = tensor([1, 1])]; + tensor var_13876 = const()[name = tensor("op_13876"), val = tensor([1, 1])]; + tensor x_25_pad_type_0 = const()[name = tensor("x_25_pad_type_0"), val = tensor("custom")]; + tensor x_25_pad_0 = const()[name = tensor("x_25_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_3_resnets_1_conv_shortcut_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(643168896))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(643322560))), name = tensor("up_blocks_3_resnets_1_conv_shortcut_weight_to_fp16_palettized"), shape = tensor([320, 640, 1, 1])]; + tensor up_blocks_3_resnets_1_conv_shortcut_bias_to_fp16 = const()[name = tensor("up_blocks_3_resnets_1_conv_shortcut_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(643322752)))]; + tensor x_25_cast = conv(bias = up_blocks_3_resnets_1_conv_shortcut_bias_to_fp16, dilations = var_13876, groups = var_12807, pad = x_25_pad_0, pad_type = x_25_pad_type_0, strides = var_13874, weight = up_blocks_3_resnets_1_conv_shortcut_weight_to_fp16_palettized, x = input_475_cast)[name = tensor("x_25_cast")]; + tensor hidden_states_303_cast = add(x = x_25_cast, y = hidden_states_301_cast)[name = tensor("hidden_states_303_cast")]; + tensor reshape_224_shape_0 = const()[name = tensor("reshape_224_shape_0"), val = tensor([2, 32, 10, 64, 64])]; + tensor reshape_224_cast = reshape(shape = reshape_224_shape_0, x = hidden_states_303_cast)[name = tensor("reshape_224_cast")]; + tensor reduce_mean_168_axes_0 = const()[name = tensor("reduce_mean_168_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_168_keep_dims_0 = const()[name = tensor("reduce_mean_168_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_168_cast = reduce_mean(axes = reduce_mean_168_axes_0, keep_dims = reduce_mean_168_keep_dims_0, x = reshape_224_cast)[name = tensor("reduce_mean_168_cast")]; + tensor sub_112_cast = sub(x = reshape_224_cast, y = reduce_mean_168_cast)[name = tensor("sub_112_cast")]; + tensor square_56_cast = square(x = sub_112_cast)[name = tensor("square_56_cast")]; + tensor reduce_mean_170_axes_0 = const()[name = tensor("reduce_mean_170_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_170_keep_dims_0 = const()[name = tensor("reduce_mean_170_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_170_cast = reduce_mean(axes = reduce_mean_170_axes_0, keep_dims = reduce_mean_170_keep_dims_0, x = square_56_cast)[name = tensor("reduce_mean_170_cast")]; + tensor add_112_y_0_to_fp16 = const()[name = tensor("add_112_y_0_to_fp16"), val = tensor(0x1.1p-20)]; + tensor add_112_cast = add(x = reduce_mean_170_cast, y = add_112_y_0_to_fp16)[name = tensor("add_112_cast")]; + tensor sqrt_56_cast = sqrt(x = add_112_cast)[name = tensor("sqrt_56_cast")]; + tensor real_div_56_cast = real_div(x = sub_112_cast, y = sqrt_56_cast)[name = tensor("real_div_56_cast")]; + tensor reshape_225_shape_0 = const()[name = tensor("reshape_225_shape_0"), val = tensor([2, 320, 64, 64])]; + tensor reshape_225_cast = reshape(shape = reshape_225_shape_0, x = real_div_56_cast)[name = tensor("reshape_225_cast")]; + tensor add_113_gamma_0_to_fp16 = const()[name = tensor("add_113_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(643323456)))]; + tensor add_113_beta_0_to_fp16 = const()[name = tensor("add_113_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(643324160)))]; + tensor add_113_epsilon_0_to_fp16 = const()[name = tensor("add_113_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_113_cast = batch_norm(beta = add_113_beta_0_to_fp16, epsilon = add_113_epsilon_0_to_fp16, gamma = add_113_gamma_0_to_fp16, mean = add_1_mean_0_to_fp16, variance = add_1_variance_0_to_fp16, x = reshape_225_cast)[name = tensor("add_113_cast")]; + tensor var_13896 = const()[name = tensor("op_13896"), val = tensor([1, 1])]; + tensor var_13898 = const()[name = tensor("op_13898"), val = tensor([1, 1])]; + tensor hidden_states_305_pad_type_0 = const()[name = tensor("hidden_states_305_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_305_pad_0 = const()[name = tensor("hidden_states_305_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_3_attentions_1_proj_in_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(643324864))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(643401728))), name = tensor("up_blocks_3_attentions_1_proj_in_weight_to_fp16_palettized"), shape = tensor([320, 320, 1, 1])]; + tensor up_blocks_3_attentions_1_proj_in_bias_to_fp16 = const()[name = tensor("up_blocks_3_attentions_1_proj_in_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(643401920)))]; + tensor hidden_states_305_cast = conv(bias = up_blocks_3_attentions_1_proj_in_bias_to_fp16, dilations = var_13898, groups = var_12807, pad = hidden_states_305_pad_0, pad_type = hidden_states_305_pad_type_0, strides = var_13896, weight = up_blocks_3_attentions_1_proj_in_weight_to_fp16_palettized, x = add_113_cast)[name = tensor("hidden_states_305_cast")]; + tensor var_13903 = const()[name = tensor("op_13903"), val = tensor([2, 320, 1, 4096])]; + tensor inputs_85_cast = reshape(shape = var_13903, x = hidden_states_305_cast)[name = tensor("inputs_85_cast")]; + tensor var_13913 = const()[name = tensor("op_13913"), val = tensor([1])]; + tensor channels_mean_85_cast = reduce_mean(axes = var_13913, keep_dims = var_12802, x = inputs_85_cast)[name = tensor("channels_mean_85_cast")]; + tensor zero_mean_85_cast = sub(x = inputs_85_cast, y = channels_mean_85_cast)[name = tensor("zero_mean_85_cast")]; + tensor zero_mean_sq_85_cast = mul(x = zero_mean_85_cast, y = zero_mean_85_cast)[name = tensor("zero_mean_sq_85_cast")]; + tensor var_13917 = const()[name = tensor("op_13917"), val = tensor([1])]; + tensor var_13918_cast = reduce_mean(axes = var_13917, keep_dims = var_12802, x = zero_mean_sq_85_cast)[name = tensor("op_13918_cast")]; + tensor var_13919_to_fp16 = const()[name = tensor("op_13919_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_13920_cast = add(x = var_13918_cast, y = var_13919_to_fp16)[name = tensor("op_13920_cast")]; + tensor denom_85_epsilon_0_to_fp16 = const()[name = tensor("denom_85_epsilon_0_to_fp16"), val = tensor(0x1p-24)]; + tensor denom_85_cast = rsqrt(epsilon = denom_85_epsilon_0_to_fp16, x = var_13920_cast)[name = tensor("denom_85_cast")]; + tensor out_85_cast = mul(x = zero_mean_85_cast, y = denom_85_cast)[name = tensor("out_85_cast")]; + tensor var_13924_to_fp16 = const()[name = tensor("op_13924_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(643402624)))]; + tensor var_13925_cast = add(x = out_85_cast, y = var_13924_to_fp16)[name = tensor("op_13925_cast")]; + tensor var_13927_to_fp16 = const()[name = tensor("op_13927_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(643403328)))]; + tensor hidden_states_307_cast = mul(x = var_13925_cast, y = var_13927_to_fp16)[name = tensor("hidden_states_307_cast")]; + tensor var_13934 = const()[name = tensor("op_13934"), val = tensor([1, 1])]; + tensor var_13936 = const()[name = tensor("op_13936"), val = tensor([1, 1])]; + tensor q_57_pad_type_0 = const()[name = tensor("q_57_pad_type_0"), val = tensor("custom")]; + tensor q_57_pad_0 = const()[name = tensor("q_57_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_3_attentions_1_transformer_blocks_0_attn1_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(643404032))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(643480896))), name = tensor("up_blocks_3_attentions_1_transformer_blocks_0_attn1_to_q_weight_to_fp16_palettized"), shape = tensor([320, 320, 1, 1])]; + tensor q_57_cast = conv(dilations = var_13936, groups = var_12807, pad = q_57_pad_0, pad_type = q_57_pad_type_0, strides = var_13934, weight = up_blocks_3_attentions_1_transformer_blocks_0_attn1_to_q_weight_to_fp16_palettized, x = hidden_states_307_cast)[name = tensor("q_57_cast")]; + tensor var_13940 = const()[name = tensor("op_13940"), val = tensor([1, 1])]; + tensor var_13942 = const()[name = tensor("op_13942"), val = tensor([1, 1])]; + tensor k_113_pad_type_0 = const()[name = tensor("k_113_pad_type_0"), val = tensor("custom")]; + tensor k_113_pad_0 = const()[name = tensor("k_113_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_3_attentions_1_transformer_blocks_0_attn1_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(643481088))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(643557952))), name = tensor("up_blocks_3_attentions_1_transformer_blocks_0_attn1_to_k_weight_to_fp16_palettized"), shape = tensor([320, 320, 1, 1])]; + tensor k_113_cast = conv(dilations = var_13942, groups = var_12807, pad = k_113_pad_0, pad_type = k_113_pad_type_0, strides = var_13940, weight = up_blocks_3_attentions_1_transformer_blocks_0_attn1_to_k_weight_to_fp16_palettized, x = hidden_states_307_cast)[name = tensor("k_113_cast")]; + tensor var_13946 = const()[name = tensor("op_13946"), val = tensor([1, 1])]; + tensor var_13948 = const()[name = tensor("op_13948"), val = tensor([1, 1])]; + tensor v_57_pad_type_0 = const()[name = tensor("v_57_pad_type_0"), val = tensor("custom")]; + tensor v_57_pad_0 = const()[name = tensor("v_57_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_3_attentions_1_transformer_blocks_0_attn1_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(643558144))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(643635008))), name = tensor("up_blocks_3_attentions_1_transformer_blocks_0_attn1_to_v_weight_to_fp16_palettized"), shape = tensor([320, 320, 1, 1])]; + tensor v_57_cast = conv(dilations = var_13948, groups = var_12807, pad = v_57_pad_0, pad_type = v_57_pad_type_0, strides = var_13946, weight = up_blocks_3_attentions_1_transformer_blocks_0_attn1_to_v_weight_to_fp16_palettized, x = hidden_states_307_cast)[name = tensor("v_57_cast")]; + tensor var_13952_begin_0 = const()[name = tensor("op_13952_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_13952_end_0 = const()[name = tensor("op_13952_end_0"), val = tensor([2, 64, 1, 4096])]; + tensor var_13952_end_mask_0 = const()[name = tensor("op_13952_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13952_cast = slice_by_index(begin = var_13952_begin_0, end = var_13952_end_0, end_mask = var_13952_end_mask_0, x = q_57_cast)[name = tensor("op_13952_cast")]; + tensor var_13956_begin_0 = const()[name = tensor("op_13956_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_13956_end_0 = const()[name = tensor("op_13956_end_0"), val = tensor([2, 128, 1, 4096])]; + tensor var_13956_end_mask_0 = const()[name = tensor("op_13956_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13956_cast = slice_by_index(begin = var_13956_begin_0, end = var_13956_end_0, end_mask = var_13956_end_mask_0, x = q_57_cast)[name = tensor("op_13956_cast")]; + tensor var_13960_begin_0 = const()[name = tensor("op_13960_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_13960_end_0 = const()[name = tensor("op_13960_end_0"), val = tensor([2, 192, 1, 4096])]; + tensor var_13960_end_mask_0 = const()[name = tensor("op_13960_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13960_cast = slice_by_index(begin = var_13960_begin_0, end = var_13960_end_0, end_mask = var_13960_end_mask_0, x = q_57_cast)[name = tensor("op_13960_cast")]; + tensor var_13964_begin_0 = const()[name = tensor("op_13964_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_13964_end_0 = const()[name = tensor("op_13964_end_0"), val = tensor([2, 256, 1, 4096])]; + tensor var_13964_end_mask_0 = const()[name = tensor("op_13964_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13964_cast = slice_by_index(begin = var_13964_begin_0, end = var_13964_end_0, end_mask = var_13964_end_mask_0, x = q_57_cast)[name = tensor("op_13964_cast")]; + tensor var_13968_begin_0 = const()[name = tensor("op_13968_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_13968_end_0 = const()[name = tensor("op_13968_end_0"), val = tensor([2, 320, 1, 4096])]; + tensor var_13968_end_mask_0 = const()[name = tensor("op_13968_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13968_cast = slice_by_index(begin = var_13968_begin_0, end = var_13968_end_0, end_mask = var_13968_end_mask_0, x = q_57_cast)[name = tensor("op_13968_cast")]; + tensor var_13971_begin_0 = const()[name = tensor("op_13971_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_13971_end_0 = const()[name = tensor("op_13971_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_13971_end_mask_0 = const()[name = tensor("op_13971_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13971_cast = slice_by_index(begin = var_13971_begin_0, end = var_13971_end_0, end_mask = var_13971_end_mask_0, x = var_13952_cast)[name = tensor("op_13971_cast")]; + tensor var_13972_begin_0 = const()[name = tensor("op_13972_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_13972_end_0 = const()[name = tensor("op_13972_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_13972_end_mask_0 = const()[name = tensor("op_13972_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13972_cast = slice_by_index(begin = var_13972_begin_0, end = var_13972_end_0, end_mask = var_13972_end_mask_0, x = var_13952_cast)[name = tensor("op_13972_cast")]; + tensor var_13973_begin_0 = const()[name = tensor("op_13973_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_13973_end_0 = const()[name = tensor("op_13973_end_0"), val = tensor([2, 64, 1, 1536])]; + tensor var_13973_end_mask_0 = const()[name = tensor("op_13973_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13973_cast = slice_by_index(begin = var_13973_begin_0, end = var_13973_end_0, end_mask = var_13973_end_mask_0, x = var_13952_cast)[name = tensor("op_13973_cast")]; + tensor var_13974_begin_0 = const()[name = tensor("op_13974_begin_0"), val = tensor([0, 0, 0, 1536])]; + tensor var_13974_end_0 = const()[name = tensor("op_13974_end_0"), val = tensor([2, 64, 1, 2048])]; + tensor var_13974_end_mask_0 = const()[name = tensor("op_13974_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13974_cast = slice_by_index(begin = var_13974_begin_0, end = var_13974_end_0, end_mask = var_13974_end_mask_0, x = var_13952_cast)[name = tensor("op_13974_cast")]; + tensor var_13975_begin_0 = const()[name = tensor("op_13975_begin_0"), val = tensor([0, 0, 0, 2048])]; + tensor var_13975_end_0 = const()[name = tensor("op_13975_end_0"), val = tensor([2, 64, 1, 2560])]; + tensor var_13975_end_mask_0 = const()[name = tensor("op_13975_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13975_cast = slice_by_index(begin = var_13975_begin_0, end = var_13975_end_0, end_mask = var_13975_end_mask_0, x = var_13952_cast)[name = tensor("op_13975_cast")]; + tensor var_13976_begin_0 = const()[name = tensor("op_13976_begin_0"), val = tensor([0, 0, 0, 2560])]; + tensor var_13976_end_0 = const()[name = tensor("op_13976_end_0"), val = tensor([2, 64, 1, 3072])]; + tensor var_13976_end_mask_0 = const()[name = tensor("op_13976_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13976_cast = slice_by_index(begin = var_13976_begin_0, end = var_13976_end_0, end_mask = var_13976_end_mask_0, x = var_13952_cast)[name = tensor("op_13976_cast")]; + tensor var_13977_begin_0 = const()[name = tensor("op_13977_begin_0"), val = tensor([0, 0, 0, 3072])]; + tensor var_13977_end_0 = const()[name = tensor("op_13977_end_0"), val = tensor([2, 64, 1, 3584])]; + tensor var_13977_end_mask_0 = const()[name = tensor("op_13977_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13977_cast = slice_by_index(begin = var_13977_begin_0, end = var_13977_end_0, end_mask = var_13977_end_mask_0, x = var_13952_cast)[name = tensor("op_13977_cast")]; + tensor var_13978_begin_0 = const()[name = tensor("op_13978_begin_0"), val = tensor([0, 0, 0, 3584])]; + tensor var_13978_end_0 = const()[name = tensor("op_13978_end_0"), val = tensor([2, 64, 1, 4096])]; + tensor var_13978_end_mask_0 = const()[name = tensor("op_13978_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13978_cast = slice_by_index(begin = var_13978_begin_0, end = var_13978_end_0, end_mask = var_13978_end_mask_0, x = var_13952_cast)[name = tensor("op_13978_cast")]; + tensor var_13979_begin_0 = const()[name = tensor("op_13979_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_13979_end_0 = const()[name = tensor("op_13979_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_13979_end_mask_0 = const()[name = tensor("op_13979_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13979_cast = slice_by_index(begin = var_13979_begin_0, end = var_13979_end_0, end_mask = var_13979_end_mask_0, x = var_13956_cast)[name = tensor("op_13979_cast")]; + tensor var_13980_begin_0 = const()[name = tensor("op_13980_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_13980_end_0 = const()[name = tensor("op_13980_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_13980_end_mask_0 = const()[name = tensor("op_13980_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13980_cast = slice_by_index(begin = var_13980_begin_0, end = var_13980_end_0, end_mask = var_13980_end_mask_0, x = var_13956_cast)[name = tensor("op_13980_cast")]; + tensor var_13981_begin_0 = const()[name = tensor("op_13981_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_13981_end_0 = const()[name = tensor("op_13981_end_0"), val = tensor([2, 64, 1, 1536])]; + tensor var_13981_end_mask_0 = const()[name = tensor("op_13981_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13981_cast = slice_by_index(begin = var_13981_begin_0, end = var_13981_end_0, end_mask = var_13981_end_mask_0, x = var_13956_cast)[name = tensor("op_13981_cast")]; + tensor var_13982_begin_0 = const()[name = tensor("op_13982_begin_0"), val = tensor([0, 0, 0, 1536])]; + tensor var_13982_end_0 = const()[name = tensor("op_13982_end_0"), val = tensor([2, 64, 1, 2048])]; + tensor var_13982_end_mask_0 = const()[name = tensor("op_13982_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13982_cast = slice_by_index(begin = var_13982_begin_0, end = var_13982_end_0, end_mask = var_13982_end_mask_0, x = var_13956_cast)[name = tensor("op_13982_cast")]; + tensor var_13983_begin_0 = const()[name = tensor("op_13983_begin_0"), val = tensor([0, 0, 0, 2048])]; + tensor var_13983_end_0 = const()[name = tensor("op_13983_end_0"), val = tensor([2, 64, 1, 2560])]; + tensor var_13983_end_mask_0 = const()[name = tensor("op_13983_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13983_cast = slice_by_index(begin = var_13983_begin_0, end = var_13983_end_0, end_mask = var_13983_end_mask_0, x = var_13956_cast)[name = tensor("op_13983_cast")]; + tensor var_13984_begin_0 = const()[name = tensor("op_13984_begin_0"), val = tensor([0, 0, 0, 2560])]; + tensor var_13984_end_0 = const()[name = tensor("op_13984_end_0"), val = tensor([2, 64, 1, 3072])]; + tensor var_13984_end_mask_0 = const()[name = tensor("op_13984_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13984_cast = slice_by_index(begin = var_13984_begin_0, end = var_13984_end_0, end_mask = var_13984_end_mask_0, x = var_13956_cast)[name = tensor("op_13984_cast")]; + tensor var_13985_begin_0 = const()[name = tensor("op_13985_begin_0"), val = tensor([0, 0, 0, 3072])]; + tensor var_13985_end_0 = const()[name = tensor("op_13985_end_0"), val = tensor([2, 64, 1, 3584])]; + tensor var_13985_end_mask_0 = const()[name = tensor("op_13985_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13985_cast = slice_by_index(begin = var_13985_begin_0, end = var_13985_end_0, end_mask = var_13985_end_mask_0, x = var_13956_cast)[name = tensor("op_13985_cast")]; + tensor var_13986_begin_0 = const()[name = tensor("op_13986_begin_0"), val = tensor([0, 0, 0, 3584])]; + tensor var_13986_end_0 = const()[name = tensor("op_13986_end_0"), val = tensor([2, 64, 1, 4096])]; + tensor var_13986_end_mask_0 = const()[name = tensor("op_13986_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13986_cast = slice_by_index(begin = var_13986_begin_0, end = var_13986_end_0, end_mask = var_13986_end_mask_0, x = var_13956_cast)[name = tensor("op_13986_cast")]; + tensor var_13987_begin_0 = const()[name = tensor("op_13987_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_13987_end_0 = const()[name = tensor("op_13987_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_13987_end_mask_0 = const()[name = tensor("op_13987_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13987_cast = slice_by_index(begin = var_13987_begin_0, end = var_13987_end_0, end_mask = var_13987_end_mask_0, x = var_13960_cast)[name = tensor("op_13987_cast")]; + tensor var_13988_begin_0 = const()[name = tensor("op_13988_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_13988_end_0 = const()[name = tensor("op_13988_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_13988_end_mask_0 = const()[name = tensor("op_13988_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13988_cast = slice_by_index(begin = var_13988_begin_0, end = var_13988_end_0, end_mask = var_13988_end_mask_0, x = var_13960_cast)[name = tensor("op_13988_cast")]; + tensor var_13989_begin_0 = const()[name = tensor("op_13989_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_13989_end_0 = const()[name = tensor("op_13989_end_0"), val = tensor([2, 64, 1, 1536])]; + tensor var_13989_end_mask_0 = const()[name = tensor("op_13989_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13989_cast = slice_by_index(begin = var_13989_begin_0, end = var_13989_end_0, end_mask = var_13989_end_mask_0, x = var_13960_cast)[name = tensor("op_13989_cast")]; + tensor var_13990_begin_0 = const()[name = tensor("op_13990_begin_0"), val = tensor([0, 0, 0, 1536])]; + tensor var_13990_end_0 = const()[name = tensor("op_13990_end_0"), val = tensor([2, 64, 1, 2048])]; + tensor var_13990_end_mask_0 = const()[name = tensor("op_13990_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13990_cast = slice_by_index(begin = var_13990_begin_0, end = var_13990_end_0, end_mask = var_13990_end_mask_0, x = var_13960_cast)[name = tensor("op_13990_cast")]; + tensor var_13991_begin_0 = const()[name = tensor("op_13991_begin_0"), val = tensor([0, 0, 0, 2048])]; + tensor var_13991_end_0 = const()[name = tensor("op_13991_end_0"), val = tensor([2, 64, 1, 2560])]; + tensor var_13991_end_mask_0 = const()[name = tensor("op_13991_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13991_cast = slice_by_index(begin = var_13991_begin_0, end = var_13991_end_0, end_mask = var_13991_end_mask_0, x = var_13960_cast)[name = tensor("op_13991_cast")]; + tensor var_13992_begin_0 = const()[name = tensor("op_13992_begin_0"), val = tensor([0, 0, 0, 2560])]; + tensor var_13992_end_0 = const()[name = tensor("op_13992_end_0"), val = tensor([2, 64, 1, 3072])]; + tensor var_13992_end_mask_0 = const()[name = tensor("op_13992_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13992_cast = slice_by_index(begin = var_13992_begin_0, end = var_13992_end_0, end_mask = var_13992_end_mask_0, x = var_13960_cast)[name = tensor("op_13992_cast")]; + tensor var_13993_begin_0 = const()[name = tensor("op_13993_begin_0"), val = tensor([0, 0, 0, 3072])]; + tensor var_13993_end_0 = const()[name = tensor("op_13993_end_0"), val = tensor([2, 64, 1, 3584])]; + tensor var_13993_end_mask_0 = const()[name = tensor("op_13993_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13993_cast = slice_by_index(begin = var_13993_begin_0, end = var_13993_end_0, end_mask = var_13993_end_mask_0, x = var_13960_cast)[name = tensor("op_13993_cast")]; + tensor var_13994_begin_0 = const()[name = tensor("op_13994_begin_0"), val = tensor([0, 0, 0, 3584])]; + tensor var_13994_end_0 = const()[name = tensor("op_13994_end_0"), val = tensor([2, 64, 1, 4096])]; + tensor var_13994_end_mask_0 = const()[name = tensor("op_13994_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13994_cast = slice_by_index(begin = var_13994_begin_0, end = var_13994_end_0, end_mask = var_13994_end_mask_0, x = var_13960_cast)[name = tensor("op_13994_cast")]; + tensor var_13995_begin_0 = const()[name = tensor("op_13995_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_13995_end_0 = const()[name = tensor("op_13995_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_13995_end_mask_0 = const()[name = tensor("op_13995_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13995_cast = slice_by_index(begin = var_13995_begin_0, end = var_13995_end_0, end_mask = var_13995_end_mask_0, x = var_13964_cast)[name = tensor("op_13995_cast")]; + tensor var_13996_begin_0 = const()[name = tensor("op_13996_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_13996_end_0 = const()[name = tensor("op_13996_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_13996_end_mask_0 = const()[name = tensor("op_13996_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13996_cast = slice_by_index(begin = var_13996_begin_0, end = var_13996_end_0, end_mask = var_13996_end_mask_0, x = var_13964_cast)[name = tensor("op_13996_cast")]; + tensor var_13997_begin_0 = const()[name = tensor("op_13997_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_13997_end_0 = const()[name = tensor("op_13997_end_0"), val = tensor([2, 64, 1, 1536])]; + tensor var_13997_end_mask_0 = const()[name = tensor("op_13997_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13997_cast = slice_by_index(begin = var_13997_begin_0, end = var_13997_end_0, end_mask = var_13997_end_mask_0, x = var_13964_cast)[name = tensor("op_13997_cast")]; + tensor var_13998_begin_0 = const()[name = tensor("op_13998_begin_0"), val = tensor([0, 0, 0, 1536])]; + tensor var_13998_end_0 = const()[name = tensor("op_13998_end_0"), val = tensor([2, 64, 1, 2048])]; + tensor var_13998_end_mask_0 = const()[name = tensor("op_13998_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13998_cast = slice_by_index(begin = var_13998_begin_0, end = var_13998_end_0, end_mask = var_13998_end_mask_0, x = var_13964_cast)[name = tensor("op_13998_cast")]; + tensor var_13999_begin_0 = const()[name = tensor("op_13999_begin_0"), val = tensor([0, 0, 0, 2048])]; + tensor var_13999_end_0 = const()[name = tensor("op_13999_end_0"), val = tensor([2, 64, 1, 2560])]; + tensor var_13999_end_mask_0 = const()[name = tensor("op_13999_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13999_cast = slice_by_index(begin = var_13999_begin_0, end = var_13999_end_0, end_mask = var_13999_end_mask_0, x = var_13964_cast)[name = tensor("op_13999_cast")]; + tensor var_14000_begin_0 = const()[name = tensor("op_14000_begin_0"), val = tensor([0, 0, 0, 2560])]; + tensor var_14000_end_0 = const()[name = tensor("op_14000_end_0"), val = tensor([2, 64, 1, 3072])]; + tensor var_14000_end_mask_0 = const()[name = tensor("op_14000_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14000_cast = slice_by_index(begin = var_14000_begin_0, end = var_14000_end_0, end_mask = var_14000_end_mask_0, x = var_13964_cast)[name = tensor("op_14000_cast")]; + tensor var_14001_begin_0 = const()[name = tensor("op_14001_begin_0"), val = tensor([0, 0, 0, 3072])]; + tensor var_14001_end_0 = const()[name = tensor("op_14001_end_0"), val = tensor([2, 64, 1, 3584])]; + tensor var_14001_end_mask_0 = const()[name = tensor("op_14001_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14001_cast = slice_by_index(begin = var_14001_begin_0, end = var_14001_end_0, end_mask = var_14001_end_mask_0, x = var_13964_cast)[name = tensor("op_14001_cast")]; + tensor var_14002_begin_0 = const()[name = tensor("op_14002_begin_0"), val = tensor([0, 0, 0, 3584])]; + tensor var_14002_end_0 = const()[name = tensor("op_14002_end_0"), val = tensor([2, 64, 1, 4096])]; + tensor var_14002_end_mask_0 = const()[name = tensor("op_14002_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14002_cast = slice_by_index(begin = var_14002_begin_0, end = var_14002_end_0, end_mask = var_14002_end_mask_0, x = var_13964_cast)[name = tensor("op_14002_cast")]; + tensor var_14003_begin_0 = const()[name = tensor("op_14003_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_14003_end_0 = const()[name = tensor("op_14003_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_14003_end_mask_0 = const()[name = tensor("op_14003_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14003_cast = slice_by_index(begin = var_14003_begin_0, end = var_14003_end_0, end_mask = var_14003_end_mask_0, x = var_13968_cast)[name = tensor("op_14003_cast")]; + tensor var_14004_begin_0 = const()[name = tensor("op_14004_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_14004_end_0 = const()[name = tensor("op_14004_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_14004_end_mask_0 = const()[name = tensor("op_14004_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14004_cast = slice_by_index(begin = var_14004_begin_0, end = var_14004_end_0, end_mask = var_14004_end_mask_0, x = var_13968_cast)[name = tensor("op_14004_cast")]; + tensor var_14005_begin_0 = const()[name = tensor("op_14005_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_14005_end_0 = const()[name = tensor("op_14005_end_0"), val = tensor([2, 64, 1, 1536])]; + tensor var_14005_end_mask_0 = const()[name = tensor("op_14005_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14005_cast = slice_by_index(begin = var_14005_begin_0, end = var_14005_end_0, end_mask = var_14005_end_mask_0, x = var_13968_cast)[name = tensor("op_14005_cast")]; + tensor var_14006_begin_0 = const()[name = tensor("op_14006_begin_0"), val = tensor([0, 0, 0, 1536])]; + tensor var_14006_end_0 = const()[name = tensor("op_14006_end_0"), val = tensor([2, 64, 1, 2048])]; + tensor var_14006_end_mask_0 = const()[name = tensor("op_14006_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14006_cast = slice_by_index(begin = var_14006_begin_0, end = var_14006_end_0, end_mask = var_14006_end_mask_0, x = var_13968_cast)[name = tensor("op_14006_cast")]; + tensor var_14007_begin_0 = const()[name = tensor("op_14007_begin_0"), val = tensor([0, 0, 0, 2048])]; + tensor var_14007_end_0 = const()[name = tensor("op_14007_end_0"), val = tensor([2, 64, 1, 2560])]; + tensor var_14007_end_mask_0 = const()[name = tensor("op_14007_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14007_cast = slice_by_index(begin = var_14007_begin_0, end = var_14007_end_0, end_mask = var_14007_end_mask_0, x = var_13968_cast)[name = tensor("op_14007_cast")]; + tensor var_14008_begin_0 = const()[name = tensor("op_14008_begin_0"), val = tensor([0, 0, 0, 2560])]; + tensor var_14008_end_0 = const()[name = tensor("op_14008_end_0"), val = tensor([2, 64, 1, 3072])]; + tensor var_14008_end_mask_0 = const()[name = tensor("op_14008_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14008_cast = slice_by_index(begin = var_14008_begin_0, end = var_14008_end_0, end_mask = var_14008_end_mask_0, x = var_13968_cast)[name = tensor("op_14008_cast")]; + tensor var_14009_begin_0 = const()[name = tensor("op_14009_begin_0"), val = tensor([0, 0, 0, 3072])]; + tensor var_14009_end_0 = const()[name = tensor("op_14009_end_0"), val = tensor([2, 64, 1, 3584])]; + tensor var_14009_end_mask_0 = const()[name = tensor("op_14009_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14009_cast = slice_by_index(begin = var_14009_begin_0, end = var_14009_end_0, end_mask = var_14009_end_mask_0, x = var_13968_cast)[name = tensor("op_14009_cast")]; + tensor var_14010_begin_0 = const()[name = tensor("op_14010_begin_0"), val = tensor([0, 0, 0, 3584])]; + tensor var_14010_end_0 = const()[name = tensor("op_14010_end_0"), val = tensor([2, 64, 1, 4096])]; + tensor var_14010_end_mask_0 = const()[name = tensor("op_14010_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14010_cast = slice_by_index(begin = var_14010_begin_0, end = var_14010_end_0, end_mask = var_14010_end_mask_0, x = var_13968_cast)[name = tensor("op_14010_cast")]; + tensor k_115_perm_0 = const()[name = tensor("k_115_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_14015_begin_0 = const()[name = tensor("op_14015_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_14015_end_0 = const()[name = tensor("op_14015_end_0"), val = tensor([2, 4096, 1, 64])]; + tensor var_14015_end_mask_0 = const()[name = tensor("op_14015_end_mask_0"), val = tensor([true, true, true, false])]; + tensor transpose_3 = transpose(perm = k_115_perm_0, x = k_113_cast)[name = tensor("transpose_3")]; + tensor var_14015_cast = slice_by_index(begin = var_14015_begin_0, end = var_14015_end_0, end_mask = var_14015_end_mask_0, x = transpose_3)[name = tensor("op_14015_cast")]; + tensor var_14019_begin_0 = const()[name = tensor("op_14019_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_14019_end_0 = const()[name = tensor("op_14019_end_0"), val = tensor([2, 4096, 1, 128])]; + tensor var_14019_end_mask_0 = const()[name = tensor("op_14019_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14019_cast = slice_by_index(begin = var_14019_begin_0, end = var_14019_end_0, end_mask = var_14019_end_mask_0, x = transpose_3)[name = tensor("op_14019_cast")]; + tensor var_14023_begin_0 = const()[name = tensor("op_14023_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_14023_end_0 = const()[name = tensor("op_14023_end_0"), val = tensor([2, 4096, 1, 192])]; + tensor var_14023_end_mask_0 = const()[name = tensor("op_14023_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14023_cast = slice_by_index(begin = var_14023_begin_0, end = var_14023_end_0, end_mask = var_14023_end_mask_0, x = transpose_3)[name = tensor("op_14023_cast")]; + tensor var_14027_begin_0 = const()[name = tensor("op_14027_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_14027_end_0 = const()[name = tensor("op_14027_end_0"), val = tensor([2, 4096, 1, 256])]; + tensor var_14027_end_mask_0 = const()[name = tensor("op_14027_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14027_cast = slice_by_index(begin = var_14027_begin_0, end = var_14027_end_0, end_mask = var_14027_end_mask_0, x = transpose_3)[name = tensor("op_14027_cast")]; + tensor var_14031_begin_0 = const()[name = tensor("op_14031_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_14031_end_0 = const()[name = tensor("op_14031_end_0"), val = tensor([2, 4096, 1, 320])]; + tensor var_14031_end_mask_0 = const()[name = tensor("op_14031_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14031_cast = slice_by_index(begin = var_14031_begin_0, end = var_14031_end_0, end_mask = var_14031_end_mask_0, x = transpose_3)[name = tensor("op_14031_cast")]; + tensor var_14033_begin_0 = const()[name = tensor("op_14033_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_14033_end_0 = const()[name = tensor("op_14033_end_0"), val = tensor([2, 64, 1, 4096])]; + tensor var_14033_end_mask_0 = const()[name = tensor("op_14033_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14033_cast = slice_by_index(begin = var_14033_begin_0, end = var_14033_end_0, end_mask = var_14033_end_mask_0, x = v_57_cast)[name = tensor("op_14033_cast")]; + tensor var_14037_begin_0 = const()[name = tensor("op_14037_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_14037_end_0 = const()[name = tensor("op_14037_end_0"), val = tensor([2, 128, 1, 4096])]; + tensor var_14037_end_mask_0 = const()[name = tensor("op_14037_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14037_cast = slice_by_index(begin = var_14037_begin_0, end = var_14037_end_0, end_mask = var_14037_end_mask_0, x = v_57_cast)[name = tensor("op_14037_cast")]; + tensor var_14041_begin_0 = const()[name = tensor("op_14041_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_14041_end_0 = const()[name = tensor("op_14041_end_0"), val = tensor([2, 192, 1, 4096])]; + tensor var_14041_end_mask_0 = const()[name = tensor("op_14041_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14041_cast = slice_by_index(begin = var_14041_begin_0, end = var_14041_end_0, end_mask = var_14041_end_mask_0, x = v_57_cast)[name = tensor("op_14041_cast")]; + tensor var_14045_begin_0 = const()[name = tensor("op_14045_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_14045_end_0 = const()[name = tensor("op_14045_end_0"), val = tensor([2, 256, 1, 4096])]; + tensor var_14045_end_mask_0 = const()[name = tensor("op_14045_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14045_cast = slice_by_index(begin = var_14045_begin_0, end = var_14045_end_0, end_mask = var_14045_end_mask_0, x = v_57_cast)[name = tensor("op_14045_cast")]; + tensor var_14049_begin_0 = const()[name = tensor("op_14049_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_14049_end_0 = const()[name = tensor("op_14049_end_0"), val = tensor([2, 320, 1, 4096])]; + tensor var_14049_end_mask_0 = const()[name = tensor("op_14049_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14049_cast = slice_by_index(begin = var_14049_begin_0, end = var_14049_end_0, end_mask = var_14049_end_mask_0, x = v_57_cast)[name = tensor("op_14049_cast")]; + tensor var_14053_equation_0 = const()[name = tensor("op_14053_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14053_cast = einsum(equation = var_14053_equation_0, values = (var_14015_cast, var_13971_cast))[name = tensor("op_14053_cast")]; + tensor var_14054_to_fp16 = const()[name = tensor("op_14054_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_881_cast = mul(x = var_14053_cast, y = var_14054_to_fp16)[name = tensor("aw_chunk_881_cast")]; + tensor var_14057_equation_0 = const()[name = tensor("op_14057_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14057_cast = einsum(equation = var_14057_equation_0, values = (var_14015_cast, var_13972_cast))[name = tensor("op_14057_cast")]; + tensor var_14058_to_fp16 = const()[name = tensor("op_14058_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_883_cast = mul(x = var_14057_cast, y = var_14058_to_fp16)[name = tensor("aw_chunk_883_cast")]; + tensor var_14061_equation_0 = const()[name = tensor("op_14061_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14061_cast = einsum(equation = var_14061_equation_0, values = (var_14015_cast, var_13973_cast))[name = tensor("op_14061_cast")]; + tensor var_14062_to_fp16 = const()[name = tensor("op_14062_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_885_cast = mul(x = var_14061_cast, y = var_14062_to_fp16)[name = tensor("aw_chunk_885_cast")]; + tensor var_14065_equation_0 = const()[name = tensor("op_14065_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14065_cast = einsum(equation = var_14065_equation_0, values = (var_14015_cast, var_13974_cast))[name = tensor("op_14065_cast")]; + tensor var_14066_to_fp16 = const()[name = tensor("op_14066_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_887_cast = mul(x = var_14065_cast, y = var_14066_to_fp16)[name = tensor("aw_chunk_887_cast")]; + tensor var_14069_equation_0 = const()[name = tensor("op_14069_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14069_cast = einsum(equation = var_14069_equation_0, values = (var_14015_cast, var_13975_cast))[name = tensor("op_14069_cast")]; + tensor var_14070_to_fp16 = const()[name = tensor("op_14070_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_889_cast = mul(x = var_14069_cast, y = var_14070_to_fp16)[name = tensor("aw_chunk_889_cast")]; + tensor var_14073_equation_0 = const()[name = tensor("op_14073_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14073_cast = einsum(equation = var_14073_equation_0, values = (var_14015_cast, var_13976_cast))[name = tensor("op_14073_cast")]; + tensor var_14074_to_fp16 = const()[name = tensor("op_14074_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_891_cast = mul(x = var_14073_cast, y = var_14074_to_fp16)[name = tensor("aw_chunk_891_cast")]; + tensor var_14077_equation_0 = const()[name = tensor("op_14077_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14077_cast = einsum(equation = var_14077_equation_0, values = (var_14015_cast, var_13977_cast))[name = tensor("op_14077_cast")]; + tensor var_14078_to_fp16 = const()[name = tensor("op_14078_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_893_cast = mul(x = var_14077_cast, y = var_14078_to_fp16)[name = tensor("aw_chunk_893_cast")]; + tensor var_14081_equation_0 = const()[name = tensor("op_14081_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14081_cast = einsum(equation = var_14081_equation_0, values = (var_14015_cast, var_13978_cast))[name = tensor("op_14081_cast")]; + tensor var_14082_to_fp16 = const()[name = tensor("op_14082_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_895_cast = mul(x = var_14081_cast, y = var_14082_to_fp16)[name = tensor("aw_chunk_895_cast")]; + tensor var_14085_equation_0 = const()[name = tensor("op_14085_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14085_cast = einsum(equation = var_14085_equation_0, values = (var_14019_cast, var_13979_cast))[name = tensor("op_14085_cast")]; + tensor var_14086_to_fp16 = const()[name = tensor("op_14086_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_897_cast = mul(x = var_14085_cast, y = var_14086_to_fp16)[name = tensor("aw_chunk_897_cast")]; + tensor var_14089_equation_0 = const()[name = tensor("op_14089_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14089_cast = einsum(equation = var_14089_equation_0, values = (var_14019_cast, var_13980_cast))[name = tensor("op_14089_cast")]; + tensor var_14090_to_fp16 = const()[name = tensor("op_14090_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_899_cast = mul(x = var_14089_cast, y = var_14090_to_fp16)[name = tensor("aw_chunk_899_cast")]; + tensor var_14093_equation_0 = const()[name = tensor("op_14093_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14093_cast = einsum(equation = var_14093_equation_0, values = (var_14019_cast, var_13981_cast))[name = tensor("op_14093_cast")]; + tensor var_14094_to_fp16 = const()[name = tensor("op_14094_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_901_cast = mul(x = var_14093_cast, y = var_14094_to_fp16)[name = tensor("aw_chunk_901_cast")]; + tensor var_14097_equation_0 = const()[name = tensor("op_14097_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14097_cast = einsum(equation = var_14097_equation_0, values = (var_14019_cast, var_13982_cast))[name = tensor("op_14097_cast")]; + tensor var_14098_to_fp16 = const()[name = tensor("op_14098_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_903_cast = mul(x = var_14097_cast, y = var_14098_to_fp16)[name = tensor("aw_chunk_903_cast")]; + tensor var_14101_equation_0 = const()[name = tensor("op_14101_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14101_cast = einsum(equation = var_14101_equation_0, values = (var_14019_cast, var_13983_cast))[name = tensor("op_14101_cast")]; + tensor var_14102_to_fp16 = const()[name = tensor("op_14102_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_905_cast = mul(x = var_14101_cast, y = var_14102_to_fp16)[name = tensor("aw_chunk_905_cast")]; + tensor var_14105_equation_0 = const()[name = tensor("op_14105_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14105_cast = einsum(equation = var_14105_equation_0, values = (var_14019_cast, var_13984_cast))[name = tensor("op_14105_cast")]; + tensor var_14106_to_fp16 = const()[name = tensor("op_14106_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_907_cast = mul(x = var_14105_cast, y = var_14106_to_fp16)[name = tensor("aw_chunk_907_cast")]; + tensor var_14109_equation_0 = const()[name = tensor("op_14109_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14109_cast = einsum(equation = var_14109_equation_0, values = (var_14019_cast, var_13985_cast))[name = tensor("op_14109_cast")]; + tensor var_14110_to_fp16 = const()[name = tensor("op_14110_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_909_cast = mul(x = var_14109_cast, y = var_14110_to_fp16)[name = tensor("aw_chunk_909_cast")]; + tensor var_14113_equation_0 = const()[name = tensor("op_14113_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14113_cast = einsum(equation = var_14113_equation_0, values = (var_14019_cast, var_13986_cast))[name = tensor("op_14113_cast")]; + tensor var_14114_to_fp16 = const()[name = tensor("op_14114_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_911_cast = mul(x = var_14113_cast, y = var_14114_to_fp16)[name = tensor("aw_chunk_911_cast")]; + tensor var_14117_equation_0 = const()[name = tensor("op_14117_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14117_cast = einsum(equation = var_14117_equation_0, values = (var_14023_cast, var_13987_cast))[name = tensor("op_14117_cast")]; + tensor var_14118_to_fp16 = const()[name = tensor("op_14118_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_913_cast = mul(x = var_14117_cast, y = var_14118_to_fp16)[name = tensor("aw_chunk_913_cast")]; + tensor var_14121_equation_0 = const()[name = tensor("op_14121_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14121_cast = einsum(equation = var_14121_equation_0, values = (var_14023_cast, var_13988_cast))[name = tensor("op_14121_cast")]; + tensor var_14122_to_fp16 = const()[name = tensor("op_14122_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_915_cast = mul(x = var_14121_cast, y = var_14122_to_fp16)[name = tensor("aw_chunk_915_cast")]; + tensor var_14125_equation_0 = const()[name = tensor("op_14125_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14125_cast = einsum(equation = var_14125_equation_0, values = (var_14023_cast, var_13989_cast))[name = tensor("op_14125_cast")]; + tensor var_14126_to_fp16 = const()[name = tensor("op_14126_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_917_cast = mul(x = var_14125_cast, y = var_14126_to_fp16)[name = tensor("aw_chunk_917_cast")]; + tensor var_14129_equation_0 = const()[name = tensor("op_14129_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14129_cast = einsum(equation = var_14129_equation_0, values = (var_14023_cast, var_13990_cast))[name = tensor("op_14129_cast")]; + tensor var_14130_to_fp16 = const()[name = tensor("op_14130_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_919_cast = mul(x = var_14129_cast, y = var_14130_to_fp16)[name = tensor("aw_chunk_919_cast")]; + tensor var_14133_equation_0 = const()[name = tensor("op_14133_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14133_cast = einsum(equation = var_14133_equation_0, values = (var_14023_cast, var_13991_cast))[name = tensor("op_14133_cast")]; + tensor var_14134_to_fp16 = const()[name = tensor("op_14134_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_921_cast = mul(x = var_14133_cast, y = var_14134_to_fp16)[name = tensor("aw_chunk_921_cast")]; + tensor var_14137_equation_0 = const()[name = tensor("op_14137_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14137_cast = einsum(equation = var_14137_equation_0, values = (var_14023_cast, var_13992_cast))[name = tensor("op_14137_cast")]; + tensor var_14138_to_fp16 = const()[name = tensor("op_14138_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_923_cast = mul(x = var_14137_cast, y = var_14138_to_fp16)[name = tensor("aw_chunk_923_cast")]; + tensor var_14141_equation_0 = const()[name = tensor("op_14141_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14141_cast = einsum(equation = var_14141_equation_0, values = (var_14023_cast, var_13993_cast))[name = tensor("op_14141_cast")]; + tensor var_14142_to_fp16 = const()[name = tensor("op_14142_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_925_cast = mul(x = var_14141_cast, y = var_14142_to_fp16)[name = tensor("aw_chunk_925_cast")]; + tensor var_14145_equation_0 = const()[name = tensor("op_14145_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14145_cast = einsum(equation = var_14145_equation_0, values = (var_14023_cast, var_13994_cast))[name = tensor("op_14145_cast")]; + tensor var_14146_to_fp16 = const()[name = tensor("op_14146_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_927_cast = mul(x = var_14145_cast, y = var_14146_to_fp16)[name = tensor("aw_chunk_927_cast")]; + tensor var_14149_equation_0 = const()[name = tensor("op_14149_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14149_cast = einsum(equation = var_14149_equation_0, values = (var_14027_cast, var_13995_cast))[name = tensor("op_14149_cast")]; + tensor var_14150_to_fp16 = const()[name = tensor("op_14150_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_929_cast = mul(x = var_14149_cast, y = var_14150_to_fp16)[name = tensor("aw_chunk_929_cast")]; + tensor var_14153_equation_0 = const()[name = tensor("op_14153_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14153_cast = einsum(equation = var_14153_equation_0, values = (var_14027_cast, var_13996_cast))[name = tensor("op_14153_cast")]; + tensor var_14154_to_fp16 = const()[name = tensor("op_14154_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_931_cast = mul(x = var_14153_cast, y = var_14154_to_fp16)[name = tensor("aw_chunk_931_cast")]; + tensor var_14157_equation_0 = const()[name = tensor("op_14157_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14157_cast = einsum(equation = var_14157_equation_0, values = (var_14027_cast, var_13997_cast))[name = tensor("op_14157_cast")]; + tensor var_14158_to_fp16 = const()[name = tensor("op_14158_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_933_cast = mul(x = var_14157_cast, y = var_14158_to_fp16)[name = tensor("aw_chunk_933_cast")]; + tensor var_14161_equation_0 = const()[name = tensor("op_14161_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14161_cast = einsum(equation = var_14161_equation_0, values = (var_14027_cast, var_13998_cast))[name = tensor("op_14161_cast")]; + tensor var_14162_to_fp16 = const()[name = tensor("op_14162_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_935_cast = mul(x = var_14161_cast, y = var_14162_to_fp16)[name = tensor("aw_chunk_935_cast")]; + tensor var_14165_equation_0 = const()[name = tensor("op_14165_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14165_cast = einsum(equation = var_14165_equation_0, values = (var_14027_cast, var_13999_cast))[name = tensor("op_14165_cast")]; + tensor var_14166_to_fp16 = const()[name = tensor("op_14166_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_937_cast = mul(x = var_14165_cast, y = var_14166_to_fp16)[name = tensor("aw_chunk_937_cast")]; + tensor var_14169_equation_0 = const()[name = tensor("op_14169_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14169_cast = einsum(equation = var_14169_equation_0, values = (var_14027_cast, var_14000_cast))[name = tensor("op_14169_cast")]; + tensor var_14170_to_fp16 = const()[name = tensor("op_14170_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_939_cast = mul(x = var_14169_cast, y = var_14170_to_fp16)[name = tensor("aw_chunk_939_cast")]; + tensor var_14173_equation_0 = const()[name = tensor("op_14173_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14173_cast = einsum(equation = var_14173_equation_0, values = (var_14027_cast, var_14001_cast))[name = tensor("op_14173_cast")]; + tensor var_14174_to_fp16 = const()[name = tensor("op_14174_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_941_cast = mul(x = var_14173_cast, y = var_14174_to_fp16)[name = tensor("aw_chunk_941_cast")]; + tensor var_14177_equation_0 = const()[name = tensor("op_14177_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14177_cast = einsum(equation = var_14177_equation_0, values = (var_14027_cast, var_14002_cast))[name = tensor("op_14177_cast")]; + tensor var_14178_to_fp16 = const()[name = tensor("op_14178_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_943_cast = mul(x = var_14177_cast, y = var_14178_to_fp16)[name = tensor("aw_chunk_943_cast")]; + tensor var_14181_equation_0 = const()[name = tensor("op_14181_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14181_cast = einsum(equation = var_14181_equation_0, values = (var_14031_cast, var_14003_cast))[name = tensor("op_14181_cast")]; + tensor var_14182_to_fp16 = const()[name = tensor("op_14182_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_945_cast = mul(x = var_14181_cast, y = var_14182_to_fp16)[name = tensor("aw_chunk_945_cast")]; + tensor var_14185_equation_0 = const()[name = tensor("op_14185_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14185_cast = einsum(equation = var_14185_equation_0, values = (var_14031_cast, var_14004_cast))[name = tensor("op_14185_cast")]; + tensor var_14186_to_fp16 = const()[name = tensor("op_14186_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_947_cast = mul(x = var_14185_cast, y = var_14186_to_fp16)[name = tensor("aw_chunk_947_cast")]; + tensor var_14189_equation_0 = const()[name = tensor("op_14189_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14189_cast = einsum(equation = var_14189_equation_0, values = (var_14031_cast, var_14005_cast))[name = tensor("op_14189_cast")]; + tensor var_14190_to_fp16 = const()[name = tensor("op_14190_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_949_cast = mul(x = var_14189_cast, y = var_14190_to_fp16)[name = tensor("aw_chunk_949_cast")]; + tensor var_14193_equation_0 = const()[name = tensor("op_14193_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14193_cast = einsum(equation = var_14193_equation_0, values = (var_14031_cast, var_14006_cast))[name = tensor("op_14193_cast")]; + tensor var_14194_to_fp16 = const()[name = tensor("op_14194_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_951_cast = mul(x = var_14193_cast, y = var_14194_to_fp16)[name = tensor("aw_chunk_951_cast")]; + tensor var_14197_equation_0 = const()[name = tensor("op_14197_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14197_cast = einsum(equation = var_14197_equation_0, values = (var_14031_cast, var_14007_cast))[name = tensor("op_14197_cast")]; + tensor var_14198_to_fp16 = const()[name = tensor("op_14198_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_953_cast = mul(x = var_14197_cast, y = var_14198_to_fp16)[name = tensor("aw_chunk_953_cast")]; + tensor var_14201_equation_0 = const()[name = tensor("op_14201_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14201_cast = einsum(equation = var_14201_equation_0, values = (var_14031_cast, var_14008_cast))[name = tensor("op_14201_cast")]; + tensor var_14202_to_fp16 = const()[name = tensor("op_14202_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_955_cast = mul(x = var_14201_cast, y = var_14202_to_fp16)[name = tensor("aw_chunk_955_cast")]; + tensor var_14205_equation_0 = const()[name = tensor("op_14205_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14205_cast = einsum(equation = var_14205_equation_0, values = (var_14031_cast, var_14009_cast))[name = tensor("op_14205_cast")]; + tensor var_14206_to_fp16 = const()[name = tensor("op_14206_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_957_cast = mul(x = var_14205_cast, y = var_14206_to_fp16)[name = tensor("aw_chunk_957_cast")]; + tensor var_14209_equation_0 = const()[name = tensor("op_14209_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14209_cast = einsum(equation = var_14209_equation_0, values = (var_14031_cast, var_14010_cast))[name = tensor("op_14209_cast")]; + tensor var_14210_to_fp16 = const()[name = tensor("op_14210_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_959_cast = mul(x = var_14209_cast, y = var_14210_to_fp16)[name = tensor("aw_chunk_959_cast")]; + tensor var_14212_cast = softmax(axis = var_12807, x = aw_chunk_881_cast)[name = tensor("op_14212_cast")]; + tensor var_14213_cast = softmax(axis = var_12807, x = aw_chunk_883_cast)[name = tensor("op_14213_cast")]; + tensor var_14214_cast = softmax(axis = var_12807, x = aw_chunk_885_cast)[name = tensor("op_14214_cast")]; + tensor var_14215_cast = softmax(axis = var_12807, x = aw_chunk_887_cast)[name = tensor("op_14215_cast")]; + tensor var_14216_cast = softmax(axis = var_12807, x = aw_chunk_889_cast)[name = tensor("op_14216_cast")]; + tensor var_14217_cast = softmax(axis = var_12807, x = aw_chunk_891_cast)[name = tensor("op_14217_cast")]; + tensor var_14218_cast = softmax(axis = var_12807, x = aw_chunk_893_cast)[name = tensor("op_14218_cast")]; + tensor var_14219_cast = softmax(axis = var_12807, x = aw_chunk_895_cast)[name = tensor("op_14219_cast")]; + tensor var_14220_cast = softmax(axis = var_12807, x = aw_chunk_897_cast)[name = tensor("op_14220_cast")]; + tensor var_14221_cast = softmax(axis = var_12807, x = aw_chunk_899_cast)[name = tensor("op_14221_cast")]; + tensor var_14222_cast = softmax(axis = var_12807, x = aw_chunk_901_cast)[name = tensor("op_14222_cast")]; + tensor var_14223_cast = softmax(axis = var_12807, x = aw_chunk_903_cast)[name = tensor("op_14223_cast")]; + tensor var_14224_cast = softmax(axis = var_12807, x = aw_chunk_905_cast)[name = tensor("op_14224_cast")]; + tensor var_14225_cast = softmax(axis = var_12807, x = aw_chunk_907_cast)[name = tensor("op_14225_cast")]; + tensor var_14226_cast = softmax(axis = var_12807, x = aw_chunk_909_cast)[name = tensor("op_14226_cast")]; + tensor var_14227_cast = softmax(axis = var_12807, x = aw_chunk_911_cast)[name = tensor("op_14227_cast")]; + tensor var_14228_cast = softmax(axis = var_12807, x = aw_chunk_913_cast)[name = tensor("op_14228_cast")]; + tensor var_14229_cast = softmax(axis = var_12807, x = aw_chunk_915_cast)[name = tensor("op_14229_cast")]; + tensor var_14230_cast = softmax(axis = var_12807, x = aw_chunk_917_cast)[name = tensor("op_14230_cast")]; + tensor var_14231_cast = softmax(axis = var_12807, x = aw_chunk_919_cast)[name = tensor("op_14231_cast")]; + tensor var_14232_cast = softmax(axis = var_12807, x = aw_chunk_921_cast)[name = tensor("op_14232_cast")]; + tensor var_14233_cast = softmax(axis = var_12807, x = aw_chunk_923_cast)[name = tensor("op_14233_cast")]; + tensor var_14234_cast = softmax(axis = var_12807, x = aw_chunk_925_cast)[name = tensor("op_14234_cast")]; + tensor var_14235_cast = softmax(axis = var_12807, x = aw_chunk_927_cast)[name = tensor("op_14235_cast")]; + tensor var_14236_cast = softmax(axis = var_12807, x = aw_chunk_929_cast)[name = tensor("op_14236_cast")]; + tensor var_14237_cast = softmax(axis = var_12807, x = aw_chunk_931_cast)[name = tensor("op_14237_cast")]; + tensor var_14238_cast = softmax(axis = var_12807, x = aw_chunk_933_cast)[name = tensor("op_14238_cast")]; + tensor var_14239_cast = softmax(axis = var_12807, x = aw_chunk_935_cast)[name = tensor("op_14239_cast")]; + tensor var_14240_cast = softmax(axis = var_12807, x = aw_chunk_937_cast)[name = tensor("op_14240_cast")]; + tensor var_14241_cast = softmax(axis = var_12807, x = aw_chunk_939_cast)[name = tensor("op_14241_cast")]; + tensor var_14242_cast = softmax(axis = var_12807, x = aw_chunk_941_cast)[name = tensor("op_14242_cast")]; + tensor var_14243_cast = softmax(axis = var_12807, x = aw_chunk_943_cast)[name = tensor("op_14243_cast")]; + tensor var_14244_cast = softmax(axis = var_12807, x = aw_chunk_945_cast)[name = tensor("op_14244_cast")]; + tensor var_14245_cast = softmax(axis = var_12807, x = aw_chunk_947_cast)[name = tensor("op_14245_cast")]; + tensor var_14246_cast = softmax(axis = var_12807, x = aw_chunk_949_cast)[name = tensor("op_14246_cast")]; + tensor var_14247_cast = softmax(axis = var_12807, x = aw_chunk_951_cast)[name = tensor("op_14247_cast")]; + tensor var_14248_cast = softmax(axis = var_12807, x = aw_chunk_953_cast)[name = tensor("op_14248_cast")]; + tensor var_14249_cast = softmax(axis = var_12807, x = aw_chunk_955_cast)[name = tensor("op_14249_cast")]; + tensor var_14250_cast = softmax(axis = var_12807, x = aw_chunk_957_cast)[name = tensor("op_14250_cast")]; + tensor var_14251_cast = softmax(axis = var_12807, x = aw_chunk_959_cast)[name = tensor("op_14251_cast")]; + tensor var_14253_equation_0 = const()[name = tensor("op_14253_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14253_cast = einsum(equation = var_14253_equation_0, values = (var_14033_cast, var_14212_cast))[name = tensor("op_14253_cast")]; + tensor var_14255_equation_0 = const()[name = tensor("op_14255_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14255_cast = einsum(equation = var_14255_equation_0, values = (var_14033_cast, var_14213_cast))[name = tensor("op_14255_cast")]; + tensor var_14257_equation_0 = const()[name = tensor("op_14257_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14257_cast = einsum(equation = var_14257_equation_0, values = (var_14033_cast, var_14214_cast))[name = tensor("op_14257_cast")]; + tensor var_14259_equation_0 = const()[name = tensor("op_14259_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14259_cast = einsum(equation = var_14259_equation_0, values = (var_14033_cast, var_14215_cast))[name = tensor("op_14259_cast")]; + tensor var_14261_equation_0 = const()[name = tensor("op_14261_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14261_cast = einsum(equation = var_14261_equation_0, values = (var_14033_cast, var_14216_cast))[name = tensor("op_14261_cast")]; + tensor var_14263_equation_0 = const()[name = tensor("op_14263_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14263_cast = einsum(equation = var_14263_equation_0, values = (var_14033_cast, var_14217_cast))[name = tensor("op_14263_cast")]; + tensor var_14265_equation_0 = const()[name = tensor("op_14265_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14265_cast = einsum(equation = var_14265_equation_0, values = (var_14033_cast, var_14218_cast))[name = tensor("op_14265_cast")]; + tensor var_14267_equation_0 = const()[name = tensor("op_14267_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14267_cast = einsum(equation = var_14267_equation_0, values = (var_14033_cast, var_14219_cast))[name = tensor("op_14267_cast")]; + tensor var_14269_equation_0 = const()[name = tensor("op_14269_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14269_cast = einsum(equation = var_14269_equation_0, values = (var_14037_cast, var_14220_cast))[name = tensor("op_14269_cast")]; + tensor var_14271_equation_0 = const()[name = tensor("op_14271_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14271_cast = einsum(equation = var_14271_equation_0, values = (var_14037_cast, var_14221_cast))[name = tensor("op_14271_cast")]; + tensor var_14273_equation_0 = const()[name = tensor("op_14273_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14273_cast = einsum(equation = var_14273_equation_0, values = (var_14037_cast, var_14222_cast))[name = tensor("op_14273_cast")]; + tensor var_14275_equation_0 = const()[name = tensor("op_14275_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14275_cast = einsum(equation = var_14275_equation_0, values = (var_14037_cast, var_14223_cast))[name = tensor("op_14275_cast")]; + tensor var_14277_equation_0 = const()[name = tensor("op_14277_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14277_cast = einsum(equation = var_14277_equation_0, values = (var_14037_cast, var_14224_cast))[name = tensor("op_14277_cast")]; + tensor var_14279_equation_0 = const()[name = tensor("op_14279_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14279_cast = einsum(equation = var_14279_equation_0, values = (var_14037_cast, var_14225_cast))[name = tensor("op_14279_cast")]; + tensor var_14281_equation_0 = const()[name = tensor("op_14281_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14281_cast = einsum(equation = var_14281_equation_0, values = (var_14037_cast, var_14226_cast))[name = tensor("op_14281_cast")]; + tensor var_14283_equation_0 = const()[name = tensor("op_14283_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14283_cast = einsum(equation = var_14283_equation_0, values = (var_14037_cast, var_14227_cast))[name = tensor("op_14283_cast")]; + tensor var_14285_equation_0 = const()[name = tensor("op_14285_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14285_cast = einsum(equation = var_14285_equation_0, values = (var_14041_cast, var_14228_cast))[name = tensor("op_14285_cast")]; + tensor var_14287_equation_0 = const()[name = tensor("op_14287_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14287_cast = einsum(equation = var_14287_equation_0, values = (var_14041_cast, var_14229_cast))[name = tensor("op_14287_cast")]; + tensor var_14289_equation_0 = const()[name = tensor("op_14289_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14289_cast = einsum(equation = var_14289_equation_0, values = (var_14041_cast, var_14230_cast))[name = tensor("op_14289_cast")]; + tensor var_14291_equation_0 = const()[name = tensor("op_14291_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14291_cast = einsum(equation = var_14291_equation_0, values = (var_14041_cast, var_14231_cast))[name = tensor("op_14291_cast")]; + tensor var_14293_equation_0 = const()[name = tensor("op_14293_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14293_cast = einsum(equation = var_14293_equation_0, values = (var_14041_cast, var_14232_cast))[name = tensor("op_14293_cast")]; + tensor var_14295_equation_0 = const()[name = tensor("op_14295_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14295_cast = einsum(equation = var_14295_equation_0, values = (var_14041_cast, var_14233_cast))[name = tensor("op_14295_cast")]; + tensor var_14297_equation_0 = const()[name = tensor("op_14297_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14297_cast = einsum(equation = var_14297_equation_0, values = (var_14041_cast, var_14234_cast))[name = tensor("op_14297_cast")]; + tensor var_14299_equation_0 = const()[name = tensor("op_14299_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14299_cast = einsum(equation = var_14299_equation_0, values = (var_14041_cast, var_14235_cast))[name = tensor("op_14299_cast")]; + tensor var_14301_equation_0 = const()[name = tensor("op_14301_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14301_cast = einsum(equation = var_14301_equation_0, values = (var_14045_cast, var_14236_cast))[name = tensor("op_14301_cast")]; + tensor var_14303_equation_0 = const()[name = tensor("op_14303_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14303_cast = einsum(equation = var_14303_equation_0, values = (var_14045_cast, var_14237_cast))[name = tensor("op_14303_cast")]; + tensor var_14305_equation_0 = const()[name = tensor("op_14305_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14305_cast = einsum(equation = var_14305_equation_0, values = (var_14045_cast, var_14238_cast))[name = tensor("op_14305_cast")]; + tensor var_14307_equation_0 = const()[name = tensor("op_14307_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14307_cast = einsum(equation = var_14307_equation_0, values = (var_14045_cast, var_14239_cast))[name = tensor("op_14307_cast")]; + tensor var_14309_equation_0 = const()[name = tensor("op_14309_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14309_cast = einsum(equation = var_14309_equation_0, values = (var_14045_cast, var_14240_cast))[name = tensor("op_14309_cast")]; + tensor var_14311_equation_0 = const()[name = tensor("op_14311_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14311_cast = einsum(equation = var_14311_equation_0, values = (var_14045_cast, var_14241_cast))[name = tensor("op_14311_cast")]; + tensor var_14313_equation_0 = const()[name = tensor("op_14313_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14313_cast = einsum(equation = var_14313_equation_0, values = (var_14045_cast, var_14242_cast))[name = tensor("op_14313_cast")]; + tensor var_14315_equation_0 = const()[name = tensor("op_14315_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14315_cast = einsum(equation = var_14315_equation_0, values = (var_14045_cast, var_14243_cast))[name = tensor("op_14315_cast")]; + tensor var_14317_equation_0 = const()[name = tensor("op_14317_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14317_cast = einsum(equation = var_14317_equation_0, values = (var_14049_cast, var_14244_cast))[name = tensor("op_14317_cast")]; + tensor var_14319_equation_0 = const()[name = tensor("op_14319_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14319_cast = einsum(equation = var_14319_equation_0, values = (var_14049_cast, var_14245_cast))[name = tensor("op_14319_cast")]; + tensor var_14321_equation_0 = const()[name = tensor("op_14321_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14321_cast = einsum(equation = var_14321_equation_0, values = (var_14049_cast, var_14246_cast))[name = tensor("op_14321_cast")]; + tensor var_14323_equation_0 = const()[name = tensor("op_14323_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14323_cast = einsum(equation = var_14323_equation_0, values = (var_14049_cast, var_14247_cast))[name = tensor("op_14323_cast")]; + tensor var_14325_equation_0 = const()[name = tensor("op_14325_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14325_cast = einsum(equation = var_14325_equation_0, values = (var_14049_cast, var_14248_cast))[name = tensor("op_14325_cast")]; + tensor var_14327_equation_0 = const()[name = tensor("op_14327_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14327_cast = einsum(equation = var_14327_equation_0, values = (var_14049_cast, var_14249_cast))[name = tensor("op_14327_cast")]; + tensor var_14329_equation_0 = const()[name = tensor("op_14329_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14329_cast = einsum(equation = var_14329_equation_0, values = (var_14049_cast, var_14250_cast))[name = tensor("op_14329_cast")]; + tensor var_14331_equation_0 = const()[name = tensor("op_14331_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14331_cast = einsum(equation = var_14331_equation_0, values = (var_14049_cast, var_14251_cast))[name = tensor("op_14331_cast")]; + tensor var_14333_interleave_0 = const()[name = tensor("op_14333_interleave_0"), val = tensor(false)]; + tensor var_14333_cast = concat(axis = var_12782, interleave = var_14333_interleave_0, values = (var_14253_cast, var_14255_cast, var_14257_cast, var_14259_cast, var_14261_cast, var_14263_cast, var_14265_cast, var_14267_cast))[name = tensor("op_14333_cast")]; + tensor var_14335_interleave_0 = const()[name = tensor("op_14335_interleave_0"), val = tensor(false)]; + tensor var_14335_cast = concat(axis = var_12782, interleave = var_14335_interleave_0, values = (var_14269_cast, var_14271_cast, var_14273_cast, var_14275_cast, var_14277_cast, var_14279_cast, var_14281_cast, var_14283_cast))[name = tensor("op_14335_cast")]; + tensor var_14337_interleave_0 = const()[name = tensor("op_14337_interleave_0"), val = tensor(false)]; + tensor var_14337_cast = concat(axis = var_12782, interleave = var_14337_interleave_0, values = (var_14285_cast, var_14287_cast, var_14289_cast, var_14291_cast, var_14293_cast, var_14295_cast, var_14297_cast, var_14299_cast))[name = tensor("op_14337_cast")]; + tensor var_14339_interleave_0 = const()[name = tensor("op_14339_interleave_0"), val = tensor(false)]; + tensor var_14339_cast = concat(axis = var_12782, interleave = var_14339_interleave_0, values = (var_14301_cast, var_14303_cast, var_14305_cast, var_14307_cast, var_14309_cast, var_14311_cast, var_14313_cast, var_14315_cast))[name = tensor("op_14339_cast")]; + tensor var_14341_interleave_0 = const()[name = tensor("op_14341_interleave_0"), val = tensor(false)]; + tensor var_14341_cast = concat(axis = var_12782, interleave = var_14341_interleave_0, values = (var_14317_cast, var_14319_cast, var_14321_cast, var_14323_cast, var_14325_cast, var_14327_cast, var_14329_cast, var_14331_cast))[name = tensor("op_14341_cast")]; + tensor input_491_interleave_0 = const()[name = tensor("input_491_interleave_0"), val = tensor(false)]; + tensor input_491_cast = concat(axis = var_12807, interleave = input_491_interleave_0, values = (var_14333_cast, var_14335_cast, var_14337_cast, var_14339_cast, var_14341_cast))[name = tensor("input_491_cast")]; + tensor var_14347 = const()[name = tensor("op_14347"), val = tensor([1, 1])]; + tensor var_14349 = const()[name = tensor("op_14349"), val = tensor([1, 1])]; + tensor var_14351_pad_type_0 = const()[name = tensor("op_14351_pad_type_0"), val = tensor("custom")]; + tensor var_14351_pad_0 = const()[name = tensor("op_14351_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_3_attentions_1_transformer_blocks_0_attn1_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(643635200))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(643712064))), name = tensor("up_blocks_3_attentions_1_transformer_blocks_0_attn1_to_out_0_weight_to_fp16_palettized"), shape = tensor([320, 320, 1, 1])]; + tensor up_blocks_3_attentions_1_transformer_blocks_0_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("up_blocks_3_attentions_1_transformer_blocks_0_attn1_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(643712256)))]; + tensor var_14351_cast = conv(bias = up_blocks_3_attentions_1_transformer_blocks_0_attn1_to_out_0_bias_to_fp16, dilations = var_14349, groups = var_12807, pad = var_14351_pad_0, pad_type = var_14351_pad_type_0, strides = var_14347, weight = up_blocks_3_attentions_1_transformer_blocks_0_attn1_to_out_0_weight_to_fp16_palettized, x = input_491_cast)[name = tensor("op_14351_cast")]; + tensor inputs_87_cast = add(x = var_14351_cast, y = inputs_85_cast)[name = tensor("inputs_87_cast")]; + tensor var_14355 = const()[name = tensor("op_14355"), val = tensor([1])]; + tensor channels_mean_87_cast = reduce_mean(axes = var_14355, keep_dims = var_12802, x = inputs_87_cast)[name = tensor("channels_mean_87_cast")]; + tensor zero_mean_87_cast = sub(x = inputs_87_cast, y = channels_mean_87_cast)[name = tensor("zero_mean_87_cast")]; + tensor zero_mean_sq_87_cast = mul(x = zero_mean_87_cast, y = zero_mean_87_cast)[name = tensor("zero_mean_sq_87_cast")]; + tensor var_14359 = const()[name = tensor("op_14359"), val = tensor([1])]; + tensor var_14360_cast = reduce_mean(axes = var_14359, keep_dims = var_12802, x = zero_mean_sq_87_cast)[name = tensor("op_14360_cast")]; + tensor var_14361_to_fp16 = const()[name = tensor("op_14361_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_14362_cast = add(x = var_14360_cast, y = var_14361_to_fp16)[name = tensor("op_14362_cast")]; + tensor denom_87_epsilon_0_to_fp16 = const()[name = tensor("denom_87_epsilon_0_to_fp16"), val = tensor(0x1p-24)]; + tensor denom_87_cast = rsqrt(epsilon = denom_87_epsilon_0_to_fp16, x = var_14362_cast)[name = tensor("denom_87_cast")]; + tensor out_87_cast = mul(x = zero_mean_87_cast, y = denom_87_cast)[name = tensor("out_87_cast")]; + tensor var_14366_to_fp16 = const()[name = tensor("op_14366_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(643712960)))]; + tensor var_14367_cast = add(x = out_87_cast, y = var_14366_to_fp16)[name = tensor("op_14367_cast")]; + tensor var_14369_to_fp16 = const()[name = tensor("op_14369_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(643713664)))]; + tensor hidden_states_309_cast = mul(x = var_14367_cast, y = var_14369_to_fp16)[name = tensor("hidden_states_309_cast")]; + tensor var_14376 = const()[name = tensor("op_14376"), val = tensor([1, 1])]; + tensor var_14378 = const()[name = tensor("op_14378"), val = tensor([1, 1])]; + tensor q_59_pad_type_0 = const()[name = tensor("q_59_pad_type_0"), val = tensor("custom")]; + tensor q_59_pad_0 = const()[name = tensor("q_59_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_3_attentions_1_transformer_blocks_0_attn2_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(643714368))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(643791232))), name = tensor("up_blocks_3_attentions_1_transformer_blocks_0_attn2_to_q_weight_to_fp16_palettized"), shape = tensor([320, 320, 1, 1])]; + tensor q_59_cast = conv(dilations = var_14378, groups = var_12807, pad = q_59_pad_0, pad_type = q_59_pad_type_0, strides = var_14376, weight = up_blocks_3_attentions_1_transformer_blocks_0_attn2_to_q_weight_to_fp16_palettized, x = hidden_states_309_cast)[name = tensor("q_59_cast")]; + tensor var_14382 = const()[name = tensor("op_14382"), val = tensor([1, 1])]; + tensor var_14384 = const()[name = tensor("op_14384"), val = tensor([1, 1])]; + tensor k_117_pad_type_0 = const()[name = tensor("k_117_pad_type_0"), val = tensor("custom")]; + tensor k_117_pad_0 = const()[name = tensor("k_117_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_3_attentions_1_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(643791424))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(644037248))), name = tensor("up_blocks_3_attentions_1_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized"), shape = tensor([320, 1024, 1, 1])]; + tensor k_117_cast = conv(dilations = var_14384, groups = var_12807, pad = k_117_pad_0, pad_type = k_117_pad_type_0, strides = var_14382, weight = up_blocks_3_attentions_1_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor("k_117_cast")]; + tensor var_14388 = const()[name = tensor("op_14388"), val = tensor([1, 1])]; + tensor var_14390 = const()[name = tensor("op_14390"), val = tensor([1, 1])]; + tensor v_59_pad_type_0 = const()[name = tensor("v_59_pad_type_0"), val = tensor("custom")]; + tensor v_59_pad_0 = const()[name = tensor("v_59_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_3_attentions_1_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(644037440))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(644283264))), name = tensor("up_blocks_3_attentions_1_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized"), shape = tensor([320, 1024, 1, 1])]; + tensor v_59_cast = conv(dilations = var_14390, groups = var_12807, pad = v_59_pad_0, pad_type = v_59_pad_type_0, strides = var_14388, weight = up_blocks_3_attentions_1_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor("v_59_cast")]; + tensor var_14394_begin_0 = const()[name = tensor("op_14394_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_14394_end_0 = const()[name = tensor("op_14394_end_0"), val = tensor([2, 64, 1, 4096])]; + tensor var_14394_end_mask_0 = const()[name = tensor("op_14394_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14394_cast = slice_by_index(begin = var_14394_begin_0, end = var_14394_end_0, end_mask = var_14394_end_mask_0, x = q_59_cast)[name = tensor("op_14394_cast")]; + tensor var_14398_begin_0 = const()[name = tensor("op_14398_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_14398_end_0 = const()[name = tensor("op_14398_end_0"), val = tensor([2, 128, 1, 4096])]; + tensor var_14398_end_mask_0 = const()[name = tensor("op_14398_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14398_cast = slice_by_index(begin = var_14398_begin_0, end = var_14398_end_0, end_mask = var_14398_end_mask_0, x = q_59_cast)[name = tensor("op_14398_cast")]; + tensor var_14402_begin_0 = const()[name = tensor("op_14402_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_14402_end_0 = const()[name = tensor("op_14402_end_0"), val = tensor([2, 192, 1, 4096])]; + tensor var_14402_end_mask_0 = const()[name = tensor("op_14402_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14402_cast = slice_by_index(begin = var_14402_begin_0, end = var_14402_end_0, end_mask = var_14402_end_mask_0, x = q_59_cast)[name = tensor("op_14402_cast")]; + tensor var_14406_begin_0 = const()[name = tensor("op_14406_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_14406_end_0 = const()[name = tensor("op_14406_end_0"), val = tensor([2, 256, 1, 4096])]; + tensor var_14406_end_mask_0 = const()[name = tensor("op_14406_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14406_cast = slice_by_index(begin = var_14406_begin_0, end = var_14406_end_0, end_mask = var_14406_end_mask_0, x = q_59_cast)[name = tensor("op_14406_cast")]; + tensor var_14410_begin_0 = const()[name = tensor("op_14410_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_14410_end_0 = const()[name = tensor("op_14410_end_0"), val = tensor([2, 320, 1, 4096])]; + tensor var_14410_end_mask_0 = const()[name = tensor("op_14410_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14410_cast = slice_by_index(begin = var_14410_begin_0, end = var_14410_end_0, end_mask = var_14410_end_mask_0, x = q_59_cast)[name = tensor("op_14410_cast")]; + tensor var_14413_begin_0 = const()[name = tensor("op_14413_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_14413_end_0 = const()[name = tensor("op_14413_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_14413_end_mask_0 = const()[name = tensor("op_14413_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14413_cast = slice_by_index(begin = var_14413_begin_0, end = var_14413_end_0, end_mask = var_14413_end_mask_0, x = var_14394_cast)[name = tensor("op_14413_cast")]; + tensor var_14414_begin_0 = const()[name = tensor("op_14414_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_14414_end_0 = const()[name = tensor("op_14414_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_14414_end_mask_0 = const()[name = tensor("op_14414_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14414_cast = slice_by_index(begin = var_14414_begin_0, end = var_14414_end_0, end_mask = var_14414_end_mask_0, x = var_14394_cast)[name = tensor("op_14414_cast")]; + tensor var_14415_begin_0 = const()[name = tensor("op_14415_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_14415_end_0 = const()[name = tensor("op_14415_end_0"), val = tensor([2, 64, 1, 1536])]; + tensor var_14415_end_mask_0 = const()[name = tensor("op_14415_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14415_cast = slice_by_index(begin = var_14415_begin_0, end = var_14415_end_0, end_mask = var_14415_end_mask_0, x = var_14394_cast)[name = tensor("op_14415_cast")]; + tensor var_14416_begin_0 = const()[name = tensor("op_14416_begin_0"), val = tensor([0, 0, 0, 1536])]; + tensor var_14416_end_0 = const()[name = tensor("op_14416_end_0"), val = tensor([2, 64, 1, 2048])]; + tensor var_14416_end_mask_0 = const()[name = tensor("op_14416_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14416_cast = slice_by_index(begin = var_14416_begin_0, end = var_14416_end_0, end_mask = var_14416_end_mask_0, x = var_14394_cast)[name = tensor("op_14416_cast")]; + tensor var_14417_begin_0 = const()[name = tensor("op_14417_begin_0"), val = tensor([0, 0, 0, 2048])]; + tensor var_14417_end_0 = const()[name = tensor("op_14417_end_0"), val = tensor([2, 64, 1, 2560])]; + tensor var_14417_end_mask_0 = const()[name = tensor("op_14417_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14417_cast = slice_by_index(begin = var_14417_begin_0, end = var_14417_end_0, end_mask = var_14417_end_mask_0, x = var_14394_cast)[name = tensor("op_14417_cast")]; + tensor var_14418_begin_0 = const()[name = tensor("op_14418_begin_0"), val = tensor([0, 0, 0, 2560])]; + tensor var_14418_end_0 = const()[name = tensor("op_14418_end_0"), val = tensor([2, 64, 1, 3072])]; + tensor var_14418_end_mask_0 = const()[name = tensor("op_14418_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14418_cast = slice_by_index(begin = var_14418_begin_0, end = var_14418_end_0, end_mask = var_14418_end_mask_0, x = var_14394_cast)[name = tensor("op_14418_cast")]; + tensor var_14419_begin_0 = const()[name = tensor("op_14419_begin_0"), val = tensor([0, 0, 0, 3072])]; + tensor var_14419_end_0 = const()[name = tensor("op_14419_end_0"), val = tensor([2, 64, 1, 3584])]; + tensor var_14419_end_mask_0 = const()[name = tensor("op_14419_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14419_cast = slice_by_index(begin = var_14419_begin_0, end = var_14419_end_0, end_mask = var_14419_end_mask_0, x = var_14394_cast)[name = tensor("op_14419_cast")]; + tensor var_14420_begin_0 = const()[name = tensor("op_14420_begin_0"), val = tensor([0, 0, 0, 3584])]; + tensor var_14420_end_0 = const()[name = tensor("op_14420_end_0"), val = tensor([2, 64, 1, 4096])]; + tensor var_14420_end_mask_0 = const()[name = tensor("op_14420_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14420_cast = slice_by_index(begin = var_14420_begin_0, end = var_14420_end_0, end_mask = var_14420_end_mask_0, x = var_14394_cast)[name = tensor("op_14420_cast")]; + tensor var_14421_begin_0 = const()[name = tensor("op_14421_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_14421_end_0 = const()[name = tensor("op_14421_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_14421_end_mask_0 = const()[name = tensor("op_14421_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14421_cast = slice_by_index(begin = var_14421_begin_0, end = var_14421_end_0, end_mask = var_14421_end_mask_0, x = var_14398_cast)[name = tensor("op_14421_cast")]; + tensor var_14422_begin_0 = const()[name = tensor("op_14422_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_14422_end_0 = const()[name = tensor("op_14422_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_14422_end_mask_0 = const()[name = tensor("op_14422_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14422_cast = slice_by_index(begin = var_14422_begin_0, end = var_14422_end_0, end_mask = var_14422_end_mask_0, x = var_14398_cast)[name = tensor("op_14422_cast")]; + tensor var_14423_begin_0 = const()[name = tensor("op_14423_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_14423_end_0 = const()[name = tensor("op_14423_end_0"), val = tensor([2, 64, 1, 1536])]; + tensor var_14423_end_mask_0 = const()[name = tensor("op_14423_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14423_cast = slice_by_index(begin = var_14423_begin_0, end = var_14423_end_0, end_mask = var_14423_end_mask_0, x = var_14398_cast)[name = tensor("op_14423_cast")]; + tensor var_14424_begin_0 = const()[name = tensor("op_14424_begin_0"), val = tensor([0, 0, 0, 1536])]; + tensor var_14424_end_0 = const()[name = tensor("op_14424_end_0"), val = tensor([2, 64, 1, 2048])]; + tensor var_14424_end_mask_0 = const()[name = tensor("op_14424_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14424_cast = slice_by_index(begin = var_14424_begin_0, end = var_14424_end_0, end_mask = var_14424_end_mask_0, x = var_14398_cast)[name = tensor("op_14424_cast")]; + tensor var_14425_begin_0 = const()[name = tensor("op_14425_begin_0"), val = tensor([0, 0, 0, 2048])]; + tensor var_14425_end_0 = const()[name = tensor("op_14425_end_0"), val = tensor([2, 64, 1, 2560])]; + tensor var_14425_end_mask_0 = const()[name = tensor("op_14425_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14425_cast = slice_by_index(begin = var_14425_begin_0, end = var_14425_end_0, end_mask = var_14425_end_mask_0, x = var_14398_cast)[name = tensor("op_14425_cast")]; + tensor var_14426_begin_0 = const()[name = tensor("op_14426_begin_0"), val = tensor([0, 0, 0, 2560])]; + tensor var_14426_end_0 = const()[name = tensor("op_14426_end_0"), val = tensor([2, 64, 1, 3072])]; + tensor var_14426_end_mask_0 = const()[name = tensor("op_14426_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14426_cast = slice_by_index(begin = var_14426_begin_0, end = var_14426_end_0, end_mask = var_14426_end_mask_0, x = var_14398_cast)[name = tensor("op_14426_cast")]; + tensor var_14427_begin_0 = const()[name = tensor("op_14427_begin_0"), val = tensor([0, 0, 0, 3072])]; + tensor var_14427_end_0 = const()[name = tensor("op_14427_end_0"), val = tensor([2, 64, 1, 3584])]; + tensor var_14427_end_mask_0 = const()[name = tensor("op_14427_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14427_cast = slice_by_index(begin = var_14427_begin_0, end = var_14427_end_0, end_mask = var_14427_end_mask_0, x = var_14398_cast)[name = tensor("op_14427_cast")]; + tensor var_14428_begin_0 = const()[name = tensor("op_14428_begin_0"), val = tensor([0, 0, 0, 3584])]; + tensor var_14428_end_0 = const()[name = tensor("op_14428_end_0"), val = tensor([2, 64, 1, 4096])]; + tensor var_14428_end_mask_0 = const()[name = tensor("op_14428_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14428_cast = slice_by_index(begin = var_14428_begin_0, end = var_14428_end_0, end_mask = var_14428_end_mask_0, x = var_14398_cast)[name = tensor("op_14428_cast")]; + tensor var_14429_begin_0 = const()[name = tensor("op_14429_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_14429_end_0 = const()[name = tensor("op_14429_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_14429_end_mask_0 = const()[name = tensor("op_14429_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14429_cast = slice_by_index(begin = var_14429_begin_0, end = var_14429_end_0, end_mask = var_14429_end_mask_0, x = var_14402_cast)[name = tensor("op_14429_cast")]; + tensor var_14430_begin_0 = const()[name = tensor("op_14430_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_14430_end_0 = const()[name = tensor("op_14430_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_14430_end_mask_0 = const()[name = tensor("op_14430_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14430_cast = slice_by_index(begin = var_14430_begin_0, end = var_14430_end_0, end_mask = var_14430_end_mask_0, x = var_14402_cast)[name = tensor("op_14430_cast")]; + tensor var_14431_begin_0 = const()[name = tensor("op_14431_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_14431_end_0 = const()[name = tensor("op_14431_end_0"), val = tensor([2, 64, 1, 1536])]; + tensor var_14431_end_mask_0 = const()[name = tensor("op_14431_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14431_cast = slice_by_index(begin = var_14431_begin_0, end = var_14431_end_0, end_mask = var_14431_end_mask_0, x = var_14402_cast)[name = tensor("op_14431_cast")]; + tensor var_14432_begin_0 = const()[name = tensor("op_14432_begin_0"), val = tensor([0, 0, 0, 1536])]; + tensor var_14432_end_0 = const()[name = tensor("op_14432_end_0"), val = tensor([2, 64, 1, 2048])]; + tensor var_14432_end_mask_0 = const()[name = tensor("op_14432_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14432_cast = slice_by_index(begin = var_14432_begin_0, end = var_14432_end_0, end_mask = var_14432_end_mask_0, x = var_14402_cast)[name = tensor("op_14432_cast")]; + tensor var_14433_begin_0 = const()[name = tensor("op_14433_begin_0"), val = tensor([0, 0, 0, 2048])]; + tensor var_14433_end_0 = const()[name = tensor("op_14433_end_0"), val = tensor([2, 64, 1, 2560])]; + tensor var_14433_end_mask_0 = const()[name = tensor("op_14433_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14433_cast = slice_by_index(begin = var_14433_begin_0, end = var_14433_end_0, end_mask = var_14433_end_mask_0, x = var_14402_cast)[name = tensor("op_14433_cast")]; + tensor var_14434_begin_0 = const()[name = tensor("op_14434_begin_0"), val = tensor([0, 0, 0, 2560])]; + tensor var_14434_end_0 = const()[name = tensor("op_14434_end_0"), val = tensor([2, 64, 1, 3072])]; + tensor var_14434_end_mask_0 = const()[name = tensor("op_14434_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14434_cast = slice_by_index(begin = var_14434_begin_0, end = var_14434_end_0, end_mask = var_14434_end_mask_0, x = var_14402_cast)[name = tensor("op_14434_cast")]; + tensor var_14435_begin_0 = const()[name = tensor("op_14435_begin_0"), val = tensor([0, 0, 0, 3072])]; + tensor var_14435_end_0 = const()[name = tensor("op_14435_end_0"), val = tensor([2, 64, 1, 3584])]; + tensor var_14435_end_mask_0 = const()[name = tensor("op_14435_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14435_cast = slice_by_index(begin = var_14435_begin_0, end = var_14435_end_0, end_mask = var_14435_end_mask_0, x = var_14402_cast)[name = tensor("op_14435_cast")]; + tensor var_14436_begin_0 = const()[name = tensor("op_14436_begin_0"), val = tensor([0, 0, 0, 3584])]; + tensor var_14436_end_0 = const()[name = tensor("op_14436_end_0"), val = tensor([2, 64, 1, 4096])]; + tensor var_14436_end_mask_0 = const()[name = tensor("op_14436_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14436_cast = slice_by_index(begin = var_14436_begin_0, end = var_14436_end_0, end_mask = var_14436_end_mask_0, x = var_14402_cast)[name = tensor("op_14436_cast")]; + tensor var_14437_begin_0 = const()[name = tensor("op_14437_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_14437_end_0 = const()[name = tensor("op_14437_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_14437_end_mask_0 = const()[name = tensor("op_14437_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14437_cast = slice_by_index(begin = var_14437_begin_0, end = var_14437_end_0, end_mask = var_14437_end_mask_0, x = var_14406_cast)[name = tensor("op_14437_cast")]; + tensor var_14438_begin_0 = const()[name = tensor("op_14438_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_14438_end_0 = const()[name = tensor("op_14438_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_14438_end_mask_0 = const()[name = tensor("op_14438_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14438_cast = slice_by_index(begin = var_14438_begin_0, end = var_14438_end_0, end_mask = var_14438_end_mask_0, x = var_14406_cast)[name = tensor("op_14438_cast")]; + tensor var_14439_begin_0 = const()[name = tensor("op_14439_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_14439_end_0 = const()[name = tensor("op_14439_end_0"), val = tensor([2, 64, 1, 1536])]; + tensor var_14439_end_mask_0 = const()[name = tensor("op_14439_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14439_cast = slice_by_index(begin = var_14439_begin_0, end = var_14439_end_0, end_mask = var_14439_end_mask_0, x = var_14406_cast)[name = tensor("op_14439_cast")]; + tensor var_14440_begin_0 = const()[name = tensor("op_14440_begin_0"), val = tensor([0, 0, 0, 1536])]; + tensor var_14440_end_0 = const()[name = tensor("op_14440_end_0"), val = tensor([2, 64, 1, 2048])]; + tensor var_14440_end_mask_0 = const()[name = tensor("op_14440_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14440_cast = slice_by_index(begin = var_14440_begin_0, end = var_14440_end_0, end_mask = var_14440_end_mask_0, x = var_14406_cast)[name = tensor("op_14440_cast")]; + tensor var_14441_begin_0 = const()[name = tensor("op_14441_begin_0"), val = tensor([0, 0, 0, 2048])]; + tensor var_14441_end_0 = const()[name = tensor("op_14441_end_0"), val = tensor([2, 64, 1, 2560])]; + tensor var_14441_end_mask_0 = const()[name = tensor("op_14441_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14441_cast = slice_by_index(begin = var_14441_begin_0, end = var_14441_end_0, end_mask = var_14441_end_mask_0, x = var_14406_cast)[name = tensor("op_14441_cast")]; + tensor var_14442_begin_0 = const()[name = tensor("op_14442_begin_0"), val = tensor([0, 0, 0, 2560])]; + tensor var_14442_end_0 = const()[name = tensor("op_14442_end_0"), val = tensor([2, 64, 1, 3072])]; + tensor var_14442_end_mask_0 = const()[name = tensor("op_14442_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14442_cast = slice_by_index(begin = var_14442_begin_0, end = var_14442_end_0, end_mask = var_14442_end_mask_0, x = var_14406_cast)[name = tensor("op_14442_cast")]; + tensor var_14443_begin_0 = const()[name = tensor("op_14443_begin_0"), val = tensor([0, 0, 0, 3072])]; + tensor var_14443_end_0 = const()[name = tensor("op_14443_end_0"), val = tensor([2, 64, 1, 3584])]; + tensor var_14443_end_mask_0 = const()[name = tensor("op_14443_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14443_cast = slice_by_index(begin = var_14443_begin_0, end = var_14443_end_0, end_mask = var_14443_end_mask_0, x = var_14406_cast)[name = tensor("op_14443_cast")]; + tensor var_14444_begin_0 = const()[name = tensor("op_14444_begin_0"), val = tensor([0, 0, 0, 3584])]; + tensor var_14444_end_0 = const()[name = tensor("op_14444_end_0"), val = tensor([2, 64, 1, 4096])]; + tensor var_14444_end_mask_0 = const()[name = tensor("op_14444_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14444_cast = slice_by_index(begin = var_14444_begin_0, end = var_14444_end_0, end_mask = var_14444_end_mask_0, x = var_14406_cast)[name = tensor("op_14444_cast")]; + tensor var_14445_begin_0 = const()[name = tensor("op_14445_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_14445_end_0 = const()[name = tensor("op_14445_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_14445_end_mask_0 = const()[name = tensor("op_14445_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14445_cast = slice_by_index(begin = var_14445_begin_0, end = var_14445_end_0, end_mask = var_14445_end_mask_0, x = var_14410_cast)[name = tensor("op_14445_cast")]; + tensor var_14446_begin_0 = const()[name = tensor("op_14446_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_14446_end_0 = const()[name = tensor("op_14446_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_14446_end_mask_0 = const()[name = tensor("op_14446_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14446_cast = slice_by_index(begin = var_14446_begin_0, end = var_14446_end_0, end_mask = var_14446_end_mask_0, x = var_14410_cast)[name = tensor("op_14446_cast")]; + tensor var_14447_begin_0 = const()[name = tensor("op_14447_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_14447_end_0 = const()[name = tensor("op_14447_end_0"), val = tensor([2, 64, 1, 1536])]; + tensor var_14447_end_mask_0 = const()[name = tensor("op_14447_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14447_cast = slice_by_index(begin = var_14447_begin_0, end = var_14447_end_0, end_mask = var_14447_end_mask_0, x = var_14410_cast)[name = tensor("op_14447_cast")]; + tensor var_14448_begin_0 = const()[name = tensor("op_14448_begin_0"), val = tensor([0, 0, 0, 1536])]; + tensor var_14448_end_0 = const()[name = tensor("op_14448_end_0"), val = tensor([2, 64, 1, 2048])]; + tensor var_14448_end_mask_0 = const()[name = tensor("op_14448_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14448_cast = slice_by_index(begin = var_14448_begin_0, end = var_14448_end_0, end_mask = var_14448_end_mask_0, x = var_14410_cast)[name = tensor("op_14448_cast")]; + tensor var_14449_begin_0 = const()[name = tensor("op_14449_begin_0"), val = tensor([0, 0, 0, 2048])]; + tensor var_14449_end_0 = const()[name = tensor("op_14449_end_0"), val = tensor([2, 64, 1, 2560])]; + tensor var_14449_end_mask_0 = const()[name = tensor("op_14449_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14449_cast = slice_by_index(begin = var_14449_begin_0, end = var_14449_end_0, end_mask = var_14449_end_mask_0, x = var_14410_cast)[name = tensor("op_14449_cast")]; + tensor var_14450_begin_0 = const()[name = tensor("op_14450_begin_0"), val = tensor([0, 0, 0, 2560])]; + tensor var_14450_end_0 = const()[name = tensor("op_14450_end_0"), val = tensor([2, 64, 1, 3072])]; + tensor var_14450_end_mask_0 = const()[name = tensor("op_14450_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14450_cast = slice_by_index(begin = var_14450_begin_0, end = var_14450_end_0, end_mask = var_14450_end_mask_0, x = var_14410_cast)[name = tensor("op_14450_cast")]; + tensor var_14451_begin_0 = const()[name = tensor("op_14451_begin_0"), val = tensor([0, 0, 0, 3072])]; + tensor var_14451_end_0 = const()[name = tensor("op_14451_end_0"), val = tensor([2, 64, 1, 3584])]; + tensor var_14451_end_mask_0 = const()[name = tensor("op_14451_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14451_cast = slice_by_index(begin = var_14451_begin_0, end = var_14451_end_0, end_mask = var_14451_end_mask_0, x = var_14410_cast)[name = tensor("op_14451_cast")]; + tensor var_14452_begin_0 = const()[name = tensor("op_14452_begin_0"), val = tensor([0, 0, 0, 3584])]; + tensor var_14452_end_0 = const()[name = tensor("op_14452_end_0"), val = tensor([2, 64, 1, 4096])]; + tensor var_14452_end_mask_0 = const()[name = tensor("op_14452_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14452_cast = slice_by_index(begin = var_14452_begin_0, end = var_14452_end_0, end_mask = var_14452_end_mask_0, x = var_14410_cast)[name = tensor("op_14452_cast")]; + tensor k_119_perm_0 = const()[name = tensor("k_119_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_14457_begin_0 = const()[name = tensor("op_14457_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_14457_end_0 = const()[name = tensor("op_14457_end_0"), val = tensor([2, 77, 1, 64])]; + tensor var_14457_end_mask_0 = const()[name = tensor("op_14457_end_mask_0"), val = tensor([true, true, true, false])]; + tensor transpose_2 = transpose(perm = k_119_perm_0, x = k_117_cast)[name = tensor("transpose_2")]; + tensor var_14457_cast = slice_by_index(begin = var_14457_begin_0, end = var_14457_end_0, end_mask = var_14457_end_mask_0, x = transpose_2)[name = tensor("op_14457_cast")]; + tensor var_14461_begin_0 = const()[name = tensor("op_14461_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_14461_end_0 = const()[name = tensor("op_14461_end_0"), val = tensor([2, 77, 1, 128])]; + tensor var_14461_end_mask_0 = const()[name = tensor("op_14461_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14461_cast = slice_by_index(begin = var_14461_begin_0, end = var_14461_end_0, end_mask = var_14461_end_mask_0, x = transpose_2)[name = tensor("op_14461_cast")]; + tensor var_14465_begin_0 = const()[name = tensor("op_14465_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_14465_end_0 = const()[name = tensor("op_14465_end_0"), val = tensor([2, 77, 1, 192])]; + tensor var_14465_end_mask_0 = const()[name = tensor("op_14465_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14465_cast = slice_by_index(begin = var_14465_begin_0, end = var_14465_end_0, end_mask = var_14465_end_mask_0, x = transpose_2)[name = tensor("op_14465_cast")]; + tensor var_14469_begin_0 = const()[name = tensor("op_14469_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_14469_end_0 = const()[name = tensor("op_14469_end_0"), val = tensor([2, 77, 1, 256])]; + tensor var_14469_end_mask_0 = const()[name = tensor("op_14469_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14469_cast = slice_by_index(begin = var_14469_begin_0, end = var_14469_end_0, end_mask = var_14469_end_mask_0, x = transpose_2)[name = tensor("op_14469_cast")]; + tensor var_14473_begin_0 = const()[name = tensor("op_14473_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_14473_end_0 = const()[name = tensor("op_14473_end_0"), val = tensor([2, 77, 1, 320])]; + tensor var_14473_end_mask_0 = const()[name = tensor("op_14473_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14473_cast = slice_by_index(begin = var_14473_begin_0, end = var_14473_end_0, end_mask = var_14473_end_mask_0, x = transpose_2)[name = tensor("op_14473_cast")]; + tensor var_14475_begin_0 = const()[name = tensor("op_14475_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_14475_end_0 = const()[name = tensor("op_14475_end_0"), val = tensor([2, 64, 1, 77])]; + tensor var_14475_end_mask_0 = const()[name = tensor("op_14475_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14475_cast = slice_by_index(begin = var_14475_begin_0, end = var_14475_end_0, end_mask = var_14475_end_mask_0, x = v_59_cast)[name = tensor("op_14475_cast")]; + tensor var_14479_begin_0 = const()[name = tensor("op_14479_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_14479_end_0 = const()[name = tensor("op_14479_end_0"), val = tensor([2, 128, 1, 77])]; + tensor var_14479_end_mask_0 = const()[name = tensor("op_14479_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14479_cast = slice_by_index(begin = var_14479_begin_0, end = var_14479_end_0, end_mask = var_14479_end_mask_0, x = v_59_cast)[name = tensor("op_14479_cast")]; + tensor var_14483_begin_0 = const()[name = tensor("op_14483_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_14483_end_0 = const()[name = tensor("op_14483_end_0"), val = tensor([2, 192, 1, 77])]; + tensor var_14483_end_mask_0 = const()[name = tensor("op_14483_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14483_cast = slice_by_index(begin = var_14483_begin_0, end = var_14483_end_0, end_mask = var_14483_end_mask_0, x = v_59_cast)[name = tensor("op_14483_cast")]; + tensor var_14487_begin_0 = const()[name = tensor("op_14487_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_14487_end_0 = const()[name = tensor("op_14487_end_0"), val = tensor([2, 256, 1, 77])]; + tensor var_14487_end_mask_0 = const()[name = tensor("op_14487_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14487_cast = slice_by_index(begin = var_14487_begin_0, end = var_14487_end_0, end_mask = var_14487_end_mask_0, x = v_59_cast)[name = tensor("op_14487_cast")]; + tensor var_14491_begin_0 = const()[name = tensor("op_14491_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_14491_end_0 = const()[name = tensor("op_14491_end_0"), val = tensor([2, 320, 1, 77])]; + tensor var_14491_end_mask_0 = const()[name = tensor("op_14491_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14491_cast = slice_by_index(begin = var_14491_begin_0, end = var_14491_end_0, end_mask = var_14491_end_mask_0, x = v_59_cast)[name = tensor("op_14491_cast")]; + tensor var_14495_equation_0 = const()[name = tensor("op_14495_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14495_cast = einsum(equation = var_14495_equation_0, values = (var_14457_cast, var_14413_cast))[name = tensor("op_14495_cast")]; + tensor var_14496_to_fp16 = const()[name = tensor("op_14496_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_961_cast = mul(x = var_14495_cast, y = var_14496_to_fp16)[name = tensor("aw_chunk_961_cast")]; + tensor var_14499_equation_0 = const()[name = tensor("op_14499_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14499_cast = einsum(equation = var_14499_equation_0, values = (var_14457_cast, var_14414_cast))[name = tensor("op_14499_cast")]; + tensor var_14500_to_fp16 = const()[name = tensor("op_14500_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_963_cast = mul(x = var_14499_cast, y = var_14500_to_fp16)[name = tensor("aw_chunk_963_cast")]; + tensor var_14503_equation_0 = const()[name = tensor("op_14503_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14503_cast = einsum(equation = var_14503_equation_0, values = (var_14457_cast, var_14415_cast))[name = tensor("op_14503_cast")]; + tensor var_14504_to_fp16 = const()[name = tensor("op_14504_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_965_cast = mul(x = var_14503_cast, y = var_14504_to_fp16)[name = tensor("aw_chunk_965_cast")]; + tensor var_14507_equation_0 = const()[name = tensor("op_14507_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14507_cast = einsum(equation = var_14507_equation_0, values = (var_14457_cast, var_14416_cast))[name = tensor("op_14507_cast")]; + tensor var_14508_to_fp16 = const()[name = tensor("op_14508_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_967_cast = mul(x = var_14507_cast, y = var_14508_to_fp16)[name = tensor("aw_chunk_967_cast")]; + tensor var_14511_equation_0 = const()[name = tensor("op_14511_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14511_cast = einsum(equation = var_14511_equation_0, values = (var_14457_cast, var_14417_cast))[name = tensor("op_14511_cast")]; + tensor var_14512_to_fp16 = const()[name = tensor("op_14512_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_969_cast = mul(x = var_14511_cast, y = var_14512_to_fp16)[name = tensor("aw_chunk_969_cast")]; + tensor var_14515_equation_0 = const()[name = tensor("op_14515_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14515_cast = einsum(equation = var_14515_equation_0, values = (var_14457_cast, var_14418_cast))[name = tensor("op_14515_cast")]; + tensor var_14516_to_fp16 = const()[name = tensor("op_14516_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_971_cast = mul(x = var_14515_cast, y = var_14516_to_fp16)[name = tensor("aw_chunk_971_cast")]; + tensor var_14519_equation_0 = const()[name = tensor("op_14519_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14519_cast = einsum(equation = var_14519_equation_0, values = (var_14457_cast, var_14419_cast))[name = tensor("op_14519_cast")]; + tensor var_14520_to_fp16 = const()[name = tensor("op_14520_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_973_cast = mul(x = var_14519_cast, y = var_14520_to_fp16)[name = tensor("aw_chunk_973_cast")]; + tensor var_14523_equation_0 = const()[name = tensor("op_14523_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14523_cast = einsum(equation = var_14523_equation_0, values = (var_14457_cast, var_14420_cast))[name = tensor("op_14523_cast")]; + tensor var_14524_to_fp16 = const()[name = tensor("op_14524_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_975_cast = mul(x = var_14523_cast, y = var_14524_to_fp16)[name = tensor("aw_chunk_975_cast")]; + tensor var_14527_equation_0 = const()[name = tensor("op_14527_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14527_cast = einsum(equation = var_14527_equation_0, values = (var_14461_cast, var_14421_cast))[name = tensor("op_14527_cast")]; + tensor var_14528_to_fp16 = const()[name = tensor("op_14528_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_977_cast = mul(x = var_14527_cast, y = var_14528_to_fp16)[name = tensor("aw_chunk_977_cast")]; + tensor var_14531_equation_0 = const()[name = tensor("op_14531_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14531_cast = einsum(equation = var_14531_equation_0, values = (var_14461_cast, var_14422_cast))[name = tensor("op_14531_cast")]; + tensor var_14532_to_fp16 = const()[name = tensor("op_14532_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_979_cast = mul(x = var_14531_cast, y = var_14532_to_fp16)[name = tensor("aw_chunk_979_cast")]; + tensor var_14535_equation_0 = const()[name = tensor("op_14535_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14535_cast = einsum(equation = var_14535_equation_0, values = (var_14461_cast, var_14423_cast))[name = tensor("op_14535_cast")]; + tensor var_14536_to_fp16 = const()[name = tensor("op_14536_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_981_cast = mul(x = var_14535_cast, y = var_14536_to_fp16)[name = tensor("aw_chunk_981_cast")]; + tensor var_14539_equation_0 = const()[name = tensor("op_14539_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14539_cast = einsum(equation = var_14539_equation_0, values = (var_14461_cast, var_14424_cast))[name = tensor("op_14539_cast")]; + tensor var_14540_to_fp16 = const()[name = tensor("op_14540_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_983_cast = mul(x = var_14539_cast, y = var_14540_to_fp16)[name = tensor("aw_chunk_983_cast")]; + tensor var_14543_equation_0 = const()[name = tensor("op_14543_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14543_cast = einsum(equation = var_14543_equation_0, values = (var_14461_cast, var_14425_cast))[name = tensor("op_14543_cast")]; + tensor var_14544_to_fp16 = const()[name = tensor("op_14544_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_985_cast = mul(x = var_14543_cast, y = var_14544_to_fp16)[name = tensor("aw_chunk_985_cast")]; + tensor var_14547_equation_0 = const()[name = tensor("op_14547_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14547_cast = einsum(equation = var_14547_equation_0, values = (var_14461_cast, var_14426_cast))[name = tensor("op_14547_cast")]; + tensor var_14548_to_fp16 = const()[name = tensor("op_14548_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_987_cast = mul(x = var_14547_cast, y = var_14548_to_fp16)[name = tensor("aw_chunk_987_cast")]; + tensor var_14551_equation_0 = const()[name = tensor("op_14551_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14551_cast = einsum(equation = var_14551_equation_0, values = (var_14461_cast, var_14427_cast))[name = tensor("op_14551_cast")]; + tensor var_14552_to_fp16 = const()[name = tensor("op_14552_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_989_cast = mul(x = var_14551_cast, y = var_14552_to_fp16)[name = tensor("aw_chunk_989_cast")]; + tensor var_14555_equation_0 = const()[name = tensor("op_14555_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14555_cast = einsum(equation = var_14555_equation_0, values = (var_14461_cast, var_14428_cast))[name = tensor("op_14555_cast")]; + tensor var_14556_to_fp16 = const()[name = tensor("op_14556_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_991_cast = mul(x = var_14555_cast, y = var_14556_to_fp16)[name = tensor("aw_chunk_991_cast")]; + tensor var_14559_equation_0 = const()[name = tensor("op_14559_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14559_cast = einsum(equation = var_14559_equation_0, values = (var_14465_cast, var_14429_cast))[name = tensor("op_14559_cast")]; + tensor var_14560_to_fp16 = const()[name = tensor("op_14560_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_993_cast = mul(x = var_14559_cast, y = var_14560_to_fp16)[name = tensor("aw_chunk_993_cast")]; + tensor var_14563_equation_0 = const()[name = tensor("op_14563_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14563_cast = einsum(equation = var_14563_equation_0, values = (var_14465_cast, var_14430_cast))[name = tensor("op_14563_cast")]; + tensor var_14564_to_fp16 = const()[name = tensor("op_14564_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_995_cast = mul(x = var_14563_cast, y = var_14564_to_fp16)[name = tensor("aw_chunk_995_cast")]; + tensor var_14567_equation_0 = const()[name = tensor("op_14567_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14567_cast = einsum(equation = var_14567_equation_0, values = (var_14465_cast, var_14431_cast))[name = tensor("op_14567_cast")]; + tensor var_14568_to_fp16 = const()[name = tensor("op_14568_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_997_cast = mul(x = var_14567_cast, y = var_14568_to_fp16)[name = tensor("aw_chunk_997_cast")]; + tensor var_14571_equation_0 = const()[name = tensor("op_14571_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14571_cast = einsum(equation = var_14571_equation_0, values = (var_14465_cast, var_14432_cast))[name = tensor("op_14571_cast")]; + tensor var_14572_to_fp16 = const()[name = tensor("op_14572_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_999_cast = mul(x = var_14571_cast, y = var_14572_to_fp16)[name = tensor("aw_chunk_999_cast")]; + tensor var_14575_equation_0 = const()[name = tensor("op_14575_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14575_cast = einsum(equation = var_14575_equation_0, values = (var_14465_cast, var_14433_cast))[name = tensor("op_14575_cast")]; + tensor var_14576_to_fp16 = const()[name = tensor("op_14576_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1001_cast = mul(x = var_14575_cast, y = var_14576_to_fp16)[name = tensor("aw_chunk_1001_cast")]; + tensor var_14579_equation_0 = const()[name = tensor("op_14579_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14579_cast = einsum(equation = var_14579_equation_0, values = (var_14465_cast, var_14434_cast))[name = tensor("op_14579_cast")]; + tensor var_14580_to_fp16 = const()[name = tensor("op_14580_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1003_cast = mul(x = var_14579_cast, y = var_14580_to_fp16)[name = tensor("aw_chunk_1003_cast")]; + tensor var_14583_equation_0 = const()[name = tensor("op_14583_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14583_cast = einsum(equation = var_14583_equation_0, values = (var_14465_cast, var_14435_cast))[name = tensor("op_14583_cast")]; + tensor var_14584_to_fp16 = const()[name = tensor("op_14584_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1005_cast = mul(x = var_14583_cast, y = var_14584_to_fp16)[name = tensor("aw_chunk_1005_cast")]; + tensor var_14587_equation_0 = const()[name = tensor("op_14587_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14587_cast = einsum(equation = var_14587_equation_0, values = (var_14465_cast, var_14436_cast))[name = tensor("op_14587_cast")]; + tensor var_14588_to_fp16 = const()[name = tensor("op_14588_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1007_cast = mul(x = var_14587_cast, y = var_14588_to_fp16)[name = tensor("aw_chunk_1007_cast")]; + tensor var_14591_equation_0 = const()[name = tensor("op_14591_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14591_cast = einsum(equation = var_14591_equation_0, values = (var_14469_cast, var_14437_cast))[name = tensor("op_14591_cast")]; + tensor var_14592_to_fp16 = const()[name = tensor("op_14592_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1009_cast = mul(x = var_14591_cast, y = var_14592_to_fp16)[name = tensor("aw_chunk_1009_cast")]; + tensor var_14595_equation_0 = const()[name = tensor("op_14595_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14595_cast = einsum(equation = var_14595_equation_0, values = (var_14469_cast, var_14438_cast))[name = tensor("op_14595_cast")]; + tensor var_14596_to_fp16 = const()[name = tensor("op_14596_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1011_cast = mul(x = var_14595_cast, y = var_14596_to_fp16)[name = tensor("aw_chunk_1011_cast")]; + tensor var_14599_equation_0 = const()[name = tensor("op_14599_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14599_cast = einsum(equation = var_14599_equation_0, values = (var_14469_cast, var_14439_cast))[name = tensor("op_14599_cast")]; + tensor var_14600_to_fp16 = const()[name = tensor("op_14600_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1013_cast = mul(x = var_14599_cast, y = var_14600_to_fp16)[name = tensor("aw_chunk_1013_cast")]; + tensor var_14603_equation_0 = const()[name = tensor("op_14603_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14603_cast = einsum(equation = var_14603_equation_0, values = (var_14469_cast, var_14440_cast))[name = tensor("op_14603_cast")]; + tensor var_14604_to_fp16 = const()[name = tensor("op_14604_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1015_cast = mul(x = var_14603_cast, y = var_14604_to_fp16)[name = tensor("aw_chunk_1015_cast")]; + tensor var_14607_equation_0 = const()[name = tensor("op_14607_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14607_cast = einsum(equation = var_14607_equation_0, values = (var_14469_cast, var_14441_cast))[name = tensor("op_14607_cast")]; + tensor var_14608_to_fp16 = const()[name = tensor("op_14608_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1017_cast = mul(x = var_14607_cast, y = var_14608_to_fp16)[name = tensor("aw_chunk_1017_cast")]; + tensor var_14611_equation_0 = const()[name = tensor("op_14611_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14611_cast = einsum(equation = var_14611_equation_0, values = (var_14469_cast, var_14442_cast))[name = tensor("op_14611_cast")]; + tensor var_14612_to_fp16 = const()[name = tensor("op_14612_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1019_cast = mul(x = var_14611_cast, y = var_14612_to_fp16)[name = tensor("aw_chunk_1019_cast")]; + tensor var_14615_equation_0 = const()[name = tensor("op_14615_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14615_cast = einsum(equation = var_14615_equation_0, values = (var_14469_cast, var_14443_cast))[name = tensor("op_14615_cast")]; + tensor var_14616_to_fp16 = const()[name = tensor("op_14616_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1021_cast = mul(x = var_14615_cast, y = var_14616_to_fp16)[name = tensor("aw_chunk_1021_cast")]; + tensor var_14619_equation_0 = const()[name = tensor("op_14619_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14619_cast = einsum(equation = var_14619_equation_0, values = (var_14469_cast, var_14444_cast))[name = tensor("op_14619_cast")]; + tensor var_14620_to_fp16 = const()[name = tensor("op_14620_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1023_cast = mul(x = var_14619_cast, y = var_14620_to_fp16)[name = tensor("aw_chunk_1023_cast")]; + tensor var_14623_equation_0 = const()[name = tensor("op_14623_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14623_cast = einsum(equation = var_14623_equation_0, values = (var_14473_cast, var_14445_cast))[name = tensor("op_14623_cast")]; + tensor var_14624_to_fp16 = const()[name = tensor("op_14624_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1025_cast = mul(x = var_14623_cast, y = var_14624_to_fp16)[name = tensor("aw_chunk_1025_cast")]; + tensor var_14627_equation_0 = const()[name = tensor("op_14627_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14627_cast = einsum(equation = var_14627_equation_0, values = (var_14473_cast, var_14446_cast))[name = tensor("op_14627_cast")]; + tensor var_14628_to_fp16 = const()[name = tensor("op_14628_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1027_cast = mul(x = var_14627_cast, y = var_14628_to_fp16)[name = tensor("aw_chunk_1027_cast")]; + tensor var_14631_equation_0 = const()[name = tensor("op_14631_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14631_cast = einsum(equation = var_14631_equation_0, values = (var_14473_cast, var_14447_cast))[name = tensor("op_14631_cast")]; + tensor var_14632_to_fp16 = const()[name = tensor("op_14632_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1029_cast = mul(x = var_14631_cast, y = var_14632_to_fp16)[name = tensor("aw_chunk_1029_cast")]; + tensor var_14635_equation_0 = const()[name = tensor("op_14635_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14635_cast = einsum(equation = var_14635_equation_0, values = (var_14473_cast, var_14448_cast))[name = tensor("op_14635_cast")]; + tensor var_14636_to_fp16 = const()[name = tensor("op_14636_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1031_cast = mul(x = var_14635_cast, y = var_14636_to_fp16)[name = tensor("aw_chunk_1031_cast")]; + tensor var_14639_equation_0 = const()[name = tensor("op_14639_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14639_cast = einsum(equation = var_14639_equation_0, values = (var_14473_cast, var_14449_cast))[name = tensor("op_14639_cast")]; + tensor var_14640_to_fp16 = const()[name = tensor("op_14640_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1033_cast = mul(x = var_14639_cast, y = var_14640_to_fp16)[name = tensor("aw_chunk_1033_cast")]; + tensor var_14643_equation_0 = const()[name = tensor("op_14643_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14643_cast = einsum(equation = var_14643_equation_0, values = (var_14473_cast, var_14450_cast))[name = tensor("op_14643_cast")]; + tensor var_14644_to_fp16 = const()[name = tensor("op_14644_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1035_cast = mul(x = var_14643_cast, y = var_14644_to_fp16)[name = tensor("aw_chunk_1035_cast")]; + tensor var_14647_equation_0 = const()[name = tensor("op_14647_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14647_cast = einsum(equation = var_14647_equation_0, values = (var_14473_cast, var_14451_cast))[name = tensor("op_14647_cast")]; + tensor var_14648_to_fp16 = const()[name = tensor("op_14648_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1037_cast = mul(x = var_14647_cast, y = var_14648_to_fp16)[name = tensor("aw_chunk_1037_cast")]; + tensor var_14651_equation_0 = const()[name = tensor("op_14651_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14651_cast = einsum(equation = var_14651_equation_0, values = (var_14473_cast, var_14452_cast))[name = tensor("op_14651_cast")]; + tensor var_14652_to_fp16 = const()[name = tensor("op_14652_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1039_cast = mul(x = var_14651_cast, y = var_14652_to_fp16)[name = tensor("aw_chunk_1039_cast")]; + tensor var_14654_cast = softmax(axis = var_12807, x = aw_chunk_961_cast)[name = tensor("op_14654_cast")]; + tensor var_14655_cast = softmax(axis = var_12807, x = aw_chunk_963_cast)[name = tensor("op_14655_cast")]; + tensor var_14656_cast = softmax(axis = var_12807, x = aw_chunk_965_cast)[name = tensor("op_14656_cast")]; + tensor var_14657_cast = softmax(axis = var_12807, x = aw_chunk_967_cast)[name = tensor("op_14657_cast")]; + tensor var_14658_cast = softmax(axis = var_12807, x = aw_chunk_969_cast)[name = tensor("op_14658_cast")]; + tensor var_14659_cast = softmax(axis = var_12807, x = aw_chunk_971_cast)[name = tensor("op_14659_cast")]; + tensor var_14660_cast = softmax(axis = var_12807, x = aw_chunk_973_cast)[name = tensor("op_14660_cast")]; + tensor var_14661_cast = softmax(axis = var_12807, x = aw_chunk_975_cast)[name = tensor("op_14661_cast")]; + tensor var_14662_cast = softmax(axis = var_12807, x = aw_chunk_977_cast)[name = tensor("op_14662_cast")]; + tensor var_14663_cast = softmax(axis = var_12807, x = aw_chunk_979_cast)[name = tensor("op_14663_cast")]; + tensor var_14664_cast = softmax(axis = var_12807, x = aw_chunk_981_cast)[name = tensor("op_14664_cast")]; + tensor var_14665_cast = softmax(axis = var_12807, x = aw_chunk_983_cast)[name = tensor("op_14665_cast")]; + tensor var_14666_cast = softmax(axis = var_12807, x = aw_chunk_985_cast)[name = tensor("op_14666_cast")]; + tensor var_14667_cast = softmax(axis = var_12807, x = aw_chunk_987_cast)[name = tensor("op_14667_cast")]; + tensor var_14668_cast = softmax(axis = var_12807, x = aw_chunk_989_cast)[name = tensor("op_14668_cast")]; + tensor var_14669_cast = softmax(axis = var_12807, x = aw_chunk_991_cast)[name = tensor("op_14669_cast")]; + tensor var_14670_cast = softmax(axis = var_12807, x = aw_chunk_993_cast)[name = tensor("op_14670_cast")]; + tensor var_14671_cast = softmax(axis = var_12807, x = aw_chunk_995_cast)[name = tensor("op_14671_cast")]; + tensor var_14672_cast = softmax(axis = var_12807, x = aw_chunk_997_cast)[name = tensor("op_14672_cast")]; + tensor var_14673_cast = softmax(axis = var_12807, x = aw_chunk_999_cast)[name = tensor("op_14673_cast")]; + tensor var_14674_cast = softmax(axis = var_12807, x = aw_chunk_1001_cast)[name = tensor("op_14674_cast")]; + tensor var_14675_cast = softmax(axis = var_12807, x = aw_chunk_1003_cast)[name = tensor("op_14675_cast")]; + tensor var_14676_cast = softmax(axis = var_12807, x = aw_chunk_1005_cast)[name = tensor("op_14676_cast")]; + tensor var_14677_cast = softmax(axis = var_12807, x = aw_chunk_1007_cast)[name = tensor("op_14677_cast")]; + tensor var_14678_cast = softmax(axis = var_12807, x = aw_chunk_1009_cast)[name = tensor("op_14678_cast")]; + tensor var_14679_cast = softmax(axis = var_12807, x = aw_chunk_1011_cast)[name = tensor("op_14679_cast")]; + tensor var_14680_cast = softmax(axis = var_12807, x = aw_chunk_1013_cast)[name = tensor("op_14680_cast")]; + tensor var_14681_cast = softmax(axis = var_12807, x = aw_chunk_1015_cast)[name = tensor("op_14681_cast")]; + tensor var_14682_cast = softmax(axis = var_12807, x = aw_chunk_1017_cast)[name = tensor("op_14682_cast")]; + tensor var_14683_cast = softmax(axis = var_12807, x = aw_chunk_1019_cast)[name = tensor("op_14683_cast")]; + tensor var_14684_cast = softmax(axis = var_12807, x = aw_chunk_1021_cast)[name = tensor("op_14684_cast")]; + tensor var_14685_cast = softmax(axis = var_12807, x = aw_chunk_1023_cast)[name = tensor("op_14685_cast")]; + tensor var_14686_cast = softmax(axis = var_12807, x = aw_chunk_1025_cast)[name = tensor("op_14686_cast")]; + tensor var_14687_cast = softmax(axis = var_12807, x = aw_chunk_1027_cast)[name = tensor("op_14687_cast")]; + tensor var_14688_cast = softmax(axis = var_12807, x = aw_chunk_1029_cast)[name = tensor("op_14688_cast")]; + tensor var_14689_cast = softmax(axis = var_12807, x = aw_chunk_1031_cast)[name = tensor("op_14689_cast")]; + tensor var_14690_cast = softmax(axis = var_12807, x = aw_chunk_1033_cast)[name = tensor("op_14690_cast")]; + tensor var_14691_cast = softmax(axis = var_12807, x = aw_chunk_1035_cast)[name = tensor("op_14691_cast")]; + tensor var_14692_cast = softmax(axis = var_12807, x = aw_chunk_1037_cast)[name = tensor("op_14692_cast")]; + tensor var_14693_cast = softmax(axis = var_12807, x = aw_chunk_1039_cast)[name = tensor("op_14693_cast")]; + tensor var_14695_equation_0 = const()[name = tensor("op_14695_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14695_cast = einsum(equation = var_14695_equation_0, values = (var_14475_cast, var_14654_cast))[name = tensor("op_14695_cast")]; + tensor var_14697_equation_0 = const()[name = tensor("op_14697_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14697_cast = einsum(equation = var_14697_equation_0, values = (var_14475_cast, var_14655_cast))[name = tensor("op_14697_cast")]; + tensor var_14699_equation_0 = const()[name = tensor("op_14699_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14699_cast = einsum(equation = var_14699_equation_0, values = (var_14475_cast, var_14656_cast))[name = tensor("op_14699_cast")]; + tensor var_14701_equation_0 = const()[name = tensor("op_14701_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14701_cast = einsum(equation = var_14701_equation_0, values = (var_14475_cast, var_14657_cast))[name = tensor("op_14701_cast")]; + tensor var_14703_equation_0 = const()[name = tensor("op_14703_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14703_cast = einsum(equation = var_14703_equation_0, values = (var_14475_cast, var_14658_cast))[name = tensor("op_14703_cast")]; + tensor var_14705_equation_0 = const()[name = tensor("op_14705_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14705_cast = einsum(equation = var_14705_equation_0, values = (var_14475_cast, var_14659_cast))[name = tensor("op_14705_cast")]; + tensor var_14707_equation_0 = const()[name = tensor("op_14707_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14707_cast = einsum(equation = var_14707_equation_0, values = (var_14475_cast, var_14660_cast))[name = tensor("op_14707_cast")]; + tensor var_14709_equation_0 = const()[name = tensor("op_14709_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14709_cast = einsum(equation = var_14709_equation_0, values = (var_14475_cast, var_14661_cast))[name = tensor("op_14709_cast")]; + tensor var_14711_equation_0 = const()[name = tensor("op_14711_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14711_cast = einsum(equation = var_14711_equation_0, values = (var_14479_cast, var_14662_cast))[name = tensor("op_14711_cast")]; + tensor var_14713_equation_0 = const()[name = tensor("op_14713_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14713_cast = einsum(equation = var_14713_equation_0, values = (var_14479_cast, var_14663_cast))[name = tensor("op_14713_cast")]; + tensor var_14715_equation_0 = const()[name = tensor("op_14715_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14715_cast = einsum(equation = var_14715_equation_0, values = (var_14479_cast, var_14664_cast))[name = tensor("op_14715_cast")]; + tensor var_14717_equation_0 = const()[name = tensor("op_14717_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14717_cast = einsum(equation = var_14717_equation_0, values = (var_14479_cast, var_14665_cast))[name = tensor("op_14717_cast")]; + tensor var_14719_equation_0 = const()[name = tensor("op_14719_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14719_cast = einsum(equation = var_14719_equation_0, values = (var_14479_cast, var_14666_cast))[name = tensor("op_14719_cast")]; + tensor var_14721_equation_0 = const()[name = tensor("op_14721_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14721_cast = einsum(equation = var_14721_equation_0, values = (var_14479_cast, var_14667_cast))[name = tensor("op_14721_cast")]; + tensor var_14723_equation_0 = const()[name = tensor("op_14723_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14723_cast = einsum(equation = var_14723_equation_0, values = (var_14479_cast, var_14668_cast))[name = tensor("op_14723_cast")]; + tensor var_14725_equation_0 = const()[name = tensor("op_14725_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14725_cast = einsum(equation = var_14725_equation_0, values = (var_14479_cast, var_14669_cast))[name = tensor("op_14725_cast")]; + tensor var_14727_equation_0 = const()[name = tensor("op_14727_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14727_cast = einsum(equation = var_14727_equation_0, values = (var_14483_cast, var_14670_cast))[name = tensor("op_14727_cast")]; + tensor var_14729_equation_0 = const()[name = tensor("op_14729_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14729_cast = einsum(equation = var_14729_equation_0, values = (var_14483_cast, var_14671_cast))[name = tensor("op_14729_cast")]; + tensor var_14731_equation_0 = const()[name = tensor("op_14731_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14731_cast = einsum(equation = var_14731_equation_0, values = (var_14483_cast, var_14672_cast))[name = tensor("op_14731_cast")]; + tensor var_14733_equation_0 = const()[name = tensor("op_14733_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14733_cast = einsum(equation = var_14733_equation_0, values = (var_14483_cast, var_14673_cast))[name = tensor("op_14733_cast")]; + tensor var_14735_equation_0 = const()[name = tensor("op_14735_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14735_cast = einsum(equation = var_14735_equation_0, values = (var_14483_cast, var_14674_cast))[name = tensor("op_14735_cast")]; + tensor var_14737_equation_0 = const()[name = tensor("op_14737_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14737_cast = einsum(equation = var_14737_equation_0, values = (var_14483_cast, var_14675_cast))[name = tensor("op_14737_cast")]; + tensor var_14739_equation_0 = const()[name = tensor("op_14739_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14739_cast = einsum(equation = var_14739_equation_0, values = (var_14483_cast, var_14676_cast))[name = tensor("op_14739_cast")]; + tensor var_14741_equation_0 = const()[name = tensor("op_14741_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14741_cast = einsum(equation = var_14741_equation_0, values = (var_14483_cast, var_14677_cast))[name = tensor("op_14741_cast")]; + tensor var_14743_equation_0 = const()[name = tensor("op_14743_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14743_cast = einsum(equation = var_14743_equation_0, values = (var_14487_cast, var_14678_cast))[name = tensor("op_14743_cast")]; + tensor var_14745_equation_0 = const()[name = tensor("op_14745_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14745_cast = einsum(equation = var_14745_equation_0, values = (var_14487_cast, var_14679_cast))[name = tensor("op_14745_cast")]; + tensor var_14747_equation_0 = const()[name = tensor("op_14747_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14747_cast = einsum(equation = var_14747_equation_0, values = (var_14487_cast, var_14680_cast))[name = tensor("op_14747_cast")]; + tensor var_14749_equation_0 = const()[name = tensor("op_14749_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14749_cast = einsum(equation = var_14749_equation_0, values = (var_14487_cast, var_14681_cast))[name = tensor("op_14749_cast")]; + tensor var_14751_equation_0 = const()[name = tensor("op_14751_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14751_cast = einsum(equation = var_14751_equation_0, values = (var_14487_cast, var_14682_cast))[name = tensor("op_14751_cast")]; + tensor var_14753_equation_0 = const()[name = tensor("op_14753_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14753_cast = einsum(equation = var_14753_equation_0, values = (var_14487_cast, var_14683_cast))[name = tensor("op_14753_cast")]; + tensor var_14755_equation_0 = const()[name = tensor("op_14755_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14755_cast = einsum(equation = var_14755_equation_0, values = (var_14487_cast, var_14684_cast))[name = tensor("op_14755_cast")]; + tensor var_14757_equation_0 = const()[name = tensor("op_14757_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14757_cast = einsum(equation = var_14757_equation_0, values = (var_14487_cast, var_14685_cast))[name = tensor("op_14757_cast")]; + tensor var_14759_equation_0 = const()[name = tensor("op_14759_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14759_cast = einsum(equation = var_14759_equation_0, values = (var_14491_cast, var_14686_cast))[name = tensor("op_14759_cast")]; + tensor var_14761_equation_0 = const()[name = tensor("op_14761_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14761_cast = einsum(equation = var_14761_equation_0, values = (var_14491_cast, var_14687_cast))[name = tensor("op_14761_cast")]; + tensor var_14763_equation_0 = const()[name = tensor("op_14763_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14763_cast = einsum(equation = var_14763_equation_0, values = (var_14491_cast, var_14688_cast))[name = tensor("op_14763_cast")]; + tensor var_14765_equation_0 = const()[name = tensor("op_14765_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14765_cast = einsum(equation = var_14765_equation_0, values = (var_14491_cast, var_14689_cast))[name = tensor("op_14765_cast")]; + tensor var_14767_equation_0 = const()[name = tensor("op_14767_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14767_cast = einsum(equation = var_14767_equation_0, values = (var_14491_cast, var_14690_cast))[name = tensor("op_14767_cast")]; + tensor var_14769_equation_0 = const()[name = tensor("op_14769_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14769_cast = einsum(equation = var_14769_equation_0, values = (var_14491_cast, var_14691_cast))[name = tensor("op_14769_cast")]; + tensor var_14771_equation_0 = const()[name = tensor("op_14771_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14771_cast = einsum(equation = var_14771_equation_0, values = (var_14491_cast, var_14692_cast))[name = tensor("op_14771_cast")]; + tensor var_14773_equation_0 = const()[name = tensor("op_14773_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14773_cast = einsum(equation = var_14773_equation_0, values = (var_14491_cast, var_14693_cast))[name = tensor("op_14773_cast")]; + tensor var_14775_interleave_0 = const()[name = tensor("op_14775_interleave_0"), val = tensor(false)]; + tensor var_14775_cast = concat(axis = var_12782, interleave = var_14775_interleave_0, values = (var_14695_cast, var_14697_cast, var_14699_cast, var_14701_cast, var_14703_cast, var_14705_cast, var_14707_cast, var_14709_cast))[name = tensor("op_14775_cast")]; + tensor var_14777_interleave_0 = const()[name = tensor("op_14777_interleave_0"), val = tensor(false)]; + tensor var_14777_cast = concat(axis = var_12782, interleave = var_14777_interleave_0, values = (var_14711_cast, var_14713_cast, var_14715_cast, var_14717_cast, var_14719_cast, var_14721_cast, var_14723_cast, var_14725_cast))[name = tensor("op_14777_cast")]; + tensor var_14779_interleave_0 = const()[name = tensor("op_14779_interleave_0"), val = tensor(false)]; + tensor var_14779_cast = concat(axis = var_12782, interleave = var_14779_interleave_0, values = (var_14727_cast, var_14729_cast, var_14731_cast, var_14733_cast, var_14735_cast, var_14737_cast, var_14739_cast, var_14741_cast))[name = tensor("op_14779_cast")]; + tensor var_14781_interleave_0 = const()[name = tensor("op_14781_interleave_0"), val = tensor(false)]; + tensor var_14781_cast = concat(axis = var_12782, interleave = var_14781_interleave_0, values = (var_14743_cast, var_14745_cast, var_14747_cast, var_14749_cast, var_14751_cast, var_14753_cast, var_14755_cast, var_14757_cast))[name = tensor("op_14781_cast")]; + tensor var_14783_interleave_0 = const()[name = tensor("op_14783_interleave_0"), val = tensor(false)]; + tensor var_14783_cast = concat(axis = var_12782, interleave = var_14783_interleave_0, values = (var_14759_cast, var_14761_cast, var_14763_cast, var_14765_cast, var_14767_cast, var_14769_cast, var_14771_cast, var_14773_cast))[name = tensor("op_14783_cast")]; + tensor input_493_interleave_0 = const()[name = tensor("input_493_interleave_0"), val = tensor(false)]; + tensor input_493_cast = concat(axis = var_12807, interleave = input_493_interleave_0, values = (var_14775_cast, var_14777_cast, var_14779_cast, var_14781_cast, var_14783_cast))[name = tensor("input_493_cast")]; + tensor var_14789 = const()[name = tensor("op_14789"), val = tensor([1, 1])]; + tensor var_14791 = const()[name = tensor("op_14791"), val = tensor([1, 1])]; + tensor var_14793_pad_type_0 = const()[name = tensor("op_14793_pad_type_0"), val = tensor("custom")]; + tensor var_14793_pad_0 = const()[name = tensor("op_14793_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_3_attentions_1_transformer_blocks_0_attn2_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(644283456))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(644360320))), name = tensor("up_blocks_3_attentions_1_transformer_blocks_0_attn2_to_out_0_weight_to_fp16_palettized"), shape = tensor([320, 320, 1, 1])]; + tensor up_blocks_3_attentions_1_transformer_blocks_0_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("up_blocks_3_attentions_1_transformer_blocks_0_attn2_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(644360512)))]; + tensor var_14793_cast = conv(bias = up_blocks_3_attentions_1_transformer_blocks_0_attn2_to_out_0_bias_to_fp16, dilations = var_14791, groups = var_12807, pad = var_14793_pad_0, pad_type = var_14793_pad_type_0, strides = var_14789, weight = up_blocks_3_attentions_1_transformer_blocks_0_attn2_to_out_0_weight_to_fp16_palettized, x = input_493_cast)[name = tensor("op_14793_cast")]; + tensor inputs_89_cast = add(x = var_14793_cast, y = inputs_87_cast)[name = tensor("inputs_89_cast")]; + tensor var_14797 = const()[name = tensor("op_14797"), val = tensor([1])]; + tensor channels_mean_89_cast = reduce_mean(axes = var_14797, keep_dims = var_12802, x = inputs_89_cast)[name = tensor("channels_mean_89_cast")]; + tensor zero_mean_89_cast = sub(x = inputs_89_cast, y = channels_mean_89_cast)[name = tensor("zero_mean_89_cast")]; + tensor zero_mean_sq_89_cast = mul(x = zero_mean_89_cast, y = zero_mean_89_cast)[name = tensor("zero_mean_sq_89_cast")]; + tensor var_14801 = const()[name = tensor("op_14801"), val = tensor([1])]; + tensor var_14802_cast = reduce_mean(axes = var_14801, keep_dims = var_12802, x = zero_mean_sq_89_cast)[name = tensor("op_14802_cast")]; + tensor var_14803_to_fp16 = const()[name = tensor("op_14803_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_14804_cast = add(x = var_14802_cast, y = var_14803_to_fp16)[name = tensor("op_14804_cast")]; + tensor denom_89_epsilon_0_to_fp16 = const()[name = tensor("denom_89_epsilon_0_to_fp16"), val = tensor(0x1p-24)]; + tensor denom_89_cast = rsqrt(epsilon = denom_89_epsilon_0_to_fp16, x = var_14804_cast)[name = tensor("denom_89_cast")]; + tensor out_89_cast = mul(x = zero_mean_89_cast, y = denom_89_cast)[name = tensor("out_89_cast")]; + tensor var_14808_to_fp16 = const()[name = tensor("op_14808_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(644361216)))]; + tensor var_14809_cast = add(x = out_89_cast, y = var_14808_to_fp16)[name = tensor("op_14809_cast")]; + tensor var_14811_to_fp16 = const()[name = tensor("op_14811_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(644361920)))]; + tensor input_495_cast = mul(x = var_14809_cast, y = var_14811_to_fp16)[name = tensor("input_495_cast")]; + tensor var_14819 = const()[name = tensor("op_14819"), val = tensor([1, 1])]; + tensor var_14821 = const()[name = tensor("op_14821"), val = tensor([1, 1])]; + tensor var_14823_pad_type_0 = const()[name = tensor("op_14823_pad_type_0"), val = tensor("custom")]; + tensor var_14823_pad_0 = const()[name = tensor("op_14823_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_3_attentions_1_transformer_blocks_0_ff_net_0_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(644362624))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(644977088))), name = tensor("up_blocks_3_attentions_1_transformer_blocks_0_ff_net_0_proj_weight_to_fp16_palettized"), shape = tensor([2560, 320, 1, 1])]; + tensor up_blocks_3_attentions_1_transformer_blocks_0_ff_net_0_proj_bias_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(644977280))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(644979264))), name = tensor("up_blocks_3_attentions_1_transformer_blocks_0_ff_net_0_proj_bias_to_fp16_palettized"), shape = tensor([2560])]; + tensor var_14823_cast = conv(bias = up_blocks_3_attentions_1_transformer_blocks_0_ff_net_0_proj_bias_to_fp16_palettized, dilations = var_14821, groups = var_12807, pad = var_14823_pad_0, pad_type = var_14823_pad_type_0, strides = var_14819, weight = up_blocks_3_attentions_1_transformer_blocks_0_ff_net_0_proj_weight_to_fp16_palettized, x = input_495_cast)[name = tensor("op_14823_cast")]; + tensor var_14824_split_sizes_0 = const()[name = tensor("op_14824_split_sizes_0"), val = tensor([1280, 1280])]; + tensor var_14824_axis_0 = const()[name = tensor("op_14824_axis_0"), val = tensor(1)]; + tensor var_14824_cast_0, tensor var_14824_cast_1 = split(axis = var_14824_axis_0, split_sizes = var_14824_split_sizes_0, x = var_14823_cast)[name = tensor("op_14824_cast")]; + tensor var_14826_mode_0 = const()[name = tensor("op_14826_mode_0"), val = tensor("EXACT")]; + tensor var_14826_cast = gelu(mode = var_14826_mode_0, x = var_14824_cast_1)[name = tensor("op_14826_cast")]; + tensor input_497_cast = mul(x = var_14824_cast_0, y = var_14826_cast)[name = tensor("input_497_cast")]; + tensor var_14830 = const()[name = tensor("op_14830"), val = tensor([1, 1])]; + tensor var_14832 = const()[name = tensor("op_14832"), val = tensor([1, 1])]; + tensor var_14834_pad_type_0 = const()[name = tensor("op_14834_pad_type_0"), val = tensor("custom")]; + tensor var_14834_pad_0 = const()[name = tensor("op_14834_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_3_attentions_1_transformer_blocks_0_ff_net_2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(644979456))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(645286720))), name = tensor("up_blocks_3_attentions_1_transformer_blocks_0_ff_net_2_weight_to_fp16_palettized"), shape = tensor([320, 1280, 1, 1])]; + tensor up_blocks_3_attentions_1_transformer_blocks_0_ff_net_2_bias_to_fp16 = const()[name = tensor("up_blocks_3_attentions_1_transformer_blocks_0_ff_net_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(645286912)))]; + tensor var_14834_cast = conv(bias = up_blocks_3_attentions_1_transformer_blocks_0_ff_net_2_bias_to_fp16, dilations = var_14832, groups = var_12807, pad = var_14834_pad_0, pad_type = var_14834_pad_type_0, strides = var_14830, weight = up_blocks_3_attentions_1_transformer_blocks_0_ff_net_2_weight_to_fp16_palettized, x = input_497_cast)[name = tensor("op_14834_cast")]; + tensor hidden_states_313_cast = add(x = var_14834_cast, y = inputs_89_cast)[name = tensor("hidden_states_313_cast")]; + tensor var_14836 = const()[name = tensor("op_14836"), val = tensor([2, 320, 64, 64])]; + tensor input_499_cast = reshape(shape = var_14836, x = hidden_states_313_cast)[name = tensor("input_499_cast")]; + tensor var_14840 = const()[name = tensor("op_14840"), val = tensor([1, 1])]; + tensor var_14842 = const()[name = tensor("op_14842"), val = tensor([1, 1])]; + tensor hidden_states_315_pad_type_0 = const()[name = tensor("hidden_states_315_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_315_pad_0 = const()[name = tensor("hidden_states_315_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_3_attentions_1_proj_out_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(645287616))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(645364480))), name = tensor("up_blocks_3_attentions_1_proj_out_weight_to_fp16_palettized"), shape = tensor([320, 320, 1, 1])]; + tensor up_blocks_3_attentions_1_proj_out_bias_to_fp16 = const()[name = tensor("up_blocks_3_attentions_1_proj_out_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(645364672)))]; + tensor hidden_states_315_cast = conv(bias = up_blocks_3_attentions_1_proj_out_bias_to_fp16, dilations = var_14842, groups = var_12807, pad = hidden_states_315_pad_0, pad_type = hidden_states_315_pad_type_0, strides = var_14840, weight = up_blocks_3_attentions_1_proj_out_weight_to_fp16_palettized, x = input_499_cast)[name = tensor("hidden_states_315_cast")]; + tensor hidden_states_317_cast = add(x = hidden_states_315_cast, y = hidden_states_303_cast)[name = tensor("hidden_states_317_cast")]; + tensor input_501_interleave_0 = const()[name = tensor("input_501_interleave_0"), val = tensor(false)]; + tensor input_501_cast = concat(axis = var_12807, interleave = input_501_interleave_0, values = (hidden_states_317_cast, input_7_cast))[name = tensor("input_501_cast")]; + tensor reshape_228_shape_0 = const()[name = tensor("reshape_228_shape_0"), val = tensor([2, 32, 20, 64, 64])]; + tensor reshape_228_cast = reshape(shape = reshape_228_shape_0, x = input_501_cast)[name = tensor("reshape_228_cast")]; + tensor reduce_mean_171_axes_0 = const()[name = tensor("reduce_mean_171_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_171_keep_dims_0 = const()[name = tensor("reduce_mean_171_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_171_cast = reduce_mean(axes = reduce_mean_171_axes_0, keep_dims = reduce_mean_171_keep_dims_0, x = reshape_228_cast)[name = tensor("reduce_mean_171_cast")]; + tensor sub_114_cast = sub(x = reshape_228_cast, y = reduce_mean_171_cast)[name = tensor("sub_114_cast")]; + tensor square_57_cast = square(x = sub_114_cast)[name = tensor("square_57_cast")]; + tensor reduce_mean_173_axes_0 = const()[name = tensor("reduce_mean_173_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_173_keep_dims_0 = const()[name = tensor("reduce_mean_173_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_173_cast = reduce_mean(axes = reduce_mean_173_axes_0, keep_dims = reduce_mean_173_keep_dims_0, x = square_57_cast)[name = tensor("reduce_mean_173_cast")]; + tensor add_114_y_0_to_fp16 = const()[name = tensor("add_114_y_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_114_cast = add(x = reduce_mean_173_cast, y = add_114_y_0_to_fp16)[name = tensor("add_114_cast")]; + tensor sqrt_57_cast = sqrt(x = add_114_cast)[name = tensor("sqrt_57_cast")]; + tensor real_div_57_cast = real_div(x = sub_114_cast, y = sqrt_57_cast)[name = tensor("real_div_57_cast")]; + tensor reshape_229_shape_0 = const()[name = tensor("reshape_229_shape_0"), val = tensor([2, 640, 64, 64])]; + tensor reshape_229_cast = reshape(shape = reshape_229_shape_0, x = real_div_57_cast)[name = tensor("reshape_229_cast")]; + tensor add_115_gamma_0_to_fp16 = const()[name = tensor("add_115_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(645365376)))]; + tensor add_115_beta_0_to_fp16 = const()[name = tensor("add_115_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(645366720)))]; + tensor add_115_epsilon_0_to_fp16 = const()[name = tensor("add_115_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_115_cast = batch_norm(beta = add_115_beta_0_to_fp16, epsilon = add_115_epsilon_0_to_fp16, gamma = add_115_gamma_0_to_fp16, mean = add_15_mean_0_to_fp16, variance = add_15_variance_0_to_fp16, x = reshape_229_cast)[name = tensor("add_115_cast")]; + tensor input_505_cast = silu(x = add_115_cast)[name = tensor("input_505_cast")]; + tensor var_14860 = const()[name = tensor("op_14860"), val = tensor([1, 1])]; + tensor var_14862 = const()[name = tensor("op_14862"), val = tensor([1, 1])]; + tensor hidden_states_319_pad_type_0 = const()[name = tensor("hidden_states_319_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_319_pad_0 = const()[name = tensor("hidden_states_319_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor up_blocks_3_resnets_2_conv1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(645368064))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(646750528))), name = tensor("up_blocks_3_resnets_2_conv1_weight_to_fp16_palettized"), shape = tensor([320, 640, 3, 3])]; + tensor up_blocks_3_resnets_2_conv1_bias_to_fp16 = const()[name = tensor("up_blocks_3_resnets_2_conv1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(646750720)))]; + tensor hidden_states_319_cast = conv(bias = up_blocks_3_resnets_2_conv1_bias_to_fp16, dilations = var_14862, groups = var_12807, pad = hidden_states_319_pad_0, pad_type = hidden_states_319_pad_type_0, strides = var_14860, weight = up_blocks_3_resnets_2_conv1_weight_to_fp16_palettized, x = input_505_cast)[name = tensor("hidden_states_319_cast")]; + tensor var_14868 = const()[name = tensor("op_14868"), val = tensor([1, 1])]; + tensor var_14870 = const()[name = tensor("op_14870"), val = tensor([1, 1])]; + tensor temb_pad_type_0 = const()[name = tensor("temb_pad_type_0"), val = tensor("custom")]; + tensor temb_pad_0 = const()[name = tensor("temb_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_3_resnets_2_time_emb_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(646751424))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(647058688))), name = tensor("up_blocks_3_resnets_2_time_emb_proj_weight_to_fp16_palettized"), shape = tensor([320, 1280, 1, 1])]; + tensor up_blocks_3_resnets_2_time_emb_proj_bias_to_fp16 = const()[name = tensor("up_blocks_3_resnets_2_time_emb_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(647058880)))]; + tensor temb_cast = conv(bias = up_blocks_3_resnets_2_time_emb_proj_bias_to_fp16, dilations = var_14870, groups = var_12807, pad = temb_pad_0, pad_type = temb_pad_type_0, strides = var_14868, weight = up_blocks_3_resnets_2_time_emb_proj_weight_to_fp16_palettized, x = input_15_cast)[name = tensor("temb_cast")]; + tensor input_509_cast = add(x = hidden_states_319_cast, y = temb_cast)[name = tensor("input_509_cast")]; + tensor reshape_232_shape_0 = const()[name = tensor("reshape_232_shape_0"), val = tensor([2, 32, 10, 64, 64])]; + tensor reshape_232_cast = reshape(shape = reshape_232_shape_0, x = input_509_cast)[name = tensor("reshape_232_cast")]; + tensor reduce_mean_174_axes_0 = const()[name = tensor("reduce_mean_174_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_174_keep_dims_0 = const()[name = tensor("reduce_mean_174_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_174_cast = reduce_mean(axes = reduce_mean_174_axes_0, keep_dims = reduce_mean_174_keep_dims_0, x = reshape_232_cast)[name = tensor("reduce_mean_174_cast")]; + tensor sub_116_cast = sub(x = reshape_232_cast, y = reduce_mean_174_cast)[name = tensor("sub_116_cast")]; + tensor square_58_cast = square(x = sub_116_cast)[name = tensor("square_58_cast")]; + tensor reduce_mean_176_axes_0 = const()[name = tensor("reduce_mean_176_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_176_keep_dims_0 = const()[name = tensor("reduce_mean_176_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_176_cast = reduce_mean(axes = reduce_mean_176_axes_0, keep_dims = reduce_mean_176_keep_dims_0, x = square_58_cast)[name = tensor("reduce_mean_176_cast")]; + tensor