program(1.0) [buildInfo = dict, tensor>({{"coremlc-component-MIL", "3304.5.2"}, {"coremlc-version", "3304.6.2"}})] { func main(tensor encoder_hidden_states, tensor sample, tensor timestep) { tensor var_20 = const()[name = tensor("op_20"), val = tensor(-1)]; tensor var_37_axes_0 = const()[name = tensor("op_37_axes_0"), val = tensor([1])]; tensor var_37_cast_fp16 = expand_dims(axes = var_37_axes_0, x = timestep)[name = tensor("op_37_cast_fp16")]; tensor var_39_to_fp16 = const()[name = tensor("op_39_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(64)))]; tensor emb_3_cast_fp16 = mul(x = var_37_cast_fp16, y = var_39_to_fp16)[name = tensor("emb_3_cast_fp16")]; tensor var_44_cast_fp16 = sin(x = emb_3_cast_fp16)[name = tensor("op_44_cast_fp16")]; tensor var_45_cast_fp16 = cos(x = emb_3_cast_fp16)[name = tensor("op_45_cast_fp16")]; tensor emb_interleave_0 = const()[name = tensor("emb_interleave_0"), val = tensor(false)]; tensor emb_cast_fp16 = concat(axis = var_20, interleave = emb_interleave_0, values = (var_44_cast_fp16, var_45_cast_fp16))[name = tensor("emb_cast_fp16")]; tensor var_49_begin_0 = const()[name = tensor("op_49_begin_0"), val = tensor([0, 160])]; tensor var_49_end_0 = const()[name = tensor("op_49_end_0"), val = tensor([2, 320])]; tensor var_49_end_mask_0 = const()[name = tensor("op_49_end_mask_0"), val = tensor([true, true])]; tensor var_49_cast_fp16 = slice_by_index(begin = var_49_begin_0, end = var_49_end_0, end_mask = var_49_end_mask_0, x = emb_cast_fp16)[name = tensor("op_49_cast_fp16")]; tensor var_51_begin_0 = const()[name = tensor("op_51_begin_0"), val = tensor([0, 0])]; tensor var_51_end_0 = const()[name = tensor("op_51_end_0"), val = tensor([2, 160])]; tensor var_51_end_mask_0 = const()[name = tensor("op_51_end_mask_0"), val = tensor([true, false])]; tensor var_51_cast_fp16 = slice_by_index(begin = var_51_begin_0, end = var_51_end_0, end_mask = var_51_end_mask_0, x = emb_cast_fp16)[name = tensor("op_51_cast_fp16")]; tensor sample_interleave_0 = const()[name = tensor("sample_interleave_0"), val = tensor(false)]; tensor sample_cast_fp16 = concat(axis = var_20, interleave = sample_interleave_0, values = (var_49_cast_fp16, var_51_cast_fp16))[name = tensor("sample_cast_fp16")]; tensor var_54 = const()[name = tensor("op_54"), val = tensor(1)]; tensor var_61_axes_0 = const()[name = tensor("op_61_axes_0"), val = tensor([-1])]; tensor var_61_cast_fp16 = expand_dims(axes = var_61_axes_0, x = sample_cast_fp16)[name = tensor("op_61_cast_fp16")]; tensor input_1_axes_0 = const()[name = tensor("input_1_axes_0"), val = tensor([-1])]; tensor input_1_cast_fp16 = expand_dims(axes = input_1_axes_0, x = var_61_cast_fp16)[name = tensor("input_1_cast_fp16")]; tensor var_65 = const()[name = tensor("op_65"), val = tensor([1, 1])]; tensor var_67 = const()[name = tensor("op_67"), 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_fp16 = conv(bias = time_embedding_linear_1_bias_to_fp16, dilations = var_67, groups = var_54, pad = input_3_pad_0, pad_type = input_3_pad_type_0, strides = var_65, weight = time_embedding_linear_1_weight_to_fp16_palettized, x = input_1_cast_fp16)[name = tensor("input_3_cast_fp16")]; tensor input_5_cast_fp16 = silu(x = input_3_cast_fp16)[name = tensor("input_5_cast_fp16")]; tensor var_73 = const()[name = tensor("op_73"), val = tensor([1, 1])]; tensor var_75 = const()[name = tensor("op_75"), 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_fp16 = conv(bias = time_embedding_linear_2_bias_to_fp16, dilations = var_75, groups = var_54, pad = input_13_pad_0, pad_type = input_13_pad_type_0, strides = var_73, weight = time_embedding_linear_2_weight_to_fp16_palettized, x = input_5_cast_fp16)[name = tensor("input_13_cast_fp16")]; tensor var_81 = const()[name = tensor("op_81"), val = tensor(1)]; tensor var_84 = const()[name = tensor("op_84"), val = tensor([1, 1])]; tensor var_86 = const()[name = tensor("op_86"), 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_fp16 = conv(bias = conv_in_bias_to_fp16, dilations = var_86, groups = var_81, pad = input_7_pad_0, pad_type = input_7_pad_type_0, strides = var_84, weight = conv_in_weight_to_fp16_palettized, x = sample)[name = tensor("input_7_cast_fp16")]; tensor var_90 = const()[name = tensor("op_90"), val = tensor(3)]; tensor var_113 = const()[name = tensor("op_113"), val = tensor(true)]; tensor var_118 = const()[name = tensor("op_118"), 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_fp16 = reshape(shape = reshape_0_shape_0, x = input_7_cast_fp16)[name = tensor("reshape_0_cast_fp16")]; 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_fp16 = reduce_mean(axes = reduce_mean_0_axes_0, keep_dims = reduce_mean_0_keep_dims_0, x = reshape_0_cast_fp16)[name = tensor("reduce_mean_0_cast_fp16")]; tensor sub_0_cast_fp16 = sub(x = reshape_0_cast_fp16, y = reduce_mean_0_cast_fp16)[name = tensor("sub_0_cast_fp16")]; tensor square_0_cast_fp16 = square(x = sub_0_cast_fp16)[name = tensor("square_0_cast_fp16")]; 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_fp16 = reduce_mean(axes = reduce_mean_2_axes_0, keep_dims = reduce_mean_2_keep_dims_0, x = square_0_cast_fp16)[name = tensor("reduce_mean_2_cast_fp16")]; 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_fp16 = add(x = reduce_mean_2_cast_fp16, y = add_0_y_0_to_fp16)[name = tensor("add_0_cast_fp16")]; tensor sqrt_0_cast_fp16 = sqrt(x = add_0_cast_fp16)[name = tensor("sqrt_0_cast_fp16")]; tensor real_div_0_cast_fp16 = real_div(x = sub_0_cast_fp16, y = sqrt_0_cast_fp16)[name = tensor("real_div_0_cast_fp16")]; tensor reshape_1_shape_0 = const()[name = tensor("reshape_1_shape_0"), val = tensor([2, 320, 64, 64])]; tensor reshape_1_cast_fp16 = reshape(shape = reshape_1_shape_0, x = real_div_0_cast_fp16)[name = tensor("reshape_1_cast_fp16")]; 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_fp16 = 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_fp16)[name = tensor("add_1_cast_fp16")]; tensor input_11_cast_fp16 = silu(x = add_1_cast_fp16)[name = tensor("input_11_cast_fp16")]; tensor var_136 = const()[name = tensor("op_136"), val = tensor([1, 1])]; tensor var_138 = const()[name = tensor("op_138"), 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_fp16 = conv(bias = down_blocks_0_resnets_0_conv1_bias_to_fp16, dilations = var_138, groups = var_118, pad = hidden_states_1_pad_0, pad_type = hidden_states_1_pad_type_0, strides = var_136, weight = down_blocks_0_resnets_0_conv1_weight_to_fp16_palettized, x = input_11_cast_fp16)[name = tensor("hidden_states_1_cast_fp16")]; tensor input_15_cast_fp16 = silu(x = input_13_cast_fp16)[name = tensor("input_15_cast_fp16")]; tensor var_144 = const()[name = tensor("op_144"), val = tensor([1, 1])]; tensor var_146 = const()[name = tensor("op_146"), 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_fp16 = conv(bias = down_blocks_0_resnets_0_time_emb_proj_bias_to_fp16, dilations = var_146, groups = var_118, pad = temb_1_pad_0, pad_type = temb_1_pad_type_0, strides = var_144, weight = down_blocks_0_resnets_0_time_emb_proj_weight_to_fp16_palettized, x = input_15_cast_fp16)[name = tensor("temb_1_cast_fp16")]; tensor input_17_cast_fp16 = add(x = hidden_states_1_cast_fp16, y = temb_1_cast_fp16)[name = tensor("input_17_cast_fp16")]; tensor reshape_4_shape_0 = const()[name = tensor("reshape_4_shape_0"), val = tensor([2, 32, 10, 64, 64])]; tensor reshape_4_cast_fp16 = reshape(shape = reshape_4_shape_0, x = input_17_cast_fp16)[name = tensor("reshape_4_cast_fp16")]; 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_fp16 = reduce_mean(axes = reduce_mean_3_axes_0, keep_dims = reduce_mean_3_keep_dims_0, x = reshape_4_cast_fp16)[name = tensor("reduce_mean_3_cast_fp16")]; tensor sub_2_cast_fp16 = sub(x = reshape_4_cast_fp16, y = reduce_mean_3_cast_fp16)[name = tensor("sub_2_cast_fp16")]; tensor square_1_cast_fp16 = square(x = sub_2_cast_fp16)[name = tensor("square_1_cast_fp16")]; 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_fp16 = reduce_mean(axes = reduce_mean_5_axes_0, keep_dims = reduce_mean_5_keep_dims_0, x = square_1_cast_fp16)[name = tensor("reduce_mean_5_cast_fp16")]; 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_fp16 = add(x = reduce_mean_5_cast_fp16, y = add_2_y_0_to_fp16)[name = tensor("add_2_cast_fp16")]; tensor sqrt_1_cast_fp16 = sqrt(x = add_2_cast_fp16)[name = tensor("sqrt_1_cast_fp16")]; tensor real_div_1_cast_fp16 = real_div(x = sub_2_cast_fp16, y = sqrt_1_cast_fp16)[name = tensor("real_div_1_cast_fp16")]; tensor reshape_5_shape_0 = const()[name = tensor("reshape_5_shape_0"), val = tensor([2, 320, 64, 64])]; tensor reshape_5_cast_fp16 = reshape(shape = reshape_5_shape_0, x = real_div_1_cast_fp16)[name = tensor("reshape_5_cast_fp16")]; 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_fp16 = 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_fp16)[name = tensor("add_3_cast_fp16")]; tensor input_21_cast_fp16 = silu(x = add_3_cast_fp16)[name = tensor("input_21_cast_fp16")]; tensor var_156 = const()[name = tensor("op_156"), val = tensor([1, 1])]; tensor var_158 = const()[name = tensor("op_158"), 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_fp16 = conv(bias = down_blocks_0_resnets_0_conv2_bias_to_fp16, dilations = var_158, groups = var_118, pad = hidden_states_3_pad_0, pad_type = hidden_states_3_pad_type_0, strides = var_156, weight = down_blocks_0_resnets_0_conv2_weight_to_fp16_palettized, x = input_21_cast_fp16)[name = tensor("hidden_states_3_cast_fp16")]; tensor hidden_states_5_cast_fp16 = add(x = input_7_cast_fp16, y = hidden_states_3_cast_fp16)[name = tensor("hidden_states_5_cast_fp16")]; tensor reshape_8_shape_0 = const()[name = tensor("reshape_8_shape_0"), val = tensor([2, 32, 10, 64, 64])]; tensor reshape_8_cast_fp16 = reshape(shape = reshape_8_shape_0, x = hidden_states_5_cast_fp16)[name = tensor("reshape_8_cast_fp16")]; 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_fp16 = reduce_mean(axes = reduce_mean_6_axes_0, keep_dims = reduce_mean_6_keep_dims_0, x = reshape_8_cast_fp16)[name = tensor("reduce_mean_6_cast_fp16")]; tensor sub_4_cast_fp16 = sub(x = reshape_8_cast_fp16, y = reduce_mean_6_cast_fp16)[name = tensor("sub_4_cast_fp16")]; tensor square_2_cast_fp16 = square(x = sub_4_cast_fp16)[name = tensor("square_2_cast_fp16")]; 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_fp16 = reduce_mean(axes = reduce_mean_8_axes_0, keep_dims = reduce_mean_8_keep_dims_0, x = square_2_cast_fp16)[name = tensor("reduce_mean_8_cast_fp16")]; 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_fp16 = add(x = reduce_mean_8_cast_fp16, y = add_4_y_0_to_fp16)[name = tensor("add_4_cast_fp16")]; tensor sqrt_2_cast_fp16 = sqrt(x = add_4_cast_fp16)[name = tensor("sqrt_2_cast_fp16")]; tensor real_div_2_cast_fp16 = real_div(x = sub_4_cast_fp16, y = sqrt_2_cast_fp16)[name = tensor("real_div_2_cast_fp16")]; tensor reshape_9_shape_0 = const()[name = tensor("reshape_9_shape_0"), val = tensor([2, 320, 64, 64])]; tensor reshape_9_cast_fp16 = reshape(shape = reshape_9_shape_0, x = real_div_2_cast_fp16)[name = tensor("reshape_9_cast_fp16")]; 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_fp16 = 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_fp16)[name = tensor("add_5_cast_fp16")]; tensor var_178 = const()[name = tensor("op_178"), val = tensor([1, 1])]; tensor var_180 = const()[name = tensor("op_180"), 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_fp16 = conv(bias = down_blocks_0_attentions_0_proj_in_bias_to_fp16, dilations = var_180, groups = var_118, pad = hidden_states_7_pad_0, pad_type = hidden_states_7_pad_type_0, strides = var_178, weight = down_blocks_0_attentions_0_proj_in_weight_to_fp16_palettized, x = add_5_cast_fp16)[name = tensor("hidden_states_7_cast_fp16")]; tensor var_185 = const()[name = tensor("op_185"), val = tensor([2, 320, 1, 4096])]; tensor inputs_1_cast_fp16 = reshape(shape = var_185, x = hidden_states_7_cast_fp16)[name = tensor("inputs_1_cast_fp16")]; tensor var_195 = const()[name = tensor("op_195"), val = tensor([1])]; tensor channels_mean_1_cast_fp16 = reduce_mean(axes = var_195, keep_dims = var_113, x = inputs_1_cast_fp16)[name = tensor("channels_mean_1_cast_fp16")]; tensor zero_mean_1_cast_fp16 = sub(x = inputs_1_cast_fp16, y = channels_mean_1_cast_fp16)[name = tensor("zero_mean_1_cast_fp16")]; tensor zero_mean_sq_1_cast_fp16 = mul(x = zero_mean_1_cast_fp16, y = zero_mean_1_cast_fp16)[name = tensor("zero_mean_sq_1_cast_fp16")]; tensor var_199 = const()[name = tensor("op_199"), val = tensor([1])]; tensor var_200_cast_fp16 = reduce_mean(axes = var_199, keep_dims = var_113, x = zero_mean_sq_1_cast_fp16)[name = tensor("op_200_cast_fp16")]; tensor var_201_to_fp16 = const()[name = tensor("op_201_to_fp16"), val = tensor(0x1.5p-17)]; tensor var_202_cast_fp16 = add(x = var_200_cast_fp16, y = var_201_to_fp16)[name = tensor("op_202_cast_fp16")]; tensor denom_1_epsilon_0_to_fp16 = const()[name = tensor("denom_1_epsilon_0_to_fp16"), val = tensor(0x1p-24)]; tensor denom_1_cast_fp16 = rsqrt(epsilon = denom_1_epsilon_0_to_fp16, x = var_202_cast_fp16)[name = tensor("denom_1_cast_fp16")]; tensor out_1_cast_fp16 = mul(x = zero_mean_1_cast_fp16, y = denom_1_cast_fp16)[name = tensor("out_1_cast_fp16")]; tensor var_206_to_fp16 = const()[name = tensor("op_206_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3327680)))]; tensor var_207_cast_fp16 = add(x = out_1_cast_fp16, y = var_206_to_fp16)[name = tensor("op_207_cast_fp16")]; tensor var_209_to_fp16 = const()[name = tensor("op_209_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3328384)))]; tensor hidden_states_9_cast_fp16 = mul(x = var_207_cast_fp16, y = var_209_to_fp16)[name = tensor("hidden_states_9_cast_fp16")]; tensor var_216 = const()[name = tensor("op_216"), val = tensor([1, 1])]; tensor var_218 = const()[name = tensor("op_218"), 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_fp16 = conv(dilations = var_218, groups = var_118, pad = q_1_pad_0, pad_type = q_1_pad_type_0, strides = var_216, weight = down_blocks_0_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16_palettized, x = hidden_states_9_cast_fp16)[name = tensor("q_1_cast_fp16")]; 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 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_fp16 = conv(dilations = var_224, groups = var_118, pad = k_1_pad_0, pad_type = k_1_pad_type_0, strides = var_222, weight = down_blocks_0_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16_palettized, x = hidden_states_9_cast_fp16)[name = tensor("k_1_cast_fp16")]; 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 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_fp16 = conv(dilations = var_230, groups = var_118, pad = v_1_pad_0, pad_type = v_1_pad_type_0, strides = var_228, weight = down_blocks_0_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16_palettized, x = hidden_states_9_cast_fp16)[name = tensor("v_1_cast_fp16")]; tensor var_234_begin_0 = const()[name = tensor("op_234_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_234_end_0 = const()[name = tensor("op_234_end_0"), val = tensor([2, 40, 1, 4096])]; tensor var_234_end_mask_0 = const()[name = tensor("op_234_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_234_cast_fp16 = slice_by_index(begin = var_234_begin_0, end = var_234_end_0, end_mask = var_234_end_mask_0, x = q_1_cast_fp16)[name = tensor("op_234_cast_fp16")]; tensor var_238_begin_0 = const()[name = tensor("op_238_begin_0"), val = tensor([0, 40, 0, 0])]; tensor var_238_end_0 = const()[name = tensor("op_238_end_0"), val = tensor([2, 80, 1, 4096])]; tensor var_238_end_mask_0 = const()[name = tensor("op_238_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_238_cast_fp16 = slice_by_index(begin = var_238_begin_0, end = var_238_end_0, end_mask = var_238_end_mask_0, x = q_1_cast_fp16)[name = tensor("op_238_cast_fp16")]; tensor var_242_begin_0 = const()[name = tensor("op_242_begin_0"), val = tensor([0, 80, 0, 0])]; tensor var_242_end_0 = const()[name = tensor("op_242_end_0"), val = tensor([2, 120, 1, 4096])]; tensor var_242_end_mask_0 = const()[name = tensor("op_242_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_242_cast_fp16 = slice_by_index(begin = var_242_begin_0, end = var_242_end_0, end_mask = var_242_end_mask_0, x = q_1_cast_fp16)[name = tensor("op_242_cast_fp16")]; tensor var_246_begin_0 = const()[name = tensor("op_246_begin_0"), val = tensor([0, 120, 0, 0])]; tensor var_246_end_0 = const()[name = tensor("op_246_end_0"), val = tensor([2, 160, 1, 4096])]; tensor var_246_end_mask_0 = const()[name = tensor("op_246_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_246_cast_fp16 = slice_by_index(begin = var_246_begin_0, end = var_246_end_0, end_mask = var_246_end_mask_0, x = q_1_cast_fp16)[name = tensor("op_246_cast_fp16")]; tensor var_250_begin_0 = const()[name = tensor("op_250_begin_0"), val = tensor([0, 160, 0, 0])]; tensor var_250_end_0 = const()[name = tensor("op_250_end_0"), val = tensor([2, 200, 1, 4096])]; tensor var_250_end_mask_0 = const()[name = tensor("op_250_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_250_cast_fp16 = slice_by_index(begin = var_250_begin_0, end = var_250_end_0, end_mask = var_250_end_mask_0, x = q_1_cast_fp16)[name = tensor("op_250_cast_fp16")]; tensor var_254_begin_0 = const()[name = tensor("op_254_begin_0"), val = tensor([0, 200, 0, 0])]; tensor var_254_end_0 = const()[name = tensor("op_254_end_0"), val = tensor([2, 240, 1, 4096])]; tensor var_254_end_mask_0 = const()[name = tensor("op_254_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_254_cast_fp16 = slice_by_index(begin = var_254_begin_0, end = var_254_end_0, end_mask = var_254_end_mask_0, x = q_1_cast_fp16)[name = tensor("op_254_cast_fp16")]; tensor var_258_begin_0 = const()[name = tensor("op_258_begin_0"), val = tensor([0, 240, 0, 0])]; tensor var_258_end_0 = const()[name = tensor("op_258_end_0"), val = tensor([2, 280, 1, 4096])]; tensor var_258_end_mask_0 = const()[name = tensor("op_258_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_258_cast_fp16 = slice_by_index(begin = var_258_begin_0, end = var_258_end_0, end_mask = var_258_end_mask_0, x = q_1_cast_fp16)[name = tensor("op_258_cast_fp16")]; tensor var_262_begin_0 = const()[name = tensor("op_262_begin_0"), val = tensor([0, 280, 0, 0])]; tensor var_262_end_0 = const()[name = tensor("op_262_end_0"), val = tensor([2, 320, 1, 4096])]; tensor var_262_end_mask_0 = const()[name = tensor("op_262_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_262_cast_fp16 = slice_by_index(begin = var_262_begin_0, end = var_262_end_0, end_mask = var_262_end_mask_0, x = q_1_cast_fp16)[name = tensor("op_262_cast_fp16")]; tensor var_265_begin_0 = const()[name = tensor("op_265_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_265_end_0 = const()[name = tensor("op_265_end_0"), val = tensor([2, 40, 1, 512])]; tensor var_265_end_mask_0 = const()[name = tensor("op_265_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_265_cast_fp16 = slice_by_index(begin = var_265_begin_0, end = var_265_end_0, end_mask = var_265_end_mask_0, x = var_234_cast_fp16)[name = tensor("op_265_cast_fp16")]; tensor var_266_begin_0 = const()[name = tensor("op_266_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_266_end_0 = const()[name = tensor("op_266_end_0"), val = tensor([2, 40, 1, 1024])]; tensor var_266_end_mask_0 = const()[name = tensor("op_266_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_266_cast_fp16 = slice_by_index(begin = var_266_begin_0, end = var_266_end_0, end_mask = var_266_end_mask_0, x = var_234_cast_fp16)[name = tensor("op_266_cast_fp16")]; tensor var_267_begin_0 = const()[name = tensor("op_267_begin_0"), val = tensor([0, 0, 0, 1024])]; tensor var_267_end_0 = const()[name = tensor("op_267_end_0"), val = tensor([2, 40, 1, 1536])]; tensor var_267_end_mask_0 = const()[name = tensor("op_267_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_267_cast_fp16 = slice_by_index(begin = var_267_begin_0, end = var_267_end_0, end_mask = var_267_end_mask_0, x = var_234_cast_fp16)[name = tensor("op_267_cast_fp16")]; tensor var_268_begin_0 = const()[name = tensor("op_268_begin_0"), val = tensor([0, 0, 0, 1536])]; tensor var_268_end_0 = const()[name = tensor("op_268_end_0"), val = tensor([2, 40, 1, 2048])]; tensor var_268_end_mask_0 = const()[name = tensor("op_268_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_268_cast_fp16 = slice_by_index(begin = var_268_begin_0, end = var_268_end_0, end_mask = var_268_end_mask_0, x = var_234_cast_fp16)[name = tensor("op_268_cast_fp16")]; tensor var_269_begin_0 = const()[name = tensor("op_269_begin_0"), val = tensor([0, 0, 0, 2048])]; tensor var_269_end_0 = const()[name = tensor("op_269_end_0"), val = tensor([2, 40, 1, 2560])]; tensor var_269_end_mask_0 = const()[name = tensor("op_269_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_269_cast_fp16 = slice_by_index(begin = var_269_begin_0, end = var_269_end_0, end_mask = var_269_end_mask_0, x = var_234_cast_fp16)[name = tensor("op_269_cast_fp16")]; tensor var_270_begin_0 = const()[name = tensor("op_270_begin_0"), val = tensor([0, 0, 0, 2560])]; tensor var_270_end_0 = const()[name = tensor("op_270_end_0"), val = tensor([2, 40, 1, 3072])]; tensor var_270_end_mask_0 = const()[name = tensor("op_270_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_270_cast_fp16 = slice_by_index(begin = var_270_begin_0, end = var_270_end_0, end_mask = var_270_end_mask_0, x = var_234_cast_fp16)[name = tensor("op_270_cast_fp16")]; tensor var_271_begin_0 = const()[name = tensor("op_271_begin_0"), val = tensor([0, 0, 0, 3072])]; tensor var_271_end_0 = const()[name = tensor("op_271_end_0"), val = tensor([2, 40, 1, 3584])]; tensor var_271_end_mask_0 = const()[name = tensor("op_271_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_271_cast_fp16 = slice_by_index(begin = var_271_begin_0, end = var_271_end_0, end_mask = var_271_end_mask_0, x = var_234_cast_fp16)[name = tensor("op_271_cast_fp16")]; tensor var_272_begin_0 = const()[name = tensor("op_272_begin_0"), val = tensor([0, 0, 0, 3584])]; tensor var_272_end_0 = const()[name = tensor("op_272_end_0"), val = tensor([2, 40, 1, 4096])]; tensor var_272_end_mask_0 = const()[name = tensor("op_272_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_272_cast_fp16 = slice_by_index(begin = var_272_begin_0, end = var_272_end_0, end_mask = var_272_end_mask_0, x = var_234_cast_fp16)[name = tensor("op_272_cast_fp16")]; tensor var_273_begin_0 = const()[name = tensor("op_273_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_273_end_0 = const()[name = tensor("op_273_end_0"), val = tensor([2, 40, 1, 512])]; tensor var_273_end_mask_0 = const()[name = tensor("op_273_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_273_cast_fp16 = slice_by_index(begin = var_273_begin_0, end = var_273_end_0, end_mask = var_273_end_mask_0, x = var_238_cast_fp16)[name = tensor("op_273_cast_fp16")]; tensor var_274_begin_0 = const()[name = tensor("op_274_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_274_end_0 = const()[name = tensor("op_274_end_0"), val = tensor([2, 40, 1, 1024])]; tensor var_274_end_mask_0 = const()[name = tensor("op_274_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_274_cast_fp16 = slice_by_index(begin = var_274_begin_0, end = var_274_end_0, end_mask = var_274_end_mask_0, x = var_238_cast_fp16)[name = tensor("op_274_cast_fp16")]; tensor var_275_begin_0 = const()[name = tensor("op_275_begin_0"), val = tensor([0, 0, 0, 1024])]; tensor var_275_end_0 = const()[name = tensor("op_275_end_0"), val = tensor([2, 40, 1, 1536])]; tensor var_275_end_mask_0 = const()[name = tensor("op_275_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_275_cast_fp16 = slice_by_index(begin = var_275_begin_0, end = var_275_end_0, end_mask = var_275_end_mask_0, x = var_238_cast_fp16)[name = tensor("op_275_cast_fp16")]; tensor var_276_begin_0 = const()[name = tensor("op_276_begin_0"), val = tensor([0, 0, 0, 1536])]; tensor var_276_end_0 = const()[name = tensor("op_276_end_0"), val = tensor([2, 40, 1, 2048])]; tensor var_276_end_mask_0 = const()[name = tensor("op_276_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_276_cast_fp16 = slice_by_index(begin = var_276_begin_0, end = var_276_end_0, end_mask = var_276_end_mask_0, x = var_238_cast_fp16)[name = tensor("op_276_cast_fp16")]; tensor var_277_begin_0 = const()[name = tensor("op_277_begin_0"), val = tensor([0, 0, 0, 2048])]; tensor var_277_end_0 = const()[name = tensor("op_277_end_0"), val = tensor([2, 40, 1, 2560])]; tensor var_277_end_mask_0 = const()[name = tensor("op_277_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_277_cast_fp16 = slice_by_index(begin = var_277_begin_0, end = var_277_end_0, end_mask = var_277_end_mask_0, x = var_238_cast_fp16)[name = tensor("op_277_cast_fp16")]; tensor var_278_begin_0 = const()[name = tensor("op_278_begin_0"), val = tensor([0, 0, 0, 2560])]; tensor var_278_end_0 = const()[name = tensor("op_278_end_0"), val = tensor([2, 40, 1, 3072])]; tensor var_278_end_mask_0 = const()[name = tensor("op_278_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_278_cast_fp16 = slice_by_index(begin = var_278_begin_0, end = var_278_end_0, end_mask = var_278_end_mask_0, x = var_238_cast_fp16)[name = tensor("op_278_cast_fp16")]; tensor var_279_begin_0 = const()[name = tensor("op_279_begin_0"), val = tensor([0, 0, 0, 3072])]; tensor var_279_end_0 = const()[name = tensor("op_279_end_0"), val = tensor([2, 40, 1, 3584])]; tensor var_279_end_mask_0 = const()[name = tensor("op_279_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_279_cast_fp16 = slice_by_index(begin = var_279_begin_0, end = var_279_end_0, end_mask = var_279_end_mask_0, x = var_238_cast_fp16)[name = tensor("op_279_cast_fp16")]; tensor var_280_begin_0 = const()[name = tensor("op_280_begin_0"), val = tensor([0, 0, 0, 3584])]; tensor var_280_end_0 = const()[name = tensor("op_280_end_0"), val = tensor([2, 40, 1, 4096])]; tensor var_280_end_mask_0 = const()[name = tensor("op_280_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_280_cast_fp16 = slice_by_index(begin = var_280_begin_0, end = var_280_end_0, end_mask = var_280_end_mask_0, x = var_238_cast_fp16)[name = tensor("op_280_cast_fp16")]; tensor var_281_begin_0 = const()[name = tensor("op_281_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_281_end_0 = const()[name = tensor("op_281_end_0"), val = tensor([2, 40, 1, 512])]; tensor var_281_end_mask_0 = const()[name = tensor("op_281_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_281_cast_fp16 = slice_by_index(begin = var_281_begin_0, end = var_281_end_0, end_mask = var_281_end_mask_0, x = var_242_cast_fp16)[name = tensor("op_281_cast_fp16")]; tensor var_282_begin_0 = const()[name = tensor("op_282_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_282_end_0 = const()[name = tensor("op_282_end_0"), val = tensor([2, 40, 1, 1024])]; tensor var_282_end_mask_0 = const()[name = tensor("op_282_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_282_cast_fp16 = slice_by_index(begin = var_282_begin_0, end = var_282_end_0, end_mask = var_282_end_mask_0, x = var_242_cast_fp16)[name = tensor("op_282_cast_fp16")]; tensor var_283_begin_0 = const()[name = tensor("op_283_begin_0"), val = tensor([0, 0, 0, 1024])]; tensor var_283_end_0 = const()[name = tensor("op_283_end_0"), val = tensor([2, 40, 1, 1536])]; tensor var_283_end_mask_0 = const()[name = tensor("op_283_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_283_cast_fp16 = slice_by_index(begin = var_283_begin_0, end = var_283_end_0, end_mask = var_283_end_mask_0, x = var_242_cast_fp16)[name = tensor("op_283_cast_fp16")]; tensor var_284_begin_0 = const()[name = tensor("op_284_begin_0"), val = tensor([0, 0, 0, 1536])]; tensor var_284_end_0 = const()[name = tensor("op_284_end_0"), val = tensor([2, 40, 1, 2048])]; tensor var_284_end_mask_0 = const()[name = tensor("op_284_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_284_cast_fp16 = slice_by_index(begin = var_284_begin_0, end = var_284_end_0, end_mask = var_284_end_mask_0, x = var_242_cast_fp16)[name = tensor("op_284_cast_fp16")]; tensor var_285_begin_0 = const()[name = tensor("op_285_begin_0"), val = tensor([0, 0, 0, 2048])]; tensor var_285_end_0 = const()[name = tensor("op_285_end_0"), val = tensor([2, 40, 1, 2560])]; tensor var_285_end_mask_0 = const()[name = tensor("op_285_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_285_cast_fp16 = slice_by_index(begin = var_285_begin_0, end = var_285_end_0, end_mask = var_285_end_mask_0, x = var_242_cast_fp16)[name = tensor("op_285_cast_fp16")]; tensor var_286_begin_0 = const()[name = tensor("op_286_begin_0"), val = tensor([0, 0, 0, 2560])]; tensor var_286_end_0 = const()[name = tensor("op_286_end_0"), val = tensor([2, 40, 1, 3072])]; tensor var_286_end_mask_0 = const()[name = tensor("op_286_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_286_cast_fp16 = slice_by_index(begin = var_286_begin_0, end = var_286_end_0, end_mask = var_286_end_mask_0, x = var_242_cast_fp16)[name = tensor("op_286_cast_fp16")]; tensor var_287_begin_0 = const()[name = tensor("op_287_begin_0"), val = tensor([0, 0, 0, 3072])]; tensor var_287_end_0 = const()[name = tensor("op_287_end_0"), val = tensor([2, 40, 1, 3584])]; tensor var_287_end_mask_0 = const()[name = tensor("op_287_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_287_cast_fp16 = slice_by_index(begin = var_287_begin_0, end = var_287_end_0, end_mask = var_287_end_mask_0, x = var_242_cast_fp16)[name = tensor("op_287_cast_fp16")]; tensor var_288_begin_0 = const()[name = tensor("op_288_begin_0"), val = tensor([0, 0, 0, 3584])]; tensor var_288_end_0 = const()[name = tensor("op_288_end_0"), val = tensor([2, 40, 1, 4096])]; tensor var_288_end_mask_0 = const()[name = tensor("op_288_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_288_cast_fp16 = slice_by_index(begin = var_288_begin_0, end = var_288_end_0, end_mask = var_288_end_mask_0, x = var_242_cast_fp16)[name = tensor("op_288_cast_fp16")]; tensor var_289_begin_0 = const()[name = tensor("op_289_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_289_end_0 = const()[name = tensor("op_289_end_0"), val = tensor([2, 40, 1, 512])]; tensor var_289_end_mask_0 = const()[name = tensor("op_289_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_289_cast_fp16 = slice_by_index(begin = var_289_begin_0, end = var_289_end_0, end_mask = var_289_end_mask_0, x = var_246_cast_fp16)[name = tensor("op_289_cast_fp16")]; tensor var_290_begin_0 = const()[name = tensor("op_290_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_290_end_0 = const()[name = tensor("op_290_end_0"), val = tensor([2, 40, 1, 1024])]; tensor var_290_end_mask_0 = const()[name = tensor("op_290_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_290_cast_fp16 = slice_by_index(begin = var_290_begin_0, end = var_290_end_0, end_mask = var_290_end_mask_0, x = var_246_cast_fp16)[name = tensor("op_290_cast_fp16")]; tensor var_291_begin_0 = const()[name = tensor("op_291_begin_0"), val = tensor([0, 0, 0, 1024])]; tensor var_291_end_0 = const()[name = tensor("op_291_end_0"), val = tensor([2, 40, 1, 1536])]; tensor var_291_end_mask_0 = const()[name = tensor("op_291_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_291_cast_fp16 = slice_by_index(begin = var_291_begin_0, end = var_291_end_0, end_mask = var_291_end_mask_0, x = var_246_cast_fp16)[name = tensor("op_291_cast_fp16")]; tensor var_292_begin_0 = const()[name = tensor("op_292_begin_0"), val = tensor([0, 0, 0, 1536])]; tensor var_292_end_0 = const()[name = tensor("op_292_end_0"), val = tensor([2, 40, 1, 2048])]; tensor var_292_end_mask_0 = const()[name = tensor("op_292_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_292_cast_fp16 = slice_by_index(begin = var_292_begin_0, end = var_292_end_0, end_mask = var_292_end_mask_0, x = var_246_cast_fp16)[name = tensor("op_292_cast_fp16")]; tensor var_293_begin_0 = const()[name = tensor("op_293_begin_0"), val = tensor([0, 0, 0, 2048])]; tensor var_293_end_0 = const()[name = tensor("op_293_end_0"), val = tensor([2, 40, 1, 2560])]; tensor var_293_end_mask_0 = const()[name = tensor("op_293_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_293_cast_fp16 = slice_by_index(begin = var_293_begin_0, end = var_293_end_0, end_mask = var_293_end_mask_0, x = var_246_cast_fp16)[name = tensor("op_293_cast_fp16")]; tensor var_294_begin_0 = const()[name = tensor("op_294_begin_0"), val = tensor([0, 0, 0, 2560])]; tensor var_294_end_0 = const()[name = tensor("op_294_end_0"), val = tensor([2, 40, 1, 3072])]; tensor var_294_end_mask_0 = const()[name = tensor("op_294_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_294_cast_fp16 = slice_by_index(begin = var_294_begin_0, end = var_294_end_0, end_mask = var_294_end_mask_0, x = var_246_cast_fp16)[name = tensor("op_294_cast_fp16")]; tensor var_295_begin_0 = const()[name = tensor("op_295_begin_0"), val = tensor([0, 0, 0, 3072])]; tensor var_295_end_0 = const()[name = tensor("op_295_end_0"), val = tensor([2, 40, 1, 3584])]; tensor var_295_end_mask_0 = const()[name = tensor("op_295_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_295_cast_fp16 = slice_by_index(begin = var_295_begin_0, end = var_295_end_0, end_mask = var_295_end_mask_0, x = var_246_cast_fp16)[name = tensor("op_295_cast_fp16")]; tensor var_296_begin_0 = const()[name = tensor("op_296_begin_0"), val = tensor([0, 0, 0, 3584])]; tensor var_296_end_0 = const()[name = tensor("op_296_end_0"), val = tensor([2, 40, 1, 4096])]; tensor var_296_end_mask_0 = const()[name = tensor("op_296_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_296_cast_fp16 = slice_by_index(begin = var_296_begin_0, end = var_296_end_0, end_mask = var_296_end_mask_0, x = var_246_cast_fp16)[name = tensor("op_296_cast_fp16")]; tensor var_297_begin_0 = const()[name = tensor("op_297_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_297_end_0 = const()[name = tensor("op_297_end_0"), val = tensor([2, 40, 1, 512])]; tensor var_297_end_mask_0 = const()[name = tensor("op_297_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_297_cast_fp16 = slice_by_index(begin = var_297_begin_0, end = var_297_end_0, end_mask = var_297_end_mask_0, x = var_250_cast_fp16)[name = tensor("op_297_cast_fp16")]; tensor var_298_begin_0 = const()[name = tensor("op_298_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_298_end_0 = const()[name = tensor("op_298_end_0"), val = tensor([2, 40, 1, 1024])]; tensor var_298_end_mask_0 = const()[name = tensor("op_298_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_298_cast_fp16 = slice_by_index(begin = var_298_begin_0, end = var_298_end_0, end_mask = var_298_end_mask_0, x = var_250_cast_fp16)[name = tensor("op_298_cast_fp16")]; tensor var_299_begin_0 = const()[name = tensor("op_299_begin_0"), val = tensor([0, 0, 0, 1024])]; tensor var_299_end_0 = const()[name = tensor("op_299_end_0"), val = tensor([2, 40, 1, 1536])]; tensor var_299_end_mask_0 = const()[name = tensor("op_299_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_299_cast_fp16 = slice_by_index(begin = var_299_begin_0, end = var_299_end_0, end_mask = var_299_end_mask_0, x = var_250_cast_fp16)[name = tensor("op_299_cast_fp16")]; tensor var_300_begin_0 = const()[name = tensor("op_300_begin_0"), val = tensor([0, 0, 0, 1536])]; tensor var_300_end_0 = const()[name = tensor("op_300_end_0"), val = tensor([2, 40, 1, 2048])]; tensor var_300_end_mask_0 = const()[name = tensor("op_300_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_300_cast_fp16 = slice_by_index(begin = var_300_begin_0, end = var_300_end_0, end_mask = var_300_end_mask_0, x = var_250_cast_fp16)[name = tensor("op_300_cast_fp16")]; tensor var_301_begin_0 = const()[name = tensor("op_301_begin_0"), val = tensor([0, 0, 0, 2048])]; tensor var_301_end_0 = const()[name = tensor("op_301_end_0"), val = tensor([2, 40, 1, 2560])]; tensor var_301_end_mask_0 = const()[name = tensor("op_301_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_301_cast_fp16 = slice_by_index(begin = var_301_begin_0, end = var_301_end_0, end_mask = var_301_end_mask_0, x = var_250_cast_fp16)[name = tensor("op_301_cast_fp16")]; tensor var_302_begin_0 = const()[name = tensor("op_302_begin_0"), val = tensor([0, 0, 0, 2560])]; tensor var_302_end_0 = const()[name = tensor("op_302_end_0"), val = tensor([2, 40, 1, 3072])]; tensor var_302_end_mask_0 = const()[name = tensor("op_302_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_302_cast_fp16 = slice_by_index(begin = var_302_begin_0, end = var_302_end_0, end_mask = var_302_end_mask_0, x = var_250_cast_fp16)[name = tensor("op_302_cast_fp16")]; tensor var_303_begin_0 = const()[name = tensor("op_303_begin_0"), val = tensor([0, 0, 0, 3072])]; tensor var_303_end_0 = const()[name = tensor("op_303_end_0"), val = tensor([2, 40, 1, 3584])]; tensor var_303_end_mask_0 = const()[name = tensor("op_303_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_303_cast_fp16 = slice_by_index(begin = var_303_begin_0, end = var_303_end_0, end_mask = var_303_end_mask_0, x = var_250_cast_fp16)[name = tensor("op_303_cast_fp16")]; tensor var_304_begin_0 = const()[name = tensor("op_304_begin_0"), val = tensor([0, 0, 0, 3584])]; tensor var_304_end_0 = const()[name = tensor("op_304_end_0"), val = tensor([2, 40, 1, 4096])]; tensor var_304_end_mask_0 = const()[name = tensor("op_304_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_304_cast_fp16 = slice_by_index(begin = var_304_begin_0, end = var_304_end_0, end_mask = var_304_end_mask_0, x = var_250_cast_fp16)[name = tensor("op_304_cast_fp16")]; tensor var_305_begin_0 = const()[name = tensor("op_305_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_305_end_0 = const()[name = tensor("op_305_end_0"), val = tensor([2, 40, 1, 512])]; tensor var_305_end_mask_0 = const()[name = tensor("op_305_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_305_cast_fp16 = slice_by_index(begin = var_305_begin_0, end = var_305_end_0, end_mask = var_305_end_mask_0, x = var_254_cast_fp16)[name = tensor("op_305_cast_fp16")]; tensor var_306_begin_0 = const()[name = tensor("op_306_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_306_end_0 = const()[name = tensor("op_306_end_0"), val = tensor([2, 40, 1, 1024])]; tensor var_306_end_mask_0 = const()[name = tensor("op_306_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_306_cast_fp16 = slice_by_index(begin = var_306_begin_0, end = var_306_end_0, end_mask = var_306_end_mask_0, x = var_254_cast_fp16)[name = tensor("op_306_cast_fp16")]; tensor var_307_begin_0 = const()[name = tensor("op_307_begin_0"), val = tensor([0, 0, 0, 1024])]; tensor var_307_end_0 = const()[name = tensor("op_307_end_0"), val = tensor([2, 40, 1, 1536])]; tensor var_307_end_mask_0 = const()[name = tensor("op_307_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_307_cast_fp16 = slice_by_index(begin = var_307_begin_0, end = var_307_end_0, end_mask = var_307_end_mask_0, x = var_254_cast_fp16)[name = tensor("op_307_cast_fp16")]; tensor var_308_begin_0 = const()[name = tensor("op_308_begin_0"), val = tensor([0, 0, 0, 1536])]; tensor var_308_end_0 = const()[name = tensor("op_308_end_0"), val = tensor([2, 40, 1, 2048])]; tensor var_308_end_mask_0 = const()[name = tensor("op_308_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_308_cast_fp16 = slice_by_index(begin = var_308_begin_0, end = var_308_end_0, end_mask = var_308_end_mask_0, x = var_254_cast_fp16)[name = tensor("op_308_cast_fp16")]; tensor var_309_begin_0 = const()[name = tensor("op_309_begin_0"), val = tensor([0, 0, 0, 2048])]; tensor var_309_end_0 = const()[name = tensor("op_309_end_0"), val = tensor([2, 40, 1, 2560])]; tensor var_309_end_mask_0 = const()[name = tensor("op_309_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_309_cast_fp16 = slice_by_index(begin = var_309_begin_0, end = var_309_end_0, end_mask = var_309_end_mask_0, x = var_254_cast_fp16)[name = tensor("op_309_cast_fp16")]; tensor var_310_begin_0 = const()[name = tensor("op_310_begin_0"), val = tensor([0, 0, 0, 2560])]; tensor var_310_end_0 = const()[name = tensor("op_310_end_0"), val = tensor([2, 40, 1, 3072])]; tensor var_310_end_mask_0 = const()[name = tensor("op_310_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_310_cast_fp16 = slice_by_index(begin = var_310_begin_0, end = var_310_end_0, end_mask = var_310_end_mask_0, x = var_254_cast_fp16)[name = tensor("op_310_cast_fp16")]; tensor var_311_begin_0 = const()[name = tensor("op_311_begin_0"), val = tensor([0, 0, 0, 3072])]; tensor var_311_end_0 = const()[name = tensor("op_311_end_0"), val = tensor([2, 40, 1, 3584])]; tensor var_311_end_mask_0 = const()[name = tensor("op_311_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_311_cast_fp16 = slice_by_index(begin = var_311_begin_0, end = var_311_end_0, end_mask = var_311_end_mask_0, x = var_254_cast_fp16)[name = tensor("op_311_cast_fp16")]; tensor var_312_begin_0 = const()[name = tensor("op_312_begin_0"), val = tensor([0, 0, 0, 3584])]; tensor var_312_end_0 = const()[name = tensor("op_312_end_0"), val = tensor([2, 40, 1, 4096])]; tensor var_312_end_mask_0 = const()[name = tensor("op_312_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_312_cast_fp16 = slice_by_index(begin = var_312_begin_0, end = var_312_end_0, end_mask = var_312_end_mask_0, x = var_254_cast_fp16)[name = tensor("op_312_cast_fp16")]; tensor var_313_begin_0 = const()[name = tensor("op_313_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_313_end_0 = const()[name = tensor("op_313_end_0"), val = tensor([2, 40, 1, 512])]; tensor var_313_end_mask_0 = const()[name = tensor("op_313_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_313_cast_fp16 = slice_by_index(begin = var_313_begin_0, end = var_313_end_0, end_mask = var_313_end_mask_0, x = var_258_cast_fp16)[name = tensor("op_313_cast_fp16")]; tensor var_314_begin_0 = const()[name = tensor("op_314_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_314_end_0 = const()[name = tensor("op_314_end_0"), val = tensor([2, 40, 1, 1024])]; tensor var_314_end_mask_0 = const()[name = tensor("op_314_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_314_cast_fp16 = slice_by_index(begin = var_314_begin_0, end = var_314_end_0, end_mask = var_314_end_mask_0, x = var_258_cast_fp16)[name = tensor("op_314_cast_fp16")]; tensor var_315_begin_0 = const()[name = tensor("op_315_begin_0"), val = tensor([0, 0, 0, 1024])]; tensor var_315_end_0 = const()[name = tensor("op_315_end_0"), val = tensor([2, 40, 1, 1536])]; tensor var_315_end_mask_0 = const()[name = tensor("op_315_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_315_cast_fp16 = slice_by_index(begin = var_315_begin_0, end = var_315_end_0, end_mask = var_315_end_mask_0, x = var_258_cast_fp16)[name = tensor("op_315_cast_fp16")]; tensor var_316_begin_0 = const()[name = tensor("op_316_begin_0"), val = tensor([0, 0, 0, 1536])]; tensor var_316_end_0 = const()[name = tensor("op_316_end_0"), val = tensor([2, 40, 1, 2048])]; tensor var_316_end_mask_0 = const()[name = tensor("op_316_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_316_cast_fp16 = slice_by_index(begin = var_316_begin_0, end = var_316_end_0, end_mask = var_316_end_mask_0, x = var_258_cast_fp16)[name = tensor("op_316_cast_fp16")]; tensor var_317_begin_0 = const()[name = tensor("op_317_begin_0"), val = tensor([0, 0, 0, 2048])]; tensor var_317_end_0 = const()[name = tensor("op_317_end_0"), val = tensor([2, 40, 1, 2560])]; tensor var_317_end_mask_0 = const()[name = tensor("op_317_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_317_cast_fp16 = slice_by_index(begin = var_317_begin_0, end = var_317_end_0, end_mask = var_317_end_mask_0, x = var_258_cast_fp16)[name = tensor("op_317_cast_fp16")]; tensor var_318_begin_0 = const()[name = tensor("op_318_begin_0"), val = tensor([0, 0, 0, 2560])]; tensor var_318_end_0 = const()[name = tensor("op_318_end_0"), val = tensor([2, 40, 1, 3072])]; tensor var_318_end_mask_0 = const()[name = tensor("op_318_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_318_cast_fp16 = slice_by_index(begin = var_318_begin_0, end = var_318_end_0, end_mask = var_318_end_mask_0, x = var_258_cast_fp16)[name = tensor("op_318_cast_fp16")]; tensor var_319_begin_0 = const()[name = tensor("op_319_begin_0"), val = tensor([0, 0, 0, 3072])]; tensor var_319_end_0 = const()[name = tensor("op_319_end_0"), val = tensor([2, 40, 1, 3584])]; tensor var_319_end_mask_0 = const()[name = tensor("op_319_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_319_cast_fp16 = slice_by_index(begin = var_319_begin_0, end = var_319_end_0, end_mask = var_319_end_mask_0, x = var_258_cast_fp16)[name = tensor("op_319_cast_fp16")]; tensor var_320_begin_0 = const()[name = tensor("op_320_begin_0"), val = tensor([0, 0, 0, 3584])]; tensor var_320_end_0 = const()[name = tensor("op_320_end_0"), val = tensor([2, 40, 1, 4096])]; tensor var_320_end_mask_0 = const()[name = tensor("op_320_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_320_cast_fp16 = slice_by_index(begin = var_320_begin_0, end = var_320_end_0, end_mask = var_320_end_mask_0, x = var_258_cast_fp16)[name = tensor("op_320_cast_fp16")]; 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, 40, 1, 512])]; tensor var_321_end_mask_0 = const()[name = tensor("op_321_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_321_cast_fp16 = slice_by_index(begin = var_321_begin_0, end = var_321_end_0, end_mask = var_321_end_mask_0, x = var_262_cast_fp16)[name = tensor("op_321_cast_fp16")]; tensor var_322_begin_0 = const()[name = tensor("op_322_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_322_end_0 = const()[name = tensor("op_322_end_0"), val = tensor([2, 40, 1, 1024])]; tensor var_322_end_mask_0 = const()[name = tensor("op_322_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_322_cast_fp16 = slice_by_index(begin = var_322_begin_0, end = var_322_end_0, end_mask = var_322_end_mask_0, x = var_262_cast_fp16)[name = tensor("op_322_cast_fp16")]; tensor var_323_begin_0 = const()[name = tensor("op_323_begin_0"), val = tensor([0, 0, 0, 1024])]; tensor var_323_end_0 = const()[name = tensor("op_323_end_0"), val = tensor([2, 40, 1, 1536])]; tensor var_323_end_mask_0 = const()[name = tensor("op_323_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_323_cast_fp16 = slice_by_index(begin = var_323_begin_0, end = var_323_end_0, end_mask = var_323_end_mask_0, x = var_262_cast_fp16)[name = tensor("op_323_cast_fp16")]; tensor var_324_begin_0 = const()[name = tensor("op_324_begin_0"), val = tensor([0, 0, 0, 1536])]; tensor var_324_end_0 = const()[name = tensor("op_324_end_0"), val = tensor([2, 40, 1, 2048])]; tensor var_324_end_mask_0 = const()[name = tensor("op_324_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_324_cast_fp16 = slice_by_index(begin = var_324_begin_0, end = var_324_end_0, end_mask = var_324_end_mask_0, x = var_262_cast_fp16)[name = tensor("op_324_cast_fp16")]; tensor var_325_begin_0 = const()[name = tensor("op_325_begin_0"), val = tensor([0, 0, 0, 2048])]; tensor var_325_end_0 = const()[name = tensor("op_325_end_0"), val = tensor([2, 40, 1, 2560])]; tensor var_325_end_mask_0 = const()[name = tensor("op_325_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_325_cast_fp16 = slice_by_index(begin = var_325_begin_0, end = var_325_end_0, end_mask = var_325_end_mask_0, x = var_262_cast_fp16)[name = tensor("op_325_cast_fp16")]; tensor var_326_begin_0 = const()[name = tensor("op_326_begin_0"), val = tensor([0, 0, 0, 2560])]; tensor var_326_end_0 = const()[name = tensor("op_326_end_0"), val = tensor([2, 40, 1, 3072])]; tensor var_326_end_mask_0 = const()[name = tensor("op_326_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_326_cast_fp16 = slice_by_index(begin = var_326_begin_0, end = var_326_end_0, end_mask = var_326_end_mask_0, x = var_262_cast_fp16)[name = tensor("op_326_cast_fp16")]; tensor var_327_begin_0 = const()[name = tensor("op_327_begin_0"), val = tensor([0, 0, 0, 3072])]; tensor var_327_end_0 = const()[name = tensor("op_327_end_0"), val = tensor([2, 40, 1, 3584])]; tensor var_327_end_mask_0 = const()[name = tensor("op_327_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_327_cast_fp16 = slice_by_index(begin = var_327_begin_0, end = var_327_end_0, end_mask = var_327_end_mask_0, x = var_262_cast_fp16)[name = tensor("op_327_cast_fp16")]; tensor var_328_begin_0 = const()[name = tensor("op_328_begin_0"), val = tensor([0, 0, 0, 3584])]; tensor var_328_end_0 = const()[name = tensor("op_328_end_0"), val = tensor([2, 40, 1, 4096])]; tensor var_328_end_mask_0 = const()[name = tensor("op_328_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_328_cast_fp16 = slice_by_index(begin = var_328_begin_0, end = var_328_end_0, end_mask = var_328_end_mask_0, x = var_262_cast_fp16)[name = tensor("op_328_cast_fp16")]; tensor k_3_perm_0 = const()[name = tensor("k_3_perm_0"), val = tensor([0, 3, 2, 1])]; tensor var_333_begin_0 = const()[name = tensor("op_333_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_333_end_0 = const()[name = tensor("op_333_end_0"), val = tensor([2, 4096, 1, 40])]; tensor var_333_end_mask_0 = const()[name = tensor("op_333_end_mask_0"), val = tensor([true, true, true, false])]; tensor transpose_17 = transpose(perm = k_3_perm_0, x = k_1_cast_fp16)[name = tensor("transpose_17")]; tensor var_333_cast_fp16 = slice_by_index(begin = var_333_begin_0, end = var_333_end_0, end_mask = var_333_end_mask_0, x = transpose_17)[name = tensor("op_333_cast_fp16")]; tensor var_337_begin_0 = const()[name = tensor("op_337_begin_0"), val = tensor([0, 0, 0, 40])]; tensor var_337_end_0 = const()[name = tensor("op_337_end_0"), val = tensor([2, 4096, 1, 80])]; tensor var_337_end_mask_0 = const()[name = tensor("op_337_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_337_cast_fp16 = slice_by_index(begin = var_337_begin_0, end = var_337_end_0, end_mask = var_337_end_mask_0, x = transpose_17)[name = tensor("op_337_cast_fp16")]; tensor var_341_begin_0 = const()[name = tensor("op_341_begin_0"), val = tensor([0, 0, 0, 80])]; tensor var_341_end_0 = const()[name = tensor("op_341_end_0"), val = tensor([2, 4096, 1, 120])]; tensor var_341_end_mask_0 = const()[name = tensor("op_341_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_341_cast_fp16 = slice_by_index(begin = var_341_begin_0, end = var_341_end_0, end_mask = var_341_end_mask_0, x = transpose_17)[name = tensor("op_341_cast_fp16")]; tensor var_345_begin_0 = const()[name = tensor("op_345_begin_0"), val = tensor([0, 0, 0, 120])]; tensor var_345_end_0 = const()[name = tensor("op_345_end_0"), val = tensor([2, 4096, 1, 160])]; tensor var_345_end_mask_0 = const()[name = tensor("op_345_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_345_cast_fp16 = slice_by_index(begin = var_345_begin_0, end = var_345_end_0, end_mask = var_345_end_mask_0, x = transpose_17)[name = tensor("op_345_cast_fp16")]; tensor var_349_begin_0 = const()[name = tensor("op_349_begin_0"), val = tensor([0, 0, 0, 160])]; tensor var_349_end_0 = const()[name = tensor("op_349_end_0"), val = tensor([2, 4096, 1, 200])]; tensor var_349_end_mask_0 = const()[name = tensor("op_349_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_349_cast_fp16 = slice_by_index(begin = var_349_begin_0, end = var_349_end_0, end_mask = var_349_end_mask_0, x = transpose_17)[name = tensor("op_349_cast_fp16")]; tensor var_353_begin_0 = const()[name = tensor("op_353_begin_0"), val = tensor([0, 0, 0, 200])]; tensor var_353_end_0 = const()[name = tensor("op_353_end_0"), val = tensor([2, 4096, 1, 240])]; tensor var_353_end_mask_0 = const()[name = tensor("op_353_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_353_cast_fp16 = slice_by_index(begin = var_353_begin_0, end = var_353_end_0, end_mask = var_353_end_mask_0, x = transpose_17)[name = tensor("op_353_cast_fp16")]; tensor var_357_begin_0 = const()[name = tensor("op_357_begin_0"), val = tensor([0, 0, 0, 240])]; tensor var_357_end_0 = const()[name = tensor("op_357_end_0"), val = tensor([2, 4096, 1, 280])]; tensor var_357_end_mask_0 = const()[name = tensor("op_357_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_357_cast_fp16 = slice_by_index(begin = var_357_begin_0, end = var_357_end_0, end_mask = var_357_end_mask_0, x = transpose_17)[name = tensor("op_357_cast_fp16")]; tensor var_361_begin_0 = const()[name = tensor("op_361_begin_0"), val = tensor([0, 0, 0, 280])]; tensor var_361_end_0 = const()[name = tensor("op_361_end_0"), val = tensor([2, 4096, 1, 320])]; tensor var_361_end_mask_0 = const()[name = tensor("op_361_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_361_cast_fp16 = slice_by_index(begin = var_361_begin_0, end = var_361_end_0, end_mask = var_361_end_mask_0, x = transpose_17)[name = tensor("op_361_cast_fp16")]; tensor var_363_begin_0 = const()[name = tensor("op_363_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_363_end_0 = const()[name = tensor("op_363_end_0"), val = tensor([2, 40, 1, 4096])]; tensor var_363_end_mask_0 = const()[name = tensor("op_363_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_363_cast_fp16 = slice_by_index(begin = var_363_begin_0, end = var_363_end_0, end_mask = var_363_end_mask_0, x = v_1_cast_fp16)[name = tensor("op_363_cast_fp16")]; tensor var_367_begin_0 = const()[name = tensor("op_367_begin_0"), val = tensor([0, 40, 0, 0])]; tensor var_367_end_0 = const()[name = tensor("op_367_end_0"), val = tensor([2, 80, 1, 4096])]; tensor var_367_end_mask_0 = const()[name = tensor("op_367_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_367_cast_fp16 = slice_by_index(begin = var_367_begin_0, end = var_367_end_0, end_mask = var_367_end_mask_0, x = v_1_cast_fp16)[name = tensor("op_367_cast_fp16")]; tensor var_371_begin_0 = const()[name = tensor("op_371_begin_0"), val = tensor([0, 80, 0, 0])]; tensor var_371_end_0 = const()[name = tensor("op_371_end_0"), val = tensor([2, 120, 1, 4096])]; tensor var_371_end_mask_0 = const()[name = tensor("op_371_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_371_cast_fp16 = slice_by_index(begin = var_371_begin_0, end = var_371_end_0, end_mask = var_371_end_mask_0, x = v_1_cast_fp16)[name = tensor("op_371_cast_fp16")]; tensor var_375_begin_0 = const()[name = tensor("op_375_begin_0"), val = tensor([0, 120, 0, 0])]; tensor var_375_end_0 = const()[name = tensor("op_375_end_0"), val = tensor([2, 160, 1, 4096])]; tensor var_375_end_mask_0 = const()[name = tensor("op_375_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_375_cast_fp16 = slice_by_index(begin = var_375_begin_0, end = var_375_end_0, end_mask = var_375_end_mask_0, x = v_1_cast_fp16)[name = tensor("op_375_cast_fp16")]; tensor var_379_begin_0 = const()[name = tensor("op_379_begin_0"), val = tensor([0, 160, 0, 0])]; tensor var_379_end_0 = const()[name = tensor("op_379_end_0"), val = tensor([2, 200, 1, 4096])]; tensor var_379_end_mask_0 = const()[name = tensor("op_379_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_379_cast_fp16 = slice_by_index(begin = var_379_begin_0, end = var_379_end_0, end_mask = var_379_end_mask_0, x = v_1_cast_fp16)[name = tensor("op_379_cast_fp16")]; tensor var_383_begin_0 = const()[name = tensor("op_383_begin_0"), val = tensor([0, 200, 0, 0])]; tensor var_383_end_0 = const()[name = tensor("op_383_end_0"), val = tensor([2, 240, 1, 4096])]; tensor var_383_end_mask_0 = const()[name = tensor("op_383_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_383_cast_fp16 = slice_by_index(begin = var_383_begin_0, end = var_383_end_0, end_mask = var_383_end_mask_0, x = v_1_cast_fp16)[name = tensor("op_383_cast_fp16")]; tensor var_387_begin_0 = const()[name = tensor("op_387_begin_0"), val = tensor([0, 240, 0, 0])]; tensor var_387_end_0 = const()[name = tensor("op_387_end_0"), val = tensor([2, 280, 1, 4096])]; tensor var_387_end_mask_0 = const()[name = tensor("op_387_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_387_cast_fp16 = slice_by_index(begin = var_387_begin_0, end = var_387_end_0, end_mask = var_387_end_mask_0, x = v_1_cast_fp16)[name = tensor("op_387_cast_fp16")]; tensor var_391_begin_0 = const()[name = tensor("op_391_begin_0"), val = tensor([0, 280, 0, 0])]; tensor var_391_end_0 = const()[name = tensor("op_391_end_0"), val = tensor([2, 320, 1, 4096])]; tensor var_391_end_mask_0 = const()[name = tensor("op_391_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_391_cast_fp16 = slice_by_index(begin = var_391_begin_0, end = var_391_end_0, end_mask = var_391_end_mask_0, x = v_1_cast_fp16)[name = tensor("op_391_cast_fp16")]; tensor var_395_equation_0 = const()[name = tensor("op_395_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_395_cast_fp16 = einsum(equation = var_395_equation_0, values = (var_333_cast_fp16, var_265_cast_fp16))[name = tensor("op_395_cast_fp16")]; tensor var_396_to_fp16 = const()[name = tensor("op_396_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_1_cast_fp16 = mul(x = var_395_cast_fp16, y = var_396_to_fp16)[name = tensor("aw_chunk_1_cast_fp16")]; tensor var_399_equation_0 = const()[name = tensor("op_399_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_399_cast_fp16 = einsum(equation = var_399_equation_0, values = (var_333_cast_fp16, var_266_cast_fp16))[name = tensor("op_399_cast_fp16")]; tensor var_400_to_fp16 = const()[name = tensor("op_400_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_3_cast_fp16 = mul(x = var_399_cast_fp16, y = var_400_to_fp16)[name = tensor("aw_chunk_3_cast_fp16")]; tensor var_403_equation_0 = const()[name = tensor("op_403_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_403_cast_fp16 = einsum(equation = var_403_equation_0, values = (var_333_cast_fp16, var_267_cast_fp16))[name = tensor("op_403_cast_fp16")]; tensor var_404_to_fp16 = const()[name = tensor("op_404_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_5_cast_fp16 = mul(x = var_403_cast_fp16, y = var_404_to_fp16)[name = tensor("aw_chunk_5_cast_fp16")]; tensor var_407_equation_0 = const()[name = tensor("op_407_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_407_cast_fp16 = einsum(equation = var_407_equation_0, values = (var_333_cast_fp16, var_268_cast_fp16))[name = tensor("op_407_cast_fp16")]; tensor var_408_to_fp16 = const()[name = tensor("op_408_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_7_cast_fp16 = mul(x = var_407_cast_fp16, y = var_408_to_fp16)[name = tensor("aw_chunk_7_cast_fp16")]; tensor var_411_equation_0 = const()[name = tensor("op_411_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_411_cast_fp16 = einsum(equation = var_411_equation_0, values = (var_333_cast_fp16, var_269_cast_fp16))[name = tensor("op_411_cast_fp16")]; tensor var_412_to_fp16 = const()[name = tensor("op_412_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_9_cast_fp16 = mul(x = var_411_cast_fp16, y = var_412_to_fp16)[name = tensor("aw_chunk_9_cast_fp16")]; tensor var_415_equation_0 = const()[name = tensor("op_415_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_415_cast_fp16 = einsum(equation = var_415_equation_0, values = (var_333_cast_fp16, var_270_cast_fp16))[name = tensor("op_415_cast_fp16")]; tensor var_416_to_fp16 = const()[name = tensor("op_416_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_11_cast_fp16 = mul(x = var_415_cast_fp16, y = var_416_to_fp16)[name = tensor("aw_chunk_11_cast_fp16")]; tensor var_419_equation_0 = const()[name = tensor("op_419_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_419_cast_fp16 = einsum(equation = var_419_equation_0, values = (var_333_cast_fp16, var_271_cast_fp16))[name = tensor("op_419_cast_fp16")]; tensor var_420_to_fp16 = const()[name = tensor("op_420_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_13_cast_fp16 = mul(x = var_419_cast_fp16, y = var_420_to_fp16)[name = tensor("aw_chunk_13_cast_fp16")]; tensor var_423_equation_0 = const()[name = tensor("op_423_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_423_cast_fp16 = einsum(equation = var_423_equation_0, values = (var_333_cast_fp16, var_272_cast_fp16))[name = tensor("op_423_cast_fp16")]; tensor var_424_to_fp16 = const()[name = tensor("op_424_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_15_cast_fp16 = mul(x = var_423_cast_fp16, y = var_424_to_fp16)[name = tensor("aw_chunk_15_cast_fp16")]; tensor var_427_equation_0 = const()[name = tensor("op_427_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_427_cast_fp16 = einsum(equation = var_427_equation_0, values = (var_337_cast_fp16, var_273_cast_fp16))[name = tensor("op_427_cast_fp16")]; tensor var_428_to_fp16 = const()[name = tensor("op_428_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_17_cast_fp16 = mul(x = var_427_cast_fp16, y = var_428_to_fp16)[name = tensor("aw_chunk_17_cast_fp16")]; tensor var_431_equation_0 = const()[name = tensor("op_431_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_431_cast_fp16 = einsum(equation = var_431_equation_0, values = (var_337_cast_fp16, var_274_cast_fp16))[name = tensor("op_431_cast_fp16")]; tensor var_432_to_fp16 = const()[name = tensor("op_432_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_19_cast_fp16 = mul(x = var_431_cast_fp16, y = var_432_to_fp16)[name = tensor("aw_chunk_19_cast_fp16")]; tensor var_435_equation_0 = const()[name = tensor("op_435_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_435_cast_fp16 = einsum(equation = var_435_equation_0, values = (var_337_cast_fp16, var_275_cast_fp16))[name = tensor("op_435_cast_fp16")]; tensor var_436_to_fp16 = const()[name = tensor("op_436_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_21_cast_fp16 = mul(x = var_435_cast_fp16, y = var_436_to_fp16)[name = tensor("aw_chunk_21_cast_fp16")]; tensor var_439_equation_0 = const()[name = tensor("op_439_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_439_cast_fp16 = einsum(equation = var_439_equation_0, values = (var_337_cast_fp16, var_276_cast_fp16))[name = tensor("op_439_cast_fp16")]; tensor var_440_to_fp16 = const()[name = tensor("op_440_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_23_cast_fp16 = mul(x = var_439_cast_fp16, y = var_440_to_fp16)[name = tensor("aw_chunk_23_cast_fp16")]; tensor var_443_equation_0 = const()[name = tensor("op_443_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_443_cast_fp16 = einsum(equation = var_443_equation_0, values = (var_337_cast_fp16, var_277_cast_fp16))[name = tensor("op_443_cast_fp16")]; tensor var_444_to_fp16 = const()[name = tensor("op_444_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_25_cast_fp16 = mul(x = var_443_cast_fp16, y = var_444_to_fp16)[name = tensor("aw_chunk_25_cast_fp16")]; tensor var_447_equation_0 = const()[name = tensor("op_447_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_447_cast_fp16 = einsum(equation = var_447_equation_0, values = (var_337_cast_fp16, var_278_cast_fp16))[name = tensor("op_447_cast_fp16")]; tensor var_448_to_fp16 = const()[name = tensor("op_448_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_27_cast_fp16 = mul(x = var_447_cast_fp16, y = var_448_to_fp16)[name = tensor("aw_chunk_27_cast_fp16")]; tensor var_451_equation_0 = const()[name = tensor("op_451_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_451_cast_fp16 = einsum(equation = var_451_equation_0, values = (var_337_cast_fp16, var_279_cast_fp16))[name = tensor("op_451_cast_fp16")]; tensor var_452_to_fp16 = const()[name = tensor("op_452_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_29_cast_fp16 = mul(x = var_451_cast_fp16, y = var_452_to_fp16)[name = tensor("aw_chunk_29_cast_fp16")]; tensor var_455_equation_0 = const()[name = tensor("op_455_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_455_cast_fp16 = einsum(equation = var_455_equation_0, values = (var_337_cast_fp16, var_280_cast_fp16))[name = tensor("op_455_cast_fp16")]; tensor var_456_to_fp16 = const()[name = tensor("op_456_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_31_cast_fp16 = mul(x = var_455_cast_fp16, y = var_456_to_fp16)[name = tensor("aw_chunk_31_cast_fp16")]; tensor var_459_equation_0 = const()[name = tensor("op_459_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_459_cast_fp16 = einsum(equation = var_459_equation_0, values = (var_341_cast_fp16, var_281_cast_fp16))[name = tensor("op_459_cast_fp16")]; tensor var_460_to_fp16 = const()[name = tensor("op_460_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_33_cast_fp16 = mul(x = var_459_cast_fp16, y = var_460_to_fp16)[name = tensor("aw_chunk_33_cast_fp16")]; tensor var_463_equation_0 = const()[name = tensor("op_463_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_463_cast_fp16 = einsum(equation = var_463_equation_0, values = (var_341_cast_fp16, var_282_cast_fp16))[name = tensor("op_463_cast_fp16")]; tensor var_464_to_fp16 = const()[name = tensor("op_464_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_35_cast_fp16 = mul(x = var_463_cast_fp16, y = var_464_to_fp16)[name = tensor("aw_chunk_35_cast_fp16")]; tensor var_467_equation_0 = const()[name = tensor("op_467_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_467_cast_fp16 = einsum(equation = var_467_equation_0, values = (var_341_cast_fp16, var_283_cast_fp16))[name = tensor("op_467_cast_fp16")]; tensor var_468_to_fp16 = const()[name = tensor("op_468_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_37_cast_fp16 = mul(x = var_467_cast_fp16, y = var_468_to_fp16)[name = tensor("aw_chunk_37_cast_fp16")]; tensor var_471_equation_0 = const()[name = tensor("op_471_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_471_cast_fp16 = einsum(equation = var_471_equation_0, values = (var_341_cast_fp16, var_284_cast_fp16))[name = tensor("op_471_cast_fp16")]; tensor var_472_to_fp16 = const()[name = tensor("op_472_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_39_cast_fp16 = mul(x = var_471_cast_fp16, y = var_472_to_fp16)[name = tensor("aw_chunk_39_cast_fp16")]; tensor var_475_equation_0 = const()[name = tensor("op_475_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_475_cast_fp16 = einsum(equation = var_475_equation_0, values = (var_341_cast_fp16, var_285_cast_fp16))[name = tensor("op_475_cast_fp16")]; tensor var_476_to_fp16 = const()[name = tensor("op_476_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_41_cast_fp16 = mul(x = var_475_cast_fp16, y = var_476_to_fp16)[name = tensor("aw_chunk_41_cast_fp16")]; tensor var_479_equation_0 = const()[name = tensor("op_479_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_479_cast_fp16 = einsum(equation = var_479_equation_0, values = (var_341_cast_fp16, var_286_cast_fp16))[name = tensor("op_479_cast_fp16")]; tensor var_480_to_fp16 = const()[name = tensor("op_480_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_43_cast_fp16 = mul(x = var_479_cast_fp16, y = var_480_to_fp16)[name = tensor("aw_chunk_43_cast_fp16")]; tensor var_483_equation_0 = const()[name = tensor("op_483_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_483_cast_fp16 = einsum(equation = var_483_equation_0, values = (var_341_cast_fp16, var_287_cast_fp16))[name = tensor("op_483_cast_fp16")]; tensor var_484_to_fp16 = const()[name = tensor("op_484_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_45_cast_fp16 = mul(x = var_483_cast_fp16, y = var_484_to_fp16)[name = tensor("aw_chunk_45_cast_fp16")]; tensor var_487_equation_0 = const()[name = tensor("op_487_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_487_cast_fp16 = einsum(equation = var_487_equation_0, values = (var_341_cast_fp16, var_288_cast_fp16))[name = tensor("op_487_cast_fp16")]; tensor var_488_to_fp16 = const()[name = tensor("op_488_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_47_cast_fp16 = mul(x = var_487_cast_fp16, y = var_488_to_fp16)[name = tensor("aw_chunk_47_cast_fp16")]; tensor var_491_equation_0 = const()[name = tensor("op_491_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_491_cast_fp16 = einsum(equation = var_491_equation_0, values = (var_345_cast_fp16, var_289_cast_fp16))[name = tensor("op_491_cast_fp16")]; tensor var_492_to_fp16 = const()[name = tensor("op_492_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_49_cast_fp16 = mul(x = var_491_cast_fp16, y = var_492_to_fp16)[name = tensor("aw_chunk_49_cast_fp16")]; tensor var_495_equation_0 = const()[name = tensor("op_495_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_495_cast_fp16 = einsum(equation = var_495_equation_0, values = (var_345_cast_fp16, var_290_cast_fp16))[name = tensor("op_495_cast_fp16")]; tensor var_496_to_fp16 = const()[name = tensor("op_496_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_51_cast_fp16 = mul(x = var_495_cast_fp16, y = var_496_to_fp16)[name = tensor("aw_chunk_51_cast_fp16")]; tensor var_499_equation_0 = const()[name = tensor("op_499_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_499_cast_fp16 = einsum(equation = var_499_equation_0, values = (var_345_cast_fp16, var_291_cast_fp16))[name = tensor("op_499_cast_fp16")]; tensor var_500_to_fp16 = const()[name = tensor("op_500_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_53_cast_fp16 = mul(x = var_499_cast_fp16, y = var_500_to_fp16)[name = tensor("aw_chunk_53_cast_fp16")]; tensor var_503_equation_0 = const()[name = tensor("op_503_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_503_cast_fp16 = einsum(equation = var_503_equation_0, values = (var_345_cast_fp16, var_292_cast_fp16))[name = tensor("op_503_cast_fp16")]; tensor var_504_to_fp16 = const()[name = tensor("op_504_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_55_cast_fp16 = mul(x = var_503_cast_fp16, y = var_504_to_fp16)[name = tensor("aw_chunk_55_cast_fp16")]; tensor var_507_equation_0 = const()[name = tensor("op_507_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_507_cast_fp16 = einsum(equation = var_507_equation_0, values = (var_345_cast_fp16, var_293_cast_fp16))[name = tensor("op_507_cast_fp16")]; tensor var_508_to_fp16 = const()[name = tensor("op_508_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_57_cast_fp16 = mul(x = var_507_cast_fp16, y = var_508_to_fp16)[name = tensor("aw_chunk_57_cast_fp16")]; tensor var_511_equation_0 = const()[name = tensor("op_511_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_511_cast_fp16 = einsum(equation = var_511_equation_0, values = (var_345_cast_fp16, var_294_cast_fp16))[name = tensor("op_511_cast_fp16")]; tensor var_512_to_fp16 = const()[name = tensor("op_512_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_59_cast_fp16 = mul(x = var_511_cast_fp16, y = var_512_to_fp16)[name = tensor("aw_chunk_59_cast_fp16")]; tensor var_515_equation_0 = const()[name = tensor("op_515_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_515_cast_fp16 = einsum(equation = var_515_equation_0, values = (var_345_cast_fp16, var_295_cast_fp16))[name = tensor("op_515_cast_fp16")]; tensor var_516_to_fp16 = const()[name = tensor("op_516_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_61_cast_fp16 = mul(x = var_515_cast_fp16, y = var_516_to_fp16)[name = tensor("aw_chunk_61_cast_fp16")]; tensor var_519_equation_0 = const()[name = tensor("op_519_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_519_cast_fp16 = einsum(equation = var_519_equation_0, values = (var_345_cast_fp16, var_296_cast_fp16))[name = tensor("op_519_cast_fp16")]; tensor var_520_to_fp16 = const()[name = tensor("op_520_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_63_cast_fp16 = mul(x = var_519_cast_fp16, y = var_520_to_fp16)[name = tensor("aw_chunk_63_cast_fp16")]; tensor var_523_equation_0 = const()[name = tensor("op_523_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_523_cast_fp16 = einsum(equation = var_523_equation_0, values = (var_349_cast_fp16, var_297_cast_fp16))[name = tensor("op_523_cast_fp16")]; tensor var_524_to_fp16 = const()[name = tensor("op_524_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_65_cast_fp16 = mul(x = var_523_cast_fp16, y = var_524_to_fp16)[name = tensor("aw_chunk_65_cast_fp16")]; tensor var_527_equation_0 = const()[name = tensor("op_527_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_527_cast_fp16 = einsum(equation = var_527_equation_0, values = (var_349_cast_fp16, var_298_cast_fp16))[name = tensor("op_527_cast_fp16")]; tensor var_528_to_fp16 = const()[name = tensor("op_528_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_67_cast_fp16 = mul(x = var_527_cast_fp16, y = var_528_to_fp16)[name = tensor("aw_chunk_67_cast_fp16")]; tensor var_531_equation_0 = const()[name = tensor("op_531_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_531_cast_fp16 = einsum(equation = var_531_equation_0, values = (var_349_cast_fp16, var_299_cast_fp16))[name = tensor("op_531_cast_fp16")]; tensor var_532_to_fp16 = const()[name = tensor("op_532_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_69_cast_fp16 = mul(x = var_531_cast_fp16, y = var_532_to_fp16)[name = tensor("aw_chunk_69_cast_fp16")]; tensor var_535_equation_0 = const()[name = tensor("op_535_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_535_cast_fp16 = einsum(equation = var_535_equation_0, values = (var_349_cast_fp16, var_300_cast_fp16))[name = tensor("op_535_cast_fp16")]; tensor var_536_to_fp16 = const()[name = tensor("op_536_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_71_cast_fp16 = mul(x = var_535_cast_fp16, y = var_536_to_fp16)[name = tensor("aw_chunk_71_cast_fp16")]; tensor var_539_equation_0 = const()[name = tensor("op_539_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_539_cast_fp16 = einsum(equation = var_539_equation_0, values = (var_349_cast_fp16, var_301_cast_fp16))[name = tensor("op_539_cast_fp16")]; tensor var_540_to_fp16 = const()[name = tensor("op_540_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_73_cast_fp16 = mul(x = var_539_cast_fp16, y = var_540_to_fp16)[name = tensor("aw_chunk_73_cast_fp16")]; tensor var_543_equation_0 = const()[name = tensor("op_543_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_543_cast_fp16 = einsum(equation = var_543_equation_0, values = (var_349_cast_fp16, var_302_cast_fp16))[name = tensor("op_543_cast_fp16")]; tensor var_544_to_fp16 = const()[name = tensor("op_544_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_75_cast_fp16 = mul(x = var_543_cast_fp16, y = var_544_to_fp16)[name = tensor("aw_chunk_75_cast_fp16")]; tensor var_547_equation_0 = const()[name = tensor("op_547_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_547_cast_fp16 = einsum(equation = var_547_equation_0, values = (var_349_cast_fp16, var_303_cast_fp16))[name = tensor("op_547_cast_fp16")]; tensor var_548_to_fp16 = const()[name = tensor("op_548_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_77_cast_fp16 = mul(x = var_547_cast_fp16, y = var_548_to_fp16)[name = tensor("aw_chunk_77_cast_fp16")]; tensor var_551_equation_0 = const()[name = tensor("op_551_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_551_cast_fp16 = einsum(equation = var_551_equation_0, values = (var_349_cast_fp16, var_304_cast_fp16))[name = tensor("op_551_cast_fp16")]; tensor var_552_to_fp16 = const()[name = tensor("op_552_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_79_cast_fp16 = mul(x = var_551_cast_fp16, y = var_552_to_fp16)[name = tensor("aw_chunk_79_cast_fp16")]; tensor var_555_equation_0 = const()[name = tensor("op_555_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_555_cast_fp16 = einsum(equation = var_555_equation_0, values = (var_353_cast_fp16, var_305_cast_fp16))[name = tensor("op_555_cast_fp16")]; tensor var_556_to_fp16 = const()[name = tensor("op_556_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_81_cast_fp16 = mul(x = var_555_cast_fp16, y = var_556_to_fp16)[name = tensor("aw_chunk_81_cast_fp16")]; tensor var_559_equation_0 = const()[name = tensor("op_559_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_559_cast_fp16 = einsum(equation = var_559_equation_0, values = (var_353_cast_fp16, var_306_cast_fp16))[name = tensor("op_559_cast_fp16")]; tensor var_560_to_fp16 = const()[name = tensor("op_560_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_83_cast_fp16 = mul(x = var_559_cast_fp16, y = var_560_to_fp16)[name = tensor("aw_chunk_83_cast_fp16")]; tensor var_563_equation_0 = const()[name = tensor("op_563_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_563_cast_fp16 = einsum(equation = var_563_equation_0, values = (var_353_cast_fp16, var_307_cast_fp16))[name = tensor("op_563_cast_fp16")]; tensor var_564_to_fp16 = const()[name = tensor("op_564_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_85_cast_fp16 = mul(x = var_563_cast_fp16, y = var_564_to_fp16)[name = tensor("aw_chunk_85_cast_fp16")]; tensor var_567_equation_0 = const()[name = tensor("op_567_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_567_cast_fp16 = einsum(equation = var_567_equation_0, values = (var_353_cast_fp16, var_308_cast_fp16))[name = tensor("op_567_cast_fp16")]; tensor var_568_to_fp16 = const()[name = tensor("op_568_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_87_cast_fp16 = mul(x = var_567_cast_fp16, y = var_568_to_fp16)[name = tensor("aw_chunk_87_cast_fp16")]; tensor var_571_equation_0 = const()[name = tensor("op_571_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_571_cast_fp16 = einsum(equation = var_571_equation_0, values = (var_353_cast_fp16, var_309_cast_fp16))[name = tensor("op_571_cast_fp16")]; tensor var_572_to_fp16 = const()[name = tensor("op_572_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_89_cast_fp16 = mul(x = var_571_cast_fp16, y = var_572_to_fp16)[name = tensor("aw_chunk_89_cast_fp16")]; tensor var_575_equation_0 = const()[name = tensor("op_575_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_575_cast_fp16 = einsum(equation = var_575_equation_0, values = (var_353_cast_fp16, var_310_cast_fp16))[name = tensor("op_575_cast_fp16")]; tensor var_576_to_fp16 = const()[name = tensor("op_576_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_91_cast_fp16 = mul(x = var_575_cast_fp16, y = var_576_to_fp16)[name = tensor("aw_chunk_91_cast_fp16")]; tensor var_579_equation_0 = const()[name = tensor("op_579_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_579_cast_fp16 = einsum(equation = var_579_equation_0, values = (var_353_cast_fp16, var_311_cast_fp16))[name = tensor("op_579_cast_fp16")]; tensor var_580_to_fp16 = const()[name = tensor("op_580_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_93_cast_fp16 = mul(x = var_579_cast_fp16, y = var_580_to_fp16)[name = tensor("aw_chunk_93_cast_fp16")]; tensor var_583_equation_0 = const()[name = tensor("op_583_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_583_cast_fp16 = einsum(equation = var_583_equation_0, values = (var_353_cast_fp16, var_312_cast_fp16))[name = tensor("op_583_cast_fp16")]; tensor var_584_to_fp16 = const()[name = tensor("op_584_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_95_cast_fp16 = mul(x = var_583_cast_fp16, y = var_584_to_fp16)[name = tensor("aw_chunk_95_cast_fp16")]; tensor var_587_equation_0 = const()[name = tensor("op_587_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_587_cast_fp16 = einsum(equation = var_587_equation_0, values = (var_357_cast_fp16, var_313_cast_fp16))[name = tensor("op_587_cast_fp16")]; tensor var_588_to_fp16 = const()[name = tensor("op_588_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_97_cast_fp16 = mul(x = var_587_cast_fp16, y = var_588_to_fp16)[name = tensor("aw_chunk_97_cast_fp16")]; tensor var_591_equation_0 = const()[name = tensor("op_591_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_591_cast_fp16 = einsum(equation = var_591_equation_0, values = (var_357_cast_fp16, var_314_cast_fp16))[name = tensor("op_591_cast_fp16")]; tensor var_592_to_fp16 = const()[name = tensor("op_592_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_99_cast_fp16 = mul(x = var_591_cast_fp16, y = var_592_to_fp16)[name = tensor("aw_chunk_99_cast_fp16")]; tensor var_595_equation_0 = const()[name = tensor("op_595_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_595_cast_fp16 = einsum(equation = var_595_equation_0, values = (var_357_cast_fp16, var_315_cast_fp16))[name = tensor("op_595_cast_fp16")]; tensor var_596_to_fp16 = const()[name = tensor("op_596_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_101_cast_fp16 = mul(x = var_595_cast_fp16, y = var_596_to_fp16)[name = tensor("aw_chunk_101_cast_fp16")]; tensor var_599_equation_0 = const()[name = tensor("op_599_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_599_cast_fp16 = einsum(equation = var_599_equation_0, values = (var_357_cast_fp16, var_316_cast_fp16))[name = tensor("op_599_cast_fp16")]; tensor var_600_to_fp16 = const()[name = tensor("op_600_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_103_cast_fp16 = mul(x = var_599_cast_fp16, y = var_600_to_fp16)[name = tensor("aw_chunk_103_cast_fp16")]; tensor var_603_equation_0 = const()[name = tensor("op_603_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_603_cast_fp16 = einsum(equation = var_603_equation_0, values = (var_357_cast_fp16, var_317_cast_fp16))[name = tensor("op_603_cast_fp16")]; tensor var_604_to_fp16 = const()[name = tensor("op_604_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_105_cast_fp16 = mul(x = var_603_cast_fp16, y = var_604_to_fp16)[name = tensor("aw_chunk_105_cast_fp16")]; tensor var_607_equation_0 = const()[name = tensor("op_607_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_607_cast_fp16 = einsum(equation = var_607_equation_0, values = (var_357_cast_fp16, var_318_cast_fp16))[name = tensor("op_607_cast_fp16")]; tensor var_608_to_fp16 = const()[name = tensor("op_608_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_107_cast_fp16 = mul(x = var_607_cast_fp16, y = var_608_to_fp16)[name = tensor("aw_chunk_107_cast_fp16")]; tensor var_611_equation_0 = const()[name = tensor("op_611_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_611_cast_fp16 = einsum(equation = var_611_equation_0, values = (var_357_cast_fp16, var_319_cast_fp16))[name = tensor("op_611_cast_fp16")]; tensor var_612_to_fp16 = const()[name = tensor("op_612_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_109_cast_fp16 = mul(x = var_611_cast_fp16, y = var_612_to_fp16)[name = tensor("aw_chunk_109_cast_fp16")]; tensor var_615_equation_0 = const()[name = tensor("op_615_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_615_cast_fp16 = einsum(equation = var_615_equation_0, values = (var_357_cast_fp16, var_320_cast_fp16))[name = tensor("op_615_cast_fp16")]; tensor var_616_to_fp16 = const()[name = tensor("op_616_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_111_cast_fp16 = mul(x = var_615_cast_fp16, y = var_616_to_fp16)[name = tensor("aw_chunk_111_cast_fp16")]; tensor var_619_equation_0 = const()[name = tensor("op_619_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_619_cast_fp16 = einsum(equation = var_619_equation_0, values = (var_361_cast_fp16, var_321_cast_fp16))[name = tensor("op_619_cast_fp16")]; tensor var_620_to_fp16 = const()[name = tensor("op_620_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_113_cast_fp16 = mul(x = var_619_cast_fp16, y = var_620_to_fp16)[name = tensor("aw_chunk_113_cast_fp16")]; tensor var_623_equation_0 = const()[name = tensor("op_623_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_623_cast_fp16 = einsum(equation = var_623_equation_0, values = (var_361_cast_fp16, var_322_cast_fp16))[name = tensor("op_623_cast_fp16")]; tensor var_624_to_fp16 = const()[name = tensor("op_624_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_115_cast_fp16 = mul(x = var_623_cast_fp16, y = var_624_to_fp16)[name = tensor("aw_chunk_115_cast_fp16")]; tensor var_627_equation_0 = const()[name = tensor("op_627_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_627_cast_fp16 = einsum(equation = var_627_equation_0, values = (var_361_cast_fp16, var_323_cast_fp16))[name = tensor("op_627_cast_fp16")]; tensor var_628_to_fp16 = const()[name = tensor("op_628_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_117_cast_fp16 = mul(x = var_627_cast_fp16, y = var_628_to_fp16)[name = tensor("aw_chunk_117_cast_fp16")]; tensor var_631_equation_0 = const()[name = tensor("op_631_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_631_cast_fp16 = einsum(equation = var_631_equation_0, values = (var_361_cast_fp16, var_324_cast_fp16))[name = tensor("op_631_cast_fp16")]; tensor var_632_to_fp16 = const()[name = tensor("op_632_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_119_cast_fp16 = mul(x = var_631_cast_fp16, y = var_632_to_fp16)[name = tensor("aw_chunk_119_cast_fp16")]; tensor var_635_equation_0 = const()[name = tensor("op_635_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_635_cast_fp16 = einsum(equation = var_635_equation_0, values = (var_361_cast_fp16, var_325_cast_fp16))[name = tensor("op_635_cast_fp16")]; tensor var_636_to_fp16 = const()[name = tensor("op_636_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_121_cast_fp16 = mul(x = var_635_cast_fp16, y = var_636_to_fp16)[name = tensor("aw_chunk_121_cast_fp16")]; tensor var_639_equation_0 = const()[name = tensor("op_639_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_639_cast_fp16 = einsum(equation = var_639_equation_0, values = (var_361_cast_fp16, var_326_cast_fp16))[name = tensor("op_639_cast_fp16")]; tensor var_640_to_fp16 = const()[name = tensor("op_640_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_123_cast_fp16 = mul(x = var_639_cast_fp16, y = var_640_to_fp16)[name = tensor("aw_chunk_123_cast_fp16")]; tensor var_643_equation_0 = const()[name = tensor("op_643_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_643_cast_fp16 = einsum(equation = var_643_equation_0, values = (var_361_cast_fp16, var_327_cast_fp16))[name = tensor("op_643_cast_fp16")]; tensor var_644_to_fp16 = const()[name = tensor("op_644_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_125_cast_fp16 = mul(x = var_643_cast_fp16, y = var_644_to_fp16)[name = tensor("aw_chunk_125_cast_fp16")]; tensor var_647_equation_0 = const()[name = tensor("op_647_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_647_cast_fp16 = einsum(equation = var_647_equation_0, values = (var_361_cast_fp16, var_328_cast_fp16))[name = tensor("op_647_cast_fp16")]; tensor var_648_to_fp16 = const()[name = tensor("op_648_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_127_cast_fp16 = mul(x = var_647_cast_fp16, y = var_648_to_fp16)[name = tensor("aw_chunk_127_cast_fp16")]; tensor var_650_cast_fp16 = softmax(axis = var_118, x = aw_chunk_1_cast_fp16)[name = tensor("op_650_cast_fp16")]; tensor var_651_cast_fp16 = softmax(axis = var_118, x = aw_chunk_3_cast_fp16)[name = tensor("op_651_cast_fp16")]; tensor var_652_cast_fp16 = softmax(axis = var_118, x = aw_chunk_5_cast_fp16)[name = tensor("op_652_cast_fp16")]; tensor var_653_cast_fp16 = softmax(axis = var_118, x = aw_chunk_7_cast_fp16)[name = tensor("op_653_cast_fp16")]; tensor var_654_cast_fp16 = softmax(axis = var_118, x = aw_chunk_9_cast_fp16)[name = tensor("op_654_cast_fp16")]; tensor var_655_cast_fp16 = softmax(axis = var_118, x = aw_chunk_11_cast_fp16)[name = tensor("op_655_cast_fp16")]; tensor var_656_cast_fp16 = softmax(axis = var_118, x = aw_chunk_13_cast_fp16)[name = tensor("op_656_cast_fp16")]; tensor var_657_cast_fp16 = softmax(axis = var_118, x = aw_chunk_15_cast_fp16)[name = tensor("op_657_cast_fp16")]; tensor var_658_cast_fp16 = softmax(axis = var_118, x = aw_chunk_17_cast_fp16)[name = tensor("op_658_cast_fp16")]; tensor var_659_cast_fp16 = softmax(axis = var_118, x = aw_chunk_19_cast_fp16)[name = tensor("op_659_cast_fp16")]; tensor var_660_cast_fp16 = softmax(axis = var_118, x = aw_chunk_21_cast_fp16)[name = tensor("op_660_cast_fp16")]; tensor var_661_cast_fp16 = softmax(axis = var_118, x = aw_chunk_23_cast_fp16)[name = tensor("op_661_cast_fp16")]; tensor var_662_cast_fp16 = softmax(axis = var_118, x = aw_chunk_25_cast_fp16)[name = tensor("op_662_cast_fp16")]; tensor var_663_cast_fp16 = softmax(axis = var_118, x = aw_chunk_27_cast_fp16)[name = tensor("op_663_cast_fp16")]; tensor var_664_cast_fp16 = softmax(axis = var_118, x = aw_chunk_29_cast_fp16)[name = tensor("op_664_cast_fp16")]; tensor var_665_cast_fp16 = softmax(axis = var_118, x = aw_chunk_31_cast_fp16)[name = tensor("op_665_cast_fp16")]; tensor var_666_cast_fp16 = softmax(axis = var_118, x = aw_chunk_33_cast_fp16)[name = tensor("op_666_cast_fp16")]; tensor var_667_cast_fp16 = softmax(axis = var_118, x = aw_chunk_35_cast_fp16)[name = tensor("op_667_cast_fp16")]; tensor var_668_cast_fp16 = softmax(axis = var_118, x = aw_chunk_37_cast_fp16)[name = tensor("op_668_cast_fp16")]; tensor var_669_cast_fp16 = softmax(axis = var_118, x = aw_chunk_39_cast_fp16)[name = tensor("op_669_cast_fp16")]; tensor var_670_cast_fp16 = softmax(axis = var_118, x = aw_chunk_41_cast_fp16)[name = tensor("op_670_cast_fp16")]; tensor var_671_cast_fp16 = softmax(axis = var_118, x = aw_chunk_43_cast_fp16)[name = tensor("op_671_cast_fp16")]; tensor var_672_cast_fp16 = softmax(axis = var_118, x = aw_chunk_45_cast_fp16)[name = tensor("op_672_cast_fp16")]; tensor var_673_cast_fp16 = softmax(axis = var_118, x = aw_chunk_47_cast_fp16)[name = tensor("op_673_cast_fp16")]; tensor var_674_cast_fp16 = softmax(axis = var_118, x = aw_chunk_49_cast_fp16)[name = tensor("op_674_cast_fp16")]; tensor var_675_cast_fp16 = softmax(axis = var_118, x = aw_chunk_51_cast_fp16)[name = tensor("op_675_cast_fp16")]; tensor var_676_cast_fp16 = softmax(axis = var_118, x = aw_chunk_53_cast_fp16)[name = tensor("op_676_cast_fp16")]; tensor var_677_cast_fp16 = softmax(axis = var_118, x = aw_chunk_55_cast_fp16)[name = tensor("op_677_cast_fp16")]; tensor var_678_cast_fp16 = softmax(axis = var_118, x = aw_chunk_57_cast_fp16)[name = tensor("op_678_cast_fp16")]; tensor var_679_cast_fp16 = softmax(axis = var_118, x = aw_chunk_59_cast_fp16)[name = tensor("op_679_cast_fp16")]; tensor var_680_cast_fp16 = softmax(axis = var_118, x = aw_chunk_61_cast_fp16)[name = tensor("op_680_cast_fp16")]; tensor var_681_cast_fp16 = softmax(axis = var_118, x = aw_chunk_63_cast_fp16)[name = tensor("op_681_cast_fp16")]; tensor var_682_cast_fp16 = softmax(axis = var_118, x = aw_chunk_65_cast_fp16)[name = tensor("op_682_cast_fp16")]; tensor var_683_cast_fp16 = softmax(axis = var_118, x = aw_chunk_67_cast_fp16)[name = tensor("op_683_cast_fp16")]; tensor var_684_cast_fp16 = softmax(axis = var_118, x = aw_chunk_69_cast_fp16)[name = tensor("op_684_cast_fp16")]; tensor var_685_cast_fp16 = softmax(axis = var_118, x = aw_chunk_71_cast_fp16)[name = tensor("op_685_cast_fp16")]; tensor var_686_cast_fp16 = softmax(axis = var_118, x = aw_chunk_73_cast_fp16)[name = tensor("op_686_cast_fp16")]; tensor var_687_cast_fp16 = softmax(axis = var_118, x = aw_chunk_75_cast_fp16)[name = tensor("op_687_cast_fp16")]; tensor var_688_cast_fp16 = softmax(axis = var_118, x = aw_chunk_77_cast_fp16)[name = tensor("op_688_cast_fp16")]; tensor var_689_cast_fp16 = softmax(axis = var_118, x = aw_chunk_79_cast_fp16)[name = tensor("op_689_cast_fp16")]; tensor var_690_cast_fp16 = softmax(axis = var_118, x = aw_chunk_81_cast_fp16)[name = tensor("op_690_cast_fp16")]; tensor var_691_cast_fp16 = softmax(axis = var_118, x = aw_chunk_83_cast_fp16)[name = tensor("op_691_cast_fp16")]; tensor var_692_cast_fp16 = softmax(axis = var_118, x = aw_chunk_85_cast_fp16)[name = tensor("op_692_cast_fp16")]; tensor var_693_cast_fp16 = softmax(axis = var_118, x = aw_chunk_87_cast_fp16)[name = tensor("op_693_cast_fp16")]; tensor var_694_cast_fp16 = softmax(axis = var_118, x = aw_chunk_89_cast_fp16)[name = tensor("op_694_cast_fp16")]; tensor var_695_cast_fp16 = softmax(axis = var_118, x = aw_chunk_91_cast_fp16)[name = tensor("op_695_cast_fp16")]; tensor var_696_cast_fp16 = softmax(axis = var_118, x = aw_chunk_93_cast_fp16)[name = tensor("op_696_cast_fp16")]; tensor var_697_cast_fp16 = softmax(axis = var_118, x = aw_chunk_95_cast_fp16)[name = tensor("op_697_cast_fp16")]; tensor var_698_cast_fp16 = softmax(axis = var_118, x = aw_chunk_97_cast_fp16)[name = tensor("op_698_cast_fp16")]; tensor var_699_cast_fp16 = softmax(axis = var_118, x = aw_chunk_99_cast_fp16)[name = tensor("op_699_cast_fp16")]; tensor var_700_cast_fp16 = softmax(axis = var_118, x = aw_chunk_101_cast_fp16)[name = tensor("op_700_cast_fp16")]; tensor var_701_cast_fp16 = softmax(axis = var_118, x = aw_chunk_103_cast_fp16)[name = tensor("op_701_cast_fp16")]; tensor var_702_cast_fp16 = softmax(axis = var_118, x = aw_chunk_105_cast_fp16)[name = tensor("op_702_cast_fp16")]; tensor var_703_cast_fp16 = softmax(axis = var_118, x = aw_chunk_107_cast_fp16)[name = tensor("op_703_cast_fp16")]; tensor var_704_cast_fp16 = softmax(axis = var_118, x = aw_chunk_109_cast_fp16)[name = tensor("op_704_cast_fp16")]; tensor var_705_cast_fp16 = softmax(axis = var_118, x = aw_chunk_111_cast_fp16)[name = tensor("op_705_cast_fp16")]; tensor var_706_cast_fp16 = softmax(axis = var_118, x = aw_chunk_113_cast_fp16)[name = tensor("op_706_cast_fp16")]; tensor var_707_cast_fp16 = softmax(axis = var_118, x = aw_chunk_115_cast_fp16)[name = tensor("op_707_cast_fp16")]; tensor var_708_cast_fp16 = softmax(axis = var_118, x = aw_chunk_117_cast_fp16)[name = tensor("op_708_cast_fp16")]; tensor var_709_cast_fp16 = softmax(axis = var_118, x = aw_chunk_119_cast_fp16)[name = tensor("op_709_cast_fp16")]; tensor var_710_cast_fp16 = softmax(axis = var_118, x = aw_chunk_121_cast_fp16)[name = tensor("op_710_cast_fp16")]; tensor var_711_cast_fp16 = softmax(axis = var_118, x = aw_chunk_123_cast_fp16)[name = tensor("op_711_cast_fp16")]; tensor var_712_cast_fp16 = softmax(axis = var_118, x = aw_chunk_125_cast_fp16)[name = tensor("op_712_cast_fp16")]; tensor var_713_cast_fp16 = softmax(axis = var_118, x = aw_chunk_127_cast_fp16)[name = tensor("op_713_cast_fp16")]; tensor var_715_equation_0 = const()[name = tensor("op_715_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_715_cast_fp16 = einsum(equation = var_715_equation_0, values = (var_363_cast_fp16, var_650_cast_fp16))[name = tensor("op_715_cast_fp16")]; tensor var_717_equation_0 = const()[name = tensor("op_717_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_717_cast_fp16 = einsum(equation = var_717_equation_0, values = (var_363_cast_fp16, var_651_cast_fp16))[name = tensor("op_717_cast_fp16")]; tensor var_719_equation_0 = const()[name = tensor("op_719_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_719_cast_fp16 = einsum(equation = var_719_equation_0, values = (var_363_cast_fp16, var_652_cast_fp16))[name = tensor("op_719_cast_fp16")]; tensor var_721_equation_0 = const()[name = tensor("op_721_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_721_cast_fp16 = einsum(equation = var_721_equation_0, values = (var_363_cast_fp16, var_653_cast_fp16))[name = tensor("op_721_cast_fp16")]; tensor var_723_equation_0 = const()[name = tensor("op_723_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_723_cast_fp16 = einsum(equation = var_723_equation_0, values = (var_363_cast_fp16, var_654_cast_fp16))[name = tensor("op_723_cast_fp16")]; tensor var_725_equation_0 = const()[name = tensor("op_725_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_725_cast_fp16 = einsum(equation = var_725_equation_0, values = (var_363_cast_fp16, var_655_cast_fp16))[name = tensor("op_725_cast_fp16")]; tensor var_727_equation_0 = const()[name = tensor("op_727_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_727_cast_fp16 = einsum(equation = var_727_equation_0, values = (var_363_cast_fp16, var_656_cast_fp16))[name = tensor("op_727_cast_fp16")]; tensor var_729_equation_0 = const()[name = tensor("op_729_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_729_cast_fp16 = einsum(equation = var_729_equation_0, values = (var_363_cast_fp16, var_657_cast_fp16))[name = tensor("op_729_cast_fp16")]; tensor var_731_equation_0 = const()[name = tensor("op_731_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_731_cast_fp16 = einsum(equation = var_731_equation_0, values = (var_367_cast_fp16, var_658_cast_fp16))[name = tensor("op_731_cast_fp16")]; tensor var_733_equation_0 = const()[name = tensor("op_733_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_733_cast_fp16 = einsum(equation = var_733_equation_0, values = (var_367_cast_fp16, var_659_cast_fp16))[name = tensor("op_733_cast_fp16")]; tensor var_735_equation_0 = const()[name = tensor("op_735_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_735_cast_fp16 = einsum(equation = var_735_equation_0, values = (var_367_cast_fp16, var_660_cast_fp16))[name = tensor("op_735_cast_fp16")]; tensor var_737_equation_0 = const()[name = tensor("op_737_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_737_cast_fp16 = einsum(equation = var_737_equation_0, values = (var_367_cast_fp16, var_661_cast_fp16))[name = tensor("op_737_cast_fp16")]; tensor var_739_equation_0 = const()[name = tensor("op_739_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_739_cast_fp16 = einsum(equation = var_739_equation_0, values = (var_367_cast_fp16, var_662_cast_fp16))[name = tensor("op_739_cast_fp16")]; tensor var_741_equation_0 = const()[name = tensor("op_741_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_741_cast_fp16 = einsum(equation = var_741_equation_0, values = (var_367_cast_fp16, var_663_cast_fp16))[name = tensor("op_741_cast_fp16")]; tensor var_743_equation_0 = const()[name = tensor("op_743_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_743_cast_fp16 = einsum(equation = var_743_equation_0, values = (var_367_cast_fp16, var_664_cast_fp16))[name = tensor("op_743_cast_fp16")]; tensor var_745_equation_0 = const()[name = tensor("op_745_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_745_cast_fp16 = einsum(equation = var_745_equation_0, values = (var_367_cast_fp16, var_665_cast_fp16))[name = tensor("op_745_cast_fp16")]; tensor var_747_equation_0 = const()[name = tensor("op_747_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_747_cast_fp16 = einsum(equation = var_747_equation_0, values = (var_371_cast_fp16, var_666_cast_fp16))[name = tensor("op_747_cast_fp16")]; tensor var_749_equation_0 = const()[name = tensor("op_749_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_749_cast_fp16 = einsum(equation = var_749_equation_0, values = (var_371_cast_fp16, var_667_cast_fp16))[name = tensor("op_749_cast_fp16")]; tensor var_751_equation_0 = const()[name = tensor("op_751_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_751_cast_fp16 = einsum(equation = var_751_equation_0, values = (var_371_cast_fp16, var_668_cast_fp16))[name = tensor("op_751_cast_fp16")]; tensor var_753_equation_0 = const()[name = tensor("op_753_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_753_cast_fp16 = einsum(equation = var_753_equation_0, values = (var_371_cast_fp16, var_669_cast_fp16))[name = tensor("op_753_cast_fp16")]; tensor var_755_equation_0 = const()[name = tensor("op_755_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_755_cast_fp16 = einsum(equation = var_755_equation_0, values = (var_371_cast_fp16, var_670_cast_fp16))[name = tensor("op_755_cast_fp16")]; tensor var_757_equation_0 = const()[name = tensor("op_757_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_757_cast_fp16 = einsum(equation = var_757_equation_0, values = (var_371_cast_fp16, var_671_cast_fp16))[name = tensor("op_757_cast_fp16")]; tensor var_759_equation_0 = const()[name = tensor("op_759_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_759_cast_fp16 = einsum(equation = var_759_equation_0, values = (var_371_cast_fp16, var_672_cast_fp16))[name = tensor("op_759_cast_fp16")]; tensor var_761_equation_0 = const()[name = tensor("op_761_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_761_cast_fp16 = einsum(equation = var_761_equation_0, values = (var_371_cast_fp16, var_673_cast_fp16))[name = tensor("op_761_cast_fp16")]; tensor var_763_equation_0 = const()[name = tensor("op_763_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_763_cast_fp16 = einsum(equation = var_763_equation_0, values = (var_375_cast_fp16, var_674_cast_fp16))[name = tensor("op_763_cast_fp16")]; tensor var_765_equation_0 = const()[name = tensor("op_765_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_765_cast_fp16 = einsum(equation = var_765_equation_0, values = (var_375_cast_fp16, var_675_cast_fp16))[name = tensor("op_765_cast_fp16")]; tensor var_767_equation_0 = const()[name = tensor("op_767_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_767_cast_fp16 = einsum(equation = var_767_equation_0, values = (var_375_cast_fp16, var_676_cast_fp16))[name = tensor("op_767_cast_fp16")]; tensor var_769_equation_0 = const()[name = tensor("op_769_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_769_cast_fp16 = einsum(equation = var_769_equation_0, values = (var_375_cast_fp16, var_677_cast_fp16))[name = tensor("op_769_cast_fp16")]; tensor var_771_equation_0 = const()[name = tensor("op_771_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_771_cast_fp16 = einsum(equation = var_771_equation_0, values = (var_375_cast_fp16, var_678_cast_fp16))[name = tensor("op_771_cast_fp16")]; tensor var_773_equation_0 = const()[name = tensor("op_773_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_773_cast_fp16 = einsum(equation = var_773_equation_0, values = (var_375_cast_fp16, var_679_cast_fp16))[name = tensor("op_773_cast_fp16")]; tensor var_775_equation_0 = const()[name = tensor("op_775_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_775_cast_fp16 = einsum(equation = var_775_equation_0, values = (var_375_cast_fp16, var_680_cast_fp16))[name = tensor("op_775_cast_fp16")]; tensor var_777_equation_0 = const()[name = tensor("op_777_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_777_cast_fp16 = einsum(equation = var_777_equation_0, values = (var_375_cast_fp16, var_681_cast_fp16))[name = tensor("op_777_cast_fp16")]; tensor var_779_equation_0 = const()[name = tensor("op_779_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_779_cast_fp16 = einsum(equation = var_779_equation_0, values = (var_379_cast_fp16, var_682_cast_fp16))[name = tensor("op_779_cast_fp16")]; tensor var_781_equation_0 = const()[name = tensor("op_781_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_781_cast_fp16 = einsum(equation = var_781_equation_0, values = (var_379_cast_fp16, var_683_cast_fp16))[name = tensor("op_781_cast_fp16")]; tensor var_783_equation_0 = const()[name = tensor("op_783_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_783_cast_fp16 = einsum(equation = var_783_equation_0, values = (var_379_cast_fp16, var_684_cast_fp16))[name = tensor("op_783_cast_fp16")]; tensor var_785_equation_0 = const()[name = tensor("op_785_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_785_cast_fp16 = einsum(equation = var_785_equation_0, values = (var_379_cast_fp16, var_685_cast_fp16))[name = tensor("op_785_cast_fp16")]; tensor var_787_equation_0 = const()[name = tensor("op_787_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_787_cast_fp16 = einsum(equation = var_787_equation_0, values = (var_379_cast_fp16, var_686_cast_fp16))[name = tensor("op_787_cast_fp16")]; tensor var_789_equation_0 = const()[name = tensor("op_789_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_789_cast_fp16 = einsum(equation = var_789_equation_0, values = (var_379_cast_fp16, var_687_cast_fp16))[name = tensor("op_789_cast_fp16")]; tensor var_791_equation_0 = const()[name = tensor("op_791_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_791_cast_fp16 = einsum(equation = var_791_equation_0, values = (var_379_cast_fp16, var_688_cast_fp16))[name = tensor("op_791_cast_fp16")]; tensor var_793_equation_0 = const()[name = tensor("op_793_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_793_cast_fp16 = einsum(equation = var_793_equation_0, values = (var_379_cast_fp16, var_689_cast_fp16))[name = tensor("op_793_cast_fp16")]; tensor var_795_equation_0 = const()[name = tensor("op_795_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_795_cast_fp16 = einsum(equation = var_795_equation_0, values = (var_383_cast_fp16, var_690_cast_fp16))[name = tensor("op_795_cast_fp16")]; tensor var_797_equation_0 = const()[name = tensor("op_797_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_797_cast_fp16 = einsum(equation = var_797_equation_0, values = (var_383_cast_fp16, var_691_cast_fp16))[name = tensor("op_797_cast_fp16")]; tensor var_799_equation_0 = const()[name = tensor("op_799_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_799_cast_fp16 = einsum(equation = var_799_equation_0, values = (var_383_cast_fp16, var_692_cast_fp16))[name = tensor("op_799_cast_fp16")]; tensor var_801_equation_0 = const()[name = tensor("op_801_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_801_cast_fp16 = einsum(equation = var_801_equation_0, values = (var_383_cast_fp16, var_693_cast_fp16))[name = tensor("op_801_cast_fp16")]; tensor var_803_equation_0 = const()[name = tensor("op_803_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_803_cast_fp16 = einsum(equation = var_803_equation_0, values = (var_383_cast_fp16, var_694_cast_fp16))[name = tensor("op_803_cast_fp16")]; tensor var_805_equation_0 = const()[name = tensor("op_805_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_805_cast_fp16 = einsum(equation = var_805_equation_0, values = (var_383_cast_fp16, var_695_cast_fp16))[name = tensor("op_805_cast_fp16")]; tensor var_807_equation_0 = const()[name = tensor("op_807_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_807_cast_fp16 = einsum(equation = var_807_equation_0, values = (var_383_cast_fp16, var_696_cast_fp16))[name = tensor("op_807_cast_fp16")]; tensor var_809_equation_0 = const()[name = tensor("op_809_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_809_cast_fp16 = einsum(equation = var_809_equation_0, values = (var_383_cast_fp16, var_697_cast_fp16))[name = tensor("op_809_cast_fp16")]; tensor var_811_equation_0 = const()[name = tensor("op_811_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_811_cast_fp16 = einsum(equation = var_811_equation_0, values = (var_387_cast_fp16, var_698_cast_fp16))[name = tensor("op_811_cast_fp16")]; tensor var_813_equation_0 = const()[name = tensor("op_813_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_813_cast_fp16 = einsum(equation = var_813_equation_0, values = (var_387_cast_fp16, var_699_cast_fp16))[name = tensor("op_813_cast_fp16")]; tensor var_815_equation_0 = const()[name = tensor("op_815_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_815_cast_fp16 = einsum(equation = var_815_equation_0, values = (var_387_cast_fp16, var_700_cast_fp16))[name = tensor("op_815_cast_fp16")]; tensor var_817_equation_0 = const()[name = tensor("op_817_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_817_cast_fp16 = einsum(equation = var_817_equation_0, values = (var_387_cast_fp16, var_701_cast_fp16))[name = tensor("op_817_cast_fp16")]; tensor var_819_equation_0 = const()[name = tensor("op_819_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_819_cast_fp16 = einsum(equation = var_819_equation_0, values = (var_387_cast_fp16, var_702_cast_fp16))[name = tensor("op_819_cast_fp16")]; tensor var_821_equation_0 = const()[name = tensor("op_821_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_821_cast_fp16 = einsum(equation = var_821_equation_0, values = (var_387_cast_fp16, var_703_cast_fp16))[name = tensor("op_821_cast_fp16")]; tensor var_823_equation_0 = const()[name = tensor("op_823_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_823_cast_fp16 = einsum(equation = var_823_equation_0, values = (var_387_cast_fp16, var_704_cast_fp16))[name = tensor("op_823_cast_fp16")]; tensor var_825_equation_0 = const()[name = tensor("op_825_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_825_cast_fp16 = einsum(equation = var_825_equation_0, values = (var_387_cast_fp16, var_705_cast_fp16))[name = tensor("op_825_cast_fp16")]; tensor var_827_equation_0 = const()[name = tensor("op_827_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_827_cast_fp16 = einsum(equation = var_827_equation_0, values = (var_391_cast_fp16, var_706_cast_fp16))[name = tensor("op_827_cast_fp16")]; tensor var_829_equation_0 = const()[name = tensor("op_829_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_829_cast_fp16 = einsum(equation = var_829_equation_0, values = (var_391_cast_fp16, var_707_cast_fp16))[name = tensor("op_829_cast_fp16")]; tensor var_831_equation_0 = const()[name = tensor("op_831_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_831_cast_fp16 = einsum(equation = var_831_equation_0, values = (var_391_cast_fp16, var_708_cast_fp16))[name = tensor("op_831_cast_fp16")]; tensor var_833_equation_0 = const()[name = tensor("op_833_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_833_cast_fp16 = einsum(equation = var_833_equation_0, values = (var_391_cast_fp16, var_709_cast_fp16))[name = tensor("op_833_cast_fp16")]; tensor var_835_equation_0 = const()[name = tensor("op_835_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_835_cast_fp16 = einsum(equation = var_835_equation_0, values = (var_391_cast_fp16, var_710_cast_fp16))[name = tensor("op_835_cast_fp16")]; tensor var_837_equation_0 = const()[name = tensor("op_837_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_837_cast_fp16 = einsum(equation = var_837_equation_0, values = (var_391_cast_fp16, var_711_cast_fp16))[name = tensor("op_837_cast_fp16")]; tensor var_839_equation_0 = const()[name = tensor("op_839_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_839_cast_fp16 = einsum(equation = var_839_equation_0, values = (var_391_cast_fp16, var_712_cast_fp16))[name = tensor("op_839_cast_fp16")]; tensor var_841_equation_0 = const()[name = tensor("op_841_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_841_cast_fp16 = einsum(equation = var_841_equation_0, values = (var_391_cast_fp16, var_713_cast_fp16))[name = tensor("op_841_cast_fp16")]; tensor var_843_interleave_0 = const()[name = tensor("op_843_interleave_0"), val = tensor(false)]; tensor var_843_cast_fp16 = concat(axis = var_90, interleave = var_843_interleave_0, values = (var_715_cast_fp16, var_717_cast_fp16, var_719_cast_fp16, var_721_cast_fp16, var_723_cast_fp16, var_725_cast_fp16, var_727_cast_fp16, var_729_cast_fp16))[name = tensor("op_843_cast_fp16")]; tensor var_845_interleave_0 = const()[name = tensor("op_845_interleave_0"), val = tensor(false)]; tensor var_845_cast_fp16 = concat(axis = var_90, interleave = var_845_interleave_0, values = (var_731_cast_fp16, var_733_cast_fp16, var_735_cast_fp16, var_737_cast_fp16, var_739_cast_fp16, var_741_cast_fp16, var_743_cast_fp16, var_745_cast_fp16))[name = tensor("op_845_cast_fp16")]; tensor var_847_interleave_0 = const()[name = tensor("op_847_interleave_0"), val = tensor(false)]; tensor var_847_cast_fp16 = concat(axis = var_90, interleave = var_847_interleave_0, values = (var_747_cast_fp16, var_749_cast_fp16, var_751_cast_fp16, var_753_cast_fp16, var_755_cast_fp16, var_757_cast_fp16, var_759_cast_fp16, var_761_cast_fp16))[name = tensor("op_847_cast_fp16")]; tensor var_849_interleave_0 = const()[name = tensor("op_849_interleave_0"), val = tensor(false)]; tensor var_849_cast_fp16 = concat(axis = var_90, interleave = var_849_interleave_0, values = (var_763_cast_fp16, var_765_cast_fp16, var_767_cast_fp16, var_769_cast_fp16, var_771_cast_fp16, var_773_cast_fp16, var_775_cast_fp16, var_777_cast_fp16))[name = tensor("op_849_cast_fp16")]; tensor var_851_interleave_0 = const()[name = tensor("op_851_interleave_0"), val = tensor(false)]; tensor var_851_cast_fp16 = concat(axis = var_90, interleave = var_851_interleave_0, values = (var_779_cast_fp16, var_781_cast_fp16, var_783_cast_fp16, var_785_cast_fp16, var_787_cast_fp16, var_789_cast_fp16, var_791_cast_fp16, var_793_cast_fp16))[name = tensor("op_851_cast_fp16")]; tensor var_853_interleave_0 = const()[name = tensor("op_853_interleave_0"), val = tensor(false)]; tensor var_853_cast_fp16 = concat(axis = var_90, interleave = var_853_interleave_0, values = (var_795_cast_fp16, var_797_cast_fp16, var_799_cast_fp16, var_801_cast_fp16, var_803_cast_fp16, var_805_cast_fp16, var_807_cast_fp16, var_809_cast_fp16))[name = tensor("op_853_cast_fp16")]; tensor var_855_interleave_0 = const()[name = tensor("op_855_interleave_0"), val = tensor(false)]; tensor var_855_cast_fp16 = concat(axis = var_90, interleave = var_855_interleave_0, values = (var_811_cast_fp16, var_813_cast_fp16, var_815_cast_fp16, var_817_cast_fp16, var_819_cast_fp16, var_821_cast_fp16, var_823_cast_fp16, var_825_cast_fp16))[name = tensor("op_855_cast_fp16")]; tensor var_857_interleave_0 = const()[name = tensor("op_857_interleave_0"), val = tensor(false)]; tensor var_857_cast_fp16 = concat(axis = var_90, interleave = var_857_interleave_0, values = (var_827_cast_fp16, var_829_cast_fp16, var_831_cast_fp16, var_833_cast_fp16, var_835_cast_fp16, var_837_cast_fp16, var_839_cast_fp16, var_841_cast_fp16))[name = tensor("op_857_cast_fp16")]; tensor input_25_interleave_0 = const()[name = tensor("input_25_interleave_0"), val = tensor(false)]; tensor input_25_cast_fp16 = concat(axis = var_118, interleave = input_25_interleave_0, values = (var_843_cast_fp16, var_845_cast_fp16, var_847_cast_fp16, var_849_cast_fp16, var_851_cast_fp16, var_853_cast_fp16, var_855_cast_fp16, var_857_cast_fp16))[name = tensor("input_25_cast_fp16")]; tensor var_863 = const()[name = tensor("op_863"), val = tensor([1, 1])]; tensor var_865 = const()[name = tensor("op_865"), val = tensor([1, 1])]; tensor var_867_pad_type_0 = const()[name = tensor("op_867_pad_type_0"), val = tensor("custom")]; tensor var_867_pad_0 = const()[name = tensor("op_867_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_867_cast_fp16 = conv(bias = down_blocks_0_attentions_0_transformer_blocks_0_attn1_to_out_0_bias_to_fp16, dilations = var_865, groups = var_118, pad = var_867_pad_0, pad_type = var_867_pad_type_0, strides = var_863, weight = down_blocks_0_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16_palettized, x = input_25_cast_fp16)[name = tensor("op_867_cast_fp16")]; tensor inputs_3_cast_fp16 = add(x = var_867_cast_fp16, y = inputs_1_cast_fp16)[name = tensor("inputs_3_cast_fp16")]; tensor var_871 = const()[name = tensor("op_871"), val = tensor([1])]; tensor channels_mean_3_cast_fp16 = reduce_mean(axes = var_871, keep_dims = var_113, x = inputs_3_cast_fp16)[name = tensor("channels_mean_3_cast_fp16")]; tensor zero_mean_3_cast_fp16 = sub(x = inputs_3_cast_fp16, y = channels_mean_3_cast_fp16)[name = tensor("zero_mean_3_cast_fp16")]; tensor zero_mean_sq_3_cast_fp16 = mul(x = zero_mean_3_cast_fp16, y = zero_mean_3_cast_fp16)[name = tensor("zero_mean_sq_3_cast_fp16")]; tensor var_875 = const()[name = tensor("op_875"), val = tensor([1])]; tensor var_876_cast_fp16 = reduce_mean(axes = var_875, keep_dims = var_113, x = zero_mean_sq_3_cast_fp16)[name = tensor("op_876_cast_fp16")]; tensor var_877_to_fp16 = const()[name = tensor("op_877_to_fp16"), val = tensor(0x1.5p-17)]; tensor var_878_cast_fp16 = add(x = var_876_cast_fp16, y = var_877_to_fp16)[name = tensor("op_878_cast_fp16")]; tensor denom_3_epsilon_0_to_fp16 = const()[name = tensor("denom_3_epsilon_0_to_fp16"), val = tensor(0x1p-24)]; tensor denom_3_cast_fp16 = rsqrt(epsilon = denom_3_epsilon_0_to_fp16, x = var_878_cast_fp16)[name = tensor("denom_3_cast_fp16")]; tensor out_3_cast_fp16 = mul(x = zero_mean_3_cast_fp16, y = denom_3_cast_fp16)[name = tensor("out_3_cast_fp16")]; tensor var_882_to_fp16 = const()[name = tensor("op_882_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3638016)))]; tensor var_883_cast_fp16 = add(x = out_3_cast_fp16, y = var_882_to_fp16)[name = tensor("op_883_cast_fp16")]; tensor