add_116_y_0_to_fp16 = const()[name = tensor("add_116_y_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_116_cast = add(x = reduce_mean_176_cast, y = add_116_y_0_to_fp16)[name = tensor("add_116_cast")]; + tensor sqrt_58_cast = sqrt(x = add_116_cast)[name = tensor("sqrt_58_cast")]; + tensor real_div_58_cast = real_div(x = sub_116_cast, y = sqrt_58_cast)[name = tensor("real_div_58_cast")]; + tensor reshape_233_shape_0 = const()[name = tensor("reshape_233_shape_0"), val = tensor([2, 320, 64, 64])]; + tensor reshape_233_cast = reshape(shape = reshape_233_shape_0, x = real_div_58_cast)[name = tensor("reshape_233_cast")]; + tensor add_117_gamma_0_to_fp16 = const()[name = tensor("add_117_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(647059584)))]; + tensor add_117_beta_0_to_fp16 = const()[name = tensor("add_117_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(647060288)))]; + tensor add_117_epsilon_0_to_fp16 = const()[name = tensor("add_117_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_117_cast = batch_norm(beta = add_117_beta_0_to_fp16, epsilon = add_117_epsilon_0_to_fp16, gamma = add_117_gamma_0_to_fp16, mean = add_1_mean_0_to_fp16, variance = add_1_variance_0_to_fp16, x = reshape_233_cast)[name = tensor("add_117_cast")]; + tensor input_513_cast = silu(x = add_117_cast)[name = tensor("input_513_cast")]; + tensor var_14880 = const()[name = tensor("op_14880"), val = tensor([1, 1])]; + tensor var_14882 = const()[name = tensor("op_14882"), val = tensor([1, 1])]; + tensor hidden_states_321_pad_type_0 = const()[name = tensor("hidden_states_321_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_321_pad_0 = const()[name = tensor("hidden_states_321_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor up_blocks_3_resnets_2_conv2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(647060992))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(647752256))), name = tensor("up_blocks_3_resnets_2_conv2_weight_to_fp16_palettized"), shape = tensor([320, 320, 3, 3])]; + tensor up_blocks_3_resnets_2_conv2_bias_to_fp16 = const()[name = tensor("up_blocks_3_resnets_2_conv2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(647752448)))]; + tensor hidden_states_321_cast = conv(bias = up_blocks_3_resnets_2_conv2_bias_to_fp16, dilations = var_14882, groups = var_12807, pad = hidden_states_321_pad_0, pad_type = hidden_states_321_pad_type_0, strides = var_14880, weight = up_blocks_3_resnets_2_conv2_weight_to_fp16_palettized, x = input_513_cast)[name = tensor("hidden_states_321_cast")]; + tensor var_14887 = const()[name = tensor("op_14887"), val = tensor([1, 1])]; + tensor var_14889 = const()[name = tensor("op_14889"), val = tensor([1, 1])]; + tensor x_pad_type_0 = const()[name = tensor("x_pad_type_0"), val = tensor("custom")]; + tensor x_pad_0 = const()[name = tensor("x_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_3_resnets_2_conv_shortcut_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(647753152))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(647906816))), name = tensor("up_blocks_3_resnets_2_conv_shortcut_weight_to_fp16_palettized"), shape = tensor([320, 640, 1, 1])]; + tensor up_blocks_3_resnets_2_conv_shortcut_bias_to_fp16 = const()[name = tensor("up_blocks_3_resnets_2_conv_shortcut_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(647907008)))]; + tensor x_cast = conv(bias = up_blocks_3_resnets_2_conv_shortcut_bias_to_fp16, dilations = var_14889, groups = var_12807, pad = x_pad_0, pad_type = x_pad_type_0, strides = var_14887, weight = up_blocks_3_resnets_2_conv_shortcut_weight_to_fp16_palettized, x = input_501_cast)[name = tensor("x_cast")]; + tensor hidden_states_323_cast = add(x = x_cast, y = hidden_states_321_cast)[name = tensor("hidden_states_323_cast")]; + tensor reshape_236_shape_0 = const()[name = tensor("reshape_236_shape_0"), val = tensor([2, 32, 10, 64, 64])]; + tensor reshape_236_cast = reshape(shape = reshape_236_shape_0, x = hidden_states_323_cast)[name = tensor("reshape_236_cast")]; + tensor reduce_mean_177_axes_0 = const()[name = tensor("reduce_mean_177_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_177_keep_dims_0 = const()[name = tensor("reduce_mean_177_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_177_cast = reduce_mean(axes = reduce_mean_177_axes_0, keep_dims = reduce_mean_177_keep_dims_0, x = reshape_236_cast)[name = tensor("reduce_mean_177_cast")]; + tensor sub_118_cast = sub(x = reshape_236_cast, y = reduce_mean_177_cast)[name = tensor("sub_118_cast")]; + tensor square_59_cast = square(x = sub_118_cast)[name = tensor("square_59_cast")]; + tensor reduce_mean_179_axes_0 = const()[name = tensor("reduce_mean_179_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_179_keep_dims_0 = const()[name = tensor("reduce_mean_179_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_179_cast = reduce_mean(axes = reduce_mean_179_axes_0, keep_dims = reduce_mean_179_keep_dims_0, x = square_59_cast)[name = tensor("reduce_mean_179_cast")]; + tensor add_118_y_0_to_fp16 = const()[name = tensor("add_118_y_0_to_fp16"), val = tensor(0x1.1p-20)]; + tensor add_118_cast = add(x = reduce_mean_179_cast, y = add_118_y_0_to_fp16)[name = tensor("add_118_cast")]; + tensor sqrt_59_cast = sqrt(x = add_118_cast)[name = tensor("sqrt_59_cast")]; + tensor real_div_59_cast = real_div(x = sub_118_cast, y = sqrt_59_cast)[name = tensor("real_div_59_cast")]; + tensor reshape_237_shape_0 = const()[name = tensor("reshape_237_shape_0"), val = tensor([2, 320, 64, 64])]; + tensor reshape_237_cast = reshape(shape = reshape_237_shape_0, x = real_div_59_cast)[name = tensor("reshape_237_cast")]; + tensor add_119_gamma_0_to_fp16 = const()[name = tensor("add_119_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(647907712)))]; + tensor add_119_beta_0_to_fp16 = const()[name = tensor("add_119_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(647908416)))]; + tensor add_119_epsilon_0_to_fp16 = const()[name = tensor("add_119_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_119_cast = batch_norm(beta = add_119_beta_0_to_fp16, epsilon = add_119_epsilon_0_to_fp16, gamma = add_119_gamma_0_to_fp16, mean = add_1_mean_0_to_fp16, variance = add_1_variance_0_to_fp16, x = reshape_237_cast)[name = tensor("add_119_cast")]; + tensor var_14909 = const()[name = tensor("op_14909"), val = tensor([1, 1])]; + tensor var_14911 = const()[name = tensor("op_14911"), val = tensor([1, 1])]; + tensor hidden_states_325_pad_type_0 = const()[name = tensor("hidden_states_325_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_325_pad_0 = const()[name = tensor("hidden_states_325_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_3_attentions_2_proj_in_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(647909120))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(647985984))), name = tensor("up_blocks_3_attentions_2_proj_in_weight_to_fp16_palettized"), shape = tensor([320, 320, 1, 1])]; + tensor up_blocks_3_attentions_2_proj_in_bias_to_fp16 = const()[name = tensor("up_blocks_3_attentions_2_proj_in_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(647986176)))]; + tensor hidden_states_325_cast = conv(bias = up_blocks_3_attentions_2_proj_in_bias_to_fp16, dilations = var_14911, groups = var_12807, pad = hidden_states_325_pad_0, pad_type = hidden_states_325_pad_type_0, strides = var_14909, weight = up_blocks_3_attentions_2_proj_in_weight_to_fp16_palettized, x = add_119_cast)[name = tensor("hidden_states_325_cast")]; + tensor var_14916 = const()[name = tensor("op_14916"), val = tensor([2, 320, 1, 4096])]; + tensor inputs_91_cast = reshape(shape = var_14916, x = hidden_states_325_cast)[name = tensor("inputs_91_cast")]; + tensor var_14926 = const()[name = tensor("op_14926"), val = tensor([1])]; + tensor channels_mean_91_cast = reduce_mean(axes = var_14926, keep_dims = var_12802, x = inputs_91_cast)[name = tensor("channels_mean_91_cast")]; + tensor zero_mean_91_cast = sub(x = inputs_91_cast, y = channels_mean_91_cast)[name = tensor("zero_mean_91_cast")]; + tensor zero_mean_sq_91_cast = mul(x = zero_mean_91_cast, y = zero_mean_91_cast)[name = tensor("zero_mean_sq_91_cast")]; + tensor var_14930 = const()[name = tensor("op_14930"), val = tensor([1])]; + tensor var_14931_cast = reduce_mean(axes = var_14930, keep_dims = var_12802, x = zero_mean_sq_91_cast)[name = tensor("op_14931_cast")]; + tensor var_14932_to_fp16 = const()[name = tensor("op_14932_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_14933_cast = add(x = var_14931_cast, y = var_14932_to_fp16)[name = tensor("op_14933_cast")]; + tensor denom_91_epsilon_0_to_fp16 = const()[name = tensor("denom_91_epsilon_0_to_fp16"), val = tensor(0x1p-24)]; + tensor denom_91_cast = rsqrt(epsilon = denom_91_epsilon_0_to_fp16, x = var_14933_cast)[name = tensor("denom_91_cast")]; + tensor out_91_cast = mul(x = zero_mean_91_cast, y = denom_91_cast)[name = tensor("out_91_cast")]; + tensor var_14937_to_fp16 = const()[name = tensor("op_14937_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(647986880)))]; + tensor var_14938_cast = add(x = out_91_cast, y = var_14937_to_fp16)[name = tensor("op_14938_cast")]; + tensor var_14940_to_fp16 = const()[name = tensor("op_14940_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(647987584)))]; + tensor hidden_states_327_cast = mul(x = var_14938_cast, y = var_14940_to_fp16)[name = tensor("hidden_states_327_cast")]; + tensor var_14947 = const()[name = tensor("op_14947"), val = tensor([1, 1])]; + tensor var_14949 = const()[name = tensor("op_14949"), val = tensor([1, 1])]; + tensor q_61_pad_type_0 = const()[name = tensor("q_61_pad_type_0"), val = tensor("custom")]; + tensor q_61_pad_0 = const()[name = tensor("q_61_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_3_attentions_2_transformer_blocks_0_attn1_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(647988288))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(648065152))), name = tensor("up_blocks_3_attentions_2_transformer_blocks_0_attn1_to_q_weight_to_fp16_palettized"), shape = tensor([320, 320, 1, 1])]; + tensor q_61_cast = conv(dilations = var_14949, groups = var_12807, pad = q_61_pad_0, pad_type = q_61_pad_type_0, strides = var_14947, weight = up_blocks_3_attentions_2_transformer_blocks_0_attn1_to_q_weight_to_fp16_palettized, x = hidden_states_327_cast)[name = tensor("q_61_cast")]; + tensor var_14953 = const()[name = tensor("op_14953"), val = tensor([1, 1])]; + tensor var_14955 = const()[name = tensor("op_14955"), val = tensor([1, 1])]; + tensor k_121_pad_type_0 = const()[name = tensor("k_121_pad_type_0"), val = tensor("custom")]; + tensor k_121_pad_0 = const()[name = tensor("k_121_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_3_attentions_2_transformer_blocks_0_attn1_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(648065344))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(648142208))), name = tensor("up_blocks_3_attentions_2_transformer_blocks_0_attn1_to_k_weight_to_fp16_palettized"), shape = tensor([320, 320, 1, 1])]; + tensor k_121_cast = conv(dilations = var_14955, groups = var_12807, pad = k_121_pad_0, pad_type = k_121_pad_type_0, strides = var_14953, weight = up_blocks_3_attentions_2_transformer_blocks_0_attn1_to_k_weight_to_fp16_palettized, x = hidden_states_327_cast)[name = tensor("k_121_cast")]; + tensor var_14959 = const()[name = tensor("op_14959"), val = tensor([1, 1])]; + tensor var_14961 = const()[name = tensor("op_14961"), val = tensor([1, 1])]; + tensor v_61_pad_type_0 = const()[name = tensor("v_61_pad_type_0"), val = tensor("custom")]; + tensor v_61_pad_0 = const()[name = tensor("v_61_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_3_attentions_2_transformer_blocks_0_attn1_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(648142400))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(648219264))), name = tensor("up_blocks_3_attentions_2_transformer_blocks_0_attn1_to_v_weight_to_fp16_palettized"), shape = tensor([320, 320, 1, 1])]; + tensor v_61_cast = conv(dilations = var_14961, groups = var_12807, pad = v_61_pad_0, pad_type = v_61_pad_type_0, strides = var_14959, weight = up_blocks_3_attentions_2_transformer_blocks_0_attn1_to_v_weight_to_fp16_palettized, x = hidden_states_327_cast)[name = tensor("v_61_cast")]; + tensor var_14965_begin_0 = const()[name = tensor("op_14965_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_14965_end_0 = const()[name = tensor("op_14965_end_0"), val = tensor([2, 64, 1, 4096])]; + tensor var_14965_end_mask_0 = const()[name = tensor("op_14965_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14965_cast = slice_by_index(begin = var_14965_begin_0, end = var_14965_end_0, end_mask = var_14965_end_mask_0, x = q_61_cast)[name = tensor("op_14965_cast")]; + tensor var_14969_begin_0 = const()[name = tensor("op_14969_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_14969_end_0 = const()[name = tensor("op_14969_end_0"), val = tensor([2, 128, 1, 4096])]; + tensor var_14969_end_mask_0 = const()[name = tensor("op_14969_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14969_cast = slice_by_index(begin = var_14969_begin_0, end = var_14969_end_0, end_mask = var_14969_end_mask_0, x = q_61_cast)[name = tensor("op_14969_cast")]; + tensor var_14973_begin_0 = const()[name = tensor("op_14973_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_14973_end_0 = const()[name = tensor("op_14973_end_0"), val = tensor([2, 192, 1, 4096])]; + tensor var_14973_end_mask_0 = const()[name = tensor("op_14973_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14973_cast = slice_by_index(begin = var_14973_begin_0, end = var_14973_end_0, end_mask = var_14973_end_mask_0, x = q_61_cast)[name = tensor("op_14973_cast")]; + tensor var_14977_begin_0 = const()[name = tensor("op_14977_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_14977_end_0 = const()[name = tensor("op_14977_end_0"), val = tensor([2, 256, 1, 4096])]; + tensor var_14977_end_mask_0 = const()[name = tensor("op_14977_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14977_cast = slice_by_index(begin = var_14977_begin_0, end = var_14977_end_0, end_mask = var_14977_end_mask_0, x = q_61_cast)[name = tensor("op_14977_cast")]; + tensor var_14981_begin_0 = const()[name = tensor("op_14981_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_14981_end_0 = const()[name = tensor("op_14981_end_0"), val = tensor([2, 320, 1, 4096])]; + tensor var_14981_end_mask_0 = const()[name = tensor("op_14981_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14981_cast = slice_by_index(begin = var_14981_begin_0, end = var_14981_end_0, end_mask = var_14981_end_mask_0, x = q_61_cast)[name = tensor("op_14981_cast")]; + tensor var_14984_begin_0 = const()[name = tensor("op_14984_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_14984_end_0 = const()[name = tensor("op_14984_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_14984_end_mask_0 = const()[name = tensor("op_14984_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14984_cast = slice_by_index(begin = var_14984_begin_0, end = var_14984_end_0, end_mask = var_14984_end_mask_0, x = var_14965_cast)[name = tensor("op_14984_cast")]; + tensor var_14985_begin_0 = const()[name = tensor("op_14985_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_14985_end_0 = const()[name = tensor("op_14985_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_14985_end_mask_0 = const()[name = tensor("op_14985_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14985_cast = slice_by_index(begin = var_14985_begin_0, end = var_14985_end_0, end_mask = var_14985_end_mask_0, x = var_14965_cast)[name = tensor("op_14985_cast")]; + tensor var_14986_begin_0 = const()[name = tensor("op_14986_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_14986_end_0 = const()[name = tensor("op_14986_end_0"), val = tensor([2, 64, 1, 1536])]; + tensor var_14986_end_mask_0 = const()[name = tensor("op_14986_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14986_cast = slice_by_index(begin = var_14986_begin_0, end = var_14986_end_0, end_mask = var_14986_end_mask_0, x = var_14965_cast)[name = tensor("op_14986_cast")]; + tensor var_14987_begin_0 = const()[name = tensor("op_14987_begin_0"), val = tensor([0, 0, 0, 1536])]; + tensor var_14987_end_0 = const()[name = tensor("op_14987_end_0"), val = tensor([2, 64, 1, 2048])]; + tensor var_14987_end_mask_0 = const()[name = tensor("op_14987_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14987_cast = slice_by_index(begin = var_14987_begin_0, end = var_14987_end_0, end_mask = var_14987_end_mask_0, x = var_14965_cast)[name = tensor("op_14987_cast")]; + tensor var_14988_begin_0 = const()[name = tensor("op_14988_begin_0"), val = tensor([0, 0, 0, 2048])]; + tensor var_14988_end_0 = const()[name = tensor("op_14988_end_0"), val = tensor([2, 64, 1, 2560])]; + tensor var_14988_end_mask_0 = const()[name = tensor("op_14988_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14988_cast = slice_by_index(begin = var_14988_begin_0, end = var_14988_end_0, end_mask = var_14988_end_mask_0, x = var_14965_cast)[name = tensor("op_14988_cast")]; + tensor var_14989_begin_0 = const()[name = tensor("op_14989_begin_0"), val = tensor([0, 0, 0, 2560])]; + tensor var_14989_end_0 = const()[name = tensor("op_14989_end_0"), val = tensor([2, 64, 1, 3072])]; + tensor var_14989_end_mask_0 = const()[name = tensor("op_14989_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14989_cast = slice_by_index(begin = var_14989_begin_0, end = var_14989_end_0, end_mask = var_14989_end_mask_0, x = var_14965_cast)[name = tensor("op_14989_cast")]; + tensor var_14990_begin_0 = const()[name = tensor("op_14990_begin_0"), val = tensor([0, 0, 0, 3072])]; + tensor var_14990_end_0 = const()[name = tensor("op_14990_end_0"), val = tensor([2, 64, 1, 3584])]; + tensor var_14990_end_mask_0 = const()[name = tensor("op_14990_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14990_cast = slice_by_index(begin = var_14990_begin_0, end = var_14990_end_0, end_mask = var_14990_end_mask_0, x = var_14965_cast)[name = tensor("op_14990_cast")]; + tensor var_14991_begin_0 = const()[name = tensor("op_14991_begin_0"), val = tensor([0, 0, 0, 3584])]; + tensor var_14991_end_0 = const()[name = tensor("op_14991_end_0"), val = tensor([2, 64, 1, 4096])]; + tensor var_14991_end_mask_0 = const()[name = tensor("op_14991_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14991_cast = slice_by_index(begin = var_14991_begin_0, end = var_14991_end_0, end_mask = var_14991_end_mask_0, x = var_14965_cast)[name = tensor("op_14991_cast")]; + tensor var_14992_begin_0 = const()[name = tensor("op_14992_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_14992_end_0 = const()[name = tensor("op_14992_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_14992_end_mask_0 = const()[name = tensor("op_14992_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14992_cast = slice_by_index(begin = var_14992_begin_0, end = var_14992_end_0, end_mask = var_14992_end_mask_0, x = var_14969_cast)[name = tensor("op_14992_cast")]; + tensor var_14993_begin_0 = const()[name = tensor("op_14993_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_14993_end_0 = const()[name = tensor("op_14993_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_14993_end_mask_0 = const()[name = tensor("op_14993_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14993_cast = slice_by_index(begin = var_14993_begin_0, end = var_14993_end_0, end_mask = var_14993_end_mask_0, x = var_14969_cast)[name = tensor("op_14993_cast")]; + tensor var_14994_begin_0 = const()[name = tensor("op_14994_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_14994_end_0 = const()[name = tensor("op_14994_end_0"), val = tensor([2, 64, 1, 1536])]; + tensor var_14994_end_mask_0 = const()[name = tensor("op_14994_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14994_cast = slice_by_index(begin = var_14994_begin_0, end = var_14994_end_0, end_mask = var_14994_end_mask_0, x = var_14969_cast)[name = tensor("op_14994_cast")]; + tensor var_14995_begin_0 = const()[name = tensor("op_14995_begin_0"), val = tensor([0, 0, 0, 1536])]; + tensor var_14995_end_0 = const()[name = tensor("op_14995_end_0"), val = tensor([2, 64, 1, 2048])]; + tensor var_14995_end_mask_0 = const()[name = tensor("op_14995_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14995_cast = slice_by_index(begin = var_14995_begin_0, end = var_14995_end_0, end_mask = var_14995_end_mask_0, x = var_14969_cast)[name = tensor("op_14995_cast")]; + tensor var_14996_begin_0 = const()[name = tensor("op_14996_begin_0"), val = tensor([0, 0, 0, 2048])]; + tensor var_14996_end_0 = const()[name = tensor("op_14996_end_0"), val = tensor([2, 64, 1, 2560])]; + tensor var_14996_end_mask_0 = const()[name = tensor("op_14996_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14996_cast = slice_by_index(begin = var_14996_begin_0, end = var_14996_end_0, end_mask = var_14996_end_mask_0, x = var_14969_cast)[name = tensor("op_14996_cast")]; + tensor var_14997_begin_0 = const()[name = tensor("op_14997_begin_0"), val = tensor([0, 0, 0, 2560])]; + tensor var_14997_end_0 = const()[name = tensor("op_14997_end_0"), val = tensor([2, 64, 1, 3072])]; + tensor var_14997_end_mask_0 = const()[name = tensor("op_14997_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14997_cast = slice_by_index(begin = var_14997_begin_0, end = var_14997_end_0, end_mask = var_14997_end_mask_0, x = var_14969_cast)[name = tensor("op_14997_cast")]; + tensor var_14998_begin_0 = const()[name = tensor("op_14998_begin_0"), val = tensor([0, 0, 0, 3072])]; + tensor var_14998_end_0 = const()[name = tensor("op_14998_end_0"), val = tensor([2, 64, 1, 3584])]; + tensor var_14998_end_mask_0 = const()[name = tensor("op_14998_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14998_cast = slice_by_index(begin = var_14998_begin_0, end = var_14998_end_0, end_mask = var_14998_end_mask_0, x = var_14969_cast)[name = tensor("op_14998_cast")]; + tensor var_14999_begin_0 = const()[name = tensor("op_14999_begin_0"), val = tensor([0, 0, 0, 3584])]; + tensor var_14999_end_0 = const()[name = tensor("op_14999_end_0"), val = tensor([2, 64, 1, 4096])]; + tensor var_14999_end_mask_0 = const()[name = tensor("op_14999_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14999_cast = slice_by_index(begin = var_14999_begin_0, end = var_14999_end_0, end_mask = var_14999_end_mask_0, x = var_14969_cast)[name = tensor("op_14999_cast")]; + tensor var_15000_begin_0 = const()[name = tensor("op_15000_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_15000_end_0 = const()[name = tensor("op_15000_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_15000_end_mask_0 = const()[name = tensor("op_15000_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15000_cast = slice_by_index(begin = var_15000_begin_0, end = var_15000_end_0, end_mask = var_15000_end_mask_0, x = var_14973_cast)[name = tensor("op_15000_cast")]; + tensor var_15001_begin_0 = const()[name = tensor("op_15001_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_15001_end_0 = const()[name = tensor("op_15001_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_15001_end_mask_0 = const()[name = tensor("op_15001_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15001_cast = slice_by_index(begin = var_15001_begin_0, end = var_15001_end_0, end_mask = var_15001_end_mask_0, x = var_14973_cast)[name = tensor("op_15001_cast")]; + tensor var_15002_begin_0 = const()[name = tensor("op_15002_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_15002_end_0 = const()[name = tensor("op_15002_end_0"), val = tensor([2, 64, 1, 1536])]; + tensor var_15002_end_mask_0 = const()[name = tensor("op_15002_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15002_cast = slice_by_index(begin = var_15002_begin_0, end = var_15002_end_0, end_mask = var_15002_end_mask_0, x = var_14973_cast)[name = tensor("op_15002_cast")]; + tensor var_15003_begin_0 = const()[name = tensor("op_15003_begin_0"), val = tensor([0, 0, 0, 1536])]; + tensor var_15003_end_0 = const()[name = tensor("op_15003_end_0"), val = tensor([2, 64, 1, 2048])]; + tensor var_15003_end_mask_0 = const()[name = tensor("op_15003_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15003_cast = slice_by_index(begin = var_15003_begin_0, end = var_15003_end_0, end_mask = var_15003_end_mask_0, x = var_14973_cast)[name = tensor("op_15003_cast")]; + tensor var_15004_begin_0 = const()[name = tensor("op_15004_begin_0"), val = tensor([0, 0, 0, 2048])]; + tensor var_15004_end_0 = const()[name = tensor("op_15004_end_0"), val = tensor([2, 64, 1, 2560])]; + tensor var_15004_end_mask_0 = const()[name = tensor("op_15004_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15004_cast = slice_by_index(begin = var_15004_begin_0, end = var_15004_end_0, end_mask = var_15004_end_mask_0, x = var_14973_cast)[name = tensor("op_15004_cast")]; + tensor var_15005_begin_0 = const()[name = tensor("op_15005_begin_0"), val = tensor([0, 0, 0, 2560])]; + tensor var_15005_end_0 = const()[name = tensor("op_15005_end_0"), val = tensor([2, 64, 1, 3072])]; + tensor var_15005_end_mask_0 = const()[name = tensor("op_15005_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15005_cast = slice_by_index(begin = var_15005_begin_0, end = var_15005_end_0, end_mask = var_15005_end_mask_0, x = var_14973_cast)[name = tensor("op_15005_cast")]; + tensor var_15006_begin_0 = const()[name = tensor("op_15006_begin_0"), val = tensor([0, 0, 0, 3072])]; + tensor var_15006_end_0 = const()[name = tensor("op_15006_end_0"), val = tensor([2, 64, 1, 3584])]; + tensor var_15006_end_mask_0 = const()[name = tensor("op_15006_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15006_cast = slice_by_index(begin = var_15006_begin_0, end = var_15006_end_0, end_mask = var_15006_end_mask_0, x = var_14973_cast)[name = tensor("op_15006_cast")]; + tensor var_15007_begin_0 = const()[name = tensor("op_15007_begin_0"), val = tensor([0, 0, 0, 3584])]; + tensor var_15007_end_0 = const()[name = tensor("op_15007_end_0"), val = tensor([2, 64, 1, 4096])]; + tensor var_15007_end_mask_0 = const()[name = tensor("op_15007_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15007_cast = slice_by_index(begin = var_15007_begin_0, end = var_15007_end_0, end_mask = var_15007_end_mask_0, x = var_14973_cast)[name = tensor("op_15007_cast")]; + tensor var_15008_begin_0 = const()[name = tensor("op_15008_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_15008_end_0 = const()[name = tensor("op_15008_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_15008_end_mask_0 = const()[name = tensor("op_15008_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15008_cast = slice_by_index(begin = var_15008_begin_0, end = var_15008_end_0, end_mask = var_15008_end_mask_0, x = var_14977_cast)[name = tensor("op_15008_cast")]; + tensor var_15009_begin_0 = const()[name = tensor("op_15009_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_15009_end_0 = const()[name = tensor("op_15009_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_15009_end_mask_0 = const()[name = tensor("op_15009_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15009_cast = slice_by_index(begin = var_15009_begin_0, end = var_15009_end_0, end_mask = var_15009_end_mask_0, x = var_14977_cast)[name = tensor("op_15009_cast")]; + tensor var_15010_begin_0 = const()[name = tensor("op_15010_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_15010_end_0 = const()[name = tensor("op_15010_end_0"), val = tensor([2, 64, 1, 1536])]; + tensor var_15010_end_mask_0 = const()[name = tensor("op_15010_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15010_cast = slice_by_index(begin = var_15010_begin_0, end = var_15010_end_0, end_mask = var_15010_end_mask_0, x = var_14977_cast)[name = tensor("op_15010_cast")]; + tensor var_15011_begin_0 = const()[name = tensor("op_15011_begin_0"), val = tensor([0, 0, 0, 1536])]; + tensor var_15011_end_0 = const()[name = tensor("op_15011_end_0"), val = tensor([2, 64, 1, 2048])]; + tensor var_15011_end_mask_0 = const()[name = tensor("op_15011_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15011_cast = slice_by_index(begin = var_15011_begin_0, end = var_15011_end_0, end_mask = var_15011_end_mask_0, x = var_14977_cast)[name = tensor("op_15011_cast")]; + tensor var_15012_begin_0 = const()[name = tensor("op_15012_begin_0"), val = tensor([0, 0, 0, 2048])]; + tensor var_15012_end_0 = const()[name = tensor("op_15012_end_0"), val = tensor([2, 64, 1, 2560])]; + tensor var_15012_end_mask_0 = const()[name = tensor("op_15012_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15012_cast = slice_by_index(begin = var_15012_begin_0, end = var_15012_end_0, end_mask = var_15012_end_mask_0, x = var_14977_cast)[name = tensor("op_15012_cast")]; + tensor var_15013_begin_0 = const()[name = tensor("op_15013_begin_0"), val = tensor([0, 0, 0, 2560])]; + tensor var_15013_end_0 = const()[name = tensor("op_15013_end_0"), val = tensor([2, 64, 1, 3072])]; + tensor var_15013_end_mask_0 = const()[name = tensor("op_15013_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15013_cast = slice_by_index(begin = var_15013_begin_0, end = var_15013_end_0, end_mask = var_15013_end_mask_0, x = var_14977_cast)[name = tensor("op_15013_cast")]; + tensor var_15014_begin_0 = const()[name = tensor("op_15014_begin_0"), val = tensor([0, 0, 0, 3072])]; + tensor var_15014_end_0 = const()[name = tensor("op_15014_end_0"), val = tensor([2, 64, 1, 3584])]; + tensor var_15014_end_mask_0 = const()[name = tensor("op_15014_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15014_cast = slice_by_index(begin = var_15014_begin_0, end = var_15014_end_0, end_mask = var_15014_end_mask_0, x = var_14977_cast)[name = tensor("op_15014_cast")]; + tensor var_15015_begin_0 = const()[name = tensor("op_15015_begin_0"), val = tensor([0, 0, 0, 3584])]; + tensor var_15015_end_0 = const()[name = tensor("op_15015_end_0"), val = tensor([2, 64, 1, 4096])]; + tensor var_15015_end_mask_0 = const()[name = tensor("op_15015_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15015_cast = slice_by_index(begin = var_15015_begin_0, end = var_15015_end_0, end_mask = var_15015_end_mask_0, x = var_14977_cast)[name = tensor("op_15015_cast")]; + tensor var_15016_begin_0 = const()[name = tensor("op_15016_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_15016_end_0 = const()[name = tensor("op_15016_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_15016_end_mask_0 = const()[name = tensor("op_15016_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15016_cast = slice_by_index(begin = var_15016_begin_0, end = var_15016_end_0, end_mask = var_15016_end_mask_0, x = var_14981_cast)[name = tensor("op_15016_cast")]; + tensor var_15017_begin_0 = const()[name = tensor("op_15017_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_15017_end_0 = const()[name = tensor("op_15017_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_15017_end_mask_0 = const()[name = tensor("op_15017_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15017_cast = slice_by_index(begin = var_15017_begin_0, end = var_15017_end_0, end_mask = var_15017_end_mask_0, x = var_14981_cast)[name = tensor("op_15017_cast")]; + tensor var_15018_begin_0 = const()[name = tensor("op_15018_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_15018_end_0 = const()[name = tensor("op_15018_end_0"), val = tensor([2, 64, 1, 1536])]; + tensor var_15018_end_mask_0 = const()[name = tensor("op_15018_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15018_cast = slice_by_index(begin = var_15018_begin_0, end = var_15018_end_0, end_mask = var_15018_end_mask_0, x = var_14981_cast)[name = tensor("op_15018_cast")]; + tensor var_15019_begin_0 = const()[name = tensor("op_15019_begin_0"), val = tensor([0, 0, 0, 1536])]; + tensor var_15019_end_0 = const()[name = tensor("op_15019_end_0"), val = tensor([2, 64, 1, 2048])]; + tensor var_15019_end_mask_0 = const()[name = tensor("op_15019_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15019_cast = slice_by_index(begin = var_15019_begin_0, end = var_15019_end_0, end_mask = var_15019_end_mask_0, x = var_14981_cast)[name = tensor("op_15019_cast")]; + tensor var_15020_begin_0 = const()[name = tensor("op_15020_begin_0"), val = tensor([0, 0, 0, 2048])]; + tensor var_15020_end_0 = const()[name = tensor("op_15020_end_0"), val = tensor([2, 64, 1, 2560])]; + tensor var_15020_end_mask_0 = const()[name = tensor("op_15020_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15020_cast = slice_by_index(begin = var_15020_begin_0, end = var_15020_end_0, end_mask = var_15020_end_mask_0, x = var_14981_cast)[name = tensor("op_15020_cast")]; + tensor var_15021_begin_0 = const()[name = tensor("op_15021_begin_0"), val = tensor([0, 0, 0, 2560])]; + tensor var_15021_end_0 = const()[name = tensor("op_15021_end_0"), val = tensor([2, 64, 1, 3072])]; + tensor var_15021_end_mask_0 = const()[name = tensor("op_15021_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15021_cast = slice_by_index(begin = var_15021_begin_0, end = var_15021_end_0, end_mask = var_15021_end_mask_0, x = var_14981_cast)[name = tensor("op_15021_cast")]; + tensor var_15022_begin_0 = const()[name = tensor("op_15022_begin_0"), val = tensor([0, 0, 0, 3072])]; + tensor var_15022_end_0 = const()[name = tensor("op_15022_end_0"), val = tensor([2, 64, 1, 3584])]; + tensor var_15022_end_mask_0 = const()[name = tensor("op_15022_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15022_cast = slice_by_index(begin = var_15022_begin_0, end = var_15022_end_0, end_mask = var_15022_end_mask_0, x = var_14981_cast)[name = tensor("op_15022_cast")]; + tensor var_15023_begin_0 = const()[name = tensor("op_15023_begin_0"), val = tensor([0, 0, 0, 3584])]; + tensor var_15023_end_0 = const()[name = tensor("op_15023_end_0"), val = tensor([2, 64, 1, 4096])]; + tensor var_15023_end_mask_0 = const()[name = tensor("op_15023_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15023_cast = slice_by_index(begin = var_15023_begin_0, end = var_15023_end_0, end_mask = var_15023_end_mask_0, x = var_14981_cast)[name = tensor("op_15023_cast")]; + tensor k_123_perm_0 = const()[name = tensor("k_123_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_15028_begin_0 = const()[name = tensor("op_15028_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_15028_end_0 = const()[name = tensor("op_15028_end_0"), val = tensor([2, 4096, 1, 64])]; + tensor var_15028_end_mask_0 = const()[name = tensor("op_15028_end_mask_0"), val = tensor([true, true, true, false])]; + tensor transpose_1 = transpose(perm = k_123_perm_0, x = k_121_cast)[name = tensor("transpose_1")]; + tensor var_15028_cast = slice_by_index(begin = var_15028_begin_0, end = var_15028_end_0, end_mask = var_15028_end_mask_0, x = transpose_1)[name = tensor("op_15028_cast")]; + tensor var_15032_begin_0 = const()[name = tensor("op_15032_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_15032_end_0 = const()[name = tensor("op_15032_end_0"), val = tensor([2, 4096, 1, 128])]; + tensor var_15032_end_mask_0 = const()[name = tensor("op_15032_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15032_cast = slice_by_index(begin = var_15032_begin_0, end = var_15032_end_0, end_mask = var_15032_end_mask_0, x = transpose_1)[name = tensor("op_15032_cast")]; + tensor var_15036_begin_0 = const()[name = tensor("op_15036_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_15036_end_0 = const()[name = tensor("op_15036_end_0"), val = tensor([2, 4096, 1, 192])]; + tensor var_15036_end_mask_0 = const()[name = tensor("op_15036_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15036_cast = slice_by_index(begin = var_15036_begin_0, end = var_15036_end_0, end_mask = var_15036_end_mask_0, x = transpose_1)[name = tensor("op_15036_cast")]; + tensor var_15040_begin_0 = const()[name = tensor("op_15040_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_15040_end_0 = const()[name = tensor("op_15040_end_0"), val = tensor([2, 4096, 1, 256])]; + tensor var_15040_end_mask_0 = const()[name = tensor("op_15040_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15040_cast = slice_by_index(begin = var_15040_begin_0, end = var_15040_end_0, end_mask = var_15040_end_mask_0, x = transpose_1)[name = tensor("op_15040_cast")]; + tensor var_15044_begin_0 = const()[name = tensor("op_15044_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_15044_end_0 = const()[name = tensor("op_15044_end_0"), val = tensor([2, 4096, 1, 320])]; + tensor var_15044_end_mask_0 = const()[name = tensor("op_15044_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15044_cast = slice_by_index(begin = var_15044_begin_0, end = var_15044_end_0, end_mask = var_15044_end_mask_0, x = transpose_1)[name = tensor("op_15044_cast")]; + tensor var_15046_begin_0 = const()[name = tensor("op_15046_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_15046_end_0 = const()[name = tensor("op_15046_end_0"), val = tensor([2, 64, 1, 4096])]; + tensor var_15046_end_mask_0 = const()[name = tensor("op_15046_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15046_cast = slice_by_index(begin = var_15046_begin_0, end = var_15046_end_0, end_mask = var_15046_end_mask_0, x = v_61_cast)[name = tensor("op_15046_cast")]; + tensor var_15050_begin_0 = const()[name = tensor("op_15050_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_15050_end_0 = const()[name = tensor("op_15050_end_0"), val = tensor([2, 128, 1, 4096])]; + tensor var_15050_end_mask_0 = const()[name = tensor("op_15050_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15050_cast = slice_by_index(begin = var_15050_begin_0, end = var_15050_end_0, end_mask = var_15050_end_mask_0, x = v_61_cast)[name = tensor("op_15050_cast")]; + tensor var_15054_begin_0 = const()[name = tensor("op_15054_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_15054_end_0 = const()[name = tensor("op_15054_end_0"), val = tensor([2, 192, 1, 4096])]; + tensor var_15054_end_mask_0 = const()[name = tensor("op_15054_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15054_cast = slice_by_index(begin = var_15054_begin_0, end = var_15054_end_0, end_mask = var_15054_end_mask_0, x = v_61_cast)[name = tensor("op_15054_cast")]; + tensor var_15058_begin_0 = const()[name = tensor("op_15058_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_15058_end_0 = const()[name = tensor("op_15058_end_0"), val = tensor([2, 256, 1, 4096])]; + tensor var_15058_end_mask_0 = const()[name = tensor("op_15058_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15058_cast = slice_by_index(begin = var_15058_begin_0, end = var_15058_end_0, end_mask = var_15058_end_mask_0, x = v_61_cast)[name = tensor("op_15058_cast")]; + tensor var_15062_begin_0 = const()[name = tensor("op_15062_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_15062_end_0 = const()[name = tensor("op_15062_end_0"), val = tensor([2, 320, 1, 4096])]; + tensor var_15062_end_mask_0 = const()[name = tensor("op_15062_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15062_cast = slice_by_index(begin = var_15062_begin_0, end = var_15062_end_0, end_mask = var_15062_end_mask_0, x = v_61_cast)[name = tensor("op_15062_cast")]; + tensor var_15066_equation_0 = const()[name = tensor("op_15066_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15066_cast = einsum(equation = var_15066_equation_0, values = (var_15028_cast, var_14984_cast))[name = tensor("op_15066_cast")]; + tensor var_15067_to_fp16 = const()[name = tensor("op_15067_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1041_cast = mul(x = var_15066_cast, y = var_15067_to_fp16)[name = tensor("aw_chunk_1041_cast")]; + tensor var_15070_equation_0 = const()[name = tensor("op_15070_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15070_cast = einsum(equation = var_15070_equation_0, values = (var_15028_cast, var_14985_cast))[name = tensor("op_15070_cast")]; + tensor var_15071_to_fp16 = const()[name = tensor("op_15071_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1043_cast = mul(x = var_15070_cast, y = var_15071_to_fp16)[name = tensor("aw_chunk_1043_cast")]; + tensor var_15074_equation_0 = const()[name = tensor("op_15074_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15074_cast = einsum(equation = var_15074_equation_0, values = (var_15028_cast, var_14986_cast))[name = tensor("op_15074_cast")]; + tensor var_15075_to_fp16 = const()[name = tensor("op_15075_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1045_cast = mul(x = var_15074_cast, y = var_15075_to_fp16)[name = tensor("aw_chunk_1045_cast")]; + tensor var_15078_equation_0 = const()[name = tensor("op_15078_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15078_cast = einsum(equation = var_15078_equation_0, values = (var_15028_cast, var_14987_cast))[name = tensor("op_15078_cast")]; + tensor var_15079_to_fp16 = const()[name = tensor("op_15079_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1047_cast = mul(x = var_15078_cast, y = var_15079_to_fp16)[name = tensor("aw_chunk_1047_cast")]; + tensor var_15082_equation_0 = const()[name = tensor("op_15082_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15082_cast = einsum(equation = var_15082_equation_0, values = (var_15028_cast, var_14988_cast))[name = tensor("op_15082_cast")]; + tensor var_15083_to_fp16 = const()[name = tensor("op_15083_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1049_cast = mul(x = var_15082_cast, y = var_15083_to_fp16)[name = tensor("aw_chunk_1049_cast")]; + tensor var_15086_equation_0 = const()[name = tensor("op_15086_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15086_cast = einsum(equation = var_15086_equation_0, values = (var_15028_cast, var_14989_cast))[name = tensor("op_15086_cast")]; + tensor var_15087_to_fp16 = const()[name = tensor("op_15087_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1051_cast = mul(x = var_15086_cast, y = var_15087_to_fp16)[name = tensor("aw_chunk_1051_cast")]; + tensor var_15090_equation_0 = const()[name = tensor("op_15090_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15090_cast = einsum(equation = var_15090_equation_0, values = (var_15028_cast, var_14990_cast))[name = tensor("op_15090_cast")]; + tensor var_15091_to_fp16 = const()[name = tensor("op_15091_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1053_cast = mul(x = var_15090_cast, y = var_15091_to_fp16)[name = tensor("aw_chunk_1053_cast")]; + tensor var_15094_equation_0 = const()[name = tensor("op_15094_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15094_cast = einsum(equation = var_15094_equation_0, values = (var_15028_cast, var_14991_cast))[name = tensor("op_15094_cast")]; + tensor var_15095_to_fp16 = const()[name = tensor("op_15095_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1055_cast = mul(x = var_15094_cast, y = var_15095_to_fp16)[name = tensor("aw_chunk_1055_cast")]; + tensor var_15098_equation_0 = const()[name = tensor("op_15098_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15098_cast = einsum(equation = var_15098_equation_0, values = (var_15032_cast, var_14992_cast))[name = tensor("op_15098_cast")]; + tensor var_15099_to_fp16 = const()[name = tensor("op_15099_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1057_cast = mul(x = var_15098_cast, y = var_15099_to_fp16)[name = tensor("aw_chunk_1057_cast")]; + tensor var_15102_equation_0 = const()[name = tensor("op_15102_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15102_cast = einsum(equation = var_15102_equation_0, values = (var_15032_cast, var_14993_cast))[name = tensor("op_15102_cast")]; + tensor var_15103_to_fp16 = const()[name = tensor("op_15103_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1059_cast = mul(x = var_15102_cast, y = var_15103_to_fp16)[name = tensor("aw_chunk_1059_cast")]; + tensor var_15106_equation_0 = const()[name = tensor("op_15106_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15106_cast = einsum(equation = var_15106_equation_0, values = (var_15032_cast, var_14994_cast))[name = tensor("op_15106_cast")]; + tensor var_15107_to_fp16 = const()[name = tensor("op_15107_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1061_cast = mul(x = var_15106_cast, y = var_15107_to_fp16)[name = tensor("aw_chunk_1061_cast")]; + tensor var_15110_equation_0 = const()[name = tensor("op_15110_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15110_cast = einsum(equation = var_15110_equation_0, values = (var_15032_cast, var_14995_cast))[name = tensor("op_15110_cast")]; + tensor var_15111_to_fp16 = const()[name = tensor("op_15111_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1063_cast = mul(x = var_15110_cast, y = var_15111_to_fp16)[name = tensor("aw_chunk_1063_cast")]; + tensor var_15114_equation_0 = const()[name = tensor("op_15114_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15114_cast = einsum(equation = var_15114_equation_0, values = (var_15032_cast, var_14996_cast))[name = tensor("op_15114_cast")]; + tensor var_15115_to_fp16 = const()[name = tensor("op_15115_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1065_cast = mul(x = var_15114_cast, y = var_15115_to_fp16)[name = tensor("aw_chunk_1065_cast")]; + tensor var_15118_equation_0 = const()[name = tensor("op_15118_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15118_cast = einsum(equation = var_15118_equation_0, values = (var_15032_cast, var_14997_cast))[name = tensor("op_15118_cast")]; + tensor var_15119_to_fp16 = const()[name = tensor("op_15119_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1067_cast = mul(x = var_15118_cast, y = var_15119_to_fp16)[name = tensor("aw_chunk_1067_cast")]; + tensor var_15122_equation_0 = const()[name = tensor("op_15122_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15122_cast = einsum(equation = var_15122_equation_0, values = (var_15032_cast, var_14998_cast))[name = tensor("op_15122_cast")]; + tensor var_15123_to_fp16 = const()[name = tensor("op_15123_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1069_cast = mul(x = var_15122_cast, y = var_15123_to_fp16)[name = tensor("aw_chunk_1069_cast")]; + tensor var_15126_equation_0 = const()[name = tensor("op_15126_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15126_cast = einsum(equation = var_15126_equation_0, values = (var_15032_cast, var_14999_cast))[name = tensor("op_15126_cast")]; + tensor var_15127_to_fp16 = const()[name = tensor("op_15127_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1071_cast = mul(x = var_15126_cast, y = var_15127_to_fp16)[name = tensor("aw_chunk_1071_cast")]; + tensor var_15130_equation_0 = const()[name = tensor("op_15130_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15130_cast = einsum(equation = var_15130_equation_0, values = (var_15036_cast, var_15000_cast))[name = tensor("op_15130_cast")]; + tensor var_15131_to_fp16 = const()[name = tensor("op_15131_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1073_cast = mul(x = var_15130_cast, y = var_15131_to_fp16)[name = tensor("aw_chunk_1073_cast")]; + tensor var_15134_equation_0 = const()[name = tensor("op_15134_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15134_cast = einsum(equation = var_15134_equation_0, values = (var_15036_cast, var_15001_cast))[name = tensor("op_15134_cast")]; + tensor var_15135_to_fp16 = const()[name = tensor("op_15135_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1075_cast = mul(x = var_15134_cast, y = var_15135_to_fp16)[name = tensor("aw_chunk_1075_cast")]; + tensor var_15138_equation_0 = const()[name = tensor("op_15138_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15138_cast = einsum(equation = var_15138_equation_0, values = (var_15036_cast, var_15002_cast))[name = tensor("op_15138_cast")]; + tensor var_15139_to_fp16 = const()[name = tensor("op_15139_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1077_cast = mul(x = var_15138_cast, y = var_15139_to_fp16)[name = tensor("aw_chunk_1077_cast")]; + tensor var_15142_equation_0 = const()[name = tensor("op_15142_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15142_cast = einsum(equation = var_15142_equation_0, values = (var_15036_cast, var_15003_cast))[name = tensor("op_15142_cast")]; + tensor var_15143_to_fp16 = const()[name = tensor("op_15143_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1079_cast = mul(x = var_15142_cast, y = var_15143_to_fp16)[name = tensor("aw_chunk_1079_cast")]; + tensor var_15146_equation_0 = const()[name = tensor("op_15146_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15146_cast = einsum(equation = var_15146_equation_0, values = (var_15036_cast, var_15004_cast))[name = tensor("op_15146_cast")]; + tensor var_15147_to_fp16 = const()[name = tensor("op_15147_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1081_cast = mul(x = var_15146_cast, y = var_15147_to_fp16)[name = tensor("aw_chunk_1081_cast")]; + tensor var_15150_equation_0 = const()[name = tensor("op_15150_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15150_cast = einsum(equation = var_15150_equation_0, values = (var_15036_cast, var_15005_cast))[name = tensor("op_15150_cast")]; + tensor var_15151_to_fp16 = const()[name = tensor("op_15151_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1083_cast = mul(x = var_15150_cast, y = var_15151_to_fp16)[name = tensor("aw_chunk_1083_cast")]; + tensor var_15154_equation_0 = const()[name = tensor("op_15154_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15154_cast = einsum(equation = var_15154_equation_0, values = (var_15036_cast, var_15006_cast))[name = tensor("op_15154_cast")]; + tensor var_15155_to_fp16 = const()[name = tensor("op_15155_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1085_cast = mul(x = var_15154_cast, y = var_15155_to_fp16)[name = tensor("aw_chunk_1085_cast")]; + tensor var_15158_equation_0 = const()[name = tensor("op_15158_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15158_cast = einsum(equation = var_15158_equation_0, values = (var_15036_cast, var_15007_cast))[name = tensor("op_15158_cast")]; + tensor var_15159_to_fp16 = const()[name = tensor("op_15159_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1087_cast = mul(x = var_15158_cast, y = var_15159_to_fp16)[name = tensor("aw_chunk_1087_cast")]; + tensor var_15162_equation_0 = const()[name = tensor("op_15162_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15162_cast = einsum(equation = var_15162_equation_0, values = (var_15040_cast, var_15008_cast))[name = tensor("op_15162_cast")]; + tensor var_15163_to_fp16 = const()[name = tensor("op_15163_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1089_cast = mul(x = var_15162_cast, y = var_15163_to_fp16)[name = tensor("aw_chunk_1089_cast")]; + tensor var_15166_equation_0 = const()[name = tensor("op_15166_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15166_cast = einsum(equation = var_15166_equation_0, values = (var_15040_cast, var_15009_cast))[name = tensor("op_15166_cast")]; + tensor var_15167_to_fp16 = const()[name = tensor("op_15167_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1091_cast = mul(x = var_15166_cast, y = var_15167_to_fp16)[name = tensor("aw_chunk_1091_cast")]; + tensor var_15170_equation_0 = const()[name = tensor("op_15170_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15170_cast = einsum(equation = var_15170_equation_0, values = (var_15040_cast, var_15010_cast))[name = tensor("op_15170_cast")]; + tensor var_15171_to_fp16 = const()[name = tensor("op_15171_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1093_cast = mul(x = var_15170_cast, y = var_15171_to_fp16)[name = tensor("aw_chunk_1093_cast")]; + tensor var_15174_equation_0 = const()[name = tensor("op_15174_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15174_cast = einsum(equation = var_15174_equation_0, values = (var_15040_cast, var_15011_cast))[name = tensor("op_15174_cast")]; + tensor var_15175_to_fp16 = const()[name = tensor("op_15175_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1095_cast = mul(x = var_15174_cast, y = var_15175_to_fp16)[name = tensor("aw_chunk_1095_cast")]; + tensor var_15178_equation_0 = const()[name = tensor("op_15178_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15178_cast = einsum(equation = var_15178_equation_0, values = (var_15040_cast, var_15012_cast))[name = tensor("op_15178_cast")]; + tensor var_15179_to_fp16 = const()[name = tensor("op_15179_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1097_cast = mul(x = var_15178_cast, y = var_15179_to_fp16)[name = tensor("aw_chunk_1097_cast")]; + tensor var_15182_equation_0 = const()[name = tensor("op_15182_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15182_cast = einsum(equation = var_15182_equation_0, values = (var_15040_cast, var_15013_cast))[name = tensor("op_15182_cast")]; + tensor var_15183_to_fp16 = const()[name = tensor("op_15183_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1099_cast = mul(x = var_15182_cast, y = var_15183_to_fp16)[name = tensor("aw_chunk_1099_cast")]; + tensor var_15186_equation_0 = const()[name = tensor("op_15186_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15186_cast = einsum(equation = var_15186_equation_0, values = (var_15040_cast, var_15014_cast))[name = tensor("op_15186_cast")]; + tensor var_15187_to_fp16 = const()[name = tensor("op_15187_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1101_cast = mul(x = var_15186_cast, y = var_15187_to_fp16)[name = tensor("aw_chunk_1101_cast")]; + tensor var_15190_equation_0 = const()[name = tensor("op_15190_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15190_cast = einsum(equation = var_15190_equation_0, values = (var_15040_cast, var_15015_cast))[name = tensor("op_15190_cast")]; + tensor var_15191_to_fp16 = const()[name = tensor("op_15191_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1103_cast = mul(x = var_15190_cast, y = var_15191_to_fp16)[name = tensor("aw_chunk_1103_cast")]; + tensor var_15194_equation_0 = const()[name = tensor("op_15194_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15194_cast = einsum(equation = var_15194_equation_0, values = (var_15044_cast, var_15016_cast))[name = tensor("op_15194_cast")]; + tensor var_15195_to_fp16 = const()[name = tensor("op_15195_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1105_cast = mul(x = var_15194_cast, y = var_15195_to_fp16)[name = tensor("aw_chunk_1105_cast")]; + tensor var_15198_equation_0 = const()[name = tensor("op_15198_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15198_cast = einsum(equation = var_15198_equation_0, values = (var_15044_cast, var_15017_cast))[name = tensor("op_15198_cast")]; + tensor var_15199_to_fp16 = const()[name = tensor("op_15199_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1107_cast = mul(x = var_15198_cast, y = var_15199_to_fp16)[name = tensor("aw_chunk_1107_cast")]; + tensor var_15202_equation_0 = const()[name = tensor("op_15202_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15202_cast = einsum(equation = var_15202_equation_0, values = (var_15044_cast, var_15018_cast))[name = tensor("op_15202_cast")]; + tensor var_15203_to_fp16 = const()[name = tensor("op_15203_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1109_cast = mul(x = var_15202_cast, y = var_15203_to_fp16)[name = tensor("aw_chunk_1109_cast")]; + tensor var_15206_equation_0 = const()[name = tensor("op_15206_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15206_cast = einsum(equation = var_15206_equation_0, values = (var_15044_cast, var_15019_cast))[name = tensor("op_15206_cast")]; + tensor var_15207_to_fp16 = const()[name = tensor("op_15207_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1111_cast = mul(x = var_15206_cast, y = var_15207_to_fp16)[name = tensor("aw_chunk_1111_cast")]; + tensor var_15210_equation_0 = const()[name = tensor("op_15210_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15210_cast = einsum(equation = var_15210_equation_0, values = (var_15044_cast, var_15020_cast))[name = tensor("op_15210_cast")]; + tensor var_15211_to_fp16 = const()[name = tensor("op_15211_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1113_cast = mul(x = var_15210_cast, y = var_15211_to_fp16)[name = tensor("aw_chunk_1113_cast")]; + tensor var_15214_equation_0 = const()[name = tensor("op_15214_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15214_cast = einsum(equation = var_15214_equation_0, values = (var_15044_cast, var_15021_cast))[name = tensor("op_15214_cast")]; + tensor var_15215_to_fp16 = const()[name = tensor("op_15215_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1115_cast = mul(x = var_15214_cast, y = var_15215_to_fp16)[name = tensor("aw_chunk_1115_cast")]; + tensor var_15218_equation_0 = const()[name = tensor("op_15218_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15218_cast = einsum(equation = var_15218_equation_0, values = (var_15044_cast, var_15022_cast))[name = tensor("op_15218_cast")]; + tensor var_15219_to_fp16 = const()[name = tensor("op_15219_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1117_cast = mul(x = var_15218_cast, y = var_15219_to_fp16)[name = tensor("aw_chunk_1117_cast")]; + tensor var_15222_equation_0 = const()[name = tensor("op_15222_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15222_cast = einsum(equation = var_15222_equation_0, values = (var_15044_cast, var_15023_cast))[name = tensor("op_15222_cast")]; + tensor var_15223_to_fp16 = const()[name = tensor("op_15223_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1119_cast = mul(x = var_15222_cast, y = var_15223_to_fp16)[name = tensor("aw_chunk_1119_cast")]; + tensor var_15225_cast = softmax(axis = var_12807, x = aw_chunk_1041_cast)[name = tensor("op_15225_cast")]; + tensor var_15226_cast = softmax(axis = var_12807, x = aw_chunk_1043_cast)[name = tensor("op_15226_cast")]; + tensor var_15227_cast = softmax(axis = var_12807, x = aw_chunk_1045_cast)[name = tensor("op_15227_cast")]; + tensor var_15228_cast = softmax(axis = var_12807, x = aw_chunk_1047_cast)[name = tensor("op_15228_cast")]; + tensor var_15229_cast = softmax(axis = var_12807, x = aw_chunk_1049_cast)[name = tensor("op_15229_cast")]; + tensor var_15230_cast = softmax(axis = var_12807, x = aw_chunk_1051_cast)[name = tensor("op_15230_cast")]; + tensor var_15231_cast = softmax(axis = var_12807, x = aw_chunk_1053_cast)[name = tensor("op_15231_cast")]; + tensor var_15232_cast = softmax(axis = var_12807, x = aw_chunk_1055_cast)[name = tensor("op_15232_cast")]; + tensor var_15233_cast = softmax(axis = var_12807, x = aw_chunk_1057_cast)[name = tensor("op_15233_cast")]; + tensor var_15234_cast = softmax(axis = var_12807, x = aw_chunk_1059_cast)[name = tensor("op_15234_cast")]; + tensor var_15235_cast = softmax(axis = var_12807, x = aw_chunk_1061_cast)[name = tensor("op_15235_cast")]; + tensor var_15236_cast = softmax(axis = var_12807, x = aw_chunk_1063_cast)[name = tensor("op_15236_cast")]; + tensor var_15237_cast = softmax(axis = var_12807, x = aw_chunk_1065_cast)[name = tensor("op_15237_cast")]; + tensor var_15238_cast = softmax(axis = var_12807, x = aw_chunk_1067_cast)[name = tensor("op_15238_cast")]; + tensor var_15239_cast = softmax(axis = var_12807, x = aw_chunk_1069_cast)[name = tensor("op_15239_cast")]; + tensor var_15240_cast = softmax(axis = var_12807, x = aw_chunk_1071_cast)[name = tensor("op_15240_cast")]; + tensor var_15241_cast = softmax(axis = var_12807, x = aw_chunk_1073_cast)[name = tensor("op_15241_cast")]; + tensor var_15242_cast = softmax(axis = var_12807, x = aw_chunk_1075_cast)[name = tensor("op_15242_cast")]; + tensor var_15243_cast = softmax(axis = var_12807, x = aw_chunk_1077_cast)[name = tensor("op_15243_cast")]; + tensor var_15244_cast = softmax(axis = var_12807, x = aw_chunk_1079_cast)[name = tensor("op_15244_cast")]; + tensor var_15245_cast = softmax(axis = var_12807, x = aw_chunk_1081_cast)[name = tensor("op_15245_cast")]; + tensor var_15246_cast = softmax(axis = var_12807, x = aw_chunk_1083_cast)[name = tensor("op_15246_cast")]; + tensor var_15247_cast = softmax(axis = var_12807, x = aw_chunk_1085_cast)[name = tensor("op_15247_cast")]; + tensor var_15248_cast = softmax(axis = var_12807, x = aw_chunk_1087_cast)[name = tensor("op_15248_cast")]; + tensor var_15249_cast = softmax(axis = var_12807, x = aw_chunk_1089_cast)[name = tensor("op_15249_cast")]; + tensor var_15250_cast = softmax(axis = var_12807, x = aw_chunk_1091_cast)[name = tensor("op_15250_cast")]; + tensor var_15251_cast = softmax(axis = var_12807, x = aw_chunk_1093_cast)[name = tensor("op_15251_cast")]; + tensor var_15252_cast = softmax(axis = var_12807, x = aw_chunk_1095_cast)[name = tensor("op_15252_cast")]; + tensor var_15253_cast = softmax(axis = var_12807, x = aw_chunk_1097_cast)[name = tensor("op_15253_cast")]; + tensor var_15254_cast = softmax(axis = var_12807, x = aw_chunk_1099_cast)[name = tensor("op_15254_cast")]; + tensor var_15255_cast = softmax(axis = var_12807, x = aw_chunk_1101_cast)[name = tensor("op_15255_cast")]; + tensor var_15256_cast = softmax(axis = var_12807, x = aw_chunk_1103_cast)[name = tensor("op_15256_cast")]; + tensor var_15257_cast = softmax(axis = var_12807, x = aw_chunk_1105_cast)[name = tensor("op_15257_cast")]; + tensor var_15258_cast = softmax(axis = var_12807, x = aw_chunk_1107_cast)[name = tensor("op_15258_cast")]; + tensor var_15259_cast = softmax(axis = var_12807, x = aw_chunk_1109_cast)[name = tensor("op_15259_cast")]; + tensor var_15260_cast = softmax(axis = var_12807, x = aw_chunk_1111_cast)[name = tensor("op_15260_cast")]; + tensor var_15261_cast = softmax(axis = var_12807, x = aw_chunk_1113_cast)[name = tensor("op_15261_cast")]; + tensor var_15262_cast = softmax(axis = var_12807, x = aw_chunk_1115_cast)[name = tensor("op_15262_cast")]; + tensor var_15263_cast = softmax(axis = var_12807, x = aw_chunk_1117_cast)[name = tensor("op_15263_cast")]; + tensor var_15264_cast = softmax(axis = var_12807, x = aw_chunk_1119_cast)[name = tensor("op_15264_cast")]; + tensor var_15266_equation_0 = const()[name = tensor("op_15266_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15266_cast = einsum(equation = var_15266_equation_0, values = (var_15046_cast, var_15225_cast))[name = tensor("op_15266_cast")]; + tensor var_15268_equation_0 = const()[name = tensor("op_15268_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15268_cast = einsum(equation = var_15268_equation_0, values = (var_15046_cast, var_15226_cast))[name = tensor("op_15268_cast")]; + tensor var_15270_equation_0 = const()[name = tensor("op_15270_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15270_cast = einsum(equation = var_15270_equation_0, values = (var_15046_cast, var_15227_cast))[name = tensor("op_15270_cast")]; + tensor var_15272_equation_0 = const()[name = tensor("op_15272_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15272_cast = einsum(equation = var_15272_equation_0, values = (var_15046_cast, var_15228_cast))[name = tensor("op_15272_cast")]; + tensor var_15274_equation_0 = const()[name = tensor("op_15274_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15274_cast = einsum(equation = var_15274_equation_0, values = (var_15046_cast, var_15229_cast))[name = tensor("op_15274_cast")]; + tensor var_15276_equation_0 = const()[name = tensor("op_15276_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15276_cast = einsum(equation = var_15276_equation_0, values = (var_15046_cast, var_15230_cast))[name = tensor("op_15276_cast")]; + tensor var_15278_equation_0 = const()[name = tensor("op_15278_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15278_cast = einsum(equation = var_15278_equation_0, values = (var_15046_cast, var_15231_cast))[name = tensor("op_15278_cast")]; + tensor var_15280_equation_0 = const()[name = tensor("op_15280_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15280_cast = einsum(equation = var_15280_equation_0, values = (var_15046_cast, var_15232_cast))[name = tensor("op_15280_cast")]; + tensor var_15282_equation_0 = const()[name = tensor("op_15282_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15282_cast = einsum(equation = var_15282_equation_0, values = (var_15050_cast, var_15233_cast))[name = tensor("op_15282_cast")]; + tensor var_15284_equation_0 = const()[name = tensor("op_15284_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15284_cast = einsum(equation = var_15284_equation_0, values = (var_15050_cast, var_15234_cast))[name = tensor("op_15284_cast")]; + tensor var_15286_equation_0 = const()[name = tensor("op_15286_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15286_cast = einsum(equation = var_15286_equation_0, values = (var_15050_cast, var_15235_cast))[name = tensor("op_15286_cast")]; + tensor var_15288_equation_0 = const()[name = tensor("op_15288_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15288_cast = einsum(equation = var_15288_equation_0, values = (var_15050_cast, var_15236_cast))[name = tensor("op_15288_cast")]; + tensor var_15290_equation_0 = const()[name = tensor("op_15290_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15290_cast = einsum(equation = var_15290_equation_0, values = (var_15050_cast, var_15237_cast))[name = tensor("op_15290_cast")]; + tensor var_15292_equation_0 = const()[name = tensor("op_15292_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15292_cast = einsum(equation = var_15292_equation_0, values = (var_15050_cast, var_15238_cast))[name = tensor("op_15292_cast")]; + tensor var_15294_equation_0 = const()[name = tensor("op_15294_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15294_cast = einsum(equation = var_15294_equation_0, values = (var_15050_cast, var_15239_cast))[name = tensor("op_15294_cast")]; + tensor var_15296_equation_0 = const()[name = tensor("op_15296_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15296_cast = einsum(equation = var_15296_equation_0, values = (var_15050_cast, var_15240_cast))[name = tensor("op_15296_cast")]; + tensor var_15298_equation_0 = const()[name = tensor("op_15298_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15298_cast = einsum(equation = var_15298_equation_0, values = (var_15054_cast, var_15241_cast))[name = tensor("op_15298_cast")]; + tensor var_15300_equation_0 = const()[name = tensor("op_15300_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15300_cast = einsum(equation = var_15300_equation_0, values = (var_15054_cast, var_15242_cast))[name = tensor("op_15300_cast")]; + tensor var_15302_equation_0 = const()[name = tensor("op_15302_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15302_cast = einsum(equation = var_15302_equation_0, values = (var_15054_cast, var_15243_cast))[name = tensor("op_15302_cast")]; + tensor var_15304_equation_0 = const()[name = tensor("op_15304_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15304_cast = einsum(equation = var_15304_equation_0, values = (var_15054_cast, var_15244_cast))[name = tensor("op_15304_cast")]; + tensor var_15306_equation_0 = const()[name = tensor("op_15306_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15306_cast = einsum(equation = var_15306_equation_0, values = (var_15054_cast, var_15245_cast))[name = tensor("op_15306_cast")]; + tensor var_15308_equation_0 = const()[name = tensor("op_15308_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15308_cast = einsum(equation = var_15308_equation_0, values = (var_15054_cast, var_15246_cast))[name = tensor("op_15308_cast")]; + tensor var_15310_equation_0 = const()[name = tensor("op_15310_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15310_cast = einsum(equation = var_15310_equation_0, values = (var_15054_cast, var_15247_cast))[name = tensor("op_15310_cast")]; + tensor var_15312_equation_0 = const()[name = tensor("op_15312_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15312_cast = einsum(equation = var_15312_equation_0, values = (var_15054_cast, var_15248_cast))[name = tensor("op_15312_cast")]; + tensor var_15314_equation_0 = const()[name = tensor("op_15314_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15314_cast = einsum(equation = var_15314_equation_0, values = (var_15058_cast, var_15249_cast))[name = tensor("op_15314_cast")]; + tensor var_15316_equation_0 = const()[name = tensor("op_15316_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15316_cast = einsum(equation = var_15316_equation_0, values = (var_15058_cast, var_15250_cast))[name = tensor("op_15316_cast")]; + tensor var_15318_equation_0 = const()[name = tensor("op_15318_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15318_cast = einsum(equation = var_15318_equation_0, values = (var_15058_cast, var_15251_cast))[name = tensor("op_15318_cast")]; + tensor var_15320_equation_0 = const()[name = tensor("op_15320_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15320_cast = einsum(equation = var_15320_equation_0, values = (var_15058_cast, var_15252_cast))[name = tensor("op_15320_cast")]; + tensor var_15322_equation_0 = const()[name = tensor("op_15322_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15322_cast = einsum(equation = var_15322_equation_0, values = (var_15058_cast, var_15253_cast))[name = tensor("op_15322_cast")]; + tensor var_15324_equation_0 = const()[name = tensor("op_15324_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15324_cast = einsum(equation = var_15324_equation_0, values = (var_15058_cast, var_15254_cast))[name = tensor("op_15324_cast")]; + tensor var_15326_equation_0 = const()[name = tensor("op_15326_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15326_cast = einsum(equation = var_15326_equation_0, values = (var_15058_cast, var_15255_cast))[name = tensor("op_15326_cast")]; + tensor var_15328_equation_0 = const()[name = tensor("op_15328_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15328_cast = einsum(equation = var_15328_equation_0, values = (var_15058_cast, var_15256_cast))[name = tensor("op_15328_cast")]; + tensor var_15330_equation_0 = const()[name = tensor("op_15330_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15330_cast = einsum(equation = var_15330_equation_0, values = (var_15062_cast, var_15257_cast))[name = tensor("op_15330_cast")]; + tensor var_15332_equation_0 = const()[name = tensor("op_15332_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15332_cast = einsum(equation = var_15332_equation_0, values = (var_15062_cast, var_15258_cast))[name = tensor("op_15332_cast")]; + tensor var_15334_equation_0 = const()[name = tensor("op_15334_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15334_cast = einsum(equation = var_15334_equation_0, values = (var_15062_cast, var_15259_cast))[name = tensor("op_15334_cast")]; + tensor var_15336_equation_0 = const()[name = tensor("op_15336_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15336_cast = einsum(equation = var_15336_equation_0, values = (var_15062_cast, var_15260_cast))[name = tensor("op_15336_cast")]; + tensor var_15338_equation_0 = const()[name = tensor("op_15338_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15338_cast = einsum(equation = var_15338_equation_0, values = (var_15062_cast, var_15261_cast))[name = tensor("op_15338_cast")]; + tensor var_15340_equation_0 = const()[name = tensor("op_15340_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15340_cast = einsum(equation = var_15340_equation_0, values = (var_15062_cast, var_15262_cast))[name = tensor("op_15340_cast")]; + tensor var_15342_equation_0 = const()[name = tensor("op_15342_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15342_cast = einsum(equation = var_15342_equation_0, values = (var_15062_cast, var_15263_cast))[name = tensor("op_15342_cast")]; + tensor var_15344_equation_0 = const()[name = tensor("op_15344_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15344_cast = einsum(equation = var_15344_equation_0, values = (var_15062_cast, var_15264_cast))[name = tensor("op_15344_cast")]; + tensor var_15346_interleave_0 = const()[name = tensor("op_15346_interleave_0"), val = tensor(false)]; + tensor var_15346_cast = concat(axis = var_12782, interleave = var_15346_interleave_0, values = (var_15266_cast, var_15268_cast, var_15270_cast, var_15272_cast, var_15274_cast, var_15276_cast, var_15278_cast, var_15280_cast))[name = tensor("op_15346_cast")]; + tensor var_15348_interleave_0 = const()[name = tensor("op_15348_interleave_0"), val = tensor(false)]; + tensor var_15348_cast = concat(axis = var_12782, interleave = var_15348_interleave_0, values = (var_15282_cast, var_15284_cast, var_15286_cast, var_15288_cast, var_15290_cast, var_15292_cast, var_15294_cast, var_15296_cast))[name = tensor("op_15348_cast")]; + tensor var_15350_interleave_0 = const()[name = tensor("op_15350_interleave_0"), val = tensor(false)]; + tensor var_15350_cast = concat(axis = var_12782, interleave = var_15350_interleave_0, values = (var_15298_cast, var_15300_cast, var_15302_cast, var_15304_cast, var_15306_cast, var_15308_cast, var_15310_cast, var_15312_cast))[name = tensor("op_15350_cast")]; + tensor var_15352_interleave_0 = const()[name = tensor("op_15352_interleave_0"), val = tensor(false)]; + tensor var_15352_cast = concat(axis = var_12782, interleave = var_15352_interleave_0, values = (var_15314_cast, var_15316_cast, var_15318_cast, var_15320_cast, var_15322_cast, var_15324_cast, var_15326_cast, var_15328_cast))[name = tensor("op_15352_cast")]; + tensor var_15354_interleave_0 = const()[name = tensor("op_15354_interleave_0"), val = tensor(false)]; + tensor var_15354_cast = concat(axis = var_12782, interleave = var_15354_interleave_0, values = (var_15330_cast, var_15332_cast, var_15334_cast, var_15336_cast, var_15338_cast, var_15340_cast, var_15342_cast, var_15344_cast))[name = tensor("op_15354_cast")]; + tensor input_517_interleave_0 = const()[name = tensor("input_517_interleave_0"), val = tensor(false)]; + tensor input_517_cast = concat(axis = var_12807, interleave = input_517_interleave_0, values = (var_15346_cast, var_15348_cast, var_15350_cast, var_15352_cast, var_15354_cast))[name = tensor("input_517_cast")]; + tensor var_15360 = const()[name = tensor("op_15360"), val = tensor([1, 1])]; + tensor var_15362 = const()[name = tensor("op_15362"), val = tensor([1, 1])]; + tensor var_15364_pad_type_0 = const()[name = tensor("op_15364_pad_type_0"), val = tensor("custom")]; + tensor var_15364_pad_0 = const()[name = tensor("op_15364_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_3_attentions_2_transformer_blocks_0_attn1_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(648219456))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(648296320))), name = tensor("up_blocks_3_attentions_2_transformer_blocks_0_attn1_to_out_0_weight_to_fp16_palettized"), shape = tensor([320, 320, 1, 1])]; + tensor up_blocks_3_attentions_2_transformer_blocks_0_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("up_blocks_3_attentions_2_transformer_blocks_0_attn1_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(648296512)))]; + tensor var_15364_cast = conv(bias = up_blocks_3_attentions_2_transformer_blocks_0_attn1_to_out_0_bias_to_fp16, dilations = var_15362, groups = var_12807, pad = var_15364_pad_0, pad_type = var_15364_pad_type_0, strides = var_15360, weight = up_blocks_3_attentions_2_transformer_blocks_0_attn1_to_out_0_weight_to_fp16_palettized, x = input_517_cast)[name = tensor("op_15364_cast")]; + tensor inputs_93_cast = add(x = var_15364_cast, y = inputs_91_cast)[name = tensor("inputs_93_cast")]; + tensor var_15368 = const()[name = tensor("op_15368"), val = tensor([1])]; + tensor channels_mean_93_cast = reduce_mean(axes = var_15368, keep_dims = var_12802, x = inputs_93_cast)[name = tensor("channels_mean_93_cast")]; + tensor zero_mean_93_cast = sub(x = inputs_93_cast, y = channels_mean_93_cast)[name = tensor("zero_mean_93_cast")]; + tensor zero_mean_sq_93_cast = mul(x = zero_mean_93_cast, y = zero_mean_93_cast)[name = tensor("zero_mean_sq_93_cast")]; + tensor var_15372 = const()[name = tensor("op_15372"), val = tensor([1])]; + tensor var_15373_cast = reduce_mean(axes = var_15372, keep_dims = var_12802, x = zero_mean_sq_93_cast)[name = tensor("op_15373_cast")]; + tensor var_15374_to_fp16 = const()[name = tensor("op_15374_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_15375_cast = add(x = var_15373_cast, y = var_15374_to_fp16)[name = tensor("op_15375_cast")]; + tensor denom_93_epsilon_0_to_fp16 = const()[name = tensor("denom_93_epsilon_0_to_fp16"), val = tensor(0x1p-24)]; + tensor denom_93_cast = rsqrt(epsilon = denom_93_epsilon_0_to_fp16, x = var_15375_cast)[name = tensor("denom_93_cast")]; + tensor out_93_cast = mul(x = zero_mean_93_cast, y = denom_93_cast)[name = tensor("out_93_cast")]; + tensor var_15379_to_fp16 = const()[name = tensor("op_15379_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(648297216)))]; + tensor