var_885_to_fp16 = const()[name = tensor("op_885_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3638720)))]; tensor hidden_states_11_cast_fp16 = mul(x = var_883_cast_fp16, y = var_885_to_fp16)[name = tensor("hidden_states_11_cast_fp16")]; tensor var_892 = const()[name = tensor("op_892"), val = tensor([1, 1])]; tensor var_894 = const()[name = tensor("op_894"), 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_fp16 = conv(dilations = var_894, groups = var_118, pad = q_3_pad_0, pad_type = q_3_pad_type_0, strides = var_892, weight = down_blocks_0_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16_palettized, x = hidden_states_11_cast_fp16)[name = tensor("q_3_cast_fp16")]; tensor var_898 = const()[name = tensor("op_898"), val = tensor([1, 1])]; tensor var_900 = const()[name = tensor("op_900"), 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(3900864))), name = tensor("down_blocks_0_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized"), shape = tensor([320, 768, 1, 1])]; tensor k_5_cast_fp16 = conv(dilations = var_900, groups = var_118, pad = k_5_pad_0, pad_type = k_5_pad_type_0, strides = var_898, 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_fp16")]; tensor var_904 = const()[name = tensor("op_904"), val = tensor([1, 1])]; tensor var_906 = const()[name = tensor("op_906"), 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(3901056))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4085440))), name = tensor("down_blocks_0_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized"), shape = tensor([320, 768, 1, 1])]; tensor v_3_cast_fp16 = conv(dilations = var_906, groups = var_118, pad = v_3_pad_0, pad_type = v_3_pad_type_0, strides = var_904, 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_fp16")]; tensor var_910_begin_0 = const()[name = tensor("op_910_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_910_end_0 = const()[name = tensor("op_910_end_0"), val = tensor([2, 40, 1, 4096])]; tensor var_910_end_mask_0 = const()[name = tensor("op_910_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_910_cast_fp16 = slice_by_index(begin = var_910_begin_0, end = var_910_end_0, end_mask = var_910_end_mask_0, x = q_3_cast_fp16)[name = tensor("op_910_cast_fp16")]; tensor var_914_begin_0 = const()[name = tensor("op_914_begin_0"), val = tensor([0, 40, 0, 0])]; tensor var_914_end_0 = const()[name = tensor("op_914_end_0"), val = tensor([2, 80, 1, 4096])]; tensor var_914_end_mask_0 = const()[name = tensor("op_914_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_914_cast_fp16 = slice_by_index(begin = var_914_begin_0, end = var_914_end_0, end_mask = var_914_end_mask_0, x = q_3_cast_fp16)[name = tensor("op_914_cast_fp16")]; tensor var_918_begin_0 = const()[name = tensor("op_918_begin_0"), val = tensor([0, 80, 0, 0])]; tensor var_918_end_0 = const()[name = tensor("op_918_end_0"), val = tensor([2, 120, 1, 4096])]; tensor var_918_end_mask_0 = const()[name = tensor("op_918_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_918_cast_fp16 = slice_by_index(begin = var_918_begin_0, end = var_918_end_0, end_mask = var_918_end_mask_0, x = q_3_cast_fp16)[name = tensor("op_918_cast_fp16")]; tensor var_922_begin_0 = const()[name = tensor("op_922_begin_0"), val = tensor([0, 120, 0, 0])]; tensor var_922_end_0 = const()[name = tensor("op_922_end_0"), val = tensor([2, 160, 1, 4096])]; tensor var_922_end_mask_0 = const()[name = tensor("op_922_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_922_cast_fp16 = slice_by_index(begin = var_922_begin_0, end = var_922_end_0, end_mask = var_922_end_mask_0, x = q_3_cast_fp16)[name = tensor("op_922_cast_fp16")]; tensor var_926_begin_0 = const()[name = tensor("op_926_begin_0"), val = tensor([0, 160, 0, 0])]; tensor var_926_end_0 = const()[name = tensor("op_926_end_0"), val = tensor([2, 200, 1, 4096])]; tensor var_926_end_mask_0 = const()[name = tensor("op_926_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_926_cast_fp16 = slice_by_index(begin = var_926_begin_0, end = var_926_end_0, end_mask = var_926_end_mask_0, x = q_3_cast_fp16)[name = tensor("op_926_cast_fp16")]; tensor var_930_begin_0 = const()[name = tensor("op_930_begin_0"), val = tensor([0, 200, 0, 0])]; tensor var_930_end_0 = const()[name = tensor("op_930_end_0"), val = tensor([2, 240, 1, 4096])]; tensor var_930_end_mask_0 = const()[name = tensor("op_930_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_930_cast_fp16 = slice_by_index(begin = var_930_begin_0, end = var_930_end_0, end_mask = var_930_end_mask_0, x = q_3_cast_fp16)[name = tensor("op_930_cast_fp16")]; tensor var_934_begin_0 = const()[name = tensor("op_934_begin_0"), val = tensor([0, 240, 0, 0])]; tensor var_934_end_0 = const()[name = tensor("op_934_end_0"), val = tensor([2, 280, 1, 4096])]; tensor var_934_end_mask_0 = const()[name = tensor("op_934_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_934_cast_fp16 = slice_by_index(begin = var_934_begin_0, end = var_934_end_0, end_mask = var_934_end_mask_0, x = q_3_cast_fp16)[name = tensor("op_934_cast_fp16")]; tensor var_938_begin_0 = const()[name = tensor("op_938_begin_0"), val = tensor([0, 280, 0, 0])]; tensor var_938_end_0 = const()[name = tensor("op_938_end_0"), val = tensor([2, 320, 1, 4096])]; tensor var_938_end_mask_0 = const()[name = tensor("op_938_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_938_cast_fp16 = slice_by_index(begin = var_938_begin_0, end = var_938_end_0, end_mask = var_938_end_mask_0, x = q_3_cast_fp16)[name = tensor("op_938_cast_fp16")]; tensor var_941_begin_0 = const()[name = tensor("op_941_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_941_end_0 = const()[name = tensor("op_941_end_0"), val = tensor([2, 40, 1, 512])]; tensor var_941_end_mask_0 = const()[name = tensor("op_941_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_941_cast_fp16 = slice_by_index(begin = var_941_begin_0, end = var_941_end_0, end_mask = var_941_end_mask_0, x = var_910_cast_fp16)[name = tensor("op_941_cast_fp16")]; tensor var_942_begin_0 = const()[name = tensor("op_942_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_942_end_0 = const()[name = tensor("op_942_end_0"), val = tensor([2, 40, 1, 1024])]; tensor var_942_end_mask_0 = const()[name = tensor("op_942_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_942_cast_fp16 = slice_by_index(begin = var_942_begin_0, end = var_942_end_0, end_mask = var_942_end_mask_0, x = var_910_cast_fp16)[name = tensor("op_942_cast_fp16")]; tensor var_943_begin_0 = const()[name = tensor("op_943_begin_0"), val = tensor([0, 0, 0, 1024])]; tensor var_943_end_0 = const()[name = tensor("op_943_end_0"), val = tensor([2, 40, 1, 1536])]; tensor var_943_end_mask_0 = const()[name = tensor("op_943_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_943_cast_fp16 = slice_by_index(begin = var_943_begin_0, end = var_943_end_0, end_mask = var_943_end_mask_0, x = var_910_cast_fp16)[name = tensor("op_943_cast_fp16")]; tensor var_944_begin_0 = const()[name = tensor("op_944_begin_0"), val = tensor([0, 0, 0, 1536])]; tensor var_944_end_0 = const()[name = tensor("op_944_end_0"), val = tensor([2, 40, 1, 2048])]; tensor var_944_end_mask_0 = const()[name = tensor("op_944_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_944_cast_fp16 = slice_by_index(begin = var_944_begin_0, end = var_944_end_0, end_mask = var_944_end_mask_0, x = var_910_cast_fp16)[name = tensor("op_944_cast_fp16")]; tensor var_945_begin_0 = const()[name = tensor("op_945_begin_0"), val = tensor([0, 0, 0, 2048])]; tensor var_945_end_0 = const()[name = tensor("op_945_end_0"), val = tensor([2, 40, 1, 2560])]; tensor var_945_end_mask_0 = const()[name = tensor("op_945_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_945_cast_fp16 = slice_by_index(begin = var_945_begin_0, end = var_945_end_0, end_mask = var_945_end_mask_0, x = var_910_cast_fp16)[name = tensor("op_945_cast_fp16")]; tensor var_946_begin_0 = const()[name = tensor("op_946_begin_0"), val = tensor([0, 0, 0, 2560])]; tensor var_946_end_0 = const()[name = tensor("op_946_end_0"), val = tensor([2, 40, 1, 3072])]; tensor var_946_end_mask_0 = const()[name = tensor("op_946_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_946_cast_fp16 = slice_by_index(begin = var_946_begin_0, end = var_946_end_0, end_mask = var_946_end_mask_0, x = var_910_cast_fp16)[name = tensor("op_946_cast_fp16")]; tensor var_947_begin_0 = const()[name = tensor("op_947_begin_0"), val = tensor([0, 0, 0, 3072])]; tensor var_947_end_0 = const()[name = tensor("op_947_end_0"), val = tensor([2, 40, 1, 3584])]; tensor var_947_end_mask_0 = const()[name = tensor("op_947_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_947_cast_fp16 = slice_by_index(begin = var_947_begin_0, end = var_947_end_0, end_mask = var_947_end_mask_0, x = var_910_cast_fp16)[name = tensor("op_947_cast_fp16")]; tensor var_948_begin_0 = const()[name = tensor("op_948_begin_0"), val = tensor([0, 0, 0, 3584])]; tensor var_948_end_0 = const()[name = tensor("op_948_end_0"), val = tensor([2, 40, 1, 4096])]; tensor var_948_end_mask_0 = const()[name = tensor("op_948_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_948_cast_fp16 = slice_by_index(begin = var_948_begin_0, end = var_948_end_0, end_mask = var_948_end_mask_0, x = var_910_cast_fp16)[name = tensor("op_948_cast_fp16")]; tensor var_949_begin_0 = const()[name = tensor("op_949_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_949_end_0 = const()[name = tensor("op_949_end_0"), val = tensor([2, 40, 1, 512])]; tensor var_949_end_mask_0 = const()[name = tensor("op_949_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_949_cast_fp16 = slice_by_index(begin = var_949_begin_0, end = var_949_end_0, end_mask = var_949_end_mask_0, x = var_914_cast_fp16)[name = tensor("op_949_cast_fp16")]; tensor var_950_begin_0 = const()[name = tensor("op_950_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_950_end_0 = const()[name = tensor("op_950_end_0"), val = tensor([2, 40, 1, 1024])]; tensor var_950_end_mask_0 = const()[name = tensor("op_950_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_950_cast_fp16 = slice_by_index(begin = var_950_begin_0, end = var_950_end_0, end_mask = var_950_end_mask_0, x = var_914_cast_fp16)[name = tensor("op_950_cast_fp16")]; tensor var_951_begin_0 = const()[name = tensor("op_951_begin_0"), val = tensor([0, 0, 0, 1024])]; tensor var_951_end_0 = const()[name = tensor("op_951_end_0"), val = tensor([2, 40, 1, 1536])]; tensor var_951_end_mask_0 = const()[name = tensor("op_951_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_951_cast_fp16 = slice_by_index(begin = var_951_begin_0, end = var_951_end_0, end_mask = var_951_end_mask_0, x = var_914_cast_fp16)[name = tensor("op_951_cast_fp16")]; tensor var_952_begin_0 = const()[name = tensor("op_952_begin_0"), val = tensor([0, 0, 0, 1536])]; tensor var_952_end_0 = const()[name = tensor("op_952_end_0"), val = tensor([2, 40, 1, 2048])]; tensor var_952_end_mask_0 = const()[name = tensor("op_952_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_952_cast_fp16 = slice_by_index(begin = var_952_begin_0, end = var_952_end_0, end_mask = var_952_end_mask_0, x = var_914_cast_fp16)[name = tensor("op_952_cast_fp16")]; tensor var_953_begin_0 = const()[name = tensor("op_953_begin_0"), val = tensor([0, 0, 0, 2048])]; tensor var_953_end_0 = const()[name = tensor("op_953_end_0"), val = tensor([2, 40, 1, 2560])]; tensor var_953_end_mask_0 = const()[name = tensor("op_953_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_953_cast_fp16 = slice_by_index(begin = var_953_begin_0, end = var_953_end_0, end_mask = var_953_end_mask_0, x = var_914_cast_fp16)[name = tensor("op_953_cast_fp16")]; tensor var_954_begin_0 = const()[name = tensor("op_954_begin_0"), val = tensor([0, 0, 0, 2560])]; tensor var_954_end_0 = const()[name = tensor("op_954_end_0"), val = tensor([2, 40, 1, 3072])]; tensor var_954_end_mask_0 = const()[name = tensor("op_954_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_954_cast_fp16 = slice_by_index(begin = var_954_begin_0, end = var_954_end_0, end_mask = var_954_end_mask_0, x = var_914_cast_fp16)[name = tensor("op_954_cast_fp16")]; tensor var_955_begin_0 = const()[name = tensor("op_955_begin_0"), val = tensor([0, 0, 0, 3072])]; tensor var_955_end_0 = const()[name = tensor("op_955_end_0"), val = tensor([2, 40, 1, 3584])]; tensor var_955_end_mask_0 = const()[name = tensor("op_955_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_955_cast_fp16 = slice_by_index(begin = var_955_begin_0, end = var_955_end_0, end_mask = var_955_end_mask_0, x = var_914_cast_fp16)[name = tensor("op_955_cast_fp16")]; tensor var_956_begin_0 = const()[name = tensor("op_956_begin_0"), val = tensor([0, 0, 0, 3584])]; tensor var_956_end_0 = const()[name = tensor("op_956_end_0"), val = tensor([2, 40, 1, 4096])]; tensor var_956_end_mask_0 = const()[name = tensor("op_956_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_956_cast_fp16 = slice_by_index(begin = var_956_begin_0, end = var_956_end_0, end_mask = var_956_end_mask_0, x = var_914_cast_fp16)[name = tensor("op_956_cast_fp16")]; tensor var_957_begin_0 = const()[name = tensor("op_957_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_957_end_0 = const()[name = tensor("op_957_end_0"), val = tensor([2, 40, 1, 512])]; tensor var_957_end_mask_0 = const()[name = tensor("op_957_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_957_cast_fp16 = slice_by_index(begin = var_957_begin_0, end = var_957_end_0, end_mask = var_957_end_mask_0, x = var_918_cast_fp16)[name = tensor("op_957_cast_fp16")]; tensor var_958_begin_0 = const()[name = tensor("op_958_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_958_end_0 = const()[name = tensor("op_958_end_0"), val = tensor([2, 40, 1, 1024])]; tensor var_958_end_mask_0 = const()[name = tensor("op_958_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_958_cast_fp16 = slice_by_index(begin = var_958_begin_0, end = var_958_end_0, end_mask = var_958_end_mask_0, x = var_918_cast_fp16)[name = tensor("op_958_cast_fp16")]; tensor var_959_begin_0 = const()[name = tensor("op_959_begin_0"), val = tensor([0, 0, 0, 1024])]; tensor var_959_end_0 = const()[name = tensor("op_959_end_0"), val = tensor([2, 40, 1, 1536])]; tensor var_959_end_mask_0 = const()[name = tensor("op_959_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_959_cast_fp16 = slice_by_index(begin = var_959_begin_0, end = var_959_end_0, end_mask = var_959_end_mask_0, x = var_918_cast_fp16)[name = tensor("op_959_cast_fp16")]; tensor var_960_begin_0 = const()[name = tensor("op_960_begin_0"), val = tensor([0, 0, 0, 1536])]; tensor var_960_end_0 = const()[name = tensor("op_960_end_0"), val = tensor([2, 40, 1, 2048])]; tensor var_960_end_mask_0 = const()[name = tensor("op_960_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_960_cast_fp16 = slice_by_index(begin = var_960_begin_0, end = var_960_end_0, end_mask = var_960_end_mask_0, x = var_918_cast_fp16)[name = tensor("op_960_cast_fp16")]; tensor var_961_begin_0 = const()[name = tensor("op_961_begin_0"), val = tensor([0, 0, 0, 2048])]; tensor var_961_end_0 = const()[name = tensor("op_961_end_0"), val = tensor([2, 40, 1, 2560])]; tensor var_961_end_mask_0 = const()[name = tensor("op_961_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_961_cast_fp16 = slice_by_index(begin = var_961_begin_0, end = var_961_end_0, end_mask = var_961_end_mask_0, x = var_918_cast_fp16)[name = tensor("op_961_cast_fp16")]; tensor var_962_begin_0 = const()[name = tensor("op_962_begin_0"), val = tensor([0, 0, 0, 2560])]; tensor var_962_end_0 = const()[name = tensor("op_962_end_0"), val = tensor([2, 40, 1, 3072])]; tensor var_962_end_mask_0 = const()[name = tensor("op_962_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_962_cast_fp16 = slice_by_index(begin = var_962_begin_0, end = var_962_end_0, end_mask = var_962_end_mask_0, x = var_918_cast_fp16)[name = tensor("op_962_cast_fp16")]; tensor var_963_begin_0 = const()[name = tensor("op_963_begin_0"), val = tensor([0, 0, 0, 3072])]; tensor var_963_end_0 = const()[name = tensor("op_963_end_0"), val = tensor([2, 40, 1, 3584])]; tensor var_963_end_mask_0 = const()[name = tensor("op_963_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_963_cast_fp16 = slice_by_index(begin = var_963_begin_0, end = var_963_end_0, end_mask = var_963_end_mask_0, x = var_918_cast_fp16)[name = tensor("op_963_cast_fp16")]; tensor var_964_begin_0 = const()[name = tensor("op_964_begin_0"), val = tensor([0, 0, 0, 3584])]; tensor var_964_end_0 = const()[name = tensor("op_964_end_0"), val = tensor([2, 40, 1, 4096])]; tensor var_964_end_mask_0 = const()[name = tensor("op_964_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_964_cast_fp16 = slice_by_index(begin = var_964_begin_0, end = var_964_end_0, end_mask = var_964_end_mask_0, x = var_918_cast_fp16)[name = tensor("op_964_cast_fp16")]; tensor var_965_begin_0 = const()[name = tensor("op_965_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_965_end_0 = const()[name = tensor("op_965_end_0"), val = tensor([2, 40, 1, 512])]; tensor var_965_end_mask_0 = const()[name = tensor("op_965_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_965_cast_fp16 = slice_by_index(begin = var_965_begin_0, end = var_965_end_0, end_mask = var_965_end_mask_0, x = var_922_cast_fp16)[name = tensor("op_965_cast_fp16")]; tensor var_966_begin_0 = const()[name = tensor("op_966_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_966_end_0 = const()[name = tensor("op_966_end_0"), val = tensor([2, 40, 1, 1024])]; tensor var_966_end_mask_0 = const()[name = tensor("op_966_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_966_cast_fp16 = slice_by_index(begin = var_966_begin_0, end = var_966_end_0, end_mask = var_966_end_mask_0, x = var_922_cast_fp16)[name = tensor("op_966_cast_fp16")]; tensor var_967_begin_0 = const()[name = tensor("op_967_begin_0"), val = tensor([0, 0, 0, 1024])]; tensor var_967_end_0 = const()[name = tensor("op_967_end_0"), val = tensor([2, 40, 1, 1536])]; tensor var_967_end_mask_0 = const()[name = tensor("op_967_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_967_cast_fp16 = slice_by_index(begin = var_967_begin_0, end = var_967_end_0, end_mask = var_967_end_mask_0, x = var_922_cast_fp16)[name = tensor("op_967_cast_fp16")]; tensor var_968_begin_0 = const()[name = tensor("op_968_begin_0"), val = tensor([0, 0, 0, 1536])]; tensor var_968_end_0 = const()[name = tensor("op_968_end_0"), val = tensor([2, 40, 1, 2048])]; tensor var_968_end_mask_0 = const()[name = tensor("op_968_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_968_cast_fp16 = slice_by_index(begin = var_968_begin_0, end = var_968_end_0, end_mask = var_968_end_mask_0, x = var_922_cast_fp16)[name = tensor("op_968_cast_fp16")]; tensor var_969_begin_0 = const()[name = tensor("op_969_begin_0"), val = tensor([0, 0, 0, 2048])]; tensor var_969_end_0 = const()[name = tensor("op_969_end_0"), val = tensor([2, 40, 1, 2560])]; tensor var_969_end_mask_0 = const()[name = tensor("op_969_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_969_cast_fp16 = slice_by_index(begin = var_969_begin_0, end = var_969_end_0, end_mask = var_969_end_mask_0, x = var_922_cast_fp16)[name = tensor("op_969_cast_fp16")]; tensor var_970_begin_0 = const()[name = tensor("op_970_begin_0"), val = tensor([0, 0, 0, 2560])]; tensor var_970_end_0 = const()[name = tensor("op_970_end_0"), val = tensor([2, 40, 1, 3072])]; tensor var_970_end_mask_0 = const()[name = tensor("op_970_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_970_cast_fp16 = slice_by_index(begin = var_970_begin_0, end = var_970_end_0, end_mask = var_970_end_mask_0, x = var_922_cast_fp16)[name = tensor("op_970_cast_fp16")]; tensor var_971_begin_0 = const()[name = tensor("op_971_begin_0"), val = tensor([0, 0, 0, 3072])]; tensor var_971_end_0 = const()[name = tensor("op_971_end_0"), val = tensor([2, 40, 1, 3584])]; tensor var_971_end_mask_0 = const()[name = tensor("op_971_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_971_cast_fp16 = slice_by_index(begin = var_971_begin_0, end = var_971_end_0, end_mask = var_971_end_mask_0, x = var_922_cast_fp16)[name = tensor("op_971_cast_fp16")]; tensor var_972_begin_0 = const()[name = tensor("op_972_begin_0"), val = tensor([0, 0, 0, 3584])]; tensor var_972_end_0 = const()[name = tensor("op_972_end_0"), val = tensor([2, 40, 1, 4096])]; tensor var_972_end_mask_0 = const()[name = tensor("op_972_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_972_cast_fp16 = slice_by_index(begin = var_972_begin_0, end = var_972_end_0, end_mask = var_972_end_mask_0, x = var_922_cast_fp16)[name = tensor("op_972_cast_fp16")]; tensor var_973_begin_0 = const()[name = tensor("op_973_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_973_end_0 = const()[name = tensor("op_973_end_0"), val = tensor([2, 40, 1, 512])]; tensor var_973_end_mask_0 = const()[name = tensor("op_973_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_973_cast_fp16 = slice_by_index(begin = var_973_begin_0, end = var_973_end_0, end_mask = var_973_end_mask_0, x = var_926_cast_fp16)[name = tensor("op_973_cast_fp16")]; tensor var_974_begin_0 = const()[name = tensor("op_974_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_974_end_0 = const()[name = tensor("op_974_end_0"), val = tensor([2, 40, 1, 1024])]; tensor var_974_end_mask_0 = const()[name = tensor("op_974_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_974_cast_fp16 = slice_by_index(begin = var_974_begin_0, end = var_974_end_0, end_mask = var_974_end_mask_0, x = var_926_cast_fp16)[name = tensor("op_974_cast_fp16")]; tensor var_975_begin_0 = const()[name = tensor("op_975_begin_0"), val = tensor([0, 0, 0, 1024])]; tensor var_975_end_0 = const()[name = tensor("op_975_end_0"), val = tensor([2, 40, 1, 1536])]; tensor var_975_end_mask_0 = const()[name = tensor("op_975_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_975_cast_fp16 = slice_by_index(begin = var_975_begin_0, end = var_975_end_0, end_mask = var_975_end_mask_0, x = var_926_cast_fp16)[name = tensor("op_975_cast_fp16")]; tensor var_976_begin_0 = const()[name = tensor("op_976_begin_0"), val = tensor([0, 0, 0, 1536])]; tensor var_976_end_0 = const()[name = tensor("op_976_end_0"), val = tensor([2, 40, 1, 2048])]; tensor var_976_end_mask_0 = const()[name = tensor("op_976_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_976_cast_fp16 = slice_by_index(begin = var_976_begin_0, end = var_976_end_0, end_mask = var_976_end_mask_0, x = var_926_cast_fp16)[name = tensor("op_976_cast_fp16")]; tensor var_977_begin_0 = const()[name = tensor("op_977_begin_0"), val = tensor([0, 0, 0, 2048])]; tensor var_977_end_0 = const()[name = tensor("op_977_end_0"), val = tensor([2, 40, 1, 2560])]; tensor var_977_end_mask_0 = const()[name = tensor("op_977_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_977_cast_fp16 = slice_by_index(begin = var_977_begin_0, end = var_977_end_0, end_mask = var_977_end_mask_0, x = var_926_cast_fp16)[name = tensor("op_977_cast_fp16")]; tensor var_978_begin_0 = const()[name = tensor("op_978_begin_0"), val = tensor([0, 0, 0, 2560])]; tensor var_978_end_0 = const()[name = tensor("op_978_end_0"), val = tensor([2, 40, 1, 3072])]; tensor var_978_end_mask_0 = const()[name = tensor("op_978_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_978_cast_fp16 = slice_by_index(begin = var_978_begin_0, end = var_978_end_0, end_mask = var_978_end_mask_0, x = var_926_cast_fp16)[name = tensor("op_978_cast_fp16")]; tensor var_979_begin_0 = const()[name = tensor("op_979_begin_0"), val = tensor([0, 0, 0, 3072])]; tensor var_979_end_0 = const()[name = tensor("op_979_end_0"), val = tensor([2, 40, 1, 3584])]; tensor var_979_end_mask_0 = const()[name = tensor("op_979_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_979_cast_fp16 = slice_by_index(begin = var_979_begin_0, end = var_979_end_0, end_mask = var_979_end_mask_0, x = var_926_cast_fp16)[name = tensor("op_979_cast_fp16")]; tensor var_980_begin_0 = const()[name = tensor("op_980_begin_0"), val = tensor([0, 0, 0, 3584])]; tensor var_980_end_0 = const()[name = tensor("op_980_end_0"), val = tensor([2, 40, 1, 4096])]; tensor var_980_end_mask_0 = const()[name = tensor("op_980_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_980_cast_fp16 = slice_by_index(begin = var_980_begin_0, end = var_980_end_0, end_mask = var_980_end_mask_0, x = var_926_cast_fp16)[name = tensor("op_980_cast_fp16")]; tensor var_981_begin_0 = const()[name = tensor("op_981_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_981_end_0 = const()[name = tensor("op_981_end_0"), val = tensor([2, 40, 1, 512])]; tensor var_981_end_mask_0 = const()[name = tensor("op_981_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_981_cast_fp16 = slice_by_index(begin = var_981_begin_0, end = var_981_end_0, end_mask = var_981_end_mask_0, x = var_930_cast_fp16)[name = tensor("op_981_cast_fp16")]; tensor var_982_begin_0 = const()[name = tensor("op_982_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_982_end_0 = const()[name = tensor("op_982_end_0"), val = tensor([2, 40, 1, 1024])]; tensor var_982_end_mask_0 = const()[name = tensor("op_982_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_982_cast_fp16 = slice_by_index(begin = var_982_begin_0, end = var_982_end_0, end_mask = var_982_end_mask_0, x = var_930_cast_fp16)[name = tensor("op_982_cast_fp16")]; tensor var_983_begin_0 = const()[name = tensor("op_983_begin_0"), val = tensor([0, 0, 0, 1024])]; tensor var_983_end_0 = const()[name = tensor("op_983_end_0"), val = tensor([2, 40, 1, 1536])]; tensor var_983_end_mask_0 = const()[name = tensor("op_983_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_983_cast_fp16 = slice_by_index(begin = var_983_begin_0, end = var_983_end_0, end_mask = var_983_end_mask_0, x = var_930_cast_fp16)[name = tensor("op_983_cast_fp16")]; tensor var_984_begin_0 = const()[name = tensor("op_984_begin_0"), val = tensor([0, 0, 0, 1536])]; tensor var_984_end_0 = const()[name = tensor("op_984_end_0"), val = tensor([2, 40, 1, 2048])]; tensor var_984_end_mask_0 = const()[name = tensor("op_984_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_984_cast_fp16 = slice_by_index(begin = var_984_begin_0, end = var_984_end_0, end_mask = var_984_end_mask_0, x = var_930_cast_fp16)[name = tensor("op_984_cast_fp16")]; tensor var_985_begin_0 = const()[name = tensor("op_985_begin_0"), val = tensor([0, 0, 0, 2048])]; tensor var_985_end_0 = const()[name = tensor("op_985_end_0"), val = tensor([2, 40, 1, 2560])]; tensor var_985_end_mask_0 = const()[name = tensor("op_985_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_985_cast_fp16 = slice_by_index(begin = var_985_begin_0, end = var_985_end_0, end_mask = var_985_end_mask_0, x = var_930_cast_fp16)[name = tensor("op_985_cast_fp16")]; tensor var_986_begin_0 = const()[name = tensor("op_986_begin_0"), val = tensor([0, 0, 0, 2560])]; tensor var_986_end_0 = const()[name = tensor("op_986_end_0"), val = tensor([2, 40, 1, 3072])]; tensor var_986_end_mask_0 = const()[name = tensor("op_986_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_986_cast_fp16 = slice_by_index(begin = var_986_begin_0, end = var_986_end_0, end_mask = var_986_end_mask_0, x = var_930_cast_fp16)[name = tensor("op_986_cast_fp16")]; tensor var_987_begin_0 = const()[name = tensor("op_987_begin_0"), val = tensor([0, 0, 0, 3072])]; tensor var_987_end_0 = const()[name = tensor("op_987_end_0"), val = tensor([2, 40, 1, 3584])]; tensor var_987_end_mask_0 = const()[name = tensor("op_987_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_987_cast_fp16 = slice_by_index(begin = var_987_begin_0, end = var_987_end_0, end_mask = var_987_end_mask_0, x = var_930_cast_fp16)[name = tensor("op_987_cast_fp16")]; tensor var_988_begin_0 = const()[name = tensor("op_988_begin_0"), val = tensor([0, 0, 0, 3584])]; tensor var_988_end_0 = const()[name = tensor("op_988_end_0"), val = tensor([2, 40, 1, 4096])]; tensor var_988_end_mask_0 = const()[name = tensor("op_988_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_988_cast_fp16 = slice_by_index(begin = var_988_begin_0, end = var_988_end_0, end_mask = var_988_end_mask_0, x = var_930_cast_fp16)[name = tensor("op_988_cast_fp16")]; tensor var_989_begin_0 = const()[name = tensor("op_989_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_989_end_0 = const()[name = tensor("op_989_end_0"), val = tensor([2, 40, 1, 512])]; tensor var_989_end_mask_0 = const()[name = tensor("op_989_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_989_cast_fp16 = slice_by_index(begin = var_989_begin_0, end = var_989_end_0, end_mask = var_989_end_mask_0, x = var_934_cast_fp16)[name = tensor("op_989_cast_fp16")]; tensor var_990_begin_0 = const()[name = tensor("op_990_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_990_end_0 = const()[name = tensor("op_990_end_0"), val = tensor([2, 40, 1, 1024])]; tensor var_990_end_mask_0 = const()[name = tensor("op_990_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_990_cast_fp16 = slice_by_index(begin = var_990_begin_0, end = var_990_end_0, end_mask = var_990_end_mask_0, x = var_934_cast_fp16)[name = tensor("op_990_cast_fp16")]; tensor var_991_begin_0 = const()[name = tensor("op_991_begin_0"), val = tensor([0, 0, 0, 1024])]; tensor var_991_end_0 = const()[name = tensor("op_991_end_0"), val = tensor([2, 40, 1, 1536])]; tensor var_991_end_mask_0 = const()[name = tensor("op_991_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_991_cast_fp16 = slice_by_index(begin = var_991_begin_0, end = var_991_end_0, end_mask = var_991_end_mask_0, x = var_934_cast_fp16)[name = tensor("op_991_cast_fp16")]; tensor var_992_begin_0 = const()[name = tensor("op_992_begin_0"), val = tensor([0, 0, 0, 1536])]; tensor var_992_end_0 = const()[name = tensor("op_992_end_0"), val = tensor([2, 40, 1, 2048])]; tensor var_992_end_mask_0 = const()[name = tensor("op_992_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_992_cast_fp16 = slice_by_index(begin = var_992_begin_0, end = var_992_end_0, end_mask = var_992_end_mask_0, x = var_934_cast_fp16)[name = tensor("op_992_cast_fp16")]; tensor var_993_begin_0 = const()[name = tensor("op_993_begin_0"), val = tensor([0, 0, 0, 2048])]; tensor var_993_end_0 = const()[name = tensor("op_993_end_0"), val = tensor([2, 40, 1, 2560])]; tensor var_993_end_mask_0 = const()[name = tensor("op_993_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_993_cast_fp16 = slice_by_index(begin = var_993_begin_0, end = var_993_end_0, end_mask = var_993_end_mask_0, x = var_934_cast_fp16)[name = tensor("op_993_cast_fp16")]; tensor var_994_begin_0 = const()[name = tensor("op_994_begin_0"), val = tensor([0, 0, 0, 2560])]; tensor var_994_end_0 = const()[name = tensor("op_994_end_0"), val = tensor([2, 40, 1, 3072])]; tensor var_994_end_mask_0 = const()[name = tensor("op_994_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_994_cast_fp16 = slice_by_index(begin = var_994_begin_0, end = var_994_end_0, end_mask = var_994_end_mask_0, x = var_934_cast_fp16)[name = tensor("op_994_cast_fp16")]; tensor var_995_begin_0 = const()[name = tensor("op_995_begin_0"), val = tensor([0, 0, 0, 3072])]; tensor var_995_end_0 = const()[name = tensor("op_995_end_0"), val = tensor([2, 40, 1, 3584])]; tensor var_995_end_mask_0 = const()[name = tensor("op_995_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_995_cast_fp16 = slice_by_index(begin = var_995_begin_0, end = var_995_end_0, end_mask = var_995_end_mask_0, x = var_934_cast_fp16)[name = tensor("op_995_cast_fp16")]; tensor var_996_begin_0 = const()[name = tensor("op_996_begin_0"), val = tensor([0, 0, 0, 3584])]; tensor var_996_end_0 = const()[name = tensor("op_996_end_0"), val = tensor([2, 40, 1, 4096])]; tensor var_996_end_mask_0 = const()[name = tensor("op_996_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_996_cast_fp16 = slice_by_index(begin = var_996_begin_0, end = var_996_end_0, end_mask = var_996_end_mask_0, x = var_934_cast_fp16)[name = tensor("op_996_cast_fp16")]; tensor var_997_begin_0 = const()[name = tensor("op_997_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_997_end_0 = const()[name = tensor("op_997_end_0"), val = tensor([2, 40, 1, 512])]; tensor var_997_end_mask_0 = const()[name = tensor("op_997_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_997_cast_fp16 = slice_by_index(begin = var_997_begin_0, end = var_997_end_0, end_mask = var_997_end_mask_0, x = var_938_cast_fp16)[name = tensor("op_997_cast_fp16")]; tensor var_998_begin_0 = const()[name = tensor("op_998_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_998_end_0 = const()[name = tensor("op_998_end_0"), val = tensor([2, 40, 1, 1024])]; tensor var_998_end_mask_0 = const()[name = tensor("op_998_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_998_cast_fp16 = slice_by_index(begin = var_998_begin_0, end = var_998_end_0, end_mask = var_998_end_mask_0, x = var_938_cast_fp16)[name = tensor("op_998_cast_fp16")]; tensor var_999_begin_0 = const()[name = tensor("op_999_begin_0"), val = tensor([0, 0, 0, 1024])]; tensor var_999_end_0 = const()[name = tensor("op_999_end_0"), val = tensor([2, 40, 1, 1536])]; tensor var_999_end_mask_0 = const()[name = tensor("op_999_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_999_cast_fp16 = slice_by_index(begin = var_999_begin_0, end = var_999_end_0, end_mask = var_999_end_mask_0, x = var_938_cast_fp16)[name = tensor("op_999_cast_fp16")]; tensor var_1000_begin_0 = const()[name = tensor("op_1000_begin_0"), val = tensor([0, 0, 0, 1536])]; tensor var_1000_end_0 = const()[name = tensor("op_1000_end_0"), val = tensor([2, 40, 1, 2048])]; tensor var_1000_end_mask_0 = const()[name = tensor("op_1000_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1000_cast_fp16 = slice_by_index(begin = var_1000_begin_0, end = var_1000_end_0, end_mask = var_1000_end_mask_0, x = var_938_cast_fp16)[name = tensor("op_1000_cast_fp16")]; tensor var_1001_begin_0 = const()[name = tensor("op_1001_begin_0"), val = tensor([0, 0, 0, 2048])]; tensor var_1001_end_0 = const()[name = tensor("op_1001_end_0"), val = tensor([2, 40, 1, 2560])]; tensor var_1001_end_mask_0 = const()[name = tensor("op_1001_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1001_cast_fp16 = slice_by_index(begin = var_1001_begin_0, end = var_1001_end_0, end_mask = var_1001_end_mask_0, x = var_938_cast_fp16)[name = tensor("op_1001_cast_fp16")]; tensor var_1002_begin_0 = const()[name = tensor("op_1002_begin_0"), val = tensor([0, 0, 0, 2560])]; tensor var_1002_end_0 = const()[name = tensor("op_1002_end_0"), val = tensor([2, 40, 1, 3072])]; tensor var_1002_end_mask_0 = const()[name = tensor("op_1002_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1002_cast_fp16 = slice_by_index(begin = var_1002_begin_0, end = var_1002_end_0, end_mask = var_1002_end_mask_0, x = var_938_cast_fp16)[name = tensor("op_1002_cast_fp16")]; tensor var_1003_begin_0 = const()[name = tensor("op_1003_begin_0"), val = tensor([0, 0, 0, 3072])]; tensor var_1003_end_0 = const()[name = tensor("op_1003_end_0"), val = tensor([2, 40, 1, 3584])]; tensor var_1003_end_mask_0 = const()[name = tensor("op_1003_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1003_cast_fp16 = slice_by_index(begin = var_1003_begin_0, end = var_1003_end_0, end_mask = var_1003_end_mask_0, x = var_938_cast_fp16)[name = tensor("op_1003_cast_fp16")]; tensor var_1004_begin_0 = const()[name = tensor("op_1004_begin_0"), val = tensor([0, 0, 0, 3584])]; tensor var_1004_end_0 = const()[name = tensor("op_1004_end_0"), val = tensor([2, 40, 1, 4096])]; tensor var_1004_end_mask_0 = const()[name = tensor("op_1004_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1004_cast_fp16 = slice_by_index(begin = var_1004_begin_0, end = var_1004_end_0, end_mask = var_1004_end_mask_0, x = var_938_cast_fp16)[name = tensor("op_1004_cast_fp16")]; tensor k_7_perm_0 = const()[name = tensor("k_7_perm_0"), val = tensor([0, 3, 2, 1])]; tensor var_1009_begin_0 = const()[name = tensor("op_1009_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_1009_end_0 = const()[name = tensor("op_1009_end_0"), val = tensor([2, 77, 1, 40])]; tensor var_1009_end_mask_0 = const()[name = tensor("op_1009_end_mask_0"), val = tensor([true, true, true, false])]; tensor transpose_16 = transpose(perm = k_7_perm_0, x = k_5_cast_fp16)[name = tensor("transpose_16")]; tensor var_1009_cast_fp16 = slice_by_index(begin = var_1009_begin_0, end = var_1009_end_0, end_mask = var_1009_end_mask_0, x = transpose_16)[name = tensor("op_1009_cast_fp16")]; tensor var_1013_begin_0 = const()[name = tensor("op_1013_begin_0"), val = tensor([0, 0, 0, 40])]; tensor var_1013_end_0 = const()[name = tensor("op_1013_end_0"), val = tensor([2, 77, 1, 80])]; tensor var_1013_end_mask_0 = const()[name = tensor("op_1013_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1013_cast_fp16 = slice_by_index(begin = var_1013_begin_0, end = var_1013_end_0, end_mask = var_1013_end_mask_0, x = transpose_16)[name = tensor("op_1013_cast_fp16")]; tensor var_1017_begin_0 = const()[name = tensor("op_1017_begin_0"), val = tensor([0, 0, 0, 80])]; tensor var_1017_end_0 = const()[name = tensor("op_1017_end_0"), val = tensor([2, 77, 1, 120])]; tensor var_1017_end_mask_0 = const()[name = tensor("op_1017_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1017_cast_fp16 = slice_by_index(begin = var_1017_begin_0, end = var_1017_end_0, end_mask = var_1017_end_mask_0, x = transpose_16)[name = tensor("op_1017_cast_fp16")]; tensor var_1021_begin_0 = const()[name = tensor("op_1021_begin_0"), val = tensor([0, 0, 0, 120])]; tensor var_1021_end_0 = const()[name = tensor("op_1021_end_0"), val = tensor([2, 77, 1, 160])]; tensor var_1021_end_mask_0 = const()[name = tensor("op_1021_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1021_cast_fp16 = slice_by_index(begin = var_1021_begin_0, end = var_1021_end_0, end_mask = var_1021_end_mask_0, x = transpose_16)[name = tensor("op_1021_cast_fp16")]; tensor var_1025_begin_0 = const()[name = tensor("op_1025_begin_0"), val = tensor([0, 0, 0, 160])]; tensor var_1025_end_0 = const()[name = tensor("op_1025_end_0"), val = tensor([2, 77, 1, 200])]; tensor var_1025_end_mask_0 = const()[name = tensor("op_1025_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1025_cast_fp16 = slice_by_index(begin = var_1025_begin_0, end = var_1025_end_0, end_mask = var_1025_end_mask_0, x = transpose_16)[name = tensor("op_1025_cast_fp16")]; tensor var_1029_begin_0 = const()[name = tensor("op_1029_begin_0"), val = tensor([0, 0, 0, 200])]; tensor var_1029_end_0 = const()[name = tensor("op_1029_end_0"), val = tensor([2, 77, 1, 240])]; tensor var_1029_end_mask_0 = const()[name = tensor("op_1029_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1029_cast_fp16 = slice_by_index(begin = var_1029_begin_0, end = var_1029_end_0, end_mask = var_1029_end_mask_0, x = transpose_16)[name = tensor("op_1029_cast_fp16")]; tensor var_1033_begin_0 = const()[name = tensor("op_1033_begin_0"), val = tensor([0, 0, 0, 240])]; tensor var_1033_end_0 = const()[name = tensor("op_1033_end_0"), val = tensor([2, 77, 1, 280])]; tensor var_1033_end_mask_0 = const()[name = tensor("op_1033_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1033_cast_fp16 = slice_by_index(begin = var_1033_begin_0, end = var_1033_end_0, end_mask = var_1033_end_mask_0, x = transpose_16)[name = tensor("op_1033_cast_fp16")]; tensor var_1037_begin_0 = const()[name = tensor("op_1037_begin_0"), val = tensor([0, 0, 0, 280])]; tensor var_1037_end_0 = const()[name = tensor("op_1037_end_0"), val = tensor([2, 77, 1, 320])]; tensor var_1037_end_mask_0 = const()[name = tensor("op_1037_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1037_cast_fp16 = slice_by_index(begin = var_1037_begin_0, end = var_1037_end_0, end_mask = var_1037_end_mask_0, x = transpose_16)[name = tensor("op_1037_cast_fp16")]; tensor var_1039_begin_0 = const()[name = tensor("op_1039_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_1039_end_0 = const()[name = tensor("op_1039_end_0"), val = tensor([2, 40, 1, 77])]; tensor var_1039_end_mask_0 = const()[name = tensor("op_1039_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_1039_cast_fp16 = slice_by_index(begin = var_1039_begin_0, end = var_1039_end_0, end_mask = var_1039_end_mask_0, x = v_3_cast_fp16)[name = tensor("op_1039_cast_fp16")]; tensor var_1043_begin_0 = const()[name = tensor("op_1043_begin_0"), val = tensor([0, 40, 0, 0])]; tensor var_1043_end_0 = const()[name = tensor("op_1043_end_0"), val = tensor([2, 80, 1, 77])]; tensor var_1043_end_mask_0 = const()[name = tensor("op_1043_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_1043_cast_fp16 = slice_by_index(begin = var_1043_begin_0, end = var_1043_end_0, end_mask = var_1043_end_mask_0, x = v_3_cast_fp16)[name = tensor("op_1043_cast_fp16")]; tensor var_1047_begin_0 = const()[name = tensor("op_1047_begin_0"), val = tensor([0, 80, 0, 0])]; tensor var_1047_end_0 = const()[name = tensor("op_1047_end_0"), val = tensor([2, 120, 1, 77])]; tensor var_1047_end_mask_0 = const()[name = tensor("op_1047_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_1047_cast_fp16 = slice_by_index(begin = var_1047_begin_0, end = var_1047_end_0, end_mask = var_1047_end_mask_0, x = v_3_cast_fp16)[name = tensor("op_1047_cast_fp16")]; tensor var_1051_begin_0 = const()[name = tensor("op_1051_begin_0"), val = tensor([0, 120, 0, 0])]; tensor var_1051_end_0 = const()[name = tensor("op_1051_end_0"), val = tensor([2, 160, 1, 77])]; tensor var_1051_end_mask_0 = const()[name = tensor("op_1051_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_1051_cast_fp16 = slice_by_index(begin = var_1051_begin_0, end = var_1051_end_0, end_mask = var_1051_end_mask_0, x = v_3_cast_fp16)[name = tensor("op_1051_cast_fp16")]; tensor var_1055_begin_0 = const()[name = tensor("op_1055_begin_0"), val = tensor([0, 160, 0, 0])]; tensor var_1055_end_0 = const()[name = tensor("op_1055_end_0"), val = tensor([2, 200, 1, 77])]; tensor var_1055_end_mask_0 = const()[name = tensor("op_1055_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_1055_cast_fp16 = slice_by_index(begin = var_1055_begin_0, end = var_1055_end_0, end_mask = var_1055_end_mask_0, x = v_3_cast_fp16)[name = tensor("op_1055_cast_fp16")]; tensor var_1059_begin_0 = const()[name = tensor("op_1059_begin_0"), val = tensor([0, 200, 0, 0])]; tensor var_1059_end_0 = const()[name = tensor("op_1059_end_0"), val = tensor([2, 240, 1, 77])]; tensor var_1059_end_mask_0 = const()[name = tensor("op_1059_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_1059_cast_fp16 = slice_by_index(begin = var_1059_begin_0, end = var_1059_end_0, end_mask = var_1059_end_mask_0, x = v_3_cast_fp16)[name = tensor("op_1059_cast_fp16")]; tensor var_1063_begin_0 = const()[name = tensor("op_1063_begin_0"), val = tensor([0, 240, 0, 0])]; tensor var_1063_end_0 = const()[name = tensor("op_1063_end_0"), val = tensor([2, 280, 1, 77])]; tensor var_1063_end_mask_0 = const()[name = tensor("op_1063_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_1063_cast_fp16 = slice_by_index(begin = var_1063_begin_0, end = var_1063_end_0, end_mask = var_1063_end_mask_0, x = v_3_cast_fp16)[name = tensor("op_1063_cast_fp16")]; tensor var_1067_begin_0 = const()[name = tensor("op_1067_begin_0"), val = tensor([0, 280, 0, 0])]; tensor var_1067_end_0 = const()[name = tensor("op_1067_end_0"), val = tensor([2, 320, 1, 77])]; tensor var_1067_end_mask_0 = const()[name = tensor("op_1067_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_1067_cast_fp16 = slice_by_index(begin = var_1067_begin_0, end = var_1067_end_0, end_mask = var_1067_end_mask_0, x = v_3_cast_fp16)[name = tensor("op_1067_cast_fp16")]; tensor var_1071_equation_0 = const()[name = tensor("op_1071_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_1071_cast_fp16 = einsum(equation = var_1071_equation_0, values = (var_1009_cast_fp16, var_941_cast_fp16))[name = tensor("op_1071_cast_fp16")]; tensor var_1072_to_fp16 = const()[name = tensor("op_1072_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_129_cast_fp16 = mul(x = var_1071_cast_fp16, y = var_1072_to_fp16)[name = tensor("aw_chunk_129_cast_fp16")]; tensor var_1075_equation_0 = const()[name = tensor("op_1075_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_1075_cast_fp16 = einsum(equation = var_1075_equation_0, values = (var_1009_cast_fp16, var_942_cast_fp16))[name = tensor("op_1075_cast_fp16")]; tensor var_1076_to_fp16 = const()[name = tensor("op_1076_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_131_cast_fp16 = mul(x = var_1075_cast_fp16, y = var_1076_to_fp16)[name = tensor("aw_chunk_131_cast_fp16")]; tensor var_1079_equation_0 = const()[name = tensor("op_1079_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_1079_cast_fp16 = einsum(equation = var_1079_equation_0, values = (var_1009_cast_fp16, var_943_cast_fp16))[name = tensor("op_1079_cast_fp16")]; tensor var_1080_to_fp16 = const()[name = tensor("op_1080_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_133_cast_fp16 = mul(x = var_1079_cast_fp16, y = var_1080_to_fp16)[name = tensor("aw_chunk_133_cast_fp16")]; tensor var_1083_equation_0 = const()[name = tensor("op_1083_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_1083_cast_fp16 = einsum(equation = var_1083_equation_0, values = (var_1009_cast_fp16, var_944_cast_fp16))[name = tensor("op_1083_cast_fp16")]; tensor var_1084_to_fp16 = const()[name = tensor("op_1084_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_135_cast_fp16 = mul(x = var_1083_cast_fp16, y = var_1084_to_fp16)[name = tensor("aw_chunk_135_cast_fp16")]; tensor var_1087_equation_0 = const()[name = tensor("op_1087_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_1087_cast_fp16 = einsum(equation = var_1087_equation_0, values = (var_1009_cast_fp16, var_945_cast_fp16))[name = tensor("op_1087_cast_fp16")]; tensor var_1088_to_fp16 = const()[name = tensor("op_1088_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_137_cast_fp16 = mul(x = var_1087_cast_fp16, y = var_1088_to_fp16)[name = tensor("aw_chunk_137_cast_fp16")]; tensor var_1091_equation_0 = const()[name = tensor("op_1091_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_1091_cast_fp16 = einsum(equation = var_1091_equation_0, values = (var_1009_cast_fp16, var_946_cast_fp16))[name = tensor("op_1091_cast_fp16")]; tensor var_1092_to_fp16 = const()[name = tensor("op_1092_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_139_cast_fp16 = mul(x = var_1091_cast_fp16, y = var_1092_to_fp16)[name = tensor("aw_chunk_139_cast_fp16")]; tensor var_1095_equation_0 = const()[name = tensor("op_1095_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_1095_cast_fp16 = einsum(equation = var_1095_equation_0, values = (var_1009_cast_fp16, var_947_cast_fp16))[name = tensor("op_1095_cast_fp16")]; tensor var_1096_to_fp16 = const()[name = tensor("op_1096_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_141_cast_fp16 = mul(x = var_1095_cast_fp16, y = var_1096_to_fp16)[name = tensor("aw_chunk_141_cast_fp16")]; tensor var_1099_equation_0 = const()[name = tensor("op_1099_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_1099_cast_fp16 = einsum(equation = var_1099_equation_0, values = (var_1009_cast_fp16, var_948_cast_fp16))[name = tensor("op_1099_cast_fp16")]; tensor var_1100_to_fp16 = const()[name = tensor("op_1100_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_143_cast_fp16 = mul(x = var_1099_cast_fp16, y = var_1100_to_fp16)[name = tensor("aw_chunk_143_cast_fp16")]; tensor var_1103_equation_0 = const()[name = tensor("op_1103_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_1103_cast_fp16 = einsum(equation = var_1103_equation_0, values = (var_1013_cast_fp16, var_949_cast_fp16))[name = tensor("op_1103_cast_fp16")]; tensor var_1104_to_fp16 = const()[name = tensor("op_1104_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_145_cast_fp16 = mul(x = var_1103_cast_fp16, y = var_1104_to_fp16)[name = tensor("aw_chunk_145_cast_fp16")]; tensor var_1107_equation_0 = const()[name = tensor("op_1107_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_1107_cast_fp16 = einsum(equation = var_1107_equation_0, values = (var_1013_cast_fp16, var_950_cast_fp16))[name = tensor("op_1107_cast_fp16")]; tensor var_1108_to_fp16 = const()[name = tensor("op_1108_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_147_cast_fp16 = mul(x = var_1107_cast_fp16, y = var_1108_to_fp16)[name = tensor("aw_chunk_147_cast_fp16")]; tensor var_1111_equation_0 = const()[name = tensor("op_1111_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_1111_cast_fp16 = einsum(equation = var_1111_equation_0, values = (var_1013_cast_fp16, var_951_cast_fp16))[name = tensor("op_1111_cast_fp16")]; tensor var_1112_to_fp16 = const()[name = tensor("op_1112_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_149_cast_fp16 = mul(x = var_1111_cast_fp16, y = var_1112_to_fp16)[name = tensor("aw_chunk_149_cast_fp16")]; tensor var_1115_equation_0 = const()[name = tensor("op_1115_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_1115_cast_fp16 = einsum(equation = var_1115_equation_0, values = (var_1013_cast_fp16, var_952_cast_fp16))[name = tensor("op_1115_cast_fp16")]; tensor var_1116_to_fp16 = const()[name = tensor("op_1116_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_151_cast_fp16 = mul(x = var_1115_cast_fp16, y = var_1116_to_fp16)[name = tensor("aw_chunk_151_cast_fp16")]; tensor var_1119_equation_0 = const()[name = tensor("op_1119_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_1119_cast_fp16 = einsum(equation = var_1119_equation_0, values = (var_1013_cast_fp16, var_953_cast_fp16))[name = tensor("op_1119_cast_fp16")]; tensor var_1120_to_fp16 = const()[name = tensor("op_1120_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_153_cast_fp16 = mul(x = var_1119_cast_fp16, y = var_1120_to_fp16)[name = tensor("aw_chunk_153_cast_fp16")]; tensor var_1123_equation_0 = const()[name = tensor("op_1123_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_1123_cast_fp16 = einsum(equation = var_1123_equation_0, values = (var_1013_cast_fp16, var_954_cast_fp16))[name = tensor("op_1123_cast_fp16")]; tensor var_1124_to_fp16 = const()[name = tensor("op_1124_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_155_cast_fp16 = mul(x = var_1123_cast_fp16, y = var_1124_to_fp16)[name = tensor("aw_chunk_155_cast_fp16")]; tensor var_1127_equation_0 = const()[name = tensor("op_1127_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_1127_cast_fp16 = einsum(equation = var_1127_equation_0, values = (var_1013_cast_fp16, var_955_cast_fp16))[name = tensor("op_1127_cast_fp16")]; tensor var_1128_to_fp16 = const()[name = tensor("op_1128_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_157_cast_fp16 = mul(x = var_1127_cast_fp16, y = var_1128_to_fp16)[name = tensor("aw_chunk_157_cast_fp16")]; tensor var_1131_equation_0 = const()[name = tensor("op_1131_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_1131_cast_fp16 = einsum(equation = var_1131_equation_0, values = (var_1013_cast_fp16, var_956_cast_fp16))[name = tensor("op_1131_cast_fp16")]; tensor var_1132_to_fp16 = const()[name = tensor("op_1132_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_159_cast_fp16 = mul(x = var_1131_cast_fp16, y = var_1132_to_fp16)[name = tensor("aw_chunk_159_cast_fp16")]; tensor var_1135_equation_0 = const()[name = tensor("op_1135_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_1135_cast_fp16 = einsum(equation = var_1135_equation_0, values = (var_1017_cast_fp16, var_957_cast_fp16))[name = tensor("op_1135_cast_fp16")]; tensor var_1136_to_fp16 = const()[name = tensor("op_1136_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_161_cast_fp16 = mul(x = var_1135_cast_fp16, y = var_1136_to_fp16)[name = tensor("aw_chunk_161_cast_fp16")]; tensor var_1139_equation_0 = const()[name = tensor("op_1139_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_1139_cast_fp16 = einsum(equation = var_1139_equation_0, values = (var_1017_cast_fp16, var_958_cast_fp16))[name = tensor("op_1139_cast_fp16")]; tensor var_1140_to_fp16 = const()[name = tensor("op_1140_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_163_cast_fp16 = mul(x = var_1139_cast_fp16, y = var_1140_to_fp16)[name = tensor("aw_chunk_163_cast_fp16")]; tensor var_1143_equation_0 = const()[name = tensor("op_1143_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_1143_cast_fp16 = einsum(equation = var_1143_equation_0, values = (var_1017_cast_fp16, var_959_cast_fp16))[name = tensor("op_1143_cast_fp16")]; tensor var_1144_to_fp16 = const()[name = tensor("op_1144_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_165_cast_fp16 = mul(x = var_1143_cast_fp16, y = var_1144_to_fp16)[name = tensor("aw_chunk_165_cast_fp16")]; tensor var_1147_equation_0 = const()[name = tensor("op_1147_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_1147_cast_fp16 = einsum(equation = var_1147_equation_0, values = (var_1017_cast_fp16, var_960_cast_fp16))[name = tensor("op_1147_cast_fp16")]; tensor var_1148_to_fp16 = const()[name = tensor("op_1148_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_167_cast_fp16 = mul(x = var_1147_cast_fp16, y = var_1148_to_fp16)[name = tensor("aw_chunk_167_cast_fp16")]; tensor var_1151_equation_0 = const()[name = tensor("op_1151_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_1151_cast_fp16 = einsum(equation = var_1151_equation_0, values = (var_1017_cast_fp16, var_961_cast_fp16))[name = tensor("op_1151_cast_fp16")]; tensor var_1152_to_fp16 = const()[name = tensor("op_1152_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_169_cast_fp16 = mul(x = var_1151_cast_fp16, y = var_1152_to_fp16)[name = tensor("aw_chunk_169_cast_fp16")]; tensor var_1155_equation_0 = const()[name = tensor("op_1155_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_1155_cast_fp16 = einsum(equation = var_1155_equation_0, values = (var_1017_cast_fp16, var_962_cast_fp16))[name = tensor("op_1155_cast_fp16")]; tensor var_1156_to_fp16 = const()[name = tensor("op_1156_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_171_cast_fp16 = mul(x = var_1155_cast_fp16, y = var_1156_to_fp16)[name = tensor("aw_chunk_171_cast_fp16")]; tensor var_1159_equation_0 = const()[name = tensor("op_1159_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_1159_cast_fp16 = einsum(equation = var_1159_equation_0, values = (var_1017_cast_fp16, var_963_cast_fp16))[name = tensor("op_1159_cast_fp16")]; tensor var_1160_to_fp16 = const()[name = tensor("op_1160_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_173_cast_fp16 = mul(x = var_1159_cast_fp16, y = var_1160_to_fp16)[name = tensor("aw_chunk_173_cast_fp16")]; tensor var_1163_equation_0 = const()[name = tensor("op_1163_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_1163_cast_fp16 = einsum(equation = var_1163_equation_0, values = (var_1017_cast_fp16, var_964_cast_fp16))[name = tensor("op_1163_cast_fp16")]; tensor var_1164_to_fp16 = const()[name = tensor("op_1164_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_175_cast_fp16 = mul(x = var_1163_cast_fp16, y = var_1164_to_fp16)[name = tensor("aw_chunk_175_cast_fp16")]; tensor var_1167_equation_0 = const()[name = tensor("op_1167_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_1167_cast_fp16 = einsum(equation = var_1167_equation_0, values = (var_1021_cast_fp16, var_965_cast_fp16))[name = tensor("op_1167_cast_fp16")]; tensor var_1168_to_fp16 = const()[name = tensor("op_1168_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_177_cast_fp16 = mul(x = var_1167_cast_fp16, y = var_1168_to_fp16)[name = tensor("aw_chunk_177_cast_fp16")]; tensor var_1171_equation_0 = const()[name = tensor("op_1171_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_1171_cast_fp16 = einsum(equation = var_1171_equation_0, values = (var_1021_cast_fp16, var_966_cast_fp16))[name = tensor("op_1171_cast_fp16")]; tensor var_1172_to_fp16 = const()[name = tensor("op_1172_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_179_cast_fp16 = mul(x = var_1171_cast_fp16, y = var_1172_to_fp16)[name = tensor("aw_chunk_179_cast_fp16")]; tensor var_1175_equation_0 = const()[name = tensor("op_1175_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_1175_cast_fp16 = einsum(equation = var_1175_equation_0, values = (var_1021_cast_fp16, var_967_cast_fp16))[name = tensor("op_1175_cast_fp16")]; tensor var_1176_to_fp16 = const()[name = tensor("op_1176_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_181_cast_fp16 = mul(x = var_1175_cast_fp16, y = var_1176_to_fp16)[name = tensor("aw_chunk_181_cast_fp16")]; tensor var_1179_equation_0 = const()[name = tensor("op_1179_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_1179_cast_fp16 = einsum(equation = var_1179_equation_0, values = (var_1021_cast_fp16, var_968_cast_fp16))[name = tensor("op_1179_cast_fp16")]; tensor var_1180_to_fp16 = const()[name = tensor("op_1180_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_183_cast_fp16 = mul(x = var_1179_cast_fp16, y = var_1180_to_fp16)[name = tensor("aw_chunk_183_cast_fp16")]; tensor var_1183_equation_0 = const()[name = tensor("op_1183_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_1183_cast_fp16 = einsum(equation = var_1183_equation_0, values = (var_1021_cast_fp16, var_969_cast_fp16))[name = tensor("op_1183_cast_fp16")]; tensor var_1184_to_fp16 = const()[name = tensor("op_1184_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_185_cast_fp16 = mul(x = var_1183_cast_fp16, y = var_1184_to_fp16)[name = tensor("aw_chunk_185_cast_fp16")]; tensor var_1187_equation_0 = const()[name = tensor("op_1187_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_1187_cast_fp16 = einsum(equation = var_1187_equation_0, values = (var_1021_cast_fp16, var_970_cast_fp16))[name = tensor("op_1187_cast_fp16")]; tensor var_1188_to_fp16 = const()[name = tensor("op_1188_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_187_cast_fp16 = mul(x = var_1187_cast_fp16, y = var_1188_to_fp16)[name = tensor("aw_chunk_187_cast_fp16")]; tensor var_1191_equation_0 = const()[name = tensor("op_1191_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_1191_cast_fp16 = einsum(equation = var_1191_equation_0, values = (var_1021_cast_fp16, var_971_cast_fp16))[name = tensor("op_1191_cast_fp16")]; tensor var_1192_to_fp16 = const()[name = tensor("op_1192_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_189_cast_fp16 = mul(x = var_1191_cast_fp16, y = var_1192_to_fp16)[name = tensor("aw_chunk_189_cast_fp16")]; tensor var_1195_equation_0 = const()[name = tensor("op_1195_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_1195_cast_fp16 = einsum(equation = var_1195_equation_0, values = (var_1021_cast_fp16, var_972_cast_fp16))[name = tensor("op_1195_cast_fp16")]; tensor var_1196_to_fp16 = const()[name = tensor("op_1196_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_191_cast_fp16 = mul(x = var_1195_cast_fp16, y = var_1196_to_fp16)[name = tensor("aw_chunk_191_cast_fp16")]; tensor var_1199_equation_0 = const()[name = tensor("op_1199_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_1199_cast_fp16 = einsum(equation = var_1199_equation_0, values = (var_1025_cast_fp16, var_973_cast_fp16))[name = tensor("op_1199_cast_fp16")]; tensor var_1200_to_fp16 = const()[name = tensor("op_1200_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_193_cast_fp16 = mul(x = var_1199_cast_fp16, y = var_1200_to_fp16)[name = tensor("aw_chunk_193_cast_fp16")]; tensor var_1203_equation_0 = const()[name = tensor("op_1203_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_1203_cast_fp16 = einsum(equation = var_1203_equation_0, values = (var_1025_cast_fp16, var_974_cast_fp16))[name = tensor("op_1203_cast_fp16")]; tensor var_1204_to_fp16 = const()[name = tensor("op_1204_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_195_cast_fp16 = mul(x = var_1203_cast_fp16, y = var_1204_to_fp16)[name = tensor("aw_chunk_195_cast_fp16")]; tensor var_1207_equation_0 = const()[name = tensor("op_1207_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_1207_cast_fp16 = einsum(equation = var_1207_equation_0, values = (var_1025_cast_fp16, var_975_cast_fp16))[name = tensor("op_1207_cast_fp16")]; tensor var_1208_to_fp16 = const()[name = tensor("op_1208_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_197_cast_fp16 = mul(x = var_1207_cast_fp16, y = var_1208_to_fp16)[name = tensor("aw_chunk_197_cast_fp16")]; tensor var_1211_equation_0 = const()[name = tensor("op_1211_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_1211_cast_fp16 = einsum(equation = var_1211_equation_0, values = (var_1025_cast_fp16, var_976_cast_fp16))[name = tensor("op_1211_cast_fp16")]; tensor var_1212_to_fp16 = const()[name = tensor("op_1212_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_199_cast_fp16 = mul(x = var_1211_cast_fp16, y = var_1212_to_fp16)[name = tensor("aw_chunk_199_cast_fp16")]; tensor var_1215_equation_0 = const()[name = tensor("op_1215_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_1215_cast_fp16 = einsum(equation = var_1215_equation_0, values = (var_1025_cast_fp16, var_977_cast_fp16))[name = tensor("op_1215_cast_fp16")]; tensor var_1216_to_fp16 = const()[name = tensor("op_1216_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_201_cast_fp16 = mul(x = var_1215_cast_fp16, y = var_1216_to_fp16)[name = tensor("aw_chunk_201_cast_fp16")]; tensor var_1219_equation_0 = const()[name = tensor("op_1219_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_1219_cast_fp16 = einsum(equation = var_1219_equation_0, values = (var_1025_cast_fp16, var_978_cast_fp16))[name = tensor("op_1219_cast_fp16")]; tensor var_1220_to_fp16 = const()[name = tensor("op_1220_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_203_cast_fp16 = mul(x = var_1219_cast_fp16, y = var_1220_to_fp16)[name = tensor("aw_chunk_203_cast_fp16")]; tensor var_1223_equation_0 = const()[name = tensor("op_1223_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_1223_cast_fp16 = einsum(equation = var_1223_equation_0, values = (var_1025_cast_fp16, var_979_cast_fp16))[name = tensor("op_1223_cast_fp16")]; tensor var_1224_to_fp16 = const()[name = tensor("op_1224_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_205_cast_fp16 = mul(x = var_1223_cast_fp16, y = var_1224_to_fp16)[name = tensor("aw_chunk_205_cast_fp16")]; tensor var_1227_equation_0 = const()[name = tensor("op_1227_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_1227_cast_fp16 = einsum(equation = var_1227_equation_0, values = (var_1025_cast_fp16, var_980_cast_fp16))[name = tensor("op_1227_cast_fp16")]; tensor var_1228_to_fp16 = const()[name = tensor("op_1228_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_207_cast_fp16 = mul(x = var_1227_cast_fp16, y = var_1228_to_fp16)[name = tensor("aw_chunk_207_cast_fp16")]; tensor var_1231_equation_0 = const()[name = tensor("op_1231_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_1231_cast_fp16 = einsum(equation = var_1231_equation_0, values = (var_1029_cast_fp16, var_981_cast_fp16))[name = tensor("op_1231_cast_fp16")]; tensor var_1232_to_fp16 = const()[name = tensor("op_1232_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_209_cast_fp16 = mul(x = var_1231_cast_fp16, y = var_1232_to_fp16)[name = tensor("aw_chunk_209_cast_fp16")]; tensor var_1235_equation_0 = const()[name = tensor("op_1235_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_1235_cast_fp16 = einsum(equation = var_1235_equation_0, values = (var_1029_cast_fp16, var_982_cast_fp16))[name = tensor("op_1235_cast_fp16")]; tensor var_1236_to_fp16 = const()[name = tensor("op_1236_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_211_cast_fp16 = mul(x = var_1235_cast_fp16, y = var_1236_to_fp16)[name = tensor("aw_chunk_211_cast_fp16")]; tensor var_1239_equation_0 = const()[name = tensor("op_1239_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_1239_cast_fp16 = einsum(equation = var_1239_equation_0, values = (var_1029_cast_fp16, var_983_cast_fp16))[name = tensor("op_1239_cast_fp16")]; tensor var_1240_to_fp16 = const()[name = tensor("op_1240_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_213_cast_fp16 = mul(x = var_1239_cast_fp16, y = var_1240_to_fp16)[name = tensor("aw_chunk_213_cast_fp16")]; tensor var_1243_equation_0 = const()[name = tensor("op_1243_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_1243_cast_fp16 = einsum(equation = var_1243_equation_0, values = (var_1029_cast_fp16, var_984_cast_fp16))[name = tensor("op_1243_cast_fp16")]; tensor var_1244_to_fp16 = const()[name = tensor("op_1244_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_215_cast_fp16 = mul(x = var_1243_cast_fp16, y = var_1244_to_fp16)[name = tensor("aw_chunk_215_cast_fp16")]; tensor var_1247_equation_0 = const()[name = tensor("op_1247_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_1247_cast_fp16 = einsum(equation = var_1247_equation_0, values = (var_1029_cast_fp16, var_985_cast_fp16))[name = tensor("op_1247_cast_fp16")]; tensor var_1248_to_fp16 = const()[name = tensor("op_1248_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_217_cast_fp16 = mul(x = var_1247_cast_fp16, y = var_1248_to_fp16)[name = tensor("aw_chunk_217_cast_fp16")]; tensor var_1251_equation_0 = const()[name = tensor("op_1251_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_1251_cast_fp16 = einsum(equation = var_1251_equation_0, values = (var_1029_cast_fp16, var_986_cast_fp16))[name = tensor("op_1251_cast_fp16")]; tensor var_1252_to_fp16 = const()[name = tensor("op_1252_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_219_cast_fp16 = mul(x = var_1251_cast_fp16, y = var_1252_to_fp16)[name = tensor("aw_chunk_219_cast_fp16")]; tensor var_1255_equation_0 = const()[name = tensor("op_1255_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_1255_cast_fp16 = einsum(equation = var_1255_equation_0, values = (var_1029_cast_fp16, var_987_cast_fp16))[name = tensor("op_1255_cast_fp16")]; tensor var_1256_to_fp16 = const()[name = tensor("op_1256_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_221_cast_fp16 = mul(x = var_1255_cast_fp16, y = var_1256_to_fp16)[name = tensor("aw_chunk_221_cast_fp16")]; tensor var_1259_equation_0 = const()[name = tensor("op_1259_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_1259_cast_fp16 = einsum(equation = var_1259_equation_0, values = (var_1029_cast_fp16, var_988_cast_fp16))[name = tensor("op_1259_cast_fp16")]; tensor var_1260_to_fp16 = const()[name = tensor("op_1260_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_223_cast_fp16 = mul(x = var_1259_cast_fp16, y = var_1260_to_fp16)[name = tensor("aw_chunk_223_cast_fp16")]; tensor var_1263_equation_0 = const()[name = tensor("op_1263_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_1263_cast_fp16 = einsum(equation = var_1263_equation_0, values = (var_1033_cast_fp16, var_989_cast_fp16))[name = tensor("op_1263_cast_fp16")]; tensor var_1264_to_fp16 = const()[name = tensor("op_1264_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_225_cast_fp16 = mul(x = var_1263_cast_fp16, y = var_1264_to_fp16)[name = tensor("aw_chunk_225_cast_fp16")]; tensor var_1267_equation_0 = const()[name = tensor("op_1267_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_1267_cast_fp16 = einsum(equation = var_1267_equation_0, values = (var_1033_cast_fp16, var_990_cast_fp16))[name = tensor("op_1267_cast_fp16")]; tensor var_1268_to_fp16 = const()[name = tensor("op_1268_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_227_cast_fp16 = mul(x = var_1267_cast_fp16, y = var_1268_to_fp16)[name = tensor("aw_chunk_227_cast_fp16")]; tensor var_1271_equation_0 = const()[name = tensor("op_1271_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_1271_cast_fp16 = einsum(equation = var_1271_equation_0, values = (var_1033_cast_fp16, var_991_cast_fp16))[name = tensor("op_1271_cast_fp16")]; tensor var_1272_to_fp16 = const()[name = tensor("op_1272_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_229_cast_fp16 = mul(x = var_1271_cast_fp16, y = var_1272_to_fp16)[name = tensor("aw_chunk_229_cast_fp16")]; tensor var_1275_equation_0 = const()[name = tensor("op_1275_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_1275_cast_fp16 = einsum(equation = var_1275_equation_0, values = (var_1033_cast_fp16, var_992_cast_fp16))[name = tensor("op_1275_cast_fp16")]; tensor var_1276_to_fp16 = const()[name = tensor("op_1276_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_231_cast_fp16 = mul(x = var_1275_cast_fp16, y = var_1276_to_fp16)[name = tensor("aw_chunk_231_cast_fp16")]; tensor var_1279_equation_0 = const()[name = tensor("op_1279_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_1279_cast_fp16 = einsum(equation = var_1279_equation_0, values = (var_1033_cast_fp16, var_993_cast_fp16))[name = tensor("op_1279_cast_fp16")]; tensor var_1280_to_fp16 = const()[name = tensor("op_1280_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_233_cast_fp16 = mul(x = var_1279_cast_fp16, y = var_1280_to_fp16)[name = tensor("aw_chunk_233_cast_fp16")]; tensor var_1283_equation_0 = const()[name = tensor("op_1283_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_1283_cast_fp16 = einsum(equation = var_1283_equation_0, values = (var_1033_cast_fp16, var_994_cast_fp16))[name = tensor("op_1283_cast_fp16")]; tensor var_1284_to_fp16 = const()[name = tensor("op_1284_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_235_cast_fp16 = mul(x = var_1283_cast_fp16, y = var_1284_to_fp16)[name = tensor("aw_chunk_235_cast_fp16")]; tensor var_1287_equation_0 = const()[name = tensor("op_1287_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_1287_cast_fp16 = einsum(equation = var_1287_equation_0, values = (var_1033_cast_fp16, var_995_cast_fp16))[name = tensor("op_1287_cast_fp16")]; tensor var_1288_to_fp16 = const()[name = tensor("op_1288_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_237_cast_fp16 = mul(x = var_1287_cast_fp16, y = var_1288_to_fp16)[name = tensor("aw_chunk_237_cast_fp16")]; tensor var_1291_equation_0 = const()[name = tensor("op_1291_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_1291_cast_fp16 = einsum(equation = var_1291_equation_0, values = (var_1033_cast_fp16, var_996_cast_fp16))[name = tensor("op_1291_cast_fp16")]; tensor var_1292_to_fp16 = const()[name = tensor("op_1292_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_239_cast_fp16 = mul(x = var_1291_cast_fp16, y = var_1292_to_fp16)[name = tensor("aw_chunk_239_cast_fp16")]; tensor var_1295_equation_0 = const()[name = tensor("op_1295_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_1295_cast_fp16 = einsum(equation = var_1295_equation_0, values = (var_1037_cast_fp16, var_997_cast_fp16))[name = tensor("op_1295_cast_fp16")]; tensor var_1296_to_fp16 = const()[name = tensor("op_1296_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_241_cast_fp16 = mul(x = var_1295_cast_fp16, y = var_1296_to_fp16)[name = tensor("aw_chunk_241_cast_fp16")]; tensor var_1299_equation_0 = const()[name = tensor("op_1299_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_1299_cast_fp16 = einsum(equation = var_1299_equation_0, values = (var_1037_cast_fp16, var_998_cast_fp16))[name = tensor("op_1299_cast_fp16")]; tensor var_1300_to_fp16 = const()[name = tensor("op_1300_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_243_cast_fp16 = mul(x = var_1299_cast_fp16, y = var_1300_to_fp16)[name = tensor("aw_chunk_243_cast_fp16")]; tensor var_1303_equation_0 = const()[name = tensor("op_1303_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_1303_cast_fp16 = einsum(equation = var_1303_equation_0, values = (var_1037_cast_fp16, var_999_cast_fp16))[name = tensor("op_1303_cast_fp16")]; tensor var_1304_to_fp16 = const()[name = tensor("op_1304_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_245_cast_fp16 = mul(x = var_1303_cast_fp16, y = var_1304_to_fp16)[name = tensor("aw_chunk_245_cast_fp16")]; tensor var_1307_equation_0 = const()[name = tensor("op_1307_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_1307_cast_fp16 = einsum(equation = var_1307_equation_0, values = (var_1037_cast_fp16, var_1000_cast_fp16))[name = tensor("op_1307_cast_fp16")]; tensor var_1308_to_fp16 = const()[name = tensor("op_1308_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_247_cast_fp16 = mul(x = var_1307_cast_fp16, y = var_1308_to_fp16)[name = tensor("aw_chunk_247_cast_fp16")]; tensor var_1311_equation_0 = const()[name = tensor("op_1311_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_1311_cast_fp16 = einsum(equation = var_1311_equation_0, values = (var_1037_cast_fp16, var_1001_cast_fp16))[name = tensor("op_1311_cast_fp16")]; tensor var_1312_to_fp16 = const()[name = tensor("op_1312_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_249_cast_fp16 = mul(x = var_1311_cast_fp16, y = var_1312_to_fp16)[name = tensor("aw_chunk_249_cast_fp16")]; tensor var_1315_equation_0 = const()[name = tensor("op_1315_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_1315_cast_fp16 = einsum(equation = var_1315_equation_0, values = (var_1037_cast_fp16, var_1002_cast_fp16))[name = tensor("op_1315_cast_fp16")]; tensor var_1316_to_fp16 = const()[name = tensor("op_1316_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_251_cast_fp16 = mul(x = var_1315_cast_fp16, y = var_1316_to_fp16)[name = tensor("aw_chunk_251_cast_fp16")]; tensor var_1319_equation_0 = const()[name = tensor("op_1319_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_1319_cast_fp16 = einsum(equation = var_1319_equation_0, values = (var_1037_cast_fp16, var_1003_cast_fp16))[name = tensor("op_1319_cast_fp16")]; tensor var_1320_to_fp16 = const()[name = tensor("op_1320_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_253_cast_fp16 = mul(x = var_1319_cast_fp16, y = var_1320_to_fp16)[name = tensor("aw_chunk_253_cast_fp16")]; tensor var_1323_equation_0 = const()[name = tensor("op_1323_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_1323_cast_fp16 = einsum(equation = var_1323_equation_0, values = (var_1037_cast_fp16, var_1004_cast_fp16))[name = tensor("op_1323_cast_fp16")]; tensor var_1324_to_fp16 = const()[name = tensor("op_1324_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_255_cast_fp16 = mul(x = var_1323_cast_fp16, y = var_1324_to_fp16)[name = tensor("aw_chunk_255_cast_fp16")]; tensor var_1326_cast_fp16 = softmax(axis = var_118, x = aw_chunk_129_cast_fp16)[name = tensor("op_1326_cast_fp16")]; tensor var_1327_cast_fp16 = softmax(axis = var_118, x = aw_chunk_131_cast_fp16)[name = tensor("op_1327_cast_fp16")]; tensor var_1328_cast_fp16 = softmax(axis = var_118, x = aw_chunk_133_cast_fp16)[name = tensor("op_1328_cast_fp16")]; tensor var_1329_cast_fp16 = softmax(axis = var_118, x = aw_chunk_135_cast_fp16)[name = tensor("op_1329_cast_fp16")]; tensor var_1330_cast_fp16 = softmax(axis = var_118, x = aw_chunk_137_cast_fp16)[name = tensor("op_1330_cast_fp16")]; tensor var_1331_cast_fp16 = softmax(axis = var_118, x = aw_chunk_139_cast_fp16)[name = tensor("op_1331_cast_fp16")]; tensor var_1332_cast_fp16 = softmax(axis = var_118, x = aw_chunk_141_cast_fp16)[name = tensor("op_1332_cast_fp16")]; tensor var_1333_cast_fp16 = softmax(axis = var_118, x = aw_chunk_143_cast_fp16)[name = tensor("op_1333_cast_fp16")]; tensor var_1334_cast_fp16 = softmax(axis = var_118, x = aw_chunk_145_cast_fp16)[name = tensor("op_1334_cast_fp16")]; tensor var_1335_cast_fp16 = softmax(axis = var_118, x = aw_chunk_147_cast_fp16)[name = tensor("op_1335_cast_fp16")]; tensor var_1336_cast_fp16 = softmax(axis = var_118, x = aw_chunk_149_cast_fp16)[name = tensor("op_1336_cast_fp16")]; tensor var_1337_cast_fp16 = softmax(axis = var_118, x = aw_chunk_151_cast_fp16)[name = tensor("op_1337_cast_fp16")]; tensor var_1338_cast_fp16 = softmax(axis = var_118, x = aw_chunk_153_cast_fp16)[name = tensor("op_1338_cast_fp16")]; tensor var_1339_cast_fp16 = softmax(axis = var_118, x = aw_chunk_155_cast_fp16)[name = tensor("op_1339_cast_fp16")]; tensor var_1340_cast_fp16 = softmax(axis = var_118, x = aw_chunk_157_cast_fp16)[name = tensor("op_1340_cast_fp16")]; tensor var_1341_cast_fp16 = softmax(axis = var_118, x = aw_chunk_159_cast_fp16)[name = tensor("op_1341_cast_fp16")]; tensor var_1342_cast_fp16 = softmax(axis = var_118, x = aw_chunk_161_cast_fp16)[name = tensor("op_1342_cast_fp16")]; tensor var_1343_cast_fp16 = softmax(axis = var_118, x = aw_chunk_163_cast_fp16)[name = tensor("op_1343_cast_fp16")]; tensor var_1344_cast_fp16 = softmax(axis = var_118, x = aw_chunk_165_cast_fp16)[name = tensor("op_1344_cast_fp16")]; tensor var_1345_cast_fp16 = softmax(axis = var_118, x = aw_chunk_167_cast_fp16)[name = tensor("op_1345_cast_fp16")]; tensor var_1346_cast_fp16 = softmax(axis = var_118, x = aw_chunk_169_cast_fp16)[name = tensor("op_1346_cast_fp16")]; tensor var_1347_cast_fp16 = softmax(axis = var_118, x = aw_chunk_171_cast_fp16)[name = tensor("op_1347_cast_fp16")]; tensor var_1348_cast_fp16 = softmax(axis = var_118, x = aw_chunk_173_cast_fp16)[name = tensor("op_1348_cast_fp16")]; tensor var_1349_cast_fp16 = softmax(axis = var_118, x = aw_chunk_175_cast_fp16)[name = tensor("op_1349_cast_fp16")]; tensor var_1350_cast_fp16 = softmax(axis = var_118, x = aw_chunk_177_cast_fp16)[name = tensor("op_1350_cast_fp16")]; tensor var_1351_cast_fp16 = softmax(axis = var_118, x = aw_chunk_179_cast_fp16)[name = tensor("op_1351_cast_fp16")]; tensor var_1352_cast_fp16 = softmax(axis = var_118, x = aw_chunk_181_cast_fp16)[name = tensor("op_1352_cast_fp16")]; tensor var_1353_cast_fp16 = softmax(axis = var_118, x = aw_chunk_183_cast_fp16)[name = tensor("op_1353_cast_fp16")]; tensor var_1354_cast_fp16 = softmax(axis = var_118, x = aw_chunk_185_cast_fp16)[name = tensor("op_1354_cast_fp16")]; tensor var_1355_cast_fp16 = softmax(axis = var_118, x = aw_chunk_187_cast_fp16)[name = tensor("op_1355_cast_fp16")]; tensor var_1356_cast_fp16 = softmax(axis = var_118, x = aw_chunk_189_cast_fp16)[name = tensor("op_1356_cast_fp16")]; tensor var_1357_cast_fp16 = softmax(axis = var_118, x = aw_chunk_191_cast_fp16)[name = tensor("op_1357_cast_fp16")]; tensor var_1358_cast_fp16 = softmax(axis = var_118, x = aw_chunk_193_cast_fp16)[name = tensor("op_1358_cast_fp16")]; tensor var_1359_cast_fp16 = softmax(axis = var_118, x = aw_chunk_195_cast_fp16)[name = tensor("op_1359_cast_fp16")]; tensor var_1360_cast_fp16 = softmax(axis = var_118, x = aw_chunk_197_cast_fp16)[name = tensor("op_1360_cast_fp16")]; tensor var_1361_cast_fp16 = softmax(axis = var_118, x = aw_chunk_199_cast_fp16)[name = tensor("op_1361_cast_fp16")]; tensor var_1362_cast_fp16 = softmax(axis = var_118, x = aw_chunk_201_cast_fp16)[name = tensor("op_1362_cast_fp16")]; tensor var_1363_cast_fp16 = softmax(axis = var_118, x = aw_chunk_203_cast_fp16)[name = tensor("op_1363_cast_fp16")]; tensor var_1364_cast_fp16 = softmax(axis = var_118, x = aw_chunk_205_cast_fp16)[name = tensor("op_1364_cast_fp16")]; tensor var_1365_cast_fp16 = softmax(axis = var_118, x = aw_chunk_207_cast_fp16)[name = tensor("op_1365_cast_fp16")]; tensor var_1366_cast_fp16 = softmax(axis = var_118, x = aw_chunk_209_cast_fp16)[name = tensor("op_1366_cast_fp16")]; tensor var_1367_cast_fp16 = softmax(axis = var_118, x = aw_chunk_211_cast_fp16)[name = tensor("op_1367_cast_fp16")]; tensor var_1368_cast_fp16 = softmax(axis = var_118, x = aw_chunk_213_cast_fp16)[name = tensor("op_1368_cast_fp16")]; tensor var_1369_cast_fp16 = softmax(axis = var_118, x = aw_chunk_215_cast_fp16)[name = tensor("op_1369_cast_fp16")]; tensor var_1370_cast_fp16 = softmax(axis = var_118, x = aw_chunk_217_cast_fp16)[name = tensor("op_1370_cast_fp16")]; tensor var_1371_cast_fp16 = softmax(axis = var_118, x = aw_chunk_219_cast_fp16)[name = tensor("op_1371_cast_fp16")]; tensor var_1372_cast_fp16 = softmax(axis = var_118, x = aw_chunk_221_cast_fp16)[name = tensor("op_1372_cast_fp16")]; tensor var_1373_cast_fp16 = softmax(axis = var_118, x = aw_chunk_223_cast_fp16)[name = tensor("op_1373_cast_fp16")]; tensor var_1374_cast_fp16 = softmax(axis = var_118, x = aw_chunk_225_cast_fp16)[name = tensor("op_1374_cast_fp16")]; tensor var_1375_cast_fp16 = softmax(axis = var_118, x = aw_chunk_227_cast_fp16)[name = tensor("op_1375_cast_fp16")]; tensor var_1376_cast_fp16 = softmax(axis = var_118, x = aw_chunk_229_cast_fp16)[name = tensor("op_1376_cast_fp16")]; tensor var_1377_cast_fp16 = softmax(axis = var_118, x = aw_chunk_231_cast_fp16)[name = tensor("op_1377_cast_fp16")]; tensor var_1378_cast_fp16 = softmax(axis = var_118, x = aw_chunk_233_cast_fp16)[name = tensor("op_1378_cast_fp16")]; tensor var_1379_cast_fp16 = softmax(axis = var_118, x = aw_chunk_235_cast_fp16)[name = tensor("op_1379_cast_fp16")]; tensor var_1380_cast_fp16 = softmax(axis = var_118, x = aw_chunk_237_cast_fp16)[name = tensor("op_1380_cast_fp16")]; tensor var_1381_cast_fp16 = softmax(axis = var_118, x = aw_chunk_239_cast_fp16)[name = tensor("op_1381_cast_fp16")]; tensor var_1382_cast_fp16 = softmax(axis = var_118, x = aw_chunk_241_cast_fp16)[name = tensor("op_1382_cast_fp16")]; tensor var_1383_cast_fp16 = softmax(axis = var_118, x = aw_chunk_243_cast_fp16)[name = tensor("op_1383_cast_fp16")]; tensor var_1384_cast_fp16 = softmax(axis = var_118, x = aw_chunk_245_cast_fp16)[name = tensor("op_1384_cast_fp16")]; tensor var_1385_cast_fp16 = softmax(axis = var_118, x = aw_chunk_247_cast_fp16)[name = tensor("op_1385_cast_fp16")]; tensor var_1386_cast_fp16 = softmax(axis = var_118, x = aw_chunk_249_cast_fp16)[name = tensor("op_1386_cast_fp16")]; tensor var_1387_cast_fp16 = softmax(axis = var_118, x = aw_chunk_251_cast_fp16)[name = tensor("op_1387_cast_fp16")]; tensor var_1388_cast_fp16 = softmax(axis = var_118, x = aw_chunk_253_cast_fp16)[name = tensor("op_1388_cast_fp16")]; tensor var_1389_cast_fp16 = softmax(axis = var_118, x = aw_chunk_255_cast_fp16)[name = tensor("op_1389_cast_fp16")]; tensor var_1391_equation_0 = const()[name = tensor("op_1391_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_1391_cast_fp16 = einsum(equation = var_1391_equation_0, values = (var_1039_cast_fp16, var_1326_cast_fp16))[name = tensor("op_1391_cast_fp16")]; tensor var_1393_equation_0 = const()[name = tensor("op_1393_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_1393_cast_fp16 = einsum(equation = var_1393_equation_0, values = (var_1039_cast_fp16, var_1327_cast_fp16))[name = tensor("op_1393_cast_fp16")]; tensor var_1395_equation_0 = const()[name = tensor("op_1395_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_1395_cast_fp16 = einsum(equation = var_1395_equation_0, values = (var_1039_cast_fp16, var_1328_cast_fp16))[name = tensor("op_1395_cast_fp16")]; tensor var_1397_equation_0 = const()[name = tensor("op_1397_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_1397_cast_fp16 = einsum(equation = var_1397_equation_0, values = (var_1039_cast_fp16, var_1329_cast_fp16))[name = tensor("op_1397_cast_fp16")]; tensor var_1399_equation_0 = const()[name = tensor("op_1399_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_1399_cast_fp16 = einsum(equation = var_1399_equation_0, values = (var_1039_cast_fp16, var_1330_cast_fp16))[name = tensor("op_1399_cast_fp16")]; tensor var_1401_equation_0 = const()[name = tensor("op_1401_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_1401_cast_fp16 = einsum(equation = var_1401_equation_0, values = (var_1039_cast_fp16, var_1331_cast_fp16))[name = tensor("op_1401_cast_fp16")]; tensor var_1403_equation_0 = const()[name = tensor("op_1403_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_1403_cast_fp16 = einsum(equation = var_1403_equation_0, values = (var_1039_cast_fp16, var_1332_cast_fp16))[name = tensor("op_1403_cast_fp16")]; tensor var_1405_equation_0 = const()[name = tensor("op_1405_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_1405_cast_fp16 = einsum(equation = var_1405_equation_0, values = (var_1039_cast_fp16, var_1333_cast_fp16))[name = tensor("op_1405_cast_fp16")]; tensor var_1407_equation_0 = const()[name = tensor("op_1407_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_1407_cast_fp16 = einsum(equation = var_1407_equation_0, values = (var_1043_cast_fp16, var_1334_cast_fp16))[name = tensor("op_1407_cast_fp16")]; tensor var_1409_equation_0 = const()[name = tensor("op_1409_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_1409_cast_fp16 = einsum(equation = var_1409_equation_0, values = (var_1043_cast_fp16, var_1335_cast_fp16))[name = tensor("op_1409_cast_fp16")]; tensor var_1411_equation_0 = const()[name = tensor("op_1411_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_1411_cast_fp16 = einsum(equation = var_1411_equation_0, values = (var_1043_cast_fp16, var_1336_cast_fp16))[name = tensor("op_1411_cast_fp16")]; tensor var_1413_equation_0 = const()[name = tensor("op_1413_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_1413_cast_fp16 = einsum(equation = var_1413_equation_0, values = (var_1043_cast_fp16, var_1337_cast_fp16))[name = tensor("op_1413_cast_fp16")]; tensor var_1415_equation_0 = const()[name = tensor("op_1415_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_1415_cast_fp16 = einsum(equation = var_1415_equation_0, values = (var_1043_cast_fp16, var_1338_cast_fp16))[name = tensor("op_1415_cast_fp16")]; tensor var_1417_equation_0 = const()[name = tensor("op_1417_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_1417_cast_fp16 = einsum(equation = var_1417_equation_0, values = (var_1043_cast_fp16, var_1339_cast_fp16))[name = tensor("op_1417_cast_fp16")]; tensor var_1419_equation_0 = const()[name = tensor("op_1419_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_1419_cast_fp16 = einsum(equation = var_1419_equation_0, values = (var_1043_cast_fp16, var_1340_cast_fp16))[name = tensor("op_1419_cast_fp16")]; tensor var_1421_equation_0 = const()[name = tensor("op_1421_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_1421_cast_fp16 = einsum(equation = var_1421_equation_0, values = (var_1043_cast_fp16, var_1341_cast_fp16))[name = tensor("op_1421_cast_fp16")]; tensor var_1423_equation_0 = const()[name = tensor("op_1423_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_1423_cast_fp16 = einsum(equation = var_1423_equation_0, values = (var_1047_cast_fp16, var_1342_cast_fp16))[name = tensor("op_1423_cast_fp16")]; tensor var_1425_equation_0 = const()[name = tensor("op_1425_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_1425_cast_fp16 = einsum(equation = var_1425_equation_0, values = (var_1047_cast_fp16, var_1343_cast_fp16))[name = tensor("op_1425_cast_fp16")]; tensor var_1427_equation_0 = const()[name = tensor("op_1427_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_1427_cast_fp16 = einsum(equation = var_1427_equation_0, values = (var_1047_cast_fp16, var_1344_cast_fp16))[name = tensor("op_1427_cast_fp16")]; tensor var_1429_equation_0 = const()[name = tensor("op_1429_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_1429_cast_fp16 = einsum(equation = var_1429_equation_0, values = (var_1047_cast_fp16, var_1345_cast_fp16))[name = tensor("op_1429_cast_fp16")]; tensor var_1431_equation_0 = const()[name = tensor("op_1431_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_1431_cast_fp16 = einsum(equation = var_1431_equation_0, values = (var_1047_cast_fp16, var_1346_cast_fp16))[name = tensor("op_1431_cast_fp16")]; tensor var_1433_equation_0 = const()[name = tensor("op_1433_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_1433_cast_fp16 = einsum(equation = var_1433_equation_0, values = (var_1047_cast_fp16, var_1347_cast_fp16))[name = tensor("op_1433_cast_fp16")]; tensor var_1435_equation_0 = const()[name = tensor("op_1435_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_1435_cast_fp16 = einsum(equation = var_1435_equation_0, values = (var_1047_cast_fp16, var_1348_cast_fp16))[name = tensor("op_1435_cast_fp16")]; tensor var_1437_equation_0 = const()[name = tensor("op_1437_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_1437_cast_fp16 = einsum(equation = var_1437_equation_0, values = (var_1047_cast_fp16, var_1349_cast_fp16))[name = tensor("op_1437_cast_fp16")]; tensor var_1439_equation_0 = const()[name = tensor("op_1439_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_1439_cast_fp16 = einsum(equation = var_1439_equation_0, values = (var_1051_cast_fp16, var_1350_cast_fp16))[name = tensor("op_1439_cast_fp16")]; tensor var_1441_equation_0 = const()[name = tensor("op_1441_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_1441_cast_fp16 = einsum(equation = var_1441_equation_0, values = (var_1051_cast_fp16, var_1351_cast_fp16))[name = tensor("op_1441_cast_fp16")]; tensor var_1443_equation_0 = const()[name = tensor("op_1443_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_1443_cast_fp16 = einsum(equation = var_1443_equation_0, values = (var_1051_cast_fp16, var_1352_cast_fp16))[name = tensor("op_1443_cast_fp16")]; tensor var_1445_equation_0 = const()[name = tensor("op_1445_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_1445_cast_fp16 = einsum(equation = var_1445_equation_0, values = (var_1051_cast_fp16, var_1353_cast_fp16))[name = tensor("op_1445_cast_fp16")]; tensor var_1447_equation_0 = const()[name = tensor("op_1447_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_1447_cast_fp16 = einsum(equation = var_1447_equation_0, values = (var_1051_cast_fp16, var_1354_cast_fp16))[name = tensor("op_1447_cast_fp16")]; tensor var_1449_equation_0 = const()[name = tensor("op_1449_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_1449_cast_fp16 = einsum(equation = var_1449_equation_0, values = (var_1051_cast_fp16, var_1355_cast_fp16))[name = tensor("op_1449_cast_fp16")]; tensor var_1451_equation_0 = const()[name = tensor("op_1451_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_1451_cast_fp16 = einsum(equation = var_1451_equation_0, values = (var_1051_cast_fp16, var_1356_cast_fp16))[name = tensor("op_1451_cast_fp16")]; tensor var_1453_equation_0 = const()[name = tensor("op_1453_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_1453_cast_fp16 = einsum(equation = var_1453_equation_0, values = (var_1051_cast_fp16, var_1357_cast_fp16))[name = tensor("op_1453_cast_fp16")]; tensor var_1455_equation_0 = const()[name = tensor("op_1455_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_1455_cast_fp16 = einsum(equation = var_1455_equation_0, values = (var_1055_cast_fp16, var_1358_cast_fp16))[name = tensor("op_1455_cast_fp16")]; tensor var_1457_equation_0 = const()[name = tensor("op_1457_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_1457_cast_fp16 = einsum(equation = var_1457_equation_0, values = (var_1055_cast_fp16, var_1359_cast_fp16))[name = tensor("op_1457_cast_fp16")]; tensor var_1459_equation_0 = const()[name = tensor("op_1459_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_1459_cast_fp16 = einsum(equation = var_1459_equation_0, values = (var_1055_cast_fp16, var_1360_cast_fp16))[name = tensor("op_1459_cast_fp16")]; tensor var_1461_equation_0 = const()[name = tensor("op_1461_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_1461_cast_fp16 = einsum(equation = var_1461_equation_0, values = (var_1055_cast_fp16, var_1361_cast_fp16))[name = tensor("op_1461_cast_fp16")]; tensor var_1463_equation_0 = const()[name = tensor("op_1463_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_1463_cast_fp16 = einsum(equation = var_1463_equation_0, values = (var_1055_cast_fp16, var_1362_cast_fp16))[name = tensor("op_1463_cast_fp16")]; tensor var_1465_equation_0 = const()[name = tensor("op_1465_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_1465_cast_fp16 = einsum(equation = var_1465_equation_0, values = (var_1055_cast_fp16, var_1363_cast_fp16))[name = tensor("op_1465_cast_fp16")]; tensor var_1467_equation_0 = const()[name = tensor("op_1467_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_1467_cast_fp16 = einsum(equation = var_1467_equation_0, values = (var_1055_cast_fp16, var_1364_cast_fp16))[name = tensor("op_1467_cast_fp16")]; tensor var_1469_equation_0 = const()[name = tensor("op_1469_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_1469_cast_fp16 = einsum(equation = var_1469_equation_0, values = (var_1055_cast_fp16, var_1365_cast_fp16))[name = tensor("op_1469_cast_fp16")]; tensor var_1471_equation_0 = const()[name = tensor("op_1471_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_1471_cast_fp16 = einsum(equation = var_1471_equation_0, values = (var_1059_cast_fp16, var_1366_cast_fp16))[name = tensor("op_1471_cast_fp16")]; tensor var_1473_equation_0 = const()[name = tensor("op_1473_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_1473_cast_fp16 = einsum(equation = var_1473_equation_0, values = (var_1059_cast_fp16, var_1367_cast_fp16))[name = tensor("op_1473_cast_fp16")]; tensor var_1475_equation_0 = const()[name = tensor("op_1475_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_1475_cast_fp16 = einsum(equation = var_1475_equation_0, values = (var_1059_cast_fp16, var_1368_cast_fp16))[name = tensor("op_1475_cast_fp16")]; tensor var_1477_equation_0 = const()[name = tensor("op_1477_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_1477_cast_fp16 = einsum(equation = var_1477_equation_0, values = (var_1059_cast_fp16, var_1369_cast_fp16))[name = tensor("op_1477_cast_fp16")]; tensor var_1479_equation_0 = const()[name = tensor("op_1479_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_1479_cast_fp16 = einsum(equation = var_1479_equation_0, values = (var_1059_cast_fp16, var_1370_cast_fp16))[name = tensor("op_1479_cast_fp16")]; tensor var_1481_equation_0 = const()[name = tensor("op_1481_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_1481_cast_fp16 = einsum(equation = var_1481_equation_0, values = (var_1059_cast_fp16, var_1371_cast_fp16))[name = tensor("op_1481_cast_fp16")]; tensor var_1483_equation_0 = const()[name = tensor("op_1483_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_1483_cast_fp16 = einsum(equation = var_1483_equation_0, values = (var_1059_cast_fp16, var_1372_cast_fp16))[name = tensor("op_1483_cast_fp16")]; tensor var_1485_equation_0 = const()[name = tensor("op_1485_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_1485_cast_fp16 = einsum(equation = var_1485_equation_0, values = (var_1059_cast_fp16, var_1373_cast_fp16))[name = tensor("op_1485_cast_fp16")]; tensor var_1487_equation_0 = const()[name = tensor("op_1487_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_1487_cast_fp16 = einsum(equation = var_1487_equation_0, values = (var_1063_cast_fp16, var_1374_cast_fp16))[name = tensor("op_1487_cast_fp16")]; tensor var_1489_equation_0 = const()[name = tensor("op_1489_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_1489_cast_fp16 = einsum(equation = var_1489_equation_0, values = (var_1063_cast_fp16, var_1375_cast_fp16))[name = tensor("op_1489_cast_fp16")]; tensor var_1491_equation_0 = const()[name = tensor("op_1491_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_1491_cast_fp16 = einsum(equation = var_1491_equation_0, values = (var_1063_cast_fp16, var_1376_cast_fp16))[name = tensor("op_1491_cast_fp16")]; tensor var_1493_equation_0 = const()[name = tensor("op_1493_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_1493_cast_fp16 = einsum(equation = var_1493_equation_0, values = (var_1063_cast_fp16, var_1377_cast_fp16))[name = tensor("op_1493_cast_fp16")]; tensor var_1495_equation_0 = const()[name = tensor("op_1495_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_1495_cast_fp16 = einsum(equation = var_1495_equation_0, values = (var_1063_cast_fp16, var_1378_cast_fp16))[name = tensor("op_1495_cast_fp16")]; tensor var_1497_equation_0 = const()[name = tensor("op_1497_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_1497_cast_fp16 = einsum(equation = var_1497_equation_0, values = (var_1063_cast_fp16, var_1379_cast_fp16))[name = tensor("op_1497_cast_fp16")]; tensor var_1499_equation_0 = const()[name = tensor("op_1499_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_1499_cast_fp16 = einsum(equation = var_1499_equation_0, values = (var_1063_cast_fp16, var_1380_cast_fp16))[name = tensor("op_1499_cast_fp16")]; tensor var_1501_equation_0 = const()[name = tensor("op_1501_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_1501_cast_fp16 = einsum(equation = var_1501_equation_0, values = (var_1063_cast_fp16, var_1381_cast_fp16))[name = tensor("op_1501_cast_fp16")]; tensor var_1503_equation_0 = const()[name = tensor("op_1503_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_1503_cast_fp16 = einsum(equation = var_1503_equation_0, values = (var_1067_cast_fp16, var_1382_cast_fp16))[name = tensor("op_1503_cast_fp16")]; tensor var_1505_equation_0 = const()[name = tensor("op_1505_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_1505_cast_fp16 = einsum(equation = var_1505_equation_0, values = (var_1067_cast_fp16, var_1383_cast_fp16))[name = tensor("op_1505_cast_fp16")]; tensor var_1507_equation_0 = const()[name = tensor("op_1507_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_1507_cast_fp16 = einsum(equation = var_1507_equation_0, values = (var_1067_cast_fp16, var_1384_cast_fp16))[name = tensor("op_1507_cast_fp16")]; tensor var_1509_equation_0 = const()[name = tensor("op_1509_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_1509_cast_fp16 = einsum(equation = var_1509_equation_0, values = (var_1067_cast_fp16, var_1385_cast_fp16))[name = tensor("op_1509_cast_fp16")]; tensor var_1511_equation_0 = const()[name = tensor("op_1511_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_1511_cast_fp16 = einsum(equation = var_1511_equation_0, values = (var_1067_cast_fp16, var_1386_cast_fp16))[name = tensor("op_1511_cast_fp16")]; tensor var_1513_equation_0 = const()[name = tensor("op_1513_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_1513_cast_fp16 = einsum(equation = var_1513_equation_0, values = (var_1067_cast_fp16, var_1387_cast_fp16))[name = tensor("op_1513_cast_fp16")]; tensor var_1515_equation_0 = const()[name = tensor("op_1515_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_1515_cast_fp16 = einsum(equation = var_1515_equation_0, values = (var_1067_cast_fp16, var_1388_cast_fp16))[name = tensor("op_1515_cast_fp16")]; tensor var_1517_equation_0 = const()[name = tensor("op_1517_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_1517_cast_fp16 = einsum(equation = var_1517_equation_0, values = (var_1067_cast_fp16, var_1389_cast_fp16))[name = tensor("op_1517_cast_fp16")]; tensor var_1519_interleave_0 = const()[name = tensor("op_1519_interleave_0"), val = tensor(false)]; tensor var_1519_cast_fp16 = concat(axis = var_90, interleave = var_1519_interleave_0, values = (var_1391_cast_fp16, var_1393_cast_fp16, var_1395_cast_fp16, var_1397_cast_fp16, var_1399_cast_fp16, var_1401_cast_fp16, var_1403_cast_fp16, var_1405_cast_fp16))[name = tensor("op_1519_cast_fp16")]; tensor var_1521_interleave_0 = const()[name = tensor("op_1521_interleave_0"), val = tensor(false)]; tensor var_1521_cast_fp16 = concat(axis = var_90, interleave = var_1521_interleave_0, values = (var_1407_cast_fp16, var_1409_cast_fp16, var_1411_cast_fp16, var_1413_cast_fp16, var_1415_cast_fp16, var_1417_cast_fp16, var_1419_cast_fp16, var_1421_cast_fp16))[name = tensor("op_1521_cast_fp16")]; tensor var_1523_interleave_0 = const()[name = tensor("op_1523_interleave_0"), val = tensor(false)]; tensor var_1523_cast_fp16 = concat(axis = var_90, interleave = var_1523_interleave_0, values = (var_1423_cast_fp16, var_1425_cast_fp16, var_1427_cast_fp16, var_1429_cast_fp16, var_1431_cast_fp16, var_1433_cast_fp16, var_1435_cast_fp16, var_1437_cast_fp16))[name = tensor("op_1523_cast_fp16")]; tensor var_1525_interleave_0 = const()[name = tensor("op_1525_interleave_0"), val = tensor(false)]; tensor var_1525_cast_fp16 = concat(axis = var_90, interleave = var_1525_interleave_0, values = (var_1439_cast_fp16, var_1441_cast_fp16, var_1443_cast_fp16, var_1445_cast_fp16, var_1447_cast_fp16, var_1449_cast_fp16, var_1451_cast_fp16, var_1453_cast_fp16))[name = tensor("op_1525_cast_fp16")]; tensor var_1527_interleave_0 = const()[name = tensor("op_1527_interleave_0"), val = tensor(false)]; tensor var_1527_cast_fp16 = concat(axis = var_90, interleave = var_1527_interleave_0, values = (var_1455_cast_fp16, var_1457_cast_fp16, var_1459_cast_fp16, var_1461_cast_fp16, var_1463_cast_fp16, var_1465_cast_fp16, var_1467_cast_fp16, var_1469_cast_fp16))[name = tensor("op_1527_cast_fp16")]; tensor var_1529_interleave_0 = const()[name = tensor("op_1529_interleave_0"), val = tensor(false)]; tensor var_1529_cast_fp16 = concat(axis = var_90, interleave = var_1529_interleave_0, values = (var_1471_cast_fp16, var_1473_cast_fp16, var_1475_cast_fp16, var_1477_cast_fp16, var_1479_cast_fp16, var_1481_cast_fp16, var_1483_cast_fp16, var_1485_cast_fp16))[name = tensor("op_1529_cast_fp16")]; tensor var_1531_interleave_0 = const()[name = tensor("op_1531_interleave_0"), val = tensor(false)]; tensor var_1531_cast_fp16 = concat(axis = var_90, interleave = var_1531_interleave_0, values = (var_1487_cast_fp16, var_1489_cast_fp16, var_1491_cast_fp16, var_1493_cast_fp16, var_1495_cast_fp16, var_1497_cast_fp16, var_1499_cast_fp16, var_1501_cast_fp16))[name = tensor("op_1531_cast_fp16")]; tensor var_1533_interleave_0 = const()[name = tensor("op_1533_interleave_0"), val = tensor(false)]; tensor var_1533_cast_fp16 = concat(axis = var_90, interleave = var_1533_interleave_0, values = (var_1503_cast_fp16, var_1505_cast_fp16, var_1507_cast_fp16, var_1509_cast_fp16, var_1511_cast_fp16, var_1513_cast_fp16, var_1515_cast_fp16, var_1517_cast_fp16))[name = tensor("op_1533_cast_fp16")]; tensor input_27_interleave_0 = const()[name = tensor("input_27_interleave_0"), val = tensor(false)]; tensor input_27_cast_fp16 = concat(axis = var_118, interleave = input_27_interleave_0, values = (var_1519_cast_fp16, var_1521_cast_fp16, var_1523_cast_fp16, var_1525_cast_fp16, var_1527_cast_fp16, var_1529_cast_fp16, var_1531_cast_fp16, var_1533_cast_fp16))[name = tensor("input_27_cast_fp16")]; tensor var_1539 = const()[name = tensor("op_1539"), val = tensor([1, 1])]; tensor var_1541 = const()[name = tensor("op_1541"), val = tensor([1, 1])]; tensor var_1543_pad_type_0 = const()[name = tensor("op_1543_pad_type_0"), val = tensor("custom")]; tensor var_1543_pad_0 = const()[name = tensor("op_1543_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(4085632))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4162496))), 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(4162688)))]; tensor var_1543_cast_fp16 = conv(bias = down_blocks_0_attentions_0_transformer_blocks_0_attn2_to_out_0_bias_to_fp16, dilations = var_1541, groups = var_118, pad = var_1543_pad_0, pad_type = var_1543_pad_type_0, strides = var_1539, weight = down_blocks_0_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16_palettized, x = input_27_cast_fp16)[name = tensor("op_1543_cast_fp16")]; tensor inputs_5_cast_fp16 = add(x = var_1543_cast_fp16, y = inputs_3_cast_fp16)[name = tensor("inputs_5_cast_fp16")]; tensor var_1547 = const()[name = tensor("op_1547"), val = tensor([1])]; tensor channels_mean_5_cast_fp16 = reduce_mean(axes = var_1547, keep_dims = var_113, x = inputs_5_cast_fp16)[name = tensor("channels_mean_5_cast_fp16")]; tensor zero_mean_5_cast_fp16 = sub(x = inputs_5_cast_fp16, y = channels_mean_5_cast_fp16)[name = tensor("zero_mean_5_cast_fp16")]; tensor zero_mean_sq_5_cast_fp16 = mul(x = zero_mean_5_cast_fp16, y = zero_mean_5_cast_fp16)[name = tensor("zero_mean_sq_5_cast_fp16")]; tensor var_1551 = const()[name = tensor("op_1551"), val = tensor([1])]; tensor var_1552_cast_fp16 = reduce_mean(axes = var_1551, keep_dims = var_113, x = zero_mean_sq_5_cast_fp16)[name = tensor("op_1552_cast_fp16")]; tensor var_1553_to_fp16 = const()[name = tensor("op_1553_to_fp16"), val = tensor(0x1.5p-17)]; tensor var_1554_cast_fp16 = add(x = var_1552_cast_fp16, y = var_1553_to_fp16)[name = tensor("op_1554_cast_fp16")]; tensor denom_5_epsilon_0_to_fp16 = const()[name = tensor("denom_5_epsilon_0_to_fp16"), val = tensor(0x1p-24)]; tensor denom_5_cast_fp16 = rsqrt(epsilon = denom_5_epsilon_0_to_fp16, x = var_1554_cast_fp16)[name = tensor("denom_5_cast_fp16")]; tensor out_5_cast_fp16 = mul(x = zero_mean_5_cast_fp16, y = denom_5_cast_fp16)[name = tensor("out_5_cast_fp16")]; tensor var_1558_to_fp16 = const()[name = tensor("op_1558_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4163392)))]; tensor var_1559_cast_fp16 = add(x = out_5_cast_fp16, y = var_1558_to_fp16)[name = tensor("op_1559_cast_fp16")]; tensor var_1561_to_fp16 = const()[name = tensor("op_1561_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4164096)))]; tensor input_29_cast_fp16 = mul(x = var_1559_cast_fp16, y = var_1561_to_fp16)[name = tensor("input_29_cast_fp16")]; tensor var_1569 = const()[name = tensor("op_1569"), val = tensor([1, 1])]; tensor var_1571 = const()[name = tensor("op_1571"), val = tensor([1, 1])]; tensor var_1573_pad_type_0 = const()[name = tensor("op_1573_pad_type_0"), val = tensor("custom")]; tensor var_1573_pad_0 = const()[name = tensor("op_1573_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(4164800))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4779264))), 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(4779456))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4781440))), name = tensor("down_blocks_0_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16_palettized"), shape = tensor([2560])]; tensor var_1573_cast_fp16 = conv(bias = down_blocks_0_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16_palettized, dilations = var_1571, groups = var_118, pad = var_1573_pad_0, pad_type = var_1573_pad_type_0, strides = var_1569, weight = down_blocks_0_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16_palettized, x = input_29_cast_fp16)[name = tensor("op_1573_cast_fp16")]; tensor var_1574_split_sizes_0 = const()[name = tensor("op_1574_split_sizes_0"), val = tensor([1280, 1280])]; tensor var_1574_axis_0 = const()[name = tensor("op_1574_axis_0"), val = tensor(1)]; tensor var_1574_cast_fp16_0, tensor var_1574_cast_fp16_1 = split(axis = var_1574_axis_0, split_sizes = var_1574_split_sizes_0, x = var_1573_cast_fp16)[name = tensor("op_1574_cast_fp16")]; tensor var_1576_mode_0 = const()[name = tensor("op_1576_mode_0"), val = tensor("EXACT")]; tensor var_1576_cast_fp16 = gelu(mode = var_1576_mode_0, x = var_1574_cast_fp16_1)[name = tensor("op_1576_cast_fp16")]; tensor input_31_cast_fp16 = mul(x = var_1574_cast_fp16_0, y = var_1576_cast_fp16)[name = tensor("input_31_cast_fp16")]; tensor var_1580 = const()[name = tensor("op_1580"), val = tensor([1, 1])]; tensor var_1582 = const()[name = tensor("op_1582"), val = tensor([1, 1])]; tensor var_1584_pad_type_0 = const()[name = tensor("op_1584_pad_type_0"), val = tensor("custom")]; tensor var_1584_pad_0 = const()[name = tensor("op_1584_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(4781632))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5088896))), 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(5089088)))]; tensor var_1584_cast_fp16 = conv(bias = down_blocks_0_attentions_0_transformer_blocks_0_ff_net_2_bias_to_fp16, dilations = var_1582, groups = var_118, pad = var_1584_pad_0, pad_type = var_1584_pad_type_0, strides = var_1580, weight = down_blocks_0_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16_palettized, x = input_31_cast_fp16)[name = tensor("op_1584_cast_fp16")]; tensor hidden_states_15_cast_fp16 = add(x = var_1584_cast_fp16, y = inputs_5_cast_fp16)[name = tensor("hidden_states_15_cast_fp16")]; tensor var_1586 = const()[name = tensor("op_1586"), val = tensor([2, 320, 64, 64])]; tensor input_33_cast_fp16 = reshape(shape = var_1586, x = hidden_states_15_cast_fp16)[name = tensor("input_33_cast_fp16")]; tensor var_1590 = const()[name = tensor("op_1590"), val = tensor([1, 1])]; tensor var_1592 = const()[name = tensor("op_1592"), 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(5089792))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5166656))), 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(5166848)))]; tensor hidden_states_17_cast_fp16 = conv(bias = down_blocks_0_attentions_0_proj_out_bias_to_fp16, dilations = var_1592, groups = var_118, pad = hidden_states_17_pad_0, pad_type = hidden_states_17_pad_type_0, strides = var_1590, weight = down_blocks_0_attentions_0_proj_out_weight_to_fp16_palettized, x = input_33_cast_fp16)[name = tensor("hidden_states_17_cast_fp16")]; tensor input_35_cast_fp16 = add(x = hidden_states_17_cast_fp16, y = hidden_states_5_cast_fp16)[name = tensor("input_35_cast_fp16")]; tensor var_1599 = const()[name = tensor("op_1599"), val = tensor([2, 2])]; tensor var_1601 = const()[name = tensor("op_1601"), val = tensor([1, 1])]; tensor input_37_pad_type_0 = const()[name = tensor("input_37_pad_type_0"), val = tensor("custom")]; tensor input_37_pad_0 = const()[name = tensor("input_37_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(5167552))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5858816))), 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(5859008)))]; tensor input_37_cast_fp16 = conv(bias = down_blocks_0_downsamplers_0_conv_bias_to_fp16, dilations = var_1601, groups = var_118, pad = input_37_pad_0, pad_type = input_37_pad_type_0, strides = var_1599, weight = down_blocks_0_downsamplers_0_conv_weight_to_fp16_palettized, x = input_35_cast_fp16)[name = tensor("input_37_cast_fp16")]; tensor var_1608 = const()[name = tensor("op_1608"), val = tensor(3)]; tensor var_1625 = const()[name = tensor("op_1625"), val = tensor(true)]; tensor var_1630 = const()[name = tensor("op_1630"), val = tensor(1)]; tensor reshape_12_shape_0 = const()[name = tensor("reshape_12_shape_0"), val = tensor([2, 32, 10, 32, 32])]; tensor reshape_12_cast_fp16 = reshape(shape = reshape_12_shape_0, x = input_37_cast_fp16)[name = tensor("reshape_12_cast_fp16")]; 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_fp16 = reduce_mean(axes = reduce_mean_9_axes_0, keep_dims = reduce_mean_9_keep_dims_0, x = reshape_12_cast_fp16)[name = tensor("reduce_mean_9_cast_fp16")]; tensor sub_6_cast_fp16 = sub(x = reshape_12_cast_fp16, y = reduce_mean_9_cast_fp16)[name = tensor("sub_6_cast_fp16")]; tensor square_3_cast_fp16 = square(x = sub_6_cast_fp16)[name = tensor("square_3_cast_fp16")]; 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_fp16 = reduce_mean(axes = reduce_mean_11_axes_0, keep_dims = reduce_mean_11_keep_dims_0, x = square_3_cast_fp16)[name = tensor("reduce_mean_11_cast_fp16")]; 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_fp16 = add(x = reduce_mean_11_cast_fp16, y = add_6_y_0_to_fp16)[name = tensor("add_6_cast_fp16")]; tensor sqrt_3_cast_fp16 = sqrt(x = add_6_cast_fp16)[name = tensor("sqrt_3_cast_fp16")]; tensor real_div_3_cast_fp16 = real_div(x = sub_6_cast_fp16, y = sqrt_3_cast_fp16)[name = tensor("real_div_3_cast_fp16")]; tensor reshape_13_shape_0 = const()[name = tensor("reshape_13_shape_0"), val = tensor([2, 320, 32, 32])]; tensor reshape_13_cast_fp16 = reshape(shape = reshape_13_shape_0, x = real_div_3_cast_fp16)[name = tensor("reshape_13_cast_fp16")]; 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(5859712)))]; 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(5860416)))]; 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_fp16 = 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_fp16)[name = tensor("add_7_cast_fp16")]; tensor input_41_cast_fp16 = silu(x = add_7_cast_fp16)[name = tensor("input_41_cast_fp16")]; tensor var_1649 = const()[name = tensor("op_1649"), val = tensor([1, 1])]; tensor var_1651 = const()[name = tensor("op_1651"), 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_1_resnets_0_conv1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5861120))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(7243584))), 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(7243776)))]; tensor hidden_states_19_cast_fp16 = conv(bias = down_blocks_1_resnets_0_conv1_bias_to_fp16, dilations = var_1651, groups = var_1630, pad = hidden_states_19_pad_0, pad_type = hidden_states_19_pad_type_0, strides = var_1649, weight = down_blocks_1_resnets_0_conv1_weight_to_fp16_palettized, x = input_41_cast_fp16)[name = tensor("hidden_states_19_cast_fp16")]; tensor var_1657 = const()[name = tensor("op_1657"), val = tensor([1, 1])]; tensor var_1659 = const()[name = tensor("op_1659"), 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_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(7245120))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(7859584))), 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(7859776)))]; tensor temb_3_cast_fp16 = conv(bias = down_blocks_1_resnets_0_time_emb_proj_bias_to_fp16, dilations = var_1659, groups = var_1630, pad = temb_3_pad_0, pad_type = temb_3_pad_type_0, strides = var_1657, weight = down_blocks_1_resnets_0_time_emb_proj_weight_to_fp16_palettized, x = input_15_cast_fp16)[name = tensor("temb_3_cast_fp16")]; tensor input_45_cast_fp16 = add(x = hidden_states_19_cast_fp16, y = temb_3_cast_fp16)[name = tensor("input_45_cast_fp16")]; tensor reshape_16_shape_0 = const()[name = tensor("reshape_16_shape_0"), val = tensor([2, 32, 20, 32, 32])]; tensor reshape_16_cast_fp16 = reshape(shape = reshape_16_shape_0, x = input_45_cast_fp16)[name = tensor("reshape_16_cast_fp16")]; 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_fp16 = reduce_mean(axes = reduce_mean_12_axes_0, keep_dims = reduce_mean_12_keep_dims_0, x = reshape_16_cast_fp16)[name = tensor("reduce_mean_12_cast_fp16")]; tensor sub_8_cast_fp16 = sub(x = reshape_16_cast_fp16, y = reduce_mean_12_cast_fp16)[name = tensor("sub_8_cast_fp16")]; tensor square_4_cast_fp16 = square(x = sub_8_cast_fp16)[name = tensor("square_4_cast_fp16")]; 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_fp16 = reduce_mean(axes = reduce_mean_14_axes_0, keep_dims = reduce_mean_14_keep_dims_0, x = square_4_cast_fp16)[name = tensor("reduce_mean_14_cast_fp16")]; 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_fp16 = add(x = reduce_mean_14_cast_fp16, y = add_8_y_0_to_fp16)[name = tensor("add_8_cast_fp16")]; tensor sqrt_4_cast_fp16 = sqrt(x = add_8_cast_fp16)[name = tensor("sqrt_4_cast_fp16")]; tensor real_div_4_cast_fp16 = real_div(x = sub_8_cast_fp16, y = sqrt_4_cast_fp16)[name = tensor("real_div_4_cast_fp16")]; tensor reshape_17_shape_0 = const()[name = tensor("reshape_17_shape_0"), val = tensor([2, 640, 32, 32])]; tensor reshape_17_cast_fp16 = reshape(shape = reshape_17_shape_0, x = real_div_4_cast_fp16)[name = tensor("reshape_17_cast_fp16")]; tensor add_9_mean_0_to_fp16 = const()[name = tensor("add_9_mean_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(7861120)))]; tensor add_9_variance_0_to_fp16 = const()[name = tensor("add_9_variance_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(7862464)))]; 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(7863808)))]; 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(7865152)))]; 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_fp16 = 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_9_mean_0_to_fp16, variance = add_9_variance_0_to_fp16, x = reshape_17_cast_fp16)[name = tensor("add_9_cast_fp16")]; tensor input_49_cast_fp16 = silu(x = add_9_cast_fp16)[name = tensor("input_49_cast_fp16")]; tensor var_1669 = const()[name = tensor("op_1669"), val = tensor([1, 1])]; tensor var_1671 = const()[name = tensor("op_1671"), 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_1_resnets_0_conv2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(7866496))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(10631360))), 