var_15380_cast = add(x = out_93_cast, y = var_15379_to_fp16)[name = tensor("op_15380_cast")]; + tensor var_15382_to_fp16 = const()[name = tensor("op_15382_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(648297920)))]; + tensor hidden_states_329_cast = mul(x = var_15380_cast, y = var_15382_to_fp16)[name = tensor("hidden_states_329_cast")]; + tensor var_15389 = const()[name = tensor("op_15389"), val = tensor([1, 1])]; + tensor var_15391 = const()[name = tensor("op_15391"), val = tensor([1, 1])]; + tensor q_pad_type_0 = const()[name = tensor("q_pad_type_0"), val = tensor("custom")]; + tensor q_pad_0 = const()[name = tensor("q_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_3_attentions_2_transformer_blocks_0_attn2_to_q_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(648298624))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(648375488))), name = tensor("up_blocks_3_attentions_2_transformer_blocks_0_attn2_to_q_weight_to_fp16_palettized"), shape = tensor([320, 320, 1, 1])]; + tensor q_cast = conv(dilations = var_15391, groups = var_12807, pad = q_pad_0, pad_type = q_pad_type_0, strides = var_15389, weight = up_blocks_3_attentions_2_transformer_blocks_0_attn2_to_q_weight_to_fp16_palettized, x = hidden_states_329_cast)[name = tensor("q_cast")]; + tensor var_15395 = const()[name = tensor("op_15395"), val = tensor([1, 1])]; + tensor var_15397 = const()[name = tensor("op_15397"), val = tensor([1, 1])]; + tensor k_125_pad_type_0 = const()[name = tensor("k_125_pad_type_0"), val = tensor("custom")]; + tensor k_125_pad_0 = const()[name = tensor("k_125_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_3_attentions_2_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(648375680))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(648621504))), name = tensor("up_blocks_3_attentions_2_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized"), shape = tensor([320, 1024, 1, 1])]; + tensor k_125_cast = conv(dilations = var_15397, groups = var_12807, pad = k_125_pad_0, pad_type = k_125_pad_type_0, strides = var_15395, weight = up_blocks_3_attentions_2_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor("k_125_cast")]; + tensor var_15401 = const()[name = tensor("op_15401"), val = tensor([1, 1])]; + tensor var_15403 = const()[name = tensor("op_15403"), val = tensor([1, 1])]; + tensor v_pad_type_0 = const()[name = tensor("v_pad_type_0"), val = tensor("custom")]; + tensor v_pad_0 = const()[name = tensor("v_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_3_attentions_2_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(648621696))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(648867520))), name = tensor("up_blocks_3_attentions_2_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized"), shape = tensor([320, 1024, 1, 1])]; + tensor v_cast = conv(dilations = var_15403, groups = var_12807, pad = v_pad_0, pad_type = v_pad_type_0, strides = var_15401, weight = up_blocks_3_attentions_2_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor("v_cast")]; + tensor var_15407_begin_0 = const()[name = tensor("op_15407_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_15407_end_0 = const()[name = tensor("op_15407_end_0"), val = tensor([2, 64, 1, 4096])]; + tensor var_15407_end_mask_0 = const()[name = tensor("op_15407_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15407_cast = slice_by_index(begin = var_15407_begin_0, end = var_15407_end_0, end_mask = var_15407_end_mask_0, x = q_cast)[name = tensor("op_15407_cast")]; + tensor var_15411_begin_0 = const()[name = tensor("op_15411_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_15411_end_0 = const()[name = tensor("op_15411_end_0"), val = tensor([2, 128, 1, 4096])]; + tensor var_15411_end_mask_0 = const()[name = tensor("op_15411_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15411_cast = slice_by_index(begin = var_15411_begin_0, end = var_15411_end_0, end_mask = var_15411_end_mask_0, x = q_cast)[name = tensor("op_15411_cast")]; + tensor var_15415_begin_0 = const()[name = tensor("op_15415_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_15415_end_0 = const()[name = tensor("op_15415_end_0"), val = tensor([2, 192, 1, 4096])]; + tensor var_15415_end_mask_0 = const()[name = tensor("op_15415_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15415_cast = slice_by_index(begin = var_15415_begin_0, end = var_15415_end_0, end_mask = var_15415_end_mask_0, x = q_cast)[name = tensor("op_15415_cast")]; + tensor var_15419_begin_0 = const()[name = tensor("op_15419_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_15419_end_0 = const()[name = tensor("op_15419_end_0"), val = tensor([2, 256, 1, 4096])]; + tensor var_15419_end_mask_0 = const()[name = tensor("op_15419_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15419_cast = slice_by_index(begin = var_15419_begin_0, end = var_15419_end_0, end_mask = var_15419_end_mask_0, x = q_cast)[name = tensor("op_15419_cast")]; + tensor var_15423_begin_0 = const()[name = tensor("op_15423_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_15423_end_0 = const()[name = tensor("op_15423_end_0"), val = tensor([2, 320, 1, 4096])]; + tensor var_15423_end_mask_0 = const()[name = tensor("op_15423_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15423_cast = slice_by_index(begin = var_15423_begin_0, end = var_15423_end_0, end_mask = var_15423_end_mask_0, x = q_cast)[name = tensor("op_15423_cast")]; + tensor var_15426_begin_0 = const()[name = tensor("op_15426_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_15426_end_0 = const()[name = tensor("op_15426_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_15426_end_mask_0 = const()[name = tensor("op_15426_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15426_cast = slice_by_index(begin = var_15426_begin_0, end = var_15426_end_0, end_mask = var_15426_end_mask_0, x = var_15407_cast)[name = tensor("op_15426_cast")]; + tensor var_15427_begin_0 = const()[name = tensor("op_15427_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_15427_end_0 = const()[name = tensor("op_15427_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_15427_end_mask_0 = const()[name = tensor("op_15427_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15427_cast = slice_by_index(begin = var_15427_begin_0, end = var_15427_end_0, end_mask = var_15427_end_mask_0, x = var_15407_cast)[name = tensor("op_15427_cast")]; + tensor var_15428_begin_0 = const()[name = tensor("op_15428_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_15428_end_0 = const()[name = tensor("op_15428_end_0"), val = tensor([2, 64, 1, 1536])]; + tensor var_15428_end_mask_0 = const()[name = tensor("op_15428_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15428_cast = slice_by_index(begin = var_15428_begin_0, end = var_15428_end_0, end_mask = var_15428_end_mask_0, x = var_15407_cast)[name = tensor("op_15428_cast")]; + tensor var_15429_begin_0 = const()[name = tensor("op_15429_begin_0"), val = tensor([0, 0, 0, 1536])]; + tensor var_15429_end_0 = const()[name = tensor("op_15429_end_0"), val = tensor([2, 64, 1, 2048])]; + tensor var_15429_end_mask_0 = const()[name = tensor("op_15429_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15429_cast = slice_by_index(begin = var_15429_begin_0, end = var_15429_end_0, end_mask = var_15429_end_mask_0, x = var_15407_cast)[name = tensor("op_15429_cast")]; + tensor var_15430_begin_0 = const()[name = tensor("op_15430_begin_0"), val = tensor([0, 0, 0, 2048])]; + tensor var_15430_end_0 = const()[name = tensor("op_15430_end_0"), val = tensor([2, 64, 1, 2560])]; + tensor var_15430_end_mask_0 = const()[name = tensor("op_15430_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15430_cast = slice_by_index(begin = var_15430_begin_0, end = var_15430_end_0, end_mask = var_15430_end_mask_0, x = var_15407_cast)[name = tensor("op_15430_cast")]; + tensor var_15431_begin_0 = const()[name = tensor("op_15431_begin_0"), val = tensor([0, 0, 0, 2560])]; + tensor var_15431_end_0 = const()[name = tensor("op_15431_end_0"), val = tensor([2, 64, 1, 3072])]; + tensor var_15431_end_mask_0 = const()[name = tensor("op_15431_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15431_cast = slice_by_index(begin = var_15431_begin_0, end = var_15431_end_0, end_mask = var_15431_end_mask_0, x = var_15407_cast)[name = tensor("op_15431_cast")]; + tensor var_15432_begin_0 = const()[name = tensor("op_15432_begin_0"), val = tensor([0, 0, 0, 3072])]; + tensor var_15432_end_0 = const()[name = tensor("op_15432_end_0"), val = tensor([2, 64, 1, 3584])]; + tensor var_15432_end_mask_0 = const()[name = tensor("op_15432_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15432_cast = slice_by_index(begin = var_15432_begin_0, end = var_15432_end_0, end_mask = var_15432_end_mask_0, x = var_15407_cast)[name = tensor("op_15432_cast")]; + tensor var_15433_begin_0 = const()[name = tensor("op_15433_begin_0"), val = tensor([0, 0, 0, 3584])]; + tensor var_15433_end_0 = const()[name = tensor("op_15433_end_0"), val = tensor([2, 64, 1, 4096])]; + tensor var_15433_end_mask_0 = const()[name = tensor("op_15433_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15433_cast = slice_by_index(begin = var_15433_begin_0, end = var_15433_end_0, end_mask = var_15433_end_mask_0, x = var_15407_cast)[name = tensor("op_15433_cast")]; + tensor var_15434_begin_0 = const()[name = tensor("op_15434_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_15434_end_0 = const()[name = tensor("op_15434_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_15434_end_mask_0 = const()[name = tensor("op_15434_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15434_cast = slice_by_index(begin = var_15434_begin_0, end = var_15434_end_0, end_mask = var_15434_end_mask_0, x = var_15411_cast)[name = tensor("op_15434_cast")]; + tensor var_15435_begin_0 = const()[name = tensor("op_15435_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_15435_end_0 = const()[name = tensor("op_15435_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_15435_end_mask_0 = const()[name = tensor("op_15435_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15435_cast = slice_by_index(begin = var_15435_begin_0, end = var_15435_end_0, end_mask = var_15435_end_mask_0, x = var_15411_cast)[name = tensor("op_15435_cast")]; + tensor var_15436_begin_0 = const()[name = tensor("op_15436_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_15436_end_0 = const()[name = tensor("op_15436_end_0"), val = tensor([2, 64, 1, 1536])]; + tensor var_15436_end_mask_0 = const()[name = tensor("op_15436_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15436_cast = slice_by_index(begin = var_15436_begin_0, end = var_15436_end_0, end_mask = var_15436_end_mask_0, x = var_15411_cast)[name = tensor("op_15436_cast")]; + tensor var_15437_begin_0 = const()[name = tensor("op_15437_begin_0"), val = tensor([0, 0, 0, 1536])]; + tensor var_15437_end_0 = const()[name = tensor("op_15437_end_0"), val = tensor([2, 64, 1, 2048])]; + tensor var_15437_end_mask_0 = const()[name = tensor("op_15437_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15437_cast = slice_by_index(begin = var_15437_begin_0, end = var_15437_end_0, end_mask = var_15437_end_mask_0, x = var_15411_cast)[name = tensor("op_15437_cast")]; + tensor var_15438_begin_0 = const()[name = tensor("op_15438_begin_0"), val = tensor([0, 0, 0, 2048])]; + tensor var_15438_end_0 = const()[name = tensor("op_15438_end_0"), val = tensor([2, 64, 1, 2560])]; + tensor var_15438_end_mask_0 = const()[name = tensor("op_15438_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15438_cast = slice_by_index(begin = var_15438_begin_0, end = var_15438_end_0, end_mask = var_15438_end_mask_0, x = var_15411_cast)[name = tensor("op_15438_cast")]; + tensor var_15439_begin_0 = const()[name = tensor("op_15439_begin_0"), val = tensor([0, 0, 0, 2560])]; + tensor var_15439_end_0 = const()[name = tensor("op_15439_end_0"), val = tensor([2, 64, 1, 3072])]; + tensor var_15439_end_mask_0 = const()[name = tensor("op_15439_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15439_cast = slice_by_index(begin = var_15439_begin_0, end = var_15439_end_0, end_mask = var_15439_end_mask_0, x = var_15411_cast)[name = tensor("op_15439_cast")]; + tensor var_15440_begin_0 = const()[name = tensor("op_15440_begin_0"), val = tensor([0, 0, 0, 3072])]; + tensor var_15440_end_0 = const()[name = tensor("op_15440_end_0"), val = tensor([2, 64, 1, 3584])]; + tensor var_15440_end_mask_0 = const()[name = tensor("op_15440_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15440_cast = slice_by_index(begin = var_15440_begin_0, end = var_15440_end_0, end_mask = var_15440_end_mask_0, x = var_15411_cast)[name = tensor("op_15440_cast")]; + tensor var_15441_begin_0 = const()[name = tensor("op_15441_begin_0"), val = tensor([0, 0, 0, 3584])]; + tensor var_15441_end_0 = const()[name = tensor("op_15441_end_0"), val = tensor([2, 64, 1, 4096])]; + tensor var_15441_end_mask_0 = const()[name = tensor("op_15441_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15441_cast = slice_by_index(begin = var_15441_begin_0, end = var_15441_end_0, end_mask = var_15441_end_mask_0, x = var_15411_cast)[name = tensor("op_15441_cast")]; + tensor var_15442_begin_0 = const()[name = tensor("op_15442_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_15442_end_0 = const()[name = tensor("op_15442_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_15442_end_mask_0 = const()[name = tensor("op_15442_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15442_cast = slice_by_index(begin = var_15442_begin_0, end = var_15442_end_0, end_mask = var_15442_end_mask_0, x = var_15415_cast)[name = tensor("op_15442_cast")]; + tensor var_15443_begin_0 = const()[name = tensor("op_15443_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_15443_end_0 = const()[name = tensor("op_15443_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_15443_end_mask_0 = const()[name = tensor("op_15443_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15443_cast = slice_by_index(begin = var_15443_begin_0, end = var_15443_end_0, end_mask = var_15443_end_mask_0, x = var_15415_cast)[name = tensor("op_15443_cast")]; + tensor var_15444_begin_0 = const()[name = tensor("op_15444_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_15444_end_0 = const()[name = tensor("op_15444_end_0"), val = tensor([2, 64, 1, 1536])]; + tensor var_15444_end_mask_0 = const()[name = tensor("op_15444_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15444_cast = slice_by_index(begin = var_15444_begin_0, end = var_15444_end_0, end_mask = var_15444_end_mask_0, x = var_15415_cast)[name = tensor("op_15444_cast")]; + tensor var_15445_begin_0 = const()[name = tensor("op_15445_begin_0"), val = tensor([0, 0, 0, 1536])]; + tensor var_15445_end_0 = const()[name = tensor("op_15445_end_0"), val = tensor([2, 64, 1, 2048])]; + tensor var_15445_end_mask_0 = const()[name = tensor("op_15445_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15445_cast = slice_by_index(begin = var_15445_begin_0, end = var_15445_end_0, end_mask = var_15445_end_mask_0, x = var_15415_cast)[name = tensor("op_15445_cast")]; + tensor var_15446_begin_0 = const()[name = tensor("op_15446_begin_0"), val = tensor([0, 0, 0, 2048])]; + tensor var_15446_end_0 = const()[name = tensor("op_15446_end_0"), val = tensor([2, 64, 1, 2560])]; + tensor var_15446_end_mask_0 = const()[name = tensor("op_15446_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15446_cast = slice_by_index(begin = var_15446_begin_0, end = var_15446_end_0, end_mask = var_15446_end_mask_0, x = var_15415_cast)[name = tensor("op_15446_cast")]; + tensor var_15447_begin_0 = const()[name = tensor("op_15447_begin_0"), val = tensor([0, 0, 0, 2560])]; + tensor var_15447_end_0 = const()[name = tensor("op_15447_end_0"), val = tensor([2, 64, 1, 3072])]; + tensor var_15447_end_mask_0 = const()[name = tensor("op_15447_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15447_cast = slice_by_index(begin = var_15447_begin_0, end = var_15447_end_0, end_mask = var_15447_end_mask_0, x = var_15415_cast)[name = tensor("op_15447_cast")]; + tensor var_15448_begin_0 = const()[name = tensor("op_15448_begin_0"), val = tensor([0, 0, 0, 3072])]; + tensor var_15448_end_0 = const()[name = tensor("op_15448_end_0"), val = tensor([2, 64, 1, 3584])]; + tensor var_15448_end_mask_0 = const()[name = tensor("op_15448_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15448_cast = slice_by_index(begin = var_15448_begin_0, end = var_15448_end_0, end_mask = var_15448_end_mask_0, x = var_15415_cast)[name = tensor("op_15448_cast")]; + tensor var_15449_begin_0 = const()[name = tensor("op_15449_begin_0"), val = tensor([0, 0, 0, 3584])]; + tensor var_15449_end_0 = const()[name = tensor("op_15449_end_0"), val = tensor([2, 64, 1, 4096])]; + tensor var_15449_end_mask_0 = const()[name = tensor("op_15449_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15449_cast = slice_by_index(begin = var_15449_begin_0, end = var_15449_end_0, end_mask = var_15449_end_mask_0, x = var_15415_cast)[name = tensor("op_15449_cast")]; + tensor var_15450_begin_0 = const()[name = tensor("op_15450_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_15450_end_0 = const()[name = tensor("op_15450_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_15450_end_mask_0 = const()[name = tensor("op_15450_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15450_cast = slice_by_index(begin = var_15450_begin_0, end = var_15450_end_0, end_mask = var_15450_end_mask_0, x = var_15419_cast)[name = tensor("op_15450_cast")]; + tensor var_15451_begin_0 = const()[name = tensor("op_15451_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_15451_end_0 = const()[name = tensor("op_15451_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_15451_end_mask_0 = const()[name = tensor("op_15451_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15451_cast = slice_by_index(begin = var_15451_begin_0, end = var_15451_end_0, end_mask = var_15451_end_mask_0, x = var_15419_cast)[name = tensor("op_15451_cast")]; + tensor var_15452_begin_0 = const()[name = tensor("op_15452_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_15452_end_0 = const()[name = tensor("op_15452_end_0"), val = tensor([2, 64, 1, 1536])]; + tensor var_15452_end_mask_0 = const()[name = tensor("op_15452_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15452_cast = slice_by_index(begin = var_15452_begin_0, end = var_15452_end_0, end_mask = var_15452_end_mask_0, x = var_15419_cast)[name = tensor("op_15452_cast")]; + tensor var_15453_begin_0 = const()[name = tensor("op_15453_begin_0"), val = tensor([0, 0, 0, 1536])]; + tensor var_15453_end_0 = const()[name = tensor("op_15453_end_0"), val = tensor([2, 64, 1, 2048])]; + tensor var_15453_end_mask_0 = const()[name = tensor("op_15453_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15453_cast = slice_by_index(begin = var_15453_begin_0, end = var_15453_end_0, end_mask = var_15453_end_mask_0, x = var_15419_cast)[name = tensor("op_15453_cast")]; + tensor var_15454_begin_0 = const()[name = tensor("op_15454_begin_0"), val = tensor([0, 0, 0, 2048])]; + tensor var_15454_end_0 = const()[name = tensor("op_15454_end_0"), val = tensor([2, 64, 1, 2560])]; + tensor var_15454_end_mask_0 = const()[name = tensor("op_15454_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15454_cast = slice_by_index(begin = var_15454_begin_0, end = var_15454_end_0, end_mask = var_15454_end_mask_0, x = var_15419_cast)[name = tensor("op_15454_cast")]; + tensor var_15455_begin_0 = const()[name = tensor("op_15455_begin_0"), val = tensor([0, 0, 0, 2560])]; + tensor var_15455_end_0 = const()[name = tensor("op_15455_end_0"), val = tensor([2, 64, 1, 3072])]; + tensor var_15455_end_mask_0 = const()[name = tensor("op_15455_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15455_cast = slice_by_index(begin = var_15455_begin_0, end = var_15455_end_0, end_mask = var_15455_end_mask_0, x = var_15419_cast)[name = tensor("op_15455_cast")]; + tensor var_15456_begin_0 = const()[name = tensor("op_15456_begin_0"), val = tensor([0, 0, 0, 3072])]; + tensor var_15456_end_0 = const()[name = tensor("op_15456_end_0"), val = tensor([2, 64, 1, 3584])]; + tensor var_15456_end_mask_0 = const()[name = tensor("op_15456_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15456_cast = slice_by_index(begin = var_15456_begin_0, end = var_15456_end_0, end_mask = var_15456_end_mask_0, x = var_15419_cast)[name = tensor("op_15456_cast")]; + tensor var_15457_begin_0 = const()[name = tensor("op_15457_begin_0"), val = tensor([0, 0, 0, 3584])]; + tensor var_15457_end_0 = const()[name = tensor("op_15457_end_0"), val = tensor([2, 64, 1, 4096])]; + tensor var_15457_end_mask_0 = const()[name = tensor("op_15457_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15457_cast = slice_by_index(begin = var_15457_begin_0, end = var_15457_end_0, end_mask = var_15457_end_mask_0, x = var_15419_cast)[name = tensor("op_15457_cast")]; + tensor var_15458_begin_0 = const()[name = tensor("op_15458_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_15458_end_0 = const()[name = tensor("op_15458_end_0"), val = tensor([2, 64, 1, 512])]; + tensor var_15458_end_mask_0 = const()[name = tensor("op_15458_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15458_cast = slice_by_index(begin = var_15458_begin_0, end = var_15458_end_0, end_mask = var_15458_end_mask_0, x = var_15423_cast)[name = tensor("op_15458_cast")]; + tensor var_15459_begin_0 = const()[name = tensor("op_15459_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_15459_end_0 = const()[name = tensor("op_15459_end_0"), val = tensor([2, 64, 1, 1024])]; + tensor var_15459_end_mask_0 = const()[name = tensor("op_15459_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15459_cast = slice_by_index(begin = var_15459_begin_0, end = var_15459_end_0, end_mask = var_15459_end_mask_0, x = var_15423_cast)[name = tensor("op_15459_cast")]; + tensor var_15460_begin_0 = const()[name = tensor("op_15460_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_15460_end_0 = const()[name = tensor("op_15460_end_0"), val = tensor([2, 64, 1, 1536])]; + tensor var_15460_end_mask_0 = const()[name = tensor("op_15460_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15460_cast = slice_by_index(begin = var_15460_begin_0, end = var_15460_end_0, end_mask = var_15460_end_mask_0, x = var_15423_cast)[name = tensor("op_15460_cast")]; + tensor var_15461_begin_0 = const()[name = tensor("op_15461_begin_0"), val = tensor([0, 0, 0, 1536])]; + tensor var_15461_end_0 = const()[name = tensor("op_15461_end_0"), val = tensor([2, 64, 1, 2048])]; + tensor var_15461_end_mask_0 = const()[name = tensor("op_15461_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15461_cast = slice_by_index(begin = var_15461_begin_0, end = var_15461_end_0, end_mask = var_15461_end_mask_0, x = var_15423_cast)[name = tensor("op_15461_cast")]; + tensor var_15462_begin_0 = const()[name = tensor("op_15462_begin_0"), val = tensor([0, 0, 0, 2048])]; + tensor var_15462_end_0 = const()[name = tensor("op_15462_end_0"), val = tensor([2, 64, 1, 2560])]; + tensor var_15462_end_mask_0 = const()[name = tensor("op_15462_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15462_cast = slice_by_index(begin = var_15462_begin_0, end = var_15462_end_0, end_mask = var_15462_end_mask_0, x = var_15423_cast)[name = tensor("op_15462_cast")]; + tensor var_15463_begin_0 = const()[name = tensor("op_15463_begin_0"), val = tensor([0, 0, 0, 2560])]; + tensor var_15463_end_0 = const()[name = tensor("op_15463_end_0"), val = tensor([2, 64, 1, 3072])]; + tensor var_15463_end_mask_0 = const()[name = tensor("op_15463_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15463_cast = slice_by_index(begin = var_15463_begin_0, end = var_15463_end_0, end_mask = var_15463_end_mask_0, x = var_15423_cast)[name = tensor("op_15463_cast")]; + tensor var_15464_begin_0 = const()[name = tensor("op_15464_begin_0"), val = tensor([0, 0, 0, 3072])]; + tensor var_15464_end_0 = const()[name = tensor("op_15464_end_0"), val = tensor([2, 64, 1, 3584])]; + tensor var_15464_end_mask_0 = const()[name = tensor("op_15464_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15464_cast = slice_by_index(begin = var_15464_begin_0, end = var_15464_end_0, end_mask = var_15464_end_mask_0, x = var_15423_cast)[name = tensor("op_15464_cast")]; + tensor var_15465_begin_0 = const()[name = tensor("op_15465_begin_0"), val = tensor([0, 0, 0, 3584])]; + tensor var_15465_end_0 = const()[name = tensor("op_15465_end_0"), val = tensor([2, 64, 1, 4096])]; + tensor var_15465_end_mask_0 = const()[name = tensor("op_15465_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15465_cast = slice_by_index(begin = var_15465_begin_0, end = var_15465_end_0, end_mask = var_15465_end_mask_0, x = var_15423_cast)[name = tensor("op_15465_cast")]; + tensor k_perm_0 = const()[name = tensor("k_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_15470_begin_0 = const()[name = tensor("op_15470_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_15470_end_0 = const()[name = tensor("op_15470_end_0"), val = tensor([2, 77, 1, 64])]; + tensor var_15470_end_mask_0 = const()[name = tensor("op_15470_end_mask_0"), val = tensor([true, true, true, false])]; + tensor transpose_0 = transpose(perm = k_perm_0, x = k_125_cast)[name = tensor("transpose_0")]; + tensor var_15470_cast = slice_by_index(begin = var_15470_begin_0, end = var_15470_end_0, end_mask = var_15470_end_mask_0, x = transpose_0)[name = tensor("op_15470_cast")]; + tensor var_15474_begin_0 = const()[name = tensor("op_15474_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_15474_end_0 = const()[name = tensor("op_15474_end_0"), val = tensor([2, 77, 1, 128])]; + tensor var_15474_end_mask_0 = const()[name = tensor("op_15474_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15474_cast = slice_by_index(begin = var_15474_begin_0, end = var_15474_end_0, end_mask = var_15474_end_mask_0, x = transpose_0)[name = tensor("op_15474_cast")]; + tensor var_15478_begin_0 = const()[name = tensor("op_15478_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_15478_end_0 = const()[name = tensor("op_15478_end_0"), val = tensor([2, 77, 1, 192])]; + tensor var_15478_end_mask_0 = const()[name = tensor("op_15478_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15478_cast = slice_by_index(begin = var_15478_begin_0, end = var_15478_end_0, end_mask = var_15478_end_mask_0, x = transpose_0)[name = tensor("op_15478_cast")]; + tensor var_15482_begin_0 = const()[name = tensor("op_15482_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_15482_end_0 = const()[name = tensor("op_15482_end_0"), val = tensor([2, 77, 1, 256])]; + tensor var_15482_end_mask_0 = const()[name = tensor("op_15482_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15482_cast = slice_by_index(begin = var_15482_begin_0, end = var_15482_end_0, end_mask = var_15482_end_mask_0, x = transpose_0)[name = tensor("op_15482_cast")]; + tensor var_15486_begin_0 = const()[name = tensor("op_15486_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_15486_end_0 = const()[name = tensor("op_15486_end_0"), val = tensor([2, 77, 1, 320])]; + tensor var_15486_end_mask_0 = const()[name = tensor("op_15486_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15486_cast = slice_by_index(begin = var_15486_begin_0, end = var_15486_end_0, end_mask = var_15486_end_mask_0, x = transpose_0)[name = tensor("op_15486_cast")]; + tensor var_15488_begin_0 = const()[name = tensor("op_15488_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_15488_end_0 = const()[name = tensor("op_15488_end_0"), val = tensor([2, 64, 1, 77])]; + tensor var_15488_end_mask_0 = const()[name = tensor("op_15488_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15488_cast = slice_by_index(begin = var_15488_begin_0, end = var_15488_end_0, end_mask = var_15488_end_mask_0, x = v_cast)[name = tensor("op_15488_cast")]; + tensor var_15492_begin_0 = const()[name = tensor("op_15492_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_15492_end_0 = const()[name = tensor("op_15492_end_0"), val = tensor([2, 128, 1, 77])]; + tensor var_15492_end_mask_0 = const()[name = tensor("op_15492_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15492_cast = slice_by_index(begin = var_15492_begin_0, end = var_15492_end_0, end_mask = var_15492_end_mask_0, x = v_cast)[name = tensor("op_15492_cast")]; + tensor var_15496_begin_0 = const()[name = tensor("op_15496_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_15496_end_0 = const()[name = tensor("op_15496_end_0"), val = tensor([2, 192, 1, 77])]; + tensor var_15496_end_mask_0 = const()[name = tensor("op_15496_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15496_cast = slice_by_index(begin = var_15496_begin_0, end = var_15496_end_0, end_mask = var_15496_end_mask_0, x = v_cast)[name = tensor("op_15496_cast")]; + tensor var_15500_begin_0 = const()[name = tensor("op_15500_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_15500_end_0 = const()[name = tensor("op_15500_end_0"), val = tensor([2, 256, 1, 77])]; + tensor var_15500_end_mask_0 = const()[name = tensor("op_15500_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15500_cast = slice_by_index(begin = var_15500_begin_0, end = var_15500_end_0, end_mask = var_15500_end_mask_0, x = v_cast)[name = tensor("op_15500_cast")]; + tensor var_15504_begin_0 = const()[name = tensor("op_15504_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_15504_end_0 = const()[name = tensor("op_15504_end_0"), val = tensor([2, 320, 1, 77])]; + tensor var_15504_end_mask_0 = const()[name = tensor("op_15504_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15504_cast = slice_by_index(begin = var_15504_begin_0, end = var_15504_end_0, end_mask = var_15504_end_mask_0, x = v_cast)[name = tensor("op_15504_cast")]; + tensor var_15508_equation_0 = const()[name = tensor("op_15508_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15508_cast = einsum(equation = var_15508_equation_0, values = (var_15470_cast, var_15426_cast))[name = tensor("op_15508_cast")]; + tensor var_15509_to_fp16 = const()[name = tensor("op_15509_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1121_cast = mul(x = var_15508_cast, y = var_15509_to_fp16)[name = tensor("aw_chunk_1121_cast")]; + tensor var_15512_equation_0 = const()[name = tensor("op_15512_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15512_cast = einsum(equation = var_15512_equation_0, values = (var_15470_cast, var_15427_cast))[name = tensor("op_15512_cast")]; + tensor var_15513_to_fp16 = const()[name = tensor("op_15513_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1123_cast = mul(x = var_15512_cast, y = var_15513_to_fp16)[name = tensor("aw_chunk_1123_cast")]; + tensor var_15516_equation_0 = const()[name = tensor("op_15516_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15516_cast = einsum(equation = var_15516_equation_0, values = (var_15470_cast, var_15428_cast))[name = tensor("op_15516_cast")]; + tensor var_15517_to_fp16 = const()[name = tensor("op_15517_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1125_cast = mul(x = var_15516_cast, y = var_15517_to_fp16)[name = tensor("aw_chunk_1125_cast")]; + tensor var_15520_equation_0 = const()[name = tensor("op_15520_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15520_cast = einsum(equation = var_15520_equation_0, values = (var_15470_cast, var_15429_cast))[name = tensor("op_15520_cast")]; + tensor var_15521_to_fp16 = const()[name = tensor("op_15521_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1127_cast = mul(x = var_15520_cast, y = var_15521_to_fp16)[name = tensor("aw_chunk_1127_cast")]; + tensor var_15524_equation_0 = const()[name = tensor("op_15524_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15524_cast = einsum(equation = var_15524_equation_0, values = (var_15470_cast, var_15430_cast))[name = tensor("op_15524_cast")]; + tensor var_15525_to_fp16 = const()[name = tensor("op_15525_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1129_cast = mul(x = var_15524_cast, y = var_15525_to_fp16)[name = tensor("aw_chunk_1129_cast")]; + tensor var_15528_equation_0 = const()[name = tensor("op_15528_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15528_cast = einsum(equation = var_15528_equation_0, values = (var_15470_cast, var_15431_cast))[name = tensor("op_15528_cast")]; + tensor var_15529_to_fp16 = const()[name = tensor("op_15529_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1131_cast = mul(x = var_15528_cast, y = var_15529_to_fp16)[name = tensor("aw_chunk_1131_cast")]; + tensor var_15532_equation_0 = const()[name = tensor("op_15532_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15532_cast = einsum(equation = var_15532_equation_0, values = (var_15470_cast, var_15432_cast))[name = tensor("op_15532_cast")]; + tensor var_15533_to_fp16 = const()[name = tensor("op_15533_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1133_cast = mul(x = var_15532_cast, y = var_15533_to_fp16)[name = tensor("aw_chunk_1133_cast")]; + tensor var_15536_equation_0 = const()[name = tensor("op_15536_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15536_cast = einsum(equation = var_15536_equation_0, values = (var_15470_cast, var_15433_cast))[name = tensor("op_15536_cast")]; + tensor var_15537_to_fp16 = const()[name = tensor("op_15537_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1135_cast = mul(x = var_15536_cast, y = var_15537_to_fp16)[name = tensor("aw_chunk_1135_cast")]; + tensor var_15540_equation_0 = const()[name = tensor("op_15540_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15540_cast = einsum(equation = var_15540_equation_0, values = (var_15474_cast, var_15434_cast))[name = tensor("op_15540_cast")]; + tensor var_15541_to_fp16 = const()[name = tensor("op_15541_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1137_cast = mul(x = var_15540_cast, y = var_15541_to_fp16)[name = tensor("aw_chunk_1137_cast")]; + tensor var_15544_equation_0 = const()[name = tensor("op_15544_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15544_cast = einsum(equation = var_15544_equation_0, values = (var_15474_cast, var_15435_cast))[name = tensor("op_15544_cast")]; + tensor var_15545_to_fp16 = const()[name = tensor("op_15545_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1139_cast = mul(x = var_15544_cast, y = var_15545_to_fp16)[name = tensor("aw_chunk_1139_cast")]; + tensor var_15548_equation_0 = const()[name = tensor("op_15548_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15548_cast = einsum(equation = var_15548_equation_0, values = (var_15474_cast, var_15436_cast))[name = tensor("op_15548_cast")]; + tensor var_15549_to_fp16 = const()[name = tensor("op_15549_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1141_cast = mul(x = var_15548_cast, y = var_15549_to_fp16)[name = tensor("aw_chunk_1141_cast")]; + tensor var_15552_equation_0 = const()[name = tensor("op_15552_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15552_cast = einsum(equation = var_15552_equation_0, values = (var_15474_cast, var_15437_cast))[name = tensor("op_15552_cast")]; + tensor var_15553_to_fp16 = const()[name = tensor("op_15553_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1143_cast = mul(x = var_15552_cast, y = var_15553_to_fp16)[name = tensor("aw_chunk_1143_cast")]; + tensor var_15556_equation_0 = const()[name = tensor("op_15556_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15556_cast = einsum(equation = var_15556_equation_0, values = (var_15474_cast, var_15438_cast))[name = tensor("op_15556_cast")]; + tensor var_15557_to_fp16 = const()[name = tensor("op_15557_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1145_cast = mul(x = var_15556_cast, y = var_15557_to_fp16)[name = tensor("aw_chunk_1145_cast")]; + tensor var_15560_equation_0 = const()[name = tensor("op_15560_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15560_cast = einsum(equation = var_15560_equation_0, values = (var_15474_cast, var_15439_cast))[name = tensor("op_15560_cast")]; + tensor var_15561_to_fp16 = const()[name = tensor("op_15561_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1147_cast = mul(x = var_15560_cast, y = var_15561_to_fp16)[name = tensor("aw_chunk_1147_cast")]; + tensor var_15564_equation_0 = const()[name = tensor("op_15564_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15564_cast = einsum(equation = var_15564_equation_0, values = (var_15474_cast, var_15440_cast))[name = tensor("op_15564_cast")]; + tensor var_15565_to_fp16 = const()[name = tensor("op_15565_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1149_cast = mul(x = var_15564_cast, y = var_15565_to_fp16)[name = tensor("aw_chunk_1149_cast")]; + tensor var_15568_equation_0 = const()[name = tensor("op_15568_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15568_cast = einsum(equation = var_15568_equation_0, values = (var_15474_cast, var_15441_cast))[name = tensor("op_15568_cast")]; + tensor var_15569_to_fp16 = const()[name = tensor("op_15569_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1151_cast = mul(x = var_15568_cast, y = var_15569_to_fp16)[name = tensor("aw_chunk_1151_cast")]; + tensor var_15572_equation_0 = const()[name = tensor("op_15572_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15572_cast = einsum(equation = var_15572_equation_0, values = (var_15478_cast, var_15442_cast))[name = tensor("op_15572_cast")]; + tensor var_15573_to_fp16 = const()[name = tensor("op_15573_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1153_cast = mul(x = var_15572_cast, y = var_15573_to_fp16)[name = tensor("aw_chunk_1153_cast")]; + tensor var_15576_equation_0 = const()[name = tensor("op_15576_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15576_cast = einsum(equation = var_15576_equation_0, values = (var_15478_cast, var_15443_cast))[name = tensor("op_15576_cast")]; + tensor var_15577_to_fp16 = const()[name = tensor("op_15577_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1155_cast = mul(x = var_15576_cast, y = var_15577_to_fp16)[name = tensor("aw_chunk_1155_cast")]; + tensor var_15580_equation_0 = const()[name = tensor("op_15580_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15580_cast = einsum(equation = var_15580_equation_0, values = (var_15478_cast, var_15444_cast))[name = tensor("op_15580_cast")]; + tensor var_15581_to_fp16 = const()[name = tensor("op_15581_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1157_cast = mul(x = var_15580_cast, y = var_15581_to_fp16)[name = tensor("aw_chunk_1157_cast")]; + tensor var_15584_equation_0 = const()[name = tensor("op_15584_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15584_cast = einsum(equation = var_15584_equation_0, values = (var_15478_cast, var_15445_cast))[name = tensor("op_15584_cast")]; + tensor var_15585_to_fp16 = const()[name = tensor("op_15585_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1159_cast = mul(x = var_15584_cast, y = var_15585_to_fp16)[name = tensor("aw_chunk_1159_cast")]; + tensor var_15588_equation_0 = const()[name = tensor("op_15588_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15588_cast = einsum(equation = var_15588_equation_0, values = (var_15478_cast, var_15446_cast))[name = tensor("op_15588_cast")]; + tensor var_15589_to_fp16 = const()[name = tensor("op_15589_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1161_cast = mul(x = var_15588_cast, y = var_15589_to_fp16)[name = tensor("aw_chunk_1161_cast")]; + tensor var_15592_equation_0 = const()[name = tensor("op_15592_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15592_cast = einsum(equation = var_15592_equation_0, values = (var_15478_cast, var_15447_cast))[name = tensor("op_15592_cast")]; + tensor var_15593_to_fp16 = const()[name = tensor("op_15593_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1163_cast = mul(x = var_15592_cast, y = var_15593_to_fp16)[name = tensor("aw_chunk_1163_cast")]; + tensor var_15596_equation_0 = const()[name = tensor("op_15596_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15596_cast = einsum(equation = var_15596_equation_0, values = (var_15478_cast, var_15448_cast))[name = tensor("op_15596_cast")]; + tensor var_15597_to_fp16 = const()[name = tensor("op_15597_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1165_cast = mul(x = var_15596_cast, y = var_15597_to_fp16)[name = tensor("aw_chunk_1165_cast")]; + tensor var_15600_equation_0 = const()[name = tensor("op_15600_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15600_cast = einsum(equation = var_15600_equation_0, values = (var_15478_cast, var_15449_cast))[name = tensor("op_15600_cast")]; + tensor var_15601_to_fp16 = const()[name = tensor("op_15601_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1167_cast = mul(x = var_15600_cast, y = var_15601_to_fp16)[name = tensor("aw_chunk_1167_cast")]; + tensor var_15604_equation_0 = const()[name = tensor("op_15604_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15604_cast = einsum(equation = var_15604_equation_0, values = (var_15482_cast, var_15450_cast))[name = tensor("op_15604_cast")]; + tensor var_15605_to_fp16 = const()[name = tensor("op_15605_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1169_cast = mul(x = var_15604_cast, y = var_15605_to_fp16)[name = tensor("aw_chunk_1169_cast")]; + tensor var_15608_equation_0 = const()[name = tensor("op_15608_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15608_cast = einsum(equation = var_15608_equation_0, values = (var_15482_cast, var_15451_cast))[name = tensor("op_15608_cast")]; + tensor var_15609_to_fp16 = const()[name = tensor("op_15609_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1171_cast = mul(x = var_15608_cast, y = var_15609_to_fp16)[name = tensor("aw_chunk_1171_cast")]; + tensor var_15612_equation_0 = const()[name = tensor("op_15612_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15612_cast = einsum(equation = var_15612_equation_0, values = (var_15482_cast, var_15452_cast))[name = tensor("op_15612_cast")]; + tensor var_15613_to_fp16 = const()[name = tensor("op_15613_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1173_cast = mul(x = var_15612_cast, y = var_15613_to_fp16)[name = tensor("aw_chunk_1173_cast")]; + tensor var_15616_equation_0 = const()[name = tensor("op_15616_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15616_cast = einsum(equation = var_15616_equation_0, values = (var_15482_cast, var_15453_cast))[name = tensor("op_15616_cast")]; + tensor var_15617_to_fp16 = const()[name = tensor("op_15617_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1175_cast = mul(x = var_15616_cast, y = var_15617_to_fp16)[name = tensor("aw_chunk_1175_cast")]; + tensor var_15620_equation_0 = const()[name = tensor("op_15620_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15620_cast = einsum(equation = var_15620_equation_0, values = (var_15482_cast, var_15454_cast))[name = tensor("op_15620_cast")]; + tensor var_15621_to_fp16 = const()[name = tensor("op_15621_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1177_cast = mul(x = var_15620_cast, y = var_15621_to_fp16)[name = tensor("aw_chunk_1177_cast")]; + tensor var_15624_equation_0 = const()[name = tensor("op_15624_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15624_cast = einsum(equation = var_15624_equation_0, values = (var_15482_cast, var_15455_cast))[name = tensor("op_15624_cast")]; + tensor var_15625_to_fp16 = const()[name = tensor("op_15625_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1179_cast = mul(x = var_15624_cast, y = var_15625_to_fp16)[name = tensor("aw_chunk_1179_cast")]; + tensor var_15628_equation_0 = const()[name = tensor("op_15628_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15628_cast = einsum(equation = var_15628_equation_0, values = (var_15482_cast, var_15456_cast))[name = tensor("op_15628_cast")]; + tensor var_15629_to_fp16 = const()[name = tensor("op_15629_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1181_cast = mul(x = var_15628_cast, y = var_15629_to_fp16)[name = tensor("aw_chunk_1181_cast")]; + tensor var_15632_equation_0 = const()[name = tensor("op_15632_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15632_cast = einsum(equation = var_15632_equation_0, values = (var_15482_cast, var_15457_cast))[name = tensor("op_15632_cast")]; + tensor var_15633_to_fp16 = const()[name = tensor("op_15633_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1183_cast = mul(x = var_15632_cast, y = var_15633_to_fp16)[name = tensor("aw_chunk_1183_cast")]; + tensor var_15636_equation_0 = const()[name = tensor("op_15636_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15636_cast = einsum(equation = var_15636_equation_0, values = (var_15486_cast, var_15458_cast))[name = tensor("op_15636_cast")]; + tensor var_15637_to_fp16 = const()[name = tensor("op_15637_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1185_cast = mul(x = var_15636_cast, y = var_15637_to_fp16)[name = tensor("aw_chunk_1185_cast")]; + tensor var_15640_equation_0 = const()[name = tensor("op_15640_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15640_cast = einsum(equation = var_15640_equation_0, values = (var_15486_cast, var_15459_cast))[name = tensor("op_15640_cast")]; + tensor var_15641_to_fp16 = const()[name = tensor("op_15641_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1187_cast = mul(x = var_15640_cast, y = var_15641_to_fp16)[name = tensor("aw_chunk_1187_cast")]; + tensor var_15644_equation_0 = const()[name = tensor("op_15644_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15644_cast = einsum(equation = var_15644_equation_0, values = (var_15486_cast, var_15460_cast))[name = tensor("op_15644_cast")]; + tensor var_15645_to_fp16 = const()[name = tensor("op_15645_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1189_cast = mul(x = var_15644_cast, y = var_15645_to_fp16)[name = tensor("aw_chunk_1189_cast")]; + tensor var_15648_equation_0 = const()[name = tensor("op_15648_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15648_cast = einsum(equation = var_15648_equation_0, values = (var_15486_cast, var_15461_cast))[name = tensor("op_15648_cast")]; + tensor var_15649_to_fp16 = const()[name = tensor("op_15649_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1191_cast = mul(x = var_15648_cast, y = var_15649_to_fp16)[name = tensor("aw_chunk_1191_cast")]; + tensor var_15652_equation_0 = const()[name = tensor("op_15652_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15652_cast = einsum(equation = var_15652_equation_0, values = (var_15486_cast, var_15462_cast))[name = tensor("op_15652_cast")]; + tensor var_15653_to_fp16 = const()[name = tensor("op_15653_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1193_cast = mul(x = var_15652_cast, y = var_15653_to_fp16)[name = tensor("aw_chunk_1193_cast")]; + tensor var_15656_equation_0 = const()[name = tensor("op_15656_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15656_cast = einsum(equation = var_15656_equation_0, values = (var_15486_cast, var_15463_cast))[name = tensor("op_15656_cast")]; + tensor var_15657_to_fp16 = const()[name = tensor("op_15657_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1195_cast = mul(x = var_15656_cast, y = var_15657_to_fp16)[name = tensor("aw_chunk_1195_cast")]; + tensor var_15660_equation_0 = const()[name = tensor("op_15660_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15660_cast = einsum(equation = var_15660_equation_0, values = (var_15486_cast, var_15464_cast))[name = tensor("op_15660_cast")]; + tensor var_15661_to_fp16 = const()[name = tensor("op_15661_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1197_cast = mul(x = var_15660_cast, y = var_15661_to_fp16)[name = tensor("aw_chunk_1197_cast")]; + tensor var_15664_equation_0 = const()[name = tensor("op_15664_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15664_cast = einsum(equation = var_15664_equation_0, values = (var_15486_cast, var_15465_cast))[name = tensor("op_15664_cast")]; + tensor var_15665_to_fp16 = const()[name = tensor("op_15665_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_cast = mul(x = var_15664_cast, y = var_15665_to_fp16)[name = tensor("aw_chunk_cast")]; + tensor var_15667_cast = softmax(axis = var_12807, x = aw_chunk_1121_cast)[name = tensor("op_15667_cast")]; + tensor var_15668_cast = softmax(axis = var_12807, x = aw_chunk_1123_cast)[name = tensor("op_15668_cast")]; + tensor var_15669_cast = softmax(axis = var_12807, x = aw_chunk_1125_cast)[name = tensor("op_15669_cast")]; + tensor var_15670_cast = softmax(axis = var_12807, x = aw_chunk_1127_cast)[name = tensor("op_15670_cast")]; + tensor var_15671_cast = softmax(axis = var_12807, x = aw_chunk_1129_cast)[name = tensor("op_15671_cast")]; + tensor var_15672_cast = softmax(axis = var_12807, x = aw_chunk_1131_cast)[name = tensor("op_15672_cast")]; + tensor var_15673_cast = softmax(axis = var_12807, x = aw_chunk_1133_cast)[name = tensor("op_15673_cast")]; + tensor var_15674_cast = softmax(axis = var_12807, x = aw_chunk_1135_cast)[name = tensor("op_15674_cast")]; + tensor var_15675_cast = softmax(axis = var_12807, x = aw_chunk_1137_cast)[name = tensor("op_15675_cast")]; + tensor var_15676_cast = softmax(axis = var_12807, x = aw_chunk_1139_cast)[name = tensor("op_15676_cast")]; + tensor var_15677_cast = softmax(axis = var_12807, x = aw_chunk_1141_cast)[name = tensor("op_15677_cast")]; + tensor var_15678_cast = softmax(axis = var_12807, x = aw_chunk_1143_cast)[name = tensor("op_15678_cast")]; + tensor var_15679_cast = softmax(axis = var_12807, x = aw_chunk_1145_cast)[name = tensor("op_15679_cast")]; + tensor var_15680_cast = softmax(axis = var_12807, x = aw_chunk_1147_cast)[name = tensor("op_15680_cast")]; + tensor var_15681_cast = softmax(axis = var_12807, x = aw_chunk_1149_cast)[name = tensor("op_15681_cast")]; + tensor var_15682_cast = softmax(axis = var_12807, x = aw_chunk_1151_cast)[name = tensor("op_15682_cast")]; + tensor var_15683_cast = softmax(axis = var_12807, x = aw_chunk_1153_cast)[name = tensor("op_15683_cast")]; + tensor var_15684_cast = softmax(axis = var_12807, x = aw_chunk_1155_cast)[name = tensor("op_15684_cast")]; + tensor var_15685_cast = softmax(axis = var_12807, x = aw_chunk_1157_cast)[name = tensor("op_15685_cast")]; + tensor var_15686_cast = softmax(axis = var_12807, x = aw_chunk_1159_cast)[name = tensor("op_15686_cast")]; + tensor var_15687_cast = softmax(axis = var_12807, x = aw_chunk_1161_cast)[name = tensor("op_15687_cast")]; + tensor var_15688_cast = softmax(axis = var_12807, x = aw_chunk_1163_cast)[name = tensor("op_15688_cast")]; + tensor var_15689_cast = softmax(axis = var_12807, x = aw_chunk_1165_cast)[name = tensor("op_15689_cast")]; + tensor var_15690_cast = softmax(axis = var_12807, x = aw_chunk_1167_cast)[name = tensor("op_15690_cast")]; + tensor var_15691_cast = softmax(axis = var_12807, x = aw_chunk_1169_cast)[name = tensor("op_15691_cast")]; + tensor var_15692_cast = softmax(axis = var_12807, x = aw_chunk_1171_cast)[name = tensor("op_15692_cast")]; + tensor var_15693_cast = softmax(axis = var_12807, x = aw_chunk_1173_cast)[name = tensor("op_15693_cast")]; + tensor var_15694_cast = softmax(axis = var_12807, x = aw_chunk_1175_cast)[name = tensor("op_15694_cast")]; + tensor var_15695_cast = softmax(axis = var_12807, x = aw_chunk_1177_cast)[name = tensor("op_15695_cast")]; + tensor var_15696_cast = softmax(axis = var_12807, x = aw_chunk_1179_cast)[name = tensor("op_15696_cast")]; + tensor var_15697_cast = softmax(axis = var_12807, x = aw_chunk_1181_cast)[name = tensor("op_15697_cast")]; + tensor var_15698_cast = softmax(axis = var_12807, x = aw_chunk_1183_cast)[name = tensor("op_15698_cast")]; + tensor var_15699_cast = softmax(axis = var_12807, x = aw_chunk_1185_cast)[name = tensor("op_15699_cast")]; + tensor var_15700_cast = softmax(axis = var_12807, x = aw_chunk_1187_cast)[name = tensor("op_15700_cast")]; + tensor var_15701_cast = softmax(axis = var_12807, x = aw_chunk_1189_cast)[name = tensor("op_15701_cast")]; + tensor var_15702_cast = softmax(axis = var_12807, x = aw_chunk_1191_cast)[name = tensor("op_15702_cast")]; + tensor var_15703_cast = softmax(axis = var_12807, x = aw_chunk_1193_cast)[name = tensor("op_15703_cast")]; + tensor var_15704_cast = softmax(axis = var_12807, x = aw_chunk_1195_cast)[name = tensor("op_15704_cast")]; + tensor var_15705_cast = softmax(axis = var_12807, x = aw_chunk_1197_cast)[name = tensor("op_15705_cast")]; + tensor var_15706_cast = softmax(axis = var_12807, x = aw_chunk_cast)[name = tensor("op_15706_cast")]; + tensor var_15708_equation_0 = const()[name = tensor("op_15708_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15708_cast = einsum(equation = var_15708_equation_0, values = (var_15488_cast, var_15667_cast))[name = tensor("op_15708_cast")]; + tensor var_15710_equation_0 = const()[name = tensor("op_15710_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15710_cast = einsum(equation = var_15710_equation_0, values = (var_15488_cast, var_15668_cast))[name = tensor("op_15710_cast")]; + tensor var_15712_equation_0 = const()[name = tensor("op_15712_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15712_cast = einsum(equation = var_15712_equation_0, values = (var_15488_cast, var_15669_cast))[name = tensor("op_15712_cast")]; + tensor var_15714_equation_0 = const()[name = tensor("op_15714_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15714_cast = einsum(equation = var_15714_equation_0, values = (var_15488_cast, var_15670_cast))[name = tensor("op_15714_cast")]; + tensor var_15716_equation_0 = const()[name = tensor("op_15716_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15716_cast = einsum(equation = var_15716_equation_0, values = (var_15488_cast, var_15671_cast))[name = tensor("op_15716_cast")]; + tensor var_15718_equation_0 = const()[name = tensor("op_15718_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15718_cast = einsum(equation = var_15718_equation_0, values = (var_15488_cast, var_15672_cast))[name = tensor("op_15718_cast")]; + tensor var_15720_equation_0 = const()[name = tensor("op_15720_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15720_cast = einsum(equation = var_15720_equation_0, values = (var_15488_cast, var_15673_cast))[name = tensor("op_15720_cast")]; + tensor var_15722_equation_0 = const()[name = tensor("op_15722_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15722_cast = einsum(equation = var_15722_equation_0, values = (var_15488_cast, var_15674_cast))[name = tensor("op_15722_cast")]; + tensor var_15724_equation_0 = const()[name = tensor("op_15724_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15724_cast = einsum(equation = var_15724_equation_0, values = (var_15492_cast, var_15675_cast))[name = tensor("op_15724_cast")]; + tensor var_15726_equation_0 = const()[name = tensor("op_15726_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15726_cast = einsum(equation = var_15726_equation_0, values = (var_15492_cast, var_15676_cast))[name = tensor("op_15726_cast")]; + tensor var_15728_equation_0 = const()[name = tensor("op_15728_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15728_cast = einsum(equation = var_15728_equation_0, values = (var_15492_cast, var_15677_cast))[name = tensor("op_15728_cast")]; + tensor var_15730_equation_0 = const()[name = tensor("op_15730_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15730_cast = einsum(equation = var_15730_equation_0, values = (var_15492_cast, var_15678_cast))[name = tensor("op_15730_cast")]; + tensor var_15732_equation_0 = const()[name = tensor("op_15732_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15732_cast = einsum(equation = var_15732_equation_0, values = (var_15492_cast, var_15679_cast))[name = tensor("op_15732_cast")]; + tensor var_15734_equation_0 = const()[name = tensor("op_15734_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15734_cast = einsum(equation = var_15734_equation_0, values = (var_15492_cast, var_15680_cast))[name = tensor("op_15734_cast")]; + tensor var_15736_equation_0 = const()[name = tensor("op_15736_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15736_cast = einsum(equation = var_15736_equation_0, values = (var_15492_cast, var_15681_cast))[name = tensor("op_15736_cast")]; + tensor var_15738_equation_0 = const()[name = tensor("op_15738_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15738_cast = einsum(equation = var_15738_equation_0, values = (var_15492_cast, var_15682_cast))[name = tensor("op_15738_cast")]; + tensor var_15740_equation_0 = const()[name = tensor("op_15740_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15740_cast = einsum(equation = var_15740_equation_0, values = (var_15496_cast, var_15683_cast))[name = tensor("op_15740_cast")]; + tensor var_15742_equation_0 = const()[name = tensor("op_15742_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15742_cast = einsum(equation = var_15742_equation_0, values = (var_15496_cast, var_15684_cast))[name = tensor("op_15742_cast")]; + tensor var_15744_equation_0 = const()[name = tensor("op_15744_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15744_cast = einsum(equation = var_15744_equation_0, values = (var_15496_cast, var_15685_cast))[name = tensor("op_15744_cast")]; + tensor var_15746_equation_0 = const()[name = tensor("op_15746_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15746_cast = einsum(equation = var_15746_equation_0, values = (var_15496_cast, var_15686_cast))[name = tensor("op_15746_cast")]; + tensor var_15748_equation_0 = const()[name = tensor("op_15748_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15748_cast = einsum(equation = var_15748_equation_0, values = (var_15496_cast, var_15687_cast))[name = tensor("op_15748_cast")]; + tensor var_15750_equation_0 = const()[name = tensor("op_15750_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15750_cast = einsum(equation = var_15750_equation_0, values = (var_15496_cast, var_15688_cast))[name = tensor("op_15750_cast")]; + tensor var_15752_equation_0 = const()[name = tensor("op_15752_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15752_cast = einsum(equation = var_15752_equation_0, values = (var_15496_cast, var_15689_cast))[name = tensor("op_15752_cast")]; + tensor var_15754_equation_0 = const()[name = tensor("op_15754_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15754_cast = einsum(equation = var_15754_equation_0, values = (var_15496_cast, var_15690_cast))[name = tensor("op_15754_cast")]; + tensor var_15756_equation_0 = const()[name = tensor("op_15756_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15756_cast = einsum(equation = var_15756_equation_0, values = (var_15500_cast, var_15691_cast))[name = tensor("op_15756_cast")]; + tensor var_15758_equation_0 = const()[name = tensor("op_15758_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15758_cast = einsum(equation = var_15758_equation_0, values = (var_15500_cast, var_15692_cast))[name = tensor("op_15758_cast")]; + tensor var_15760_equation_0 = const()[name = tensor("op_15760_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15760_cast = einsum(equation = var_15760_equation_0, values = (var_15500_cast, var_15693_cast))[name = tensor("op_15760_cast")]; + tensor var_15762_equation_0 = const()[name = tensor("op_15762_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15762_cast = einsum(equation = var_15762_equation_0, values = (var_15500_cast, var_15694_cast))[name = tensor("op_15762_cast")]; + tensor var_15764_equation_0 = const()[name = tensor("op_15764_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15764_cast = einsum(equation = var_15764_equation_0, values = (var_15500_cast, var_15695_cast))[name = tensor("op_15764_cast")]; + tensor var_15766_equation_0 = const()[name = tensor("op_15766_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15766_cast = einsum(equation = var_15766_equation_0, values = (var_15500_cast, var_15696_cast))[name = tensor("op_15766_cast")]; + tensor var_15768_equation_0 = const()[name = tensor("op_15768_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15768_cast = einsum(equation = var_15768_equation_0, values = (var_15500_cast, var_15697_cast))[name = tensor("op_15768_cast")]; + tensor var_15770_equation_0 = const()[name = tensor("op_15770_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15770_cast = einsum(equation = var_15770_equation_0, values = (var_15500_cast, var_15698_cast))[name = tensor("op_15770_cast")]; + tensor var_15772_equation_0 = const()[name = tensor("op_15772_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15772_cast = einsum(equation = var_15772_equation_0, values = (var_15504_cast, var_15699_cast))[name = tensor("op_15772_cast")]; + tensor var_15774_equation_0 = const()[name = tensor("op_15774_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15774_cast = einsum(equation = var_15774_equation_0, values = (var_15504_cast, var_15700_cast))[name = tensor("op_15774_cast")]; + tensor var_15776_equation_0 = const()[name = tensor("op_15776_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15776_cast = einsum(equation = var_15776_equation_0, values = (var_15504_cast, var_15701_cast))[name = tensor("op_15776_cast")]; + tensor var_15778_equation_0 = const()[name = tensor("op_15778_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15778_cast = einsum(equation = var_15778_equation_0, values = (var_15504_cast, var_15702_cast))[name = tensor("op_15778_cast")]; + tensor var_15780_equation_0 = const()[name = tensor("op_15780_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15780_cast = einsum(equation = var_15780_equation_0, values = (var_15504_cast, var_15703_cast))[name = tensor("op_15780_cast")]; + tensor var_15782_equation_0 = const()[name = tensor("op_15782_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15782_cast = einsum(equation = var_15782_equation_0, values = (var_15504_cast, var_15704_cast))[name = tensor("op_15782_cast")]; + tensor var_15784_equation_0 = const()[name = tensor("op_15784_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15784_cast = einsum(equation = var_15784_equation_0, values = (var_15504_cast, var_15705_cast))[name = tensor("op_15784_cast")]; + tensor var_15786_equation_0 = const()[name = tensor("op_15786_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_15786_cast = einsum(equation = var_15786_equation_0, values = (var_15504_cast, var_15706_cast))[name = tensor("op_15786_cast")]; + tensor var_15788_interleave_0 = const()[name = tensor("op_15788_interleave_0"), val = tensor(false)]; + tensor var_15788_cast = concat(axis = var_12782, interleave = var_15788_interleave_0, values = (var_15708_cast, var_15710_cast, var_15712_cast, var_15714_cast, var_15716_cast, var_15718_cast, var_15720_cast, var_15722_cast))[name = tensor("op_15788_cast")]; + tensor var_15790_interleave_0 = const()[name = tensor("op_15790_interleave_0"), val = tensor(false)]; + tensor var_15790_cast = concat(axis = var_12782, interleave = var_15790_interleave_0, values = (var_15724_cast, var_15726_cast, var_15728_cast, var_15730_cast, var_15732_cast, var_15734_cast, var_15736_cast, var_15738_cast))[name = tensor("op_15790_cast")]; + tensor var_15792_interleave_0 = const()[name = tensor("op_15792_interleave_0"), val = tensor(false)]; + tensor var_15792_cast = concat(axis = var_12782, interleave = var_15792_interleave_0, values = (var_15740_cast, var_15742_cast, var_15744_cast, var_15746_cast, var_15748_cast, var_15750_cast, var_15752_cast, var_15754_cast))[name = tensor("op_15792_cast")]; + tensor var_15794_interleave_0 = const()[name = tensor("op_15794_interleave_0"), val = tensor(false)]; + tensor var_15794_cast = concat(axis = var_12782, interleave = var_15794_interleave_0, values = (var_15756_cast, var_15758_cast, var_15760_cast, var_15762_cast, var_15764_cast, var_15766_cast, var_15768_cast, var_15770_cast))[name = tensor("op_15794_cast")]; + tensor var_15796_interleave_0 = const()[name = tensor("op_15796_interleave_0"), val = tensor(false)]; + tensor var_15796_cast = concat(axis = var_12782, interleave = var_15796_interleave_0, values = (var_15772_cast, var_15774_cast, var_15776_cast, var_15778_cast, var_15780_cast, var_15782_cast, var_15784_cast, var_15786_cast))[name = tensor("op_15796_cast")]; + tensor input_519_interleave_0 = const()[name = tensor("input_519_interleave_0"), val = tensor(false)]; + tensor input_519_cast = concat(axis = var_12807, interleave = input_519_interleave_0, values = (var_15788_cast, var_15790_cast, var_15792_cast, var_15794_cast, var_15796_cast))[name = tensor("input_519_cast")]; + tensor var_15802 = const()[name = tensor("op_15802"), val = tensor([1, 1])]; + tensor var_15804 = const()[name = tensor("op_15804"), val = tensor([1, 1])]; + tensor var_15806_pad_type_0 = const()[name = tensor("op_15806_pad_type_0"), val = tensor("custom")]; + tensor var_15806_pad_0 = const()[name = tensor("op_15806_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_3_attentions_2_transformer_blocks_0_attn2_to_out_0_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(648867712))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(648944576))), name = tensor("up_blocks_3_attentions_2_transformer_blocks_0_attn2_to_out_0_weight_to_fp16_palettized"), shape = tensor([320, 320, 1, 1])]; + tensor up_blocks_3_attentions_2_transformer_blocks_0_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("up_blocks_3_attentions_2_transformer_blocks_0_attn2_to_out_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(648944768)))]; + tensor var_15806_cast = conv(bias = up_blocks_3_attentions_2_transformer_blocks_0_attn2_to_out_0_bias_to_fp16, dilations = var_15804, groups = var_12807, pad = var_15806_pad_0, pad_type = var_15806_pad_type_0, strides = var_15802, weight = up_blocks_3_attentions_2_transformer_blocks_0_attn2_to_out_0_weight_to_fp16_palettized, x = input_519_cast)[name = tensor("op_15806_cast")]; + tensor inputs_cast = add(x = var_15806_cast, y = inputs_93_cast)[name = tensor("inputs_cast")]; + tensor var_15810 = const()[name = tensor("op_15810"), val = tensor([1])]; + tensor channels_mean_cast = reduce_mean(axes = var_15810, keep_dims = var_12802, x = inputs_cast)[name = tensor("channels_mean_cast")]; + tensor zero_mean_cast = sub(x = inputs_cast, y = channels_mean_cast)[name = tensor("zero_mean_cast")]; + tensor zero_mean_sq_cast = mul(x = zero_mean_cast, y = zero_mean_cast)[name = tensor("zero_mean_sq_cast")]; + tensor var_15814 = const()[name = tensor("op_15814"), val = tensor([1])]; + tensor var_15815_cast = reduce_mean(axes = var_15814, keep_dims = var_12802, x = zero_mean_sq_cast)[name = tensor("op_15815_cast")]; + tensor var_15816_to_fp16 = const()[name = tensor("op_15816_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_15817_cast = add(x = var_15815_cast, y = var_15816_to_fp16)[name = tensor("op_15817_cast")]; + tensor denom_epsilon_0_to_fp16 = const()[name = tensor("denom_epsilon_0_to_fp16"), val = tensor(0x1p-24)]; + tensor denom_cast = rsqrt(epsilon = denom_epsilon_0_to_fp16, x = var_15817_cast)[name = tensor("denom_cast")]; + tensor out_cast = mul(x = zero_mean_cast, y = denom_cast)[name = tensor("out_cast")]; + tensor var_15821_to_fp16 = const()[name = tensor("op_15821_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(648945472)))]; + tensor var_15822_cast = add(x = out_cast, y = var_15821_to_fp16)[name = tensor("op_15822_cast")]; + tensor var_15824_to_fp16 = const()[name = tensor("op_15824_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(648946176)))]; + tensor input_521_cast = mul(x = var_15822_cast, y = var_15824_to_fp16)[name = tensor("input_521_cast")]; + tensor var_15832 = const()[name = tensor("op_15832"), val = tensor([1, 1])]; + tensor var_15834 = const()[name = tensor("op_15834"), val = tensor([1, 1])]; + tensor var_15836_pad_type_0 = const()[name = tensor("op_15836_pad_type_0"), val = tensor("custom")]; + tensor var_15836_pad_0 = const()[name = tensor("op_15836_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_3_attentions_2_transformer_blocks_0_ff_net_0_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(648946880))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(649561344))), name = tensor("up_blocks_3_attentions_2_transformer_blocks_0_ff_net_0_proj_weight_to_fp16_palettized"), shape = tensor([2560, 320, 1, 1])]; + tensor up_blocks_3_attentions_2_transformer_blocks_0_ff_net_0_proj_bias_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(649561536))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(649563520))), name = tensor("up_blocks_3_attentions_2_transformer_blocks_0_ff_net_0_proj_bias_to_fp16_palettized"), shape = tensor([2560])]; + tensor var_15836_cast = conv(bias = up_blocks_3_attentions_2_transformer_blocks_0_ff_net_0_proj_bias_to_fp16_palettized, dilations = var_15834, groups = var_12807, pad = var_15836_pad_0, pad_type = var_15836_pad_type_0, strides = var_15832, weight = up_blocks_3_attentions_2_transformer_blocks_0_ff_net_0_proj_weight_to_fp16_palettized, x = input_521_cast)[name = tensor("op_15836_cast")]; + tensor var_15837_split_sizes_0 = const()[name = tensor("op_15837_split_sizes_0"), val = tensor([1280, 1280])]; + tensor var_15837_axis_0 = const()[name = tensor("op_15837_axis_0"), val = tensor(1)]; + tensor var_15837_cast_0, tensor var_15837_cast_1 = split(axis = var_15837_axis_0, split_sizes = var_15837_split_sizes_0, x = var_15836_cast)[name = tensor("op_15837_cast")]; + tensor var_15839_mode_0 = const()[name = tensor("op_15839_mode_0"), val = tensor("EXACT")]; + tensor var_15839_cast = gelu(mode = var_15839_mode_0, x = var_15837_cast_1)[name = tensor("op_15839_cast")]; + tensor input_523_cast = mul(x = var_15837_cast_0, y = var_15839_cast)[name = tensor("input_523_cast")]; + tensor var_15843 = const()[name = tensor("op_15843"), val = tensor([1, 1])]; + tensor var_15845 = const()[name = tensor("op_15845"), val = tensor([1, 1])]; + tensor var_15847_pad_type_0 = const()[name = tensor("op_15847_pad_type_0"), val = tensor("custom")]; + tensor var_15847_pad_0 = const()[name = tensor("op_15847_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_3_attentions_2_transformer_blocks_0_ff_net_2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(649563712))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(649870976))), name = tensor("up_blocks_3_attentions_2_transformer_blocks_0_ff_net_2_weight_to_fp16_palettized"), shape = tensor([320, 1280, 1, 1])]; + tensor up_blocks_3_attentions_2_transformer_blocks_0_ff_net_2_bias_to_fp16 = const()[name = tensor("up_blocks_3_attentions_2_transformer_blocks_0_ff_net_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(649871168)))]; + tensor var_15847_cast = conv(bias = up_blocks_3_attentions_2_transformer_blocks_0_ff_net_2_bias_to_fp16, dilations = var_15845, groups = var_12807, pad = var_15847_pad_0, pad_type = var_15847_pad_type_0, strides = var_15843, weight = up_blocks_3_attentions_2_transformer_blocks_0_ff_net_2_weight_to_fp16_palettized, x = input_523_cast)[name = tensor("op_15847_cast")]; + tensor hidden_states_333_cast = add(x = var_15847_cast, y = inputs_cast)[name = tensor("hidden_states_333_cast")]; + tensor var_15849 = const()[name = tensor("op_15849"), val = tensor([2, 320, 64, 64])]; + tensor input_525_cast = reshape(shape = var_15849, x = hidden_states_333_cast)[name = tensor("input_525_cast")]; + tensor var_15853 = const()[name = tensor("op_15853"), val = tensor([1, 1])]; + tensor var_15855 = const()[name = tensor("op_15855"), val = tensor([1, 1])]; + tensor hidden_states_pad_type_0 = const()[name = tensor("hidden_states_pad_type_0"), val = tensor("custom")]; + tensor hidden_states_pad_0 = const()[name = tensor("hidden_states_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor up_blocks_3_attentions_2_proj_out_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(649871872))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(649948736))), name = tensor("up_blocks_3_attentions_2_proj_out_weight_to_fp16_palettized"), shape = tensor([320, 320, 1, 1])]; + tensor up_blocks_3_attentions_2_proj_out_bias_to_fp16 = const()[name = tensor("up_blocks_3_attentions_2_proj_out_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(649948928)))]; + tensor hidden_states_cast = conv(bias = up_blocks_3_attentions_2_proj_out_bias_to_fp16, dilations = var_15855, groups = var_12807, pad = hidden_states_pad_0, pad_type = hidden_states_pad_type_0, strides = var_15853, weight = up_blocks_3_attentions_2_proj_out_weight_to_fp16_palettized, x = input_525_cast)[name = tensor("hidden_states_cast")]; + tensor input_527_cast = add(x = hidden_states_cast, y = hidden_states_323_cast)[name = tensor("input_527_cast")]; + tensor reshape_240_shape_0 = const()[name = tensor("reshape_240_shape_0"), val = tensor([2, 32, 10, 64, 64])]; + tensor reshape_240_cast = reshape(shape = reshape_240_shape_0, x = input_527_cast)[name = tensor("reshape_240_cast")]; + tensor reduce_mean_180_axes_0 = const()[name = tensor("reduce_mean_180_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_180_keep_dims_0 = const()[name = tensor("reduce_mean_180_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_180_cast = reduce_mean(axes = reduce_mean_180_axes_0, keep_dims = reduce_mean_180_keep_dims_0, x = reshape_240_cast)[name = tensor("reduce_mean_180_cast")]; + tensor sub_120_cast = sub(x = reshape_240_cast, y = reduce_mean_180_cast)[name = tensor("sub_120_cast")]; + tensor square_60_cast = square(x = sub_120_cast)[name = tensor("square_60_cast")]; + tensor reduce_mean_182_axes_0 = const()[name = tensor("reduce_mean_182_axes_0"), val = tensor([2, 3, 4])]; + tensor reduce_mean_182_keep_dims_0 = const()[name = tensor("reduce_mean_182_keep_dims_0"), val = tensor(true)]; + tensor reduce_mean_182_cast = reduce_mean(axes = reduce_mean_182_axes_0, keep_dims = reduce_mean_182_keep_dims_0, x = square_60_cast)[name = tensor("reduce_mean_182_cast")]; + tensor add_120_y_0_to_fp16 = const()[name = tensor("add_120_y_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_120_cast = add(x = reduce_mean_182_cast, y = add_120_y_0_to_fp16)[name = tensor("add_120_cast")]; + tensor sqrt_60_cast = sqrt(x = add_120_cast)[name = tensor("sqrt_60_cast")]; + tensor real_div_60_cast = real_div(x = sub_120_cast, y = sqrt_60_cast)[name = tensor("real_div_60_cast")]; + tensor reshape_241_shape_0 = const()[name = tensor("reshape_241_shape_0"), val = tensor([2, 320, 64, 64])]; + tensor reshape_241_cast = reshape(shape = reshape_241_shape_0, x = real_div_60_cast)[name = tensor("reshape_241_cast")]; + tensor add_121_gamma_0_to_fp16 = const()[name = tensor("add_121_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(649949632)))]; + tensor add_121_beta_0_to_fp16 = const()[name = tensor("add_121_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(649950336)))]; + tensor add_121_epsilon_0_to_fp16 = const()[name = tensor("add_121_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor add_121_cast = batch_norm(beta = add_121_beta_0_to_fp16, epsilon = add_121_epsilon_0_to_fp16, gamma = add_121_gamma_0_to_fp16, mean = add_1_mean_0_to_fp16, variance = add_1_variance_0_to_fp16, x = reshape_241_cast)[name = tensor("add_121_cast")]; + tensor input_cast = silu(x = add_121_cast)[name = tensor("input_cast")]; + tensor var_15869 = const()[name = tensor("op_15869"), val = tensor(1)]; + tensor var_15872 = const()[name = tensor("op_15872"), val = tensor([1, 1])]; + tensor var_15874 = const()[name = tensor("op_15874"), val = tensor([1, 1])]; + tensor var_15876_pad_type_0 = const()[name = tensor("op_15876_pad_type_0"), val = tensor("custom")]; + tensor var_15876_pad_0 = const()[name = tensor("op_15876_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor conv_out_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(649951040))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(649959744))), name = tensor("conv_out_weight_to_fp16_palettized"), shape = tensor([4, 320, 3, 3])]; + tensor conv_out_bias_to_fp16 = const()[name = tensor("conv_out_bias_to_fp16"), val = tensor([-0x1.39p-9, 0x1.9f8p-9, 0x1.c64p-12, 0x1.d6cp-10])]; + tensor var_15876_cast = conv(bias = conv_out_bias_to_fp16, dilations = var_15874, groups = var_15869, pad = var_15876_pad_0, pad_type = var_15876_pad_type_0, strides = var_15872, weight = conv_out_weight_to_fp16_palettized, x = input_cast)[name = tensor("op_15876_cast")]; + tensor var_15876_cast_to_fp32_dtype_0 = const()[name = tensor("op_15876_cast_to_fp32_dtype_0"), val = tensor("fp32")]; + tensor noise_pred = cast(dtype = var_15876_cast_to_fp32_dtype_0, x = var_15876_cast)[name = tensor("cast_0")]; + } -> (noise_pred); +} \ No newline at end of file