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(10631552)))]; tensor hidden_states_21_cast_fp16 = conv(bias = down_blocks_1_resnets_0_conv2_bias_to_fp16, dilations = var_1671, groups = var_1630, pad = hidden_states_21_pad_0, pad_type = hidden_states_21_pad_type_0, strides = var_1669, weight = down_blocks_1_resnets_0_conv2_weight_to_fp16_palettized, x = input_49_cast_fp16)[name = tensor("hidden_states_21_cast_fp16")]; tensor var_1676 = const()[name = tensor("op_1676"), val = tensor([1, 1])]; tensor var_1678 = const()[name = tensor("op_1678"), 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(10632896))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(10786560))), 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(10786752)))]; tensor x_1_cast_fp16 = conv(bias = down_blocks_1_resnets_0_conv_shortcut_bias_to_fp16, dilations = var_1678, groups = var_1630, pad = x_1_pad_0, pad_type = x_1_pad_type_0, strides = var_1676, weight = down_blocks_1_resnets_0_conv_shortcut_weight_to_fp16_palettized, x = input_37_cast_fp16)[name = tensor("x_1_cast_fp16")]; tensor hidden_states_23_cast_fp16 = add(x = x_1_cast_fp16, y = hidden_states_21_cast_fp16)[name = tensor("hidden_states_23_cast_fp16")]; tensor reshape_20_shape_0 = const()[name = tensor("reshape_20_shape_0"), val = tensor([2, 32, 20, 32, 32])]; tensor reshape_20_cast_fp16 = reshape(shape = reshape_20_shape_0, x = hidden_states_23_cast_fp16)[name = tensor("reshape_20_cast_fp16")]; 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_fp16 = reduce_mean(axes = reduce_mean_15_axes_0, keep_dims = reduce_mean_15_keep_dims_0, x = reshape_20_cast_fp16)[name = tensor("reduce_mean_15_cast_fp16")]; tensor sub_10_cast_fp16 = sub(x = reshape_20_cast_fp16, y = reduce_mean_15_cast_fp16)[name = tensor("sub_10_cast_fp16")]; tensor square_5_cast_fp16 = square(x = sub_10_cast_fp16)[name = tensor("square_5_cast_fp16")]; 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_fp16 = reduce_mean(axes = reduce_mean_17_axes_0, keep_dims = reduce_mean_17_keep_dims_0, x = square_5_cast_fp16)[name = tensor("reduce_mean_17_cast_fp16")]; 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_fp16 = add(x = reduce_mean_17_cast_fp16, y = add_10_y_0_to_fp16)[name = tensor("add_10_cast_fp16")]; tensor sqrt_5_cast_fp16 = sqrt(x = add_10_cast_fp16)[name = tensor("sqrt_5_cast_fp16")]; tensor real_div_5_cast_fp16 = real_div(x = sub_10_cast_fp16, y = sqrt_5_cast_fp16)[name = tensor("real_div_5_cast_fp16")]; tensor reshape_21_shape_0 = const()[name = tensor("reshape_21_shape_0"), val = tensor([2, 640, 32, 32])]; tensor reshape_21_cast_fp16 = reshape(shape = reshape_21_shape_0, x = real_div_5_cast_fp16)[name = tensor("reshape_21_cast_fp16")]; 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(10788096)))]; 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(10789440)))]; 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_fp16 = 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_9_mean_0_to_fp16, variance = add_9_variance_0_to_fp16, x = reshape_21_cast_fp16)[name = tensor("add_11_cast_fp16")]; tensor var_1698 = const()[name = tensor("op_1698"), val = tensor([1, 1])]; tensor var_1700 = const()[name = tensor("op_1700"), 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_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(10790784))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(11098048))), 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(11098240)))]; tensor hidden_states_25_cast_fp16 = conv(bias = down_blocks_1_attentions_0_proj_in_bias_to_fp16, dilations = var_1700, groups = var_1630, pad = hidden_states_25_pad_0, pad_type = hidden_states_25_pad_type_0, strides = var_1698, weight = down_blocks_1_attentions_0_proj_in_weight_to_fp16_palettized, x = add_11_cast_fp16)[name = tensor("hidden_states_25_cast_fp16")]; tensor var_1705 = const()[name = tensor("op_1705"), val = tensor([2, 640, 1, 1024])]; tensor inputs_7_cast_fp16 = reshape(shape = var_1705, x = hidden_states_25_cast_fp16)[name = tensor("inputs_7_cast_fp16")]; tensor var_1715 = const()[name = tensor("op_1715"), val = tensor([1])]; tensor channels_mean_7_cast_fp16 = reduce_mean(axes = var_1715, keep_dims = var_1625, x = inputs_7_cast_fp16)[name = tensor("channels_mean_7_cast_fp16")]; tensor zero_mean_7_cast_fp16 = sub(x = inputs_7_cast_fp16, y = channels_mean_7_cast_fp16)[name = tensor("zero_mean_7_cast_fp16")]; tensor zero_mean_sq_7_cast_fp16 = mul(x = zero_mean_7_cast_fp16, y = zero_mean_7_cast_fp16)[name = tensor("zero_mean_sq_7_cast_fp16")]; tensor var_1719 = const()[name = tensor("op_1719"), val = tensor([1])]; tensor var_1720_cast_fp16 = reduce_mean(axes = var_1719, keep_dims = var_1625, x = zero_mean_sq_7_cast_fp16)[name = tensor("op_1720_cast_fp16")]; tensor var_1721_to_fp16 = const()[name = tensor("op_1721_to_fp16"), val = tensor(0x1.5p-17)]; tensor var_1722_cast_fp16 = add(x = var_1720_cast_fp16, y = var_1721_to_fp16)[name = tensor("op_1722_cast_fp16")]; tensor denom_7_epsilon_0_to_fp16 = const()[name = tensor("denom_7_epsilon_0_to_fp16"), val = tensor(0x1p-24)]; tensor denom_7_cast_fp16 = rsqrt(epsilon = denom_7_epsilon_0_to_fp16, x = var_1722_cast_fp16)[name = tensor("denom_7_cast_fp16")]; tensor out_7_cast_fp16 = mul(x = zero_mean_7_cast_fp16, y = denom_7_cast_fp16)[name = tensor("out_7_cast_fp16")]; tensor var_1726_to_fp16 = const()[name = tensor("op_1726_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(11099584)))]; tensor var_1727_cast_fp16 = add(x = out_7_cast_fp16, y = var_1726_to_fp16)[name = tensor("op_1727_cast_fp16")]; tensor var_1729_to_fp16 = const()[name = tensor("op_1729_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(11100928)))]; tensor hidden_states_27_cast_fp16 = mul(x = var_1727_cast_fp16, y = var_1729_to_fp16)[name = tensor("hidden_states_27_cast_fp16")]; tensor var_1736 = const()[name = tensor("op_1736"), val = tensor([1, 1])]; tensor var_1738 = const()[name = tensor("op_1738"), 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_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(11102272))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(11409536))), 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_5_cast_fp16 = conv(dilations = var_1738, groups = var_1630, pad = q_5_pad_0, pad_type = q_5_pad_type_0, strides = var_1736, weight = down_blocks_1_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16_palettized, x = hidden_states_27_cast_fp16)[name = tensor("q_5_cast_fp16")]; tensor var_1742 = const()[name = tensor("op_1742"), val = tensor([1, 1])]; tensor var_1744 = const()[name = tensor("op_1744"), 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_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(11409728))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(11716992))), 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_9_cast_fp16 = conv(dilations = var_1744, groups = var_1630, pad = k_9_pad_0, pad_type = k_9_pad_type_0, strides = var_1742, weight = down_blocks_1_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16_palettized, x = hidden_states_27_cast_fp16)[name = tensor("k_9_cast_fp16")]; tensor var_1748 = const()[name = tensor("op_1748"), val = tensor([1, 1])]; tensor var_1750 = const()[name = tensor("op_1750"), 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_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(11717184))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(12024448))), 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_5_cast_fp16 = conv(dilations = var_1750, groups = var_1630, pad = v_5_pad_0, pad_type = v_5_pad_type_0, strides = var_1748, weight = down_blocks_1_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16_palettized, x = hidden_states_27_cast_fp16)[name = tensor("v_5_cast_fp16")]; tensor var_1754_begin_0 = const()[name = tensor("op_1754_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_1754_end_0 = const()[name = tensor("op_1754_end_0"), val = tensor([2, 80, 1, 1024])]; tensor var_1754_end_mask_0 = const()[name = tensor("op_1754_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_1754_cast_fp16 = slice_by_index(begin = var_1754_begin_0, end = var_1754_end_0, end_mask = var_1754_end_mask_0, x = q_5_cast_fp16)[name = tensor("op_1754_cast_fp16")]; tensor var_1758_begin_0 = const()[name = tensor("op_1758_begin_0"), val = tensor([0, 80, 0, 0])]; tensor var_1758_end_0 = const()[name = tensor("op_1758_end_0"), val = tensor([2, 160, 1, 1024])]; tensor var_1758_end_mask_0 = const()[name = tensor("op_1758_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_1758_cast_fp16 = slice_by_index(begin = var_1758_begin_0, end = var_1758_end_0, end_mask = var_1758_end_mask_0, x = q_5_cast_fp16)[name = tensor("op_1758_cast_fp16")]; tensor var_1762_begin_0 = const()[name = tensor("op_1762_begin_0"), val = tensor([0, 160, 0, 0])]; tensor var_1762_end_0 = const()[name = tensor("op_1762_end_0"), val = tensor([2, 240, 1, 1024])]; tensor var_1762_end_mask_0 = const()[name = tensor("op_1762_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_1762_cast_fp16 = slice_by_index(begin = var_1762_begin_0, end = var_1762_end_0, end_mask = var_1762_end_mask_0, x = q_5_cast_fp16)[name = tensor("op_1762_cast_fp16")]; tensor var_1766_begin_0 = const()[name = tensor("op_1766_begin_0"), val = tensor([0, 240, 0, 0])]; tensor var_1766_end_0 = const()[name = tensor("op_1766_end_0"), val = tensor([2, 320, 1, 1024])]; tensor var_1766_end_mask_0 = const()[name = tensor("op_1766_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_1766_cast_fp16 = slice_by_index(begin = var_1766_begin_0, end = var_1766_end_0, end_mask = var_1766_end_mask_0, x = q_5_cast_fp16)[name = tensor("op_1766_cast_fp16")]; tensor var_1770_begin_0 = const()[name = tensor("op_1770_begin_0"), val = tensor([0, 320, 0, 0])]; tensor var_1770_end_0 = const()[name = tensor("op_1770_end_0"), val = tensor([2, 400, 1, 1024])]; tensor var_1770_end_mask_0 = const()[name = tensor("op_1770_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_1770_cast_fp16 = slice_by_index(begin = var_1770_begin_0, end = var_1770_end_0, end_mask = var_1770_end_mask_0, x = q_5_cast_fp16)[name = tensor("op_1770_cast_fp16")]; tensor var_1774_begin_0 = const()[name = tensor("op_1774_begin_0"), val = tensor([0, 400, 0, 0])]; tensor var_1774_end_0 = const()[name = tensor("op_1774_end_0"), val = tensor([2, 480, 1, 1024])]; tensor var_1774_end_mask_0 = const()[name = tensor("op_1774_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_1774_cast_fp16 = slice_by_index(begin = var_1774_begin_0, end = var_1774_end_0, end_mask = var_1774_end_mask_0, x = q_5_cast_fp16)[name = tensor("op_1774_cast_fp16")]; tensor var_1778_begin_0 = const()[name = tensor("op_1778_begin_0"), val = tensor([0, 480, 0, 0])]; tensor var_1778_end_0 = const()[name = tensor("op_1778_end_0"), val = tensor([2, 560, 1, 1024])]; tensor var_1778_end_mask_0 = const()[name = tensor("op_1778_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_1778_cast_fp16 = slice_by_index(begin = var_1778_begin_0, end = var_1778_end_0, end_mask = var_1778_end_mask_0, x = q_5_cast_fp16)[name = tensor("op_1778_cast_fp16")]; tensor var_1782_begin_0 = const()[name = tensor("op_1782_begin_0"), val = tensor([0, 560, 0, 0])]; tensor var_1782_end_0 = const()[name = tensor("op_1782_end_0"), val = tensor([2, 640, 1, 1024])]; tensor var_1782_end_mask_0 = const()[name = tensor("op_1782_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_1782_cast_fp16 = slice_by_index(begin = var_1782_begin_0, end = var_1782_end_0, end_mask = var_1782_end_mask_0, x = q_5_cast_fp16)[name = tensor("op_1782_cast_fp16")]; tensor var_1785_begin_0 = const()[name = tensor("op_1785_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_1785_end_0 = const()[name = tensor("op_1785_end_0"), val = tensor([2, 80, 1, 512])]; tensor var_1785_end_mask_0 = const()[name = tensor("op_1785_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1785_cast_fp16 = slice_by_index(begin = var_1785_begin_0, end = var_1785_end_0, end_mask = var_1785_end_mask_0, x = var_1754_cast_fp16)[name = tensor("op_1785_cast_fp16")]; tensor var_1786_begin_0 = const()[name = tensor("op_1786_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_1786_end_0 = const()[name = tensor("op_1786_end_0"), val = tensor([2, 80, 1, 1024])]; tensor var_1786_end_mask_0 = const()[name = tensor("op_1786_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1786_cast_fp16 = slice_by_index(begin = var_1786_begin_0, end = var_1786_end_0, end_mask = var_1786_end_mask_0, x = var_1754_cast_fp16)[name = tensor("op_1786_cast_fp16")]; tensor var_1787_begin_0 = const()[name = tensor("op_1787_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_1787_end_0 = const()[name = tensor("op_1787_end_0"), val = tensor([2, 80, 1, 512])]; tensor var_1787_end_mask_0 = const()[name = tensor("op_1787_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1787_cast_fp16 = slice_by_index(begin = var_1787_begin_0, end = var_1787_end_0, end_mask = var_1787_end_mask_0, x = var_1758_cast_fp16)[name = tensor("op_1787_cast_fp16")]; tensor var_1788_begin_0 = const()[name = tensor("op_1788_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_1788_end_0 = const()[name = tensor("op_1788_end_0"), val = tensor([2, 80, 1, 1024])]; tensor var_1788_end_mask_0 = const()[name = tensor("op_1788_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1788_cast_fp16 = slice_by_index(begin = var_1788_begin_0, end = var_1788_end_0, end_mask = var_1788_end_mask_0, x = var_1758_cast_fp16)[name = tensor("op_1788_cast_fp16")]; tensor var_1789_begin_0 = const()[name = tensor("op_1789_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_1789_end_0 = const()[name = tensor("op_1789_end_0"), val = tensor([2, 80, 1, 512])]; tensor var_1789_end_mask_0 = const()[name = tensor("op_1789_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1789_cast_fp16 = slice_by_index(begin = var_1789_begin_0, end = var_1789_end_0, end_mask = var_1789_end_mask_0, x = var_1762_cast_fp16)[name = tensor("op_1789_cast_fp16")]; tensor var_1790_begin_0 = const()[name = tensor("op_1790_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_1790_end_0 = const()[name = tensor("op_1790_end_0"), val = tensor([2, 80, 1, 1024])]; tensor var_1790_end_mask_0 = const()[name = tensor("op_1790_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1790_cast_fp16 = slice_by_index(begin = var_1790_begin_0, end = var_1790_end_0, end_mask = var_1790_end_mask_0, x = var_1762_cast_fp16)[name = tensor("op_1790_cast_fp16")]; tensor var_1791_begin_0 = const()[name = tensor("op_1791_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_1791_end_0 = const()[name = tensor("op_1791_end_0"), val = tensor([2, 80, 1, 512])]; tensor var_1791_end_mask_0 = const()[name = tensor("op_1791_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1791_cast_fp16 = slice_by_index(begin = var_1791_begin_0, end = var_1791_end_0, end_mask = var_1791_end_mask_0, x = var_1766_cast_fp16)[name = tensor("op_1791_cast_fp16")]; tensor var_1792_begin_0 = const()[name = tensor("op_1792_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_1792_end_0 = const()[name = tensor("op_1792_end_0"), val = tensor([2, 80, 1, 1024])]; tensor var_1792_end_mask_0 = const()[name = tensor("op_1792_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1792_cast_fp16 = slice_by_index(begin = var_1792_begin_0, end = var_1792_end_0, end_mask = var_1792_end_mask_0, x = var_1766_cast_fp16)[name = tensor("op_1792_cast_fp16")]; tensor var_1793_begin_0 = const()[name = tensor("op_1793_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_1793_end_0 = const()[name = tensor("op_1793_end_0"), val = tensor([2, 80, 1, 512])]; tensor var_1793_end_mask_0 = const()[name = tensor("op_1793_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1793_cast_fp16 = slice_by_index(begin = var_1793_begin_0, end = var_1793_end_0, end_mask = var_1793_end_mask_0, x = var_1770_cast_fp16)[name = tensor("op_1793_cast_fp16")]; tensor var_1794_begin_0 = const()[name = tensor("op_1794_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_1794_end_0 = const()[name = tensor("op_1794_end_0"), val = tensor([2, 80, 1, 1024])]; tensor var_1794_end_mask_0 = const()[name = tensor("op_1794_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1794_cast_fp16 = slice_by_index(begin = var_1794_begin_0, end = var_1794_end_0, end_mask = var_1794_end_mask_0, x = var_1770_cast_fp16)[name = tensor("op_1794_cast_fp16")]; tensor var_1795_begin_0 = const()[name = tensor("op_1795_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_1795_end_0 = const()[name = tensor("op_1795_end_0"), val = tensor([2, 80, 1, 512])]; tensor var_1795_end_mask_0 = const()[name = tensor("op_1795_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1795_cast_fp16 = slice_by_index(begin = var_1795_begin_0, end = var_1795_end_0, end_mask = var_1795_end_mask_0, x = var_1774_cast_fp16)[name = tensor("op_1795_cast_fp16")]; tensor var_1796_begin_0 = const()[name = tensor("op_1796_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_1796_end_0 = const()[name = tensor("op_1796_end_0"), val = tensor([2, 80, 1, 1024])]; tensor var_1796_end_mask_0 = const()[name = tensor("op_1796_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1796_cast_fp16 = slice_by_index(begin = var_1796_begin_0, end = var_1796_end_0, end_mask = var_1796_end_mask_0, x = var_1774_cast_fp16)[name = tensor("op_1796_cast_fp16")]; tensor var_1797_begin_0 = const()[name = tensor("op_1797_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_1797_end_0 = const()[name = tensor("op_1797_end_0"), val = tensor([2, 80, 1, 512])]; tensor var_1797_end_mask_0 = const()[name = tensor("op_1797_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1797_cast_fp16 = slice_by_index(begin = var_1797_begin_0, end = var_1797_end_0, end_mask = var_1797_end_mask_0, x = var_1778_cast_fp16)[name = tensor("op_1797_cast_fp16")]; tensor var_1798_begin_0 = const()[name = tensor("op_1798_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_1798_end_0 = const()[name = tensor("op_1798_end_0"), val = tensor([2, 80, 1, 1024])]; tensor var_1798_end_mask_0 = const()[name = tensor("op_1798_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1798_cast_fp16 = slice_by_index(begin = var_1798_begin_0, end = var_1798_end_0, end_mask = var_1798_end_mask_0, x = var_1778_cast_fp16)[name = tensor("op_1798_cast_fp16")]; tensor var_1799_begin_0 = const()[name = tensor("op_1799_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_1799_end_0 = const()[name = tensor("op_1799_end_0"), val = tensor([2, 80, 1, 512])]; tensor var_1799_end_mask_0 = const()[name = tensor("op_1799_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1799_cast_fp16 = slice_by_index(begin = var_1799_begin_0, end = var_1799_end_0, end_mask = var_1799_end_mask_0, x = var_1782_cast_fp16)[name = tensor("op_1799_cast_fp16")]; tensor var_1800_begin_0 = const()[name = tensor("op_1800_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_1800_end_0 = const()[name = tensor("op_1800_end_0"), val = tensor([2, 80, 1, 1024])]; tensor var_1800_end_mask_0 = const()[name = tensor("op_1800_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1800_cast_fp16 = slice_by_index(begin = var_1800_begin_0, end = var_1800_end_0, end_mask = var_1800_end_mask_0, x = var_1782_cast_fp16)[name = tensor("op_1800_cast_fp16")]; tensor k_11_perm_0 = const()[name = tensor("k_11_perm_0"), val = tensor([0, 3, 2, 1])]; tensor var_1805_begin_0 = const()[name = tensor("op_1805_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_1805_end_0 = const()[name = tensor("op_1805_end_0"), val = tensor([2, 1024, 1, 80])]; tensor var_1805_end_mask_0 = const()[name = tensor("op_1805_end_mask_0"), val = tensor([true, true, true, false])]; tensor transpose_15 = transpose(perm = k_11_perm_0, x = k_9_cast_fp16)[name = tensor("transpose_15")]; tensor var_1805_cast_fp16 = slice_by_index(begin = var_1805_begin_0, end = var_1805_end_0, end_mask = var_1805_end_mask_0, x = transpose_15)[name = tensor("op_1805_cast_fp16")]; tensor var_1809_begin_0 = const()[name = tensor("op_1809_begin_0"), val = tensor([0, 0, 0, 80])]; tensor var_1809_end_0 = const()[name = tensor("op_1809_end_0"), val = tensor([2, 1024, 1, 160])]; tensor var_1809_end_mask_0 = const()[name = tensor("op_1809_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1809_cast_fp16 = slice_by_index(begin = var_1809_begin_0, end = var_1809_end_0, end_mask = var_1809_end_mask_0, x = transpose_15)[name = tensor("op_1809_cast_fp16")]; tensor var_1813_begin_0 = const()[name = tensor("op_1813_begin_0"), val = tensor([0, 0, 0, 160])]; tensor var_1813_end_0 = const()[name = tensor("op_1813_end_0"), val = tensor([2, 1024, 1, 240])]; tensor var_1813_end_mask_0 = const()[name = tensor("op_1813_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1813_cast_fp16 = slice_by_index(begin = var_1813_begin_0, end = var_1813_end_0, end_mask = var_1813_end_mask_0, x = transpose_15)[name = tensor("op_1813_cast_fp16")]; tensor var_1817_begin_0 = const()[name = tensor("op_1817_begin_0"), val = tensor([0, 0, 0, 240])]; tensor var_1817_end_0 = const()[name = tensor("op_1817_end_0"), val = tensor([2, 1024, 1, 320])]; tensor var_1817_end_mask_0 = const()[name = tensor("op_1817_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1817_cast_fp16 = slice_by_index(begin = var_1817_begin_0, end = var_1817_end_0, end_mask = var_1817_end_mask_0, x = transpose_15)[name = tensor("op_1817_cast_fp16")]; tensor var_1821_begin_0 = const()[name = tensor("op_1821_begin_0"), val = tensor([0, 0, 0, 320])]; tensor var_1821_end_0 = const()[name = tensor("op_1821_end_0"), val = tensor([2, 1024, 1, 400])]; tensor var_1821_end_mask_0 = const()[name = tensor("op_1821_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1821_cast_fp16 = slice_by_index(begin = var_1821_begin_0, end = var_1821_end_0, end_mask = var_1821_end_mask_0, x = transpose_15)[name = tensor("op_1821_cast_fp16")]; tensor var_1825_begin_0 = const()[name = tensor("op_1825_begin_0"), val = tensor([0, 0, 0, 400])]; tensor var_1825_end_0 = const()[name = tensor("op_1825_end_0"), val = tensor([2, 1024, 1, 480])]; tensor var_1825_end_mask_0 = const()[name = tensor("op_1825_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1825_cast_fp16 = slice_by_index(begin = var_1825_begin_0, end = var_1825_end_0, end_mask = var_1825_end_mask_0, x = transpose_15)[name = tensor("op_1825_cast_fp16")]; tensor var_1829_begin_0 = const()[name = tensor("op_1829_begin_0"), val = tensor([0, 0, 0, 480])]; tensor var_1829_end_0 = const()[name = tensor("op_1829_end_0"), val = tensor([2, 1024, 1, 560])]; tensor var_1829_end_mask_0 = const()[name = tensor("op_1829_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1829_cast_fp16 = slice_by_index(begin = var_1829_begin_0, end = var_1829_end_0, end_mask = var_1829_end_mask_0, x = transpose_15)[name = tensor("op_1829_cast_fp16")]; tensor var_1833_begin_0 = const()[name = tensor("op_1833_begin_0"), val = tensor([0, 0, 0, 560])]; tensor var_1833_end_0 = const()[name = tensor("op_1833_end_0"), val = tensor([2, 1024, 1, 640])]; tensor var_1833_end_mask_0 = const()[name = tensor("op_1833_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1833_cast_fp16 = slice_by_index(begin = var_1833_begin_0, end = var_1833_end_0, end_mask = var_1833_end_mask_0, x = transpose_15)[name = tensor("op_1833_cast_fp16")]; tensor var_1835_begin_0 = const()[name = tensor("op_1835_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_1835_end_0 = const()[name = tensor("op_1835_end_0"), val = tensor([2, 80, 1, 1024])]; tensor var_1835_end_mask_0 = const()[name = tensor("op_1835_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_1835_cast_fp16 = slice_by_index(begin = var_1835_begin_0, end = var_1835_end_0, end_mask = var_1835_end_mask_0, x = v_5_cast_fp16)[name = tensor("op_1835_cast_fp16")]; tensor var_1839_begin_0 = const()[name = tensor("op_1839_begin_0"), val = tensor([0, 80, 0, 0])]; tensor var_1839_end_0 = const()[name = tensor("op_1839_end_0"), val = tensor([2, 160, 1, 1024])]; tensor var_1839_end_mask_0 = const()[name = tensor("op_1839_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_1839_cast_fp16 = slice_by_index(begin = var_1839_begin_0, end = var_1839_end_0, end_mask = var_1839_end_mask_0, x = v_5_cast_fp16)[name = tensor("op_1839_cast_fp16")]; tensor var_1843_begin_0 = const()[name = tensor("op_1843_begin_0"), val = tensor([0, 160, 0, 0])]; tensor var_1843_end_0 = const()[name = tensor("op_1843_end_0"), val = tensor([2, 240, 1, 1024])]; tensor var_1843_end_mask_0 = const()[name = tensor("op_1843_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_1843_cast_fp16 = slice_by_index(begin = var_1843_begin_0, end = var_1843_end_0, end_mask = var_1843_end_mask_0, x = v_5_cast_fp16)[name = tensor("op_1843_cast_fp16")]; tensor var_1847_begin_0 = const()[name = tensor("op_1847_begin_0"), val = tensor([0, 240, 0, 0])]; tensor var_1847_end_0 = const()[name = tensor("op_1847_end_0"), val = tensor([2, 320, 1, 1024])]; tensor var_1847_end_mask_0 = const()[name = tensor("op_1847_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_1847_cast_fp16 = slice_by_index(begin = var_1847_begin_0, end = var_1847_end_0, end_mask = var_1847_end_mask_0, x = v_5_cast_fp16)[name = tensor("op_1847_cast_fp16")]; tensor var_1851_begin_0 = const()[name = tensor("op_1851_begin_0"), val = tensor([0, 320, 0, 0])]; tensor var_1851_end_0 = const()[name = tensor("op_1851_end_0"), val = tensor([2, 400, 1, 1024])]; tensor var_1851_end_mask_0 = const()[name = tensor("op_1851_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_1851_cast_fp16 = slice_by_index(begin = var_1851_begin_0, end = var_1851_end_0, end_mask = var_1851_end_mask_0, x = v_5_cast_fp16)[name = tensor("op_1851_cast_fp16")]; tensor var_1855_begin_0 = const()[name = tensor("op_1855_begin_0"), val = tensor([0, 400, 0, 0])]; tensor var_1855_end_0 = const()[name = tensor("op_1855_end_0"), val = tensor([2, 480, 1, 1024])]; tensor var_1855_end_mask_0 = const()[name = tensor("op_1855_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_1855_cast_fp16 = slice_by_index(begin = var_1855_begin_0, end = var_1855_end_0, end_mask = var_1855_end_mask_0, x = v_5_cast_fp16)[name = tensor("op_1855_cast_fp16")]; tensor var_1859_begin_0 = const()[name = tensor("op_1859_begin_0"), val = tensor([0, 480, 0, 0])]; tensor var_1859_end_0 = const()[name = tensor("op_1859_end_0"), val = tensor([2, 560, 1, 1024])]; tensor var_1859_end_mask_0 = const()[name = tensor("op_1859_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_1859_cast_fp16 = slice_by_index(begin = var_1859_begin_0, end = var_1859_end_0, end_mask = var_1859_end_mask_0, x = v_5_cast_fp16)[name = tensor("op_1859_cast_fp16")]; tensor var_1863_begin_0 = const()[name = tensor("op_1863_begin_0"), val = tensor([0, 560, 0, 0])]; tensor var_1863_end_0 = const()[name = tensor("op_1863_end_0"), val = tensor([2, 640, 1, 1024])]; tensor var_1863_end_mask_0 = const()[name = tensor("op_1863_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_1863_cast_fp16 = slice_by_index(begin = var_1863_begin_0, end = var_1863_end_0, end_mask = var_1863_end_mask_0, x = v_5_cast_fp16)[name = tensor("op_1863_cast_fp16")]; tensor var_1867_equation_0 = const()[name = tensor("op_1867_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_1867_cast_fp16 = einsum(equation = var_1867_equation_0, values = (var_1805_cast_fp16, var_1785_cast_fp16))[name = tensor("op_1867_cast_fp16")]; tensor var_1868_to_fp16 = const()[name = tensor("op_1868_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_257_cast_fp16 = mul(x = var_1867_cast_fp16, y = var_1868_to_fp16)[name = tensor("aw_chunk_257_cast_fp16")]; tensor var_1871_equation_0 = const()[name = tensor("op_1871_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_1871_cast_fp16 = einsum(equation = var_1871_equation_0, values = (var_1805_cast_fp16, var_1786_cast_fp16))[name = tensor("op_1871_cast_fp16")]; tensor var_1872_to_fp16 = const()[name = tensor("op_1872_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_259_cast_fp16 = mul(x = var_1871_cast_fp16, y = var_1872_to_fp16)[name = tensor("aw_chunk_259_cast_fp16")]; tensor var_1875_equation_0 = const()[name = tensor("op_1875_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_1875_cast_fp16 = einsum(equation = var_1875_equation_0, values = (var_1809_cast_fp16, var_1787_cast_fp16))[name = tensor("op_1875_cast_fp16")]; tensor var_1876_to_fp16 = const()[name = tensor("op_1876_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_261_cast_fp16 = mul(x = var_1875_cast_fp16, y = var_1876_to_fp16)[name = tensor("aw_chunk_261_cast_fp16")]; tensor var_1879_equation_0 = const()[name = tensor("op_1879_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_1879_cast_fp16 = einsum(equation = var_1879_equation_0, values = (var_1809_cast_fp16, var_1788_cast_fp16))[name = tensor("op_1879_cast_fp16")]; tensor var_1880_to_fp16 = const()[name = tensor("op_1880_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_263_cast_fp16 = mul(x = var_1879_cast_fp16, y = var_1880_to_fp16)[name = tensor("aw_chunk_263_cast_fp16")]; tensor var_1883_equation_0 = const()[name = tensor("op_1883_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_1883_cast_fp16 = einsum(equation = var_1883_equation_0, values = (var_1813_cast_fp16, var_1789_cast_fp16))[name = tensor("op_1883_cast_fp16")]; tensor var_1884_to_fp16 = const()[name = tensor("op_1884_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_265_cast_fp16 = mul(x = var_1883_cast_fp16, y = var_1884_to_fp16)[name = tensor("aw_chunk_265_cast_fp16")]; tensor var_1887_equation_0 = const()[name = tensor("op_1887_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_1887_cast_fp16 = einsum(equation = var_1887_equation_0, values = (var_1813_cast_fp16, var_1790_cast_fp16))[name = tensor("op_1887_cast_fp16")]; tensor var_1888_to_fp16 = const()[name = tensor("op_1888_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_267_cast_fp16 = mul(x = var_1887_cast_fp16, y = var_1888_to_fp16)[name = tensor("aw_chunk_267_cast_fp16")]; tensor var_1891_equation_0 = const()[name = tensor("op_1891_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_1891_cast_fp16 = einsum(equation = var_1891_equation_0, values = (var_1817_cast_fp16, var_1791_cast_fp16))[name = tensor("op_1891_cast_fp16")]; tensor var_1892_to_fp16 = const()[name = tensor("op_1892_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_269_cast_fp16 = mul(x = var_1891_cast_fp16, y = var_1892_to_fp16)[name = tensor("aw_chunk_269_cast_fp16")]; tensor var_1895_equation_0 = const()[name = tensor("op_1895_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_1895_cast_fp16 = einsum(equation = var_1895_equation_0, values = (var_1817_cast_fp16, var_1792_cast_fp16))[name = tensor("op_1895_cast_fp16")]; tensor var_1896_to_fp16 = const()[name = tensor("op_1896_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_271_cast_fp16 = mul(x = var_1895_cast_fp16, y = var_1896_to_fp16)[name = tensor("aw_chunk_271_cast_fp16")]; tensor var_1899_equation_0 = const()[name = tensor("op_1899_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_1899_cast_fp16 = einsum(equation = var_1899_equation_0, values = (var_1821_cast_fp16, var_1793_cast_fp16))[name = tensor("op_1899_cast_fp16")]; tensor var_1900_to_fp16 = const()[name = tensor("op_1900_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_273_cast_fp16 = mul(x = var_1899_cast_fp16, y = var_1900_to_fp16)[name = tensor("aw_chunk_273_cast_fp16")]; tensor var_1903_equation_0 = const()[name = tensor("op_1903_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_1903_cast_fp16 = einsum(equation = var_1903_equation_0, values = (var_1821_cast_fp16, var_1794_cast_fp16))[name = tensor("op_1903_cast_fp16")]; tensor var_1904_to_fp16 = const()[name = tensor("op_1904_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_275_cast_fp16 = mul(x = var_1903_cast_fp16, y = var_1904_to_fp16)[name = tensor("aw_chunk_275_cast_fp16")]; tensor var_1907_equation_0 = const()[name = tensor("op_1907_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_1907_cast_fp16 = einsum(equation = var_1907_equation_0, values = (var_1825_cast_fp16, var_1795_cast_fp16))[name = tensor("op_1907_cast_fp16")]; tensor var_1908_to_fp16 = const()[name = tensor("op_1908_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_277_cast_fp16 = mul(x = var_1907_cast_fp16, y = var_1908_to_fp16)[name = tensor("aw_chunk_277_cast_fp16")]; tensor var_1911_equation_0 = const()[name = tensor("op_1911_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_1911_cast_fp16 = einsum(equation = var_1911_equation_0, values = (var_1825_cast_fp16, var_1796_cast_fp16))[name = tensor("op_1911_cast_fp16")]; tensor var_1912_to_fp16 = const()[name = tensor("op_1912_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_279_cast_fp16 = mul(x = var_1911_cast_fp16, y = var_1912_to_fp16)[name = tensor("aw_chunk_279_cast_fp16")]; tensor var_1915_equation_0 = const()[name = tensor("op_1915_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_1915_cast_fp16 = einsum(equation = var_1915_equation_0, values = (var_1829_cast_fp16, var_1797_cast_fp16))[name = tensor("op_1915_cast_fp16")]; tensor var_1916_to_fp16 = const()[name = tensor("op_1916_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_281_cast_fp16 = mul(x = var_1915_cast_fp16, y = var_1916_to_fp16)[name = tensor("aw_chunk_281_cast_fp16")]; tensor var_1919_equation_0 = const()[name = tensor("op_1919_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_1919_cast_fp16 = einsum(equation = var_1919_equation_0, values = (var_1829_cast_fp16, var_1798_cast_fp16))[name = tensor("op_1919_cast_fp16")]; tensor var_1920_to_fp16 = const()[name = tensor("op_1920_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_283_cast_fp16 = mul(x = var_1919_cast_fp16, y = var_1920_to_fp16)[name = tensor("aw_chunk_283_cast_fp16")]; tensor var_1923_equation_0 = const()[name = tensor("op_1923_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_1923_cast_fp16 = einsum(equation = var_1923_equation_0, values = (var_1833_cast_fp16, var_1799_cast_fp16))[name = tensor("op_1923_cast_fp16")]; tensor var_1924_to_fp16 = const()[name = tensor("op_1924_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_285_cast_fp16 = mul(x = var_1923_cast_fp16, y = var_1924_to_fp16)[name = tensor("aw_chunk_285_cast_fp16")]; tensor var_1927_equation_0 = const()[name = tensor("op_1927_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_1927_cast_fp16 = einsum(equation = var_1927_equation_0, values = (var_1833_cast_fp16, var_1800_cast_fp16))[name = tensor("op_1927_cast_fp16")]; tensor var_1928_to_fp16 = const()[name = tensor("op_1928_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_287_cast_fp16 = mul(x = var_1927_cast_fp16, y = var_1928_to_fp16)[name = tensor("aw_chunk_287_cast_fp16")]; tensor var_1930_cast_fp16 = softmax(axis = var_1630, x = aw_chunk_257_cast_fp16)[name = tensor("op_1930_cast_fp16")]; tensor var_1931_cast_fp16 = softmax(axis = var_1630, x = aw_chunk_259_cast_fp16)[name = tensor("op_1931_cast_fp16")]; tensor var_1932_cast_fp16 = softmax(axis = var_1630, x = aw_chunk_261_cast_fp16)[name = tensor("op_1932_cast_fp16")]; tensor var_1933_cast_fp16 = softmax(axis = var_1630, x = aw_chunk_263_cast_fp16)[name = tensor("op_1933_cast_fp16")]; tensor var_1934_cast_fp16 = softmax(axis = var_1630, x = aw_chunk_265_cast_fp16)[name = tensor("op_1934_cast_fp16")]; tensor var_1935_cast_fp16 = softmax(axis = var_1630, x = aw_chunk_267_cast_fp16)[name = tensor("op_1935_cast_fp16")]; tensor var_1936_cast_fp16 = softmax(axis = var_1630, x = aw_chunk_269_cast_fp16)[name = tensor("op_1936_cast_fp16")]; tensor var_1937_cast_fp16 = softmax(axis = var_1630, x = aw_chunk_271_cast_fp16)[name = tensor("op_1937_cast_fp16")]; tensor var_1938_cast_fp16 = softmax(axis = var_1630, x = aw_chunk_273_cast_fp16)[name = tensor("op_1938_cast_fp16")]; tensor var_1939_cast_fp16 = softmax(axis = var_1630, x = aw_chunk_275_cast_fp16)[name = tensor("op_1939_cast_fp16")]; tensor var_1940_cast_fp16 = softmax(axis = var_1630, x = aw_chunk_277_cast_fp16)[name = tensor("op_1940_cast_fp16")]; tensor var_1941_cast_fp16 = softmax(axis = var_1630, x = aw_chunk_279_cast_fp16)[name = tensor("op_1941_cast_fp16")]; tensor var_1942_cast_fp16 = softmax(axis = var_1630, x = aw_chunk_281_cast_fp16)[name = tensor("op_1942_cast_fp16")]; tensor var_1943_cast_fp16 = softmax(axis = var_1630, x = aw_chunk_283_cast_fp16)[name = tensor("op_1943_cast_fp16")]; tensor var_1944_cast_fp16 = softmax(axis = var_1630, x = aw_chunk_285_cast_fp16)[name = tensor("op_1944_cast_fp16")]; tensor var_1945_cast_fp16 = softmax(axis = var_1630, x = aw_chunk_287_cast_fp16)[name = tensor("op_1945_cast_fp16")]; tensor var_1947_equation_0 = const()[name = tensor("op_1947_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_1947_cast_fp16 = einsum(equation = var_1947_equation_0, values = (var_1835_cast_fp16, var_1930_cast_fp16))[name = tensor("op_1947_cast_fp16")]; tensor var_1949_equation_0 = const()[name = tensor("op_1949_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_1949_cast_fp16 = einsum(equation = var_1949_equation_0, values = (var_1835_cast_fp16, var_1931_cast_fp16))[name = tensor("op_1949_cast_fp16")]; tensor var_1951_equation_0 = const()[name = tensor("op_1951_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_1951_cast_fp16 = einsum(equation = var_1951_equation_0, values = (var_1839_cast_fp16, var_1932_cast_fp16))[name = tensor("op_1951_cast_fp16")]; tensor var_1953_equation_0 = const()[name = tensor("op_1953_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_1953_cast_fp16 = einsum(equation = var_1953_equation_0, values = (var_1839_cast_fp16, var_1933_cast_fp16))[name = tensor("op_1953_cast_fp16")]; tensor var_1955_equation_0 = const()[name = tensor("op_1955_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_1955_cast_fp16 = einsum(equation = var_1955_equation_0, values = (var_1843_cast_fp16, var_1934_cast_fp16))[name = tensor("op_1955_cast_fp16")]; tensor var_1957_equation_0 = const()[name = tensor("op_1957_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_1957_cast_fp16 = einsum(equation = var_1957_equation_0, values = (var_1843_cast_fp16, var_1935_cast_fp16))[name = tensor("op_1957_cast_fp16")]; tensor var_1959_equation_0 = const()[name = tensor("op_1959_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_1959_cast_fp16 = einsum(equation = var_1959_equation_0, values = (var_1847_cast_fp16, var_1936_cast_fp16))[name = tensor("op_1959_cast_fp16")]; tensor var_1961_equation_0 = const()[name = tensor("op_1961_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_1961_cast_fp16 = einsum(equation = var_1961_equation_0, values = (var_1847_cast_fp16, var_1937_cast_fp16))[name = tensor("op_1961_cast_fp16")]; tensor var_1963_equation_0 = const()[name = tensor("op_1963_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_1963_cast_fp16 = einsum(equation = var_1963_equation_0, values = (var_1851_cast_fp16, var_1938_cast_fp16))[name = tensor("op_1963_cast_fp16")]; tensor var_1965_equation_0 = const()[name = tensor("op_1965_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_1965_cast_fp16 = einsum(equation = var_1965_equation_0, values = (var_1851_cast_fp16, var_1939_cast_fp16))[name = tensor("op_1965_cast_fp16")]; tensor var_1967_equation_0 = const()[name = tensor("op_1967_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_1967_cast_fp16 = einsum(equation = var_1967_equation_0, values = (var_1855_cast_fp16, var_1940_cast_fp16))[name = tensor("op_1967_cast_fp16")]; tensor var_1969_equation_0 = const()[name = tensor("op_1969_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_1969_cast_fp16 = einsum(equation = var_1969_equation_0, values = (var_1855_cast_fp16, var_1941_cast_fp16))[name = tensor("op_1969_cast_fp16")]; tensor var_1971_equation_0 = const()[name = tensor("op_1971_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_1971_cast_fp16 = einsum(equation = var_1971_equation_0, values = (var_1859_cast_fp16, var_1942_cast_fp16))[name = tensor("op_1971_cast_fp16")]; tensor var_1973_equation_0 = const()[name = tensor("op_1973_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_1973_cast_fp16 = einsum(equation = var_1973_equation_0, values = (var_1859_cast_fp16, var_1943_cast_fp16))[name = tensor("op_1973_cast_fp16")]; tensor var_1975_equation_0 = const()[name = tensor("op_1975_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_1975_cast_fp16 = einsum(equation = var_1975_equation_0, values = (var_1863_cast_fp16, var_1944_cast_fp16))[name = tensor("op_1975_cast_fp16")]; tensor var_1977_equation_0 = const()[name = tensor("op_1977_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_1977_cast_fp16 = einsum(equation = var_1977_equation_0, values = (var_1863_cast_fp16, var_1945_cast_fp16))[name = tensor("op_1977_cast_fp16")]; tensor var_1979_interleave_0 = const()[name = tensor("op_1979_interleave_0"), val = tensor(false)]; tensor var_1979_cast_fp16 = concat(axis = var_1608, interleave = var_1979_interleave_0, values = (var_1947_cast_fp16, var_1949_cast_fp16))[name = tensor("op_1979_cast_fp16")]; tensor var_1981_interleave_0 = const()[name = tensor("op_1981_interleave_0"), val = tensor(false)]; tensor var_1981_cast_fp16 = concat(axis = var_1608, interleave = var_1981_interleave_0, values = (var_1951_cast_fp16, var_1953_cast_fp16))[name = tensor("op_1981_cast_fp16")]; tensor var_1983_interleave_0 = const()[name = tensor("op_1983_interleave_0"), val = tensor(false)]; tensor var_1983_cast_fp16 = concat(axis = var_1608, interleave = var_1983_interleave_0, values = (var_1955_cast_fp16, var_1957_cast_fp16))[name = tensor("op_1983_cast_fp16")]; tensor var_1985_interleave_0 = const()[name = tensor("op_1985_interleave_0"), val = tensor(false)]; tensor var_1985_cast_fp16 = concat(axis = var_1608, interleave = var_1985_interleave_0, values = (var_1959_cast_fp16, var_1961_cast_fp16))[name = tensor("op_1985_cast_fp16")]; tensor var_1987_interleave_0 = const()[name = tensor("op_1987_interleave_0"), val = tensor(false)]; tensor var_1987_cast_fp16 = concat(axis = var_1608, interleave = var_1987_interleave_0, values = (var_1963_cast_fp16, var_1965_cast_fp16))[name = tensor("op_1987_cast_fp16")]; tensor var_1989_interleave_0 = const()[name = tensor("op_1989_interleave_0"), val = tensor(false)]; tensor var_1989_cast_fp16 = concat(axis = var_1608, interleave = var_1989_interleave_0, values = (var_1967_cast_fp16, var_1969_cast_fp16))[name = tensor("op_1989_cast_fp16")]; tensor var_1991_interleave_0 = const()[name = tensor("op_1991_interleave_0"), val = tensor(false)]; tensor var_1991_cast_fp16 = concat(axis = var_1608, interleave = var_1991_interleave_0, values = (var_1971_cast_fp16, var_1973_cast_fp16))[name = tensor("op_1991_cast_fp16")]; tensor var_1993_interleave_0 = const()[name = tensor("op_1993_interleave_0"), val = tensor(false)]; tensor var_1993_cast_fp16 = concat(axis = var_1608, interleave = var_1993_interleave_0, values = (var_1975_cast_fp16, var_1977_cast_fp16))[name = tensor("op_1993_cast_fp16")]; tensor input_53_interleave_0 = const()[name = tensor("input_53_interleave_0"), val = tensor(false)]; tensor input_53_cast_fp16 = concat(axis = var_1630, interleave = input_53_interleave_0, values = (var_1979_cast_fp16, var_1981_cast_fp16, var_1983_cast_fp16, var_1985_cast_fp16, var_1987_cast_fp16, var_1989_cast_fp16, var_1991_cast_fp16, var_1993_cast_fp16))[name = tensor("input_53_cast_fp16")]; tensor var_1999 = const()[name = tensor("op_1999"), val = tensor([1, 1])]; tensor var_2001 = const()[name = tensor("op_2001"), val = tensor([1, 1])]; tensor var_2003_pad_type_0 = const()[name = tensor("op_2003_pad_type_0"), val = tensor("custom")]; tensor var_2003_pad_0 = const()[name = tensor("op_2003_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(12024640))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(12331904))), 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(12332096)))]; tensor var_2003_cast_fp16 = conv(bias = down_blocks_1_attentions_0_transformer_blocks_0_attn1_to_out_0_bias_to_fp16, dilations = var_2001, groups = var_1630, pad = var_2003_pad_0, pad_type = var_2003_pad_type_0, strides = var_1999, weight = down_blocks_1_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16_palettized, x = input_53_cast_fp16)[name = tensor("op_2003_cast_fp16")]; tensor inputs_9_cast_fp16 = add(x = var_2003_cast_fp16, y = inputs_7_cast_fp16)[name = tensor("inputs_9_cast_fp16")]; tensor var_2007 = const()[name = tensor("op_2007"), val = tensor([1])]; tensor channels_mean_9_cast_fp16 = reduce_mean(axes = var_2007, keep_dims = var_1625, x = inputs_9_cast_fp16)[name = tensor("channels_mean_9_cast_fp16")]; tensor zero_mean_9_cast_fp16 = sub(x = inputs_9_cast_fp16, y = channels_mean_9_cast_fp16)[name = tensor("zero_mean_9_cast_fp16")]; tensor zero_mean_sq_9_cast_fp16 = mul(x = zero_mean_9_cast_fp16, y = zero_mean_9_cast_fp16)[name = tensor("zero_mean_sq_9_cast_fp16")]; tensor var_2011 = const()[name = tensor("op_2011"), val = tensor([1])]; tensor var_2012_cast_fp16 = reduce_mean(axes = var_2011, keep_dims = var_1625, x = zero_mean_sq_9_cast_fp16)[name = tensor("op_2012_cast_fp16")]; tensor var_2013_to_fp16 = const()[name = tensor("op_2013_to_fp16"), val = tensor(0x1.5p-17)]; tensor var_2014_cast_fp16 = add(x = var_2012_cast_fp16, y = var_2013_to_fp16)[name = tensor("op_2014_cast_fp16")]; tensor denom_9_epsilon_0_to_fp16 = const()[name = tensor("denom_9_epsilon_0_to_fp16"), val = tensor(0x1p-24)]; tensor denom_9_cast_fp16 = rsqrt(epsilon = denom_9_epsilon_0_to_fp16, x = var_2014_cast_fp16)[name = tensor("denom_9_cast_fp16")]; tensor out_9_cast_fp16 = mul(x = zero_mean_9_cast_fp16, y = denom_9_cast_fp16)[name = tensor("out_9_cast_fp16")]; tensor var_2018_to_fp16 = const()[name = tensor("op_2018_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(12333440)))]; tensor var_2019_cast_fp16 = add(x = out_9_cast_fp16, y = var_2018_to_fp16)[name = tensor("op_2019_cast_fp16")]; tensor var_2021_to_fp16 = const()[name = tensor("op_2021_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(12334784)))]; tensor hidden_states_29_cast_fp16 = mul(x = var_2019_cast_fp16, y = var_2021_to_fp16)[name = tensor("hidden_states_29_cast_fp16")]; tensor var_2028 = const()[name = tensor("op_2028"), val = tensor([1, 1])]; tensor var_2030 = const()[name = tensor("op_2030"), 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_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(12336128))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(12643392))), 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_7_cast_fp16 = conv(dilations = var_2030, groups = var_1630, pad = q_7_pad_0, pad_type = q_7_pad_type_0, strides = var_2028, weight = down_blocks_1_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16_palettized, x = hidden_states_29_cast_fp16)[name = tensor("q_7_cast_fp16")]; tensor var_2034 = const()[name = tensor("op_2034"), val = tensor([1, 1])]; tensor var_2036 = const()[name = tensor("op_2036"), 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_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(12643584))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(13012288))), name = tensor("down_blocks_1_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized"), shape = tensor([640, 768, 1, 1])]; tensor k_13_cast_fp16 = conv(dilations = var_2036, groups = var_1630, pad = k_13_pad_0, pad_type = k_13_pad_type_0, strides = var_2034, weight = down_blocks_1_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor("k_13_cast_fp16")]; tensor var_2040 = const()[name = tensor("op_2040"), val = tensor([1, 1])]; tensor var_2042 = const()[name = tensor("op_2042"), 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_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(13012480))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(13381184))), name = tensor("down_blocks_1_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized"), shape = tensor([640, 768, 1, 1])]; tensor v_7_cast_fp16 = conv(dilations = var_2042, groups = var_1630, pad = v_7_pad_0, pad_type = v_7_pad_type_0, strides = var_2040, weight = down_blocks_1_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor("v_7_cast_fp16")]; tensor var_2046_begin_0 = const()[name = tensor("op_2046_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_2046_end_0 = const()[name = tensor("op_2046_end_0"), val = tensor([2, 80, 1, 1024])]; tensor var_2046_end_mask_0 = const()[name = tensor("op_2046_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_2046_cast_fp16 = slice_by_index(begin = var_2046_begin_0, end = var_2046_end_0, end_mask = var_2046_end_mask_0, x = q_7_cast_fp16)[name = tensor("op_2046_cast_fp16")]; tensor var_2050_begin_0 = const()[name = tensor("op_2050_begin_0"), val = tensor([0, 80, 0, 0])]; tensor var_2050_end_0 = const()[name = tensor("op_2050_end_0"), val = tensor([2, 160, 1, 1024])]; tensor var_2050_end_mask_0 = const()[name = tensor("op_2050_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_2050_cast_fp16 = slice_by_index(begin = var_2050_begin_0, end = var_2050_end_0, end_mask = var_2050_end_mask_0, x = q_7_cast_fp16)[name = tensor("op_2050_cast_fp16")]; tensor var_2054_begin_0 = const()[name = tensor("op_2054_begin_0"), val = tensor([0, 160, 0, 0])]; tensor var_2054_end_0 = const()[name = tensor("op_2054_end_0"), val = tensor([2, 240, 1, 1024])]; tensor var_2054_end_mask_0 = const()[name = tensor("op_2054_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_2054_cast_fp16 = slice_by_index(begin = var_2054_begin_0, end = var_2054_end_0, end_mask = var_2054_end_mask_0, x = q_7_cast_fp16)[name = tensor("op_2054_cast_fp16")]; tensor var_2058_begin_0 = const()[name = tensor("op_2058_begin_0"), val = tensor([0, 240, 0, 0])]; tensor var_2058_end_0 = const()[name = tensor("op_2058_end_0"), val = tensor([2, 320, 1, 1024])]; tensor var_2058_end_mask_0 = const()[name = tensor("op_2058_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_2058_cast_fp16 = slice_by_index(begin = var_2058_begin_0, end = var_2058_end_0, end_mask = var_2058_end_mask_0, x = q_7_cast_fp16)[name = tensor("op_2058_cast_fp16")]; tensor var_2062_begin_0 = const()[name = tensor("op_2062_begin_0"), val = tensor([0, 320, 0, 0])]; tensor var_2062_end_0 = const()[name = tensor("op_2062_end_0"), val = tensor([2, 400, 1, 1024])]; tensor var_2062_end_mask_0 = const()[name = tensor("op_2062_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_2062_cast_fp16 = slice_by_index(begin = var_2062_begin_0, end = var_2062_end_0, end_mask = var_2062_end_mask_0, x = q_7_cast_fp16)[name = tensor("op_2062_cast_fp16")]; tensor var_2066_begin_0 = const()[name = tensor("op_2066_begin_0"), val = tensor([0, 400, 0, 0])]; tensor var_2066_end_0 = const()[name = tensor("op_2066_end_0"), val = tensor([2, 480, 1, 1024])]; tensor var_2066_end_mask_0 = const()[name = tensor("op_2066_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_2066_cast_fp16 = slice_by_index(begin = var_2066_begin_0, end = var_2066_end_0, end_mask = var_2066_end_mask_0, x = q_7_cast_fp16)[name = tensor("op_2066_cast_fp16")]; tensor var_2070_begin_0 = const()[name = tensor("op_2070_begin_0"), val = tensor([0, 480, 0, 0])]; tensor var_2070_end_0 = const()[name = tensor("op_2070_end_0"), val = tensor([2, 560, 1, 1024])]; tensor var_2070_end_mask_0 = const()[name = tensor("op_2070_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_2070_cast_fp16 = slice_by_index(begin = var_2070_begin_0, end = var_2070_end_0, end_mask = var_2070_end_mask_0, x = q_7_cast_fp16)[name = tensor("op_2070_cast_fp16")]; tensor var_2074_begin_0 = const()[name = tensor("op_2074_begin_0"), val = tensor([0, 560, 0, 0])]; tensor var_2074_end_0 = const()[name = tensor("op_2074_end_0"), val = tensor([2, 640, 1, 1024])]; tensor var_2074_end_mask_0 = const()[name = tensor("op_2074_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_2074_cast_fp16 = slice_by_index(begin = var_2074_begin_0, end = var_2074_end_0, end_mask = var_2074_end_mask_0, x = q_7_cast_fp16)[name = tensor("op_2074_cast_fp16")]; tensor var_2077_begin_0 = const()[name = tensor("op_2077_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_2077_end_0 = const()[name = tensor("op_2077_end_0"), val = tensor([2, 80, 1, 512])]; tensor var_2077_end_mask_0 = const()[name = tensor("op_2077_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2077_cast_fp16 = slice_by_index(begin = var_2077_begin_0, end = var_2077_end_0, end_mask = var_2077_end_mask_0, x = var_2046_cast_fp16)[name = tensor("op_2077_cast_fp16")]; tensor var_2078_begin_0 = const()[name = tensor("op_2078_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_2078_end_0 = const()[name = tensor("op_2078_end_0"), val = tensor([2, 80, 1, 1024])]; tensor var_2078_end_mask_0 = const()[name = tensor("op_2078_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2078_cast_fp16 = slice_by_index(begin = var_2078_begin_0, end = var_2078_end_0, end_mask = var_2078_end_mask_0, x = var_2046_cast_fp16)[name = tensor("op_2078_cast_fp16")]; tensor var_2079_begin_0 = const()[name = tensor("op_2079_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_2079_end_0 = const()[name = tensor("op_2079_end_0"), val = tensor([2, 80, 1, 512])]; tensor var_2079_end_mask_0 = const()[name = tensor("op_2079_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2079_cast_fp16 = slice_by_index(begin = var_2079_begin_0, end = var_2079_end_0, end_mask = var_2079_end_mask_0, x = var_2050_cast_fp16)[name = tensor("op_2079_cast_fp16")]; tensor var_2080_begin_0 = const()[name = tensor("op_2080_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_2080_end_0 = const()[name = tensor("op_2080_end_0"), val = tensor([2, 80, 1, 1024])]; tensor var_2080_end_mask_0 = const()[name = tensor("op_2080_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2080_cast_fp16 = slice_by_index(begin = var_2080_begin_0, end = var_2080_end_0, end_mask = var_2080_end_mask_0, x = var_2050_cast_fp16)[name = tensor("op_2080_cast_fp16")]; tensor var_2081_begin_0 = const()[name = tensor("op_2081_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_2081_end_0 = const()[name = tensor("op_2081_end_0"), val = tensor([2, 80, 1, 512])]; tensor var_2081_end_mask_0 = const()[name = tensor("op_2081_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2081_cast_fp16 = slice_by_index(begin = var_2081_begin_0, end = var_2081_end_0, end_mask = var_2081_end_mask_0, x = var_2054_cast_fp16)[name = tensor("op_2081_cast_fp16")]; tensor var_2082_begin_0 = const()[name = tensor("op_2082_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_2082_end_0 = const()[name = tensor("op_2082_end_0"), val = tensor([2, 80, 1, 1024])]; tensor var_2082_end_mask_0 = const()[name = tensor("op_2082_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2082_cast_fp16 = slice_by_index(begin = var_2082_begin_0, end = var_2082_end_0, end_mask = var_2082_end_mask_0, x = var_2054_cast_fp16)[name = tensor("op_2082_cast_fp16")]; tensor var_2083_begin_0 = const()[name = tensor("op_2083_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_2083_end_0 = const()[name = tensor("op_2083_end_0"), val = tensor([2, 80, 1, 512])]; tensor var_2083_end_mask_0 = const()[name = tensor("op_2083_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2083_cast_fp16 = slice_by_index(begin = var_2083_begin_0, end = var_2083_end_0, end_mask = var_2083_end_mask_0, x = var_2058_cast_fp16)[name = tensor("op_2083_cast_fp16")]; tensor var_2084_begin_0 = const()[name = tensor("op_2084_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_2084_end_0 = const()[name = tensor("op_2084_end_0"), val = tensor([2, 80, 1, 1024])]; tensor var_2084_end_mask_0 = const()[name = tensor("op_2084_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2084_cast_fp16 = slice_by_index(begin = var_2084_begin_0, end = var_2084_end_0, end_mask = var_2084_end_mask_0, x = var_2058_cast_fp16)[name = tensor("op_2084_cast_fp16")]; tensor var_2085_begin_0 = const()[name = tensor("op_2085_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_2085_end_0 = const()[name = tensor("op_2085_end_0"), val = tensor([2, 80, 1, 512])]; tensor var_2085_end_mask_0 = const()[name = tensor("op_2085_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2085_cast_fp16 = slice_by_index(begin = var_2085_begin_0, end = var_2085_end_0, end_mask = var_2085_end_mask_0, x = var_2062_cast_fp16)[name = tensor("op_2085_cast_fp16")]; tensor var_2086_begin_0 = const()[name = tensor("op_2086_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_2086_end_0 = const()[name = tensor("op_2086_end_0"), val = tensor([2, 80, 1, 1024])]; tensor var_2086_end_mask_0 = const()[name = tensor("op_2086_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2086_cast_fp16 = slice_by_index(begin = var_2086_begin_0, end = var_2086_end_0, end_mask = var_2086_end_mask_0, x = var_2062_cast_fp16)[name = tensor("op_2086_cast_fp16")]; tensor var_2087_begin_0 = const()[name = tensor("op_2087_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_2087_end_0 = const()[name = tensor("op_2087_end_0"), val = tensor([2, 80, 1, 512])]; tensor var_2087_end_mask_0 = const()[name = tensor("op_2087_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2087_cast_fp16 = slice_by_index(begin = var_2087_begin_0, end = var_2087_end_0, end_mask = var_2087_end_mask_0, x = var_2066_cast_fp16)[name = tensor("op_2087_cast_fp16")]; tensor var_2088_begin_0 = const()[name = tensor("op_2088_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_2088_end_0 = const()[name = tensor("op_2088_end_0"), val = tensor([2, 80, 1, 1024])]; tensor var_2088_end_mask_0 = const()[name = tensor("op_2088_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2088_cast_fp16 = slice_by_index(begin = var_2088_begin_0, end = var_2088_end_0, end_mask = var_2088_end_mask_0, x = var_2066_cast_fp16)[name = tensor("op_2088_cast_fp16")]; tensor var_2089_begin_0 = const()[name = tensor("op_2089_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_2089_end_0 = const()[name = tensor("op_2089_end_0"), val = tensor([2, 80, 1, 512])]; tensor var_2089_end_mask_0 = const()[name = tensor("op_2089_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2089_cast_fp16 = slice_by_index(begin = var_2089_begin_0, end = var_2089_end_0, end_mask = var_2089_end_mask_0, x = var_2070_cast_fp16)[name = tensor("op_2089_cast_fp16")]; tensor var_2090_begin_0 = const()[name = tensor("op_2090_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_2090_end_0 = const()[name = tensor("op_2090_end_0"), val = tensor([2, 80, 1, 1024])]; tensor var_2090_end_mask_0 = const()[name = tensor("op_2090_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2090_cast_fp16 = slice_by_index(begin = var_2090_begin_0, end = var_2090_end_0, end_mask = var_2090_end_mask_0, x = var_2070_cast_fp16)[name = tensor("op_2090_cast_fp16")]; tensor var_2091_begin_0 = const()[name = tensor("op_2091_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_2091_end_0 = const()[name = tensor("op_2091_end_0"), val = tensor([2, 80, 1, 512])]; tensor var_2091_end_mask_0 = const()[name = tensor("op_2091_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2091_cast_fp16 = slice_by_index(begin = var_2091_begin_0, end = var_2091_end_0, end_mask = var_2091_end_mask_0, x = var_2074_cast_fp16)[name = tensor("op_2091_cast_fp16")]; tensor var_2092_begin_0 = const()[name = tensor("op_2092_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_2092_end_0 = const()[name = tensor("op_2092_end_0"), val = tensor([2, 80, 1, 1024])]; tensor var_2092_end_mask_0 = const()[name = tensor("op_2092_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2092_cast_fp16 = slice_by_index(begin = var_2092_begin_0, end = var_2092_end_0, end_mask = var_2092_end_mask_0, x = var_2074_cast_fp16)[name = tensor("op_2092_cast_fp16")]; tensor k_15_perm_0 = const()[name = tensor("k_15_perm_0"), val = tensor([0, 3, 2, 1])]; tensor var_2097_begin_0 = const()[name = tensor("op_2097_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_2097_end_0 = const()[name = tensor("op_2097_end_0"), val = tensor([2, 77, 1, 80])]; tensor var_2097_end_mask_0 = const()[name = tensor("op_2097_end_mask_0"), val = tensor([true, true, true, false])]; tensor transpose_14 = transpose(perm = k_15_perm_0, x = k_13_cast_fp16)[name = tensor("transpose_14")]; tensor var_2097_cast_fp16 = slice_by_index(begin = var_2097_begin_0, end = var_2097_end_0, end_mask = var_2097_end_mask_0, x = transpose_14)[name = tensor("op_2097_cast_fp16")]; tensor var_2101_begin_0 = const()[name = tensor("op_2101_begin_0"), val = tensor([0, 0, 0, 80])]; tensor var_2101_end_0 = const()[name = tensor("op_2101_end_0"), val = tensor([2, 77, 1, 160])]; tensor var_2101_end_mask_0 = const()[name = tensor("op_2101_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2101_cast_fp16 = slice_by_index(begin = var_2101_begin_0, end = var_2101_end_0, end_mask = var_2101_end_mask_0, x = transpose_14)[name = tensor("op_2101_cast_fp16")]; tensor var_2105_begin_0 = const()[name = tensor("op_2105_begin_0"), val = tensor([0, 0, 0, 160])]; tensor var_2105_end_0 = const()[name = tensor("op_2105_end_0"), val = tensor([2, 77, 1, 240])]; tensor var_2105_end_mask_0 = const()[name = tensor("op_2105_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2105_cast_fp16 = slice_by_index(begin = var_2105_begin_0, end = var_2105_end_0, end_mask = var_2105_end_mask_0, x = transpose_14)[name = tensor("op_2105_cast_fp16")]; tensor var_2109_begin_0 = const()[name = tensor("op_2109_begin_0"), val = tensor([0, 0, 0, 240])]; tensor var_2109_end_0 = const()[name = tensor("op_2109_end_0"), val = tensor([2, 77, 1, 320])]; tensor var_2109_end_mask_0 = const()[name = tensor("op_2109_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2109_cast_fp16 = slice_by_index(begin = var_2109_begin_0, end = var_2109_end_0, end_mask = var_2109_end_mask_0, x = transpose_14)[name = tensor("op_2109_cast_fp16")]; tensor var_2113_begin_0 = const()[name = tensor("op_2113_begin_0"), val = tensor([0, 0, 0, 320])]; tensor var_2113_end_0 = const()[name = tensor("op_2113_end_0"), val = tensor([2, 77, 1, 400])]; tensor var_2113_end_mask_0 = const()[name = tensor("op_2113_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2113_cast_fp16 = slice_by_index(begin = var_2113_begin_0, end = var_2113_end_0, end_mask = var_2113_end_mask_0, x = transpose_14)[name = tensor("op_2113_cast_fp16")]; tensor var_2117_begin_0 = const()[name = tensor("op_2117_begin_0"), val = tensor([0, 0, 0, 400])]; tensor var_2117_end_0 = const()[name = tensor("op_2117_end_0"), val = tensor([2, 77, 1, 480])]; tensor var_2117_end_mask_0 = const()[name = tensor("op_2117_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2117_cast_fp16 = slice_by_index(begin = var_2117_begin_0, end = var_2117_end_0, end_mask = var_2117_end_mask_0, x = transpose_14)[name = tensor("op_2117_cast_fp16")]; tensor var_2121_begin_0 = const()[name = tensor("op_2121_begin_0"), val = tensor([0, 0, 0, 480])]; tensor var_2121_end_0 = const()[name = tensor("op_2121_end_0"), val = tensor([2, 77, 1, 560])]; tensor var_2121_end_mask_0 = const()[name = tensor("op_2121_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2121_cast_fp16 = slice_by_index(begin = var_2121_begin_0, end = var_2121_end_0, end_mask = var_2121_end_mask_0, x = transpose_14)[name = tensor("op_2121_cast_fp16")]; tensor var_2125_begin_0 = const()[name = tensor("op_2125_begin_0"), val = tensor([0, 0, 0, 560])]; tensor var_2125_end_0 = const()[name = tensor("op_2125_end_0"), val = tensor([2, 77, 1, 640])]; tensor var_2125_end_mask_0 = const()[name = tensor("op_2125_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2125_cast_fp16 = slice_by_index(begin = var_2125_begin_0, end = var_2125_end_0, end_mask = var_2125_end_mask_0, x = transpose_14)[name = tensor("op_2125_cast_fp16")]; tensor var_2127_begin_0 = const()[name = tensor("op_2127_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_2127_end_0 = const()[name = tensor("op_2127_end_0"), val = tensor([2, 80, 1, 77])]; tensor var_2127_end_mask_0 = const()[name = tensor("op_2127_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_2127_cast_fp16 = slice_by_index(begin = var_2127_begin_0, end = var_2127_end_0, end_mask = var_2127_end_mask_0, x = v_7_cast_fp16)[name = tensor("op_2127_cast_fp16")]; tensor var_2131_begin_0 = const()[name = tensor("op_2131_begin_0"), val = tensor([0, 80, 0, 0])]; tensor var_2131_end_0 = const()[name = tensor("op_2131_end_0"), val = tensor([2, 160, 1, 77])]; tensor var_2131_end_mask_0 = const()[name = tensor("op_2131_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_2131_cast_fp16 = slice_by_index(begin = var_2131_begin_0, end = var_2131_end_0, end_mask = var_2131_end_mask_0, x = v_7_cast_fp16)[name = tensor("op_2131_cast_fp16")]; tensor var_2135_begin_0 = const()[name = tensor("op_2135_begin_0"), val = tensor([0, 160, 0, 0])]; tensor var_2135_end_0 = const()[name = tensor("op_2135_end_0"), val = tensor([2, 240, 1, 77])]; tensor var_2135_end_mask_0 = const()[name = tensor("op_2135_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_2135_cast_fp16 = slice_by_index(begin = var_2135_begin_0, end = var_2135_end_0, end_mask = var_2135_end_mask_0, x = v_7_cast_fp16)[name = tensor("op_2135_cast_fp16")]; tensor var_2139_begin_0 = const()[name = tensor("op_2139_begin_0"), val = tensor([0, 240, 0, 0])]; tensor var_2139_end_0 = const()[name = tensor("op_2139_end_0"), val = tensor([2, 320, 1, 77])]; tensor var_2139_end_mask_0 = const()[name = tensor("op_2139_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_2139_cast_fp16 = slice_by_index(begin = var_2139_begin_0, end = var_2139_end_0, end_mask = var_2139_end_mask_0, x = v_7_cast_fp16)[name = tensor("op_2139_cast_fp16")]; tensor var_2143_begin_0 = const()[name = tensor("op_2143_begin_0"), val = tensor([0, 320, 0, 0])]; tensor var_2143_end_0 = const()[name = tensor("op_2143_end_0"), val = tensor([2, 400, 1, 77])]; tensor var_2143_end_mask_0 = const()[name = tensor("op_2143_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_2143_cast_fp16 = slice_by_index(begin = var_2143_begin_0, end = var_2143_end_0, end_mask = var_2143_end_mask_0, x = v_7_cast_fp16)[name = tensor("op_2143_cast_fp16")]; tensor var_2147_begin_0 = const()[name = tensor("op_2147_begin_0"), val = tensor([0, 400, 0, 0])]; tensor var_2147_end_0 = const()[name = tensor("op_2147_end_0"), val = tensor([2, 480, 1, 77])]; tensor var_2147_end_mask_0 = const()[name = tensor("op_2147_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_2147_cast_fp16 = slice_by_index(begin = var_2147_begin_0, end = var_2147_end_0, end_mask = var_2147_end_mask_0, x = v_7_cast_fp16)[name = tensor("op_2147_cast_fp16")]; tensor var_2151_begin_0 = const()[name = tensor("op_2151_begin_0"), val = tensor([0, 480, 0, 0])]; tensor var_2151_end_0 = const()[name = tensor("op_2151_end_0"), val = tensor([2, 560, 1, 77])]; tensor var_2151_end_mask_0 = const()[name = tensor("op_2151_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_2151_cast_fp16 = slice_by_index(begin = var_2151_begin_0, end = var_2151_end_0, end_mask = var_2151_end_mask_0, x = v_7_cast_fp16)[name = tensor("op_2151_cast_fp16")]; tensor var_2155_begin_0 = const()[name = tensor("op_2155_begin_0"), val = tensor([0, 560, 0, 0])]; tensor var_2155_end_0 = const()[name = tensor("op_2155_end_0"), val = tensor([2, 640, 1, 77])]; tensor var_2155_end_mask_0 = const()[name = tensor("op_2155_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_2155_cast_fp16 = slice_by_index(begin = var_2155_begin_0, end = var_2155_end_0, end_mask = var_2155_end_mask_0, x = v_7_cast_fp16)[name = tensor("op_2155_cast_fp16")]; tensor var_2159_equation_0 = const()[name = tensor("op_2159_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_2159_cast_fp16 = einsum(equation = var_2159_equation_0, values = (var_2097_cast_fp16, var_2077_cast_fp16))[name = tensor("op_2159_cast_fp16")]; tensor var_2160_to_fp16 = const()[name = tensor("op_2160_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_289_cast_fp16 = mul(x = var_2159_cast_fp16, y = var_2160_to_fp16)[name = tensor("aw_chunk_289_cast_fp16")]; tensor var_2163_equation_0 = const()[name = tensor("op_2163_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_2163_cast_fp16 = einsum(equation = var_2163_equation_0, values = (var_2097_cast_fp16, var_2078_cast_fp16))[name = tensor("op_2163_cast_fp16")]; tensor var_2164_to_fp16 = const()[name = tensor("op_2164_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_291_cast_fp16 = mul(x = var_2163_cast_fp16, y = var_2164_to_fp16)[name = tensor("aw_chunk_291_cast_fp16")]; tensor var_2167_equation_0 = const()[name = tensor("op_2167_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_2167_cast_fp16 = einsum(equation = var_2167_equation_0, values = (var_2101_cast_fp16, var_2079_cast_fp16))[name = tensor("op_2167_cast_fp16")]; tensor var_2168_to_fp16 = const()[name = tensor("op_2168_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_293_cast_fp16 = mul(x = var_2167_cast_fp16, y = var_2168_to_fp16)[name = tensor("aw_chunk_293_cast_fp16")]; tensor var_2171_equation_0 = const()[name = tensor("op_2171_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_2171_cast_fp16 = einsum(equation = var_2171_equation_0, values = (var_2101_cast_fp16, var_2080_cast_fp16))[name = tensor("op_2171_cast_fp16")]; tensor var_2172_to_fp16 = const()[name = tensor("op_2172_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_295_cast_fp16 = mul(x = var_2171_cast_fp16, y = var_2172_to_fp16)[name = tensor("aw_chunk_295_cast_fp16")]; tensor var_2175_equation_0 = const()[name = tensor("op_2175_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_2175_cast_fp16 = einsum(equation = var_2175_equation_0, values = (var_2105_cast_fp16, var_2081_cast_fp16))[name = tensor("op_2175_cast_fp16")]; tensor var_2176_to_fp16 = const()[name = tensor("op_2176_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_297_cast_fp16 = mul(x = var_2175_cast_fp16, y = var_2176_to_fp16)[name = tensor("aw_chunk_297_cast_fp16")]; tensor var_2179_equation_0 = const()[name = tensor("op_2179_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_2179_cast_fp16 = einsum(equation = var_2179_equation_0, values = (var_2105_cast_fp16, var_2082_cast_fp16))[name = tensor("op_2179_cast_fp16")]; tensor var_2180_to_fp16 = const()[name = tensor("op_2180_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_299_cast_fp16 = mul(x = var_2179_cast_fp16, y = var_2180_to_fp16)[name = tensor("aw_chunk_299_cast_fp16")]; tensor var_2183_equation_0 = const()[name = tensor("op_2183_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_2183_cast_fp16 = einsum(equation = var_2183_equation_0, values = (var_2109_cast_fp16, var_2083_cast_fp16))[name = tensor("op_2183_cast_fp16")]; tensor var_2184_to_fp16 = const()[name = tensor("op_2184_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_301_cast_fp16 = mul(x = var_2183_cast_fp16, y = var_2184_to_fp16)[name = tensor("aw_chunk_301_cast_fp16")]; tensor var_2187_equation_0 = const()[name = tensor("op_2187_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_2187_cast_fp16 = einsum(equation = var_2187_equation_0, values = (var_2109_cast_fp16, var_2084_cast_fp16))[name = tensor("op_2187_cast_fp16")]; tensor var_2188_to_fp16 = const()[name = tensor("op_2188_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_303_cast_fp16 = mul(x = var_2187_cast_fp16, y = var_2188_to_fp16)[name = tensor("aw_chunk_303_cast_fp16")]; tensor var_2191_equation_0 = const()[name = tensor("op_2191_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_2191_cast_fp16 = einsum(equation = var_2191_equation_0, values = (var_2113_cast_fp16, var_2085_cast_fp16))[name = tensor("op_2191_cast_fp16")]; tensor var_2192_to_fp16 = const()[name = tensor("op_2192_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_305_cast_fp16 = mul(x = var_2191_cast_fp16, y = var_2192_to_fp16)[name = tensor("aw_chunk_305_cast_fp16")]; tensor var_2195_equation_0 = const()[name = tensor("op_2195_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_2195_cast_fp16 = einsum(equation = var_2195_equation_0, values = (var_2113_cast_fp16, var_2086_cast_fp16))[name = tensor("op_2195_cast_fp16")]; tensor var_2196_to_fp16 = const()[name = tensor("op_2196_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_307_cast_fp16 = mul(x = var_2195_cast_fp16, y = var_2196_to_fp16)[name = tensor("aw_chunk_307_cast_fp16")]; tensor var_2199_equation_0 = const()[name = tensor("op_2199_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_2199_cast_fp16 = einsum(equation = var_2199_equation_0, values = (var_2117_cast_fp16, var_2087_cast_fp16))[name = tensor("op_2199_cast_fp16")]; tensor var_2200_to_fp16 = const()[name = tensor("op_2200_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_309_cast_fp16 = mul(x = var_2199_cast_fp16, y = var_2200_to_fp16)[name = tensor("aw_chunk_309_cast_fp16")]; tensor var_2203_equation_0 = const()[name = tensor("op_2203_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_2203_cast_fp16 = einsum(equation = var_2203_equation_0, values = (var_2117_cast_fp16, var_2088_cast_fp16))[name = tensor("op_2203_cast_fp16")]; tensor var_2204_to_fp16 = const()[name = tensor("op_2204_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_311_cast_fp16 = mul(x = var_2203_cast_fp16, y = var_2204_to_fp16)[name = tensor("aw_chunk_311_cast_fp16")]; tensor var_2207_equation_0 = const()[name = tensor("op_2207_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_2207_cast_fp16 = einsum(equation = var_2207_equation_0, values = (var_2121_cast_fp16, var_2089_cast_fp16))[name = tensor("op_2207_cast_fp16")]; tensor var_2208_to_fp16 = const()[name = tensor("op_2208_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_313_cast_fp16 = mul(x = var_2207_cast_fp16, y = var_2208_to_fp16)[name = tensor("aw_chunk_313_cast_fp16")]; tensor var_2211_equation_0 = const()[name = tensor("op_2211_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_2211_cast_fp16 = einsum(equation = var_2211_equation_0, values = (var_2121_cast_fp16, var_2090_cast_fp16))[name = tensor("op_2211_cast_fp16")]; tensor var_2212_to_fp16 = const()[name = tensor("op_2212_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_315_cast_fp16 = mul(x = var_2211_cast_fp16, y = var_2212_to_fp16)[name = tensor("aw_chunk_315_cast_fp16")]; tensor var_2215_equation_0 = const()[name = tensor("op_2215_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_2215_cast_fp16 = einsum(equation = var_2215_equation_0, values = (var_2125_cast_fp16, var_2091_cast_fp16))[name = tensor("op_2215_cast_fp16")]; tensor var_2216_to_fp16 = const()[name = tensor("op_2216_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_317_cast_fp16 = mul(x = var_2215_cast_fp16, y = var_2216_to_fp16)[name = tensor("aw_chunk_317_cast_fp16")]; tensor var_2219_equation_0 = const()[name = tensor("op_2219_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_2219_cast_fp16 = einsum(equation = var_2219_equation_0, values = (var_2125_cast_fp16, var_2092_cast_fp16))[name = tensor("op_2219_cast_fp16")]; tensor var_2220_to_fp16 = const()[name = tensor("op_2220_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_319_cast_fp16 = mul(x = var_2219_cast_fp16, y = var_2220_to_fp16)[name = tensor("aw_chunk_319_cast_fp16")]; tensor var_2222_cast_fp16 = softmax(axis = var_1630, x = aw_chunk_289_cast_fp16)[name = tensor("op_2222_cast_fp16")]; tensor var_2223_cast_fp16 = softmax(axis = var_1630, x = aw_chunk_291_cast_fp16)[name = tensor("op_2223_cast_fp16")]; tensor var_2224_cast_fp16 = softmax(axis = var_1630, x = aw_chunk_293_cast_fp16)[name = tensor("op_2224_cast_fp16")]; tensor var_2225_cast_fp16 = softmax(axis = var_1630, x = aw_chunk_295_cast_fp16)[name = tensor("op_2225_cast_fp16")]; tensor var_2226_cast_fp16 = softmax(axis = var_1630, x = aw_chunk_297_cast_fp16)[name = tensor("op_2226_cast_fp16")]; tensor var_2227_cast_fp16 = softmax(axis = var_1630, x = aw_chunk_299_cast_fp16)[name = tensor("op_2227_cast_fp16")]; tensor var_2228_cast_fp16 = softmax(axis = var_1630, x = aw_chunk_301_cast_fp16)[name = tensor("op_2228_cast_fp16")]; tensor var_2229_cast_fp16 = softmax(axis = var_1630, x = aw_chunk_303_cast_fp16)[name = tensor("op_2229_cast_fp16")]; tensor var_2230_cast_fp16 = softmax(axis = var_1630, x = aw_chunk_305_cast_fp16)[name = tensor("op_2230_cast_fp16")]; tensor var_2231_cast_fp16 = softmax(axis = var_1630, x = aw_chunk_307_cast_fp16)[name = tensor("op_2231_cast_fp16")]; tensor var_2232_cast_fp16 = softmax(axis = var_1630, x = aw_chunk_309_cast_fp16)[name = tensor("op_2232_cast_fp16")]; tensor var_2233_cast_fp16 = softmax(axis = var_1630, x = aw_chunk_311_cast_fp16)[name = tensor("op_2233_cast_fp16")]; tensor var_2234_cast_fp16 = softmax(axis = var_1630, x = aw_chunk_313_cast_fp16)[name = tensor("op_2234_cast_fp16")]; tensor var_2235_cast_fp16 = softmax(axis = var_1630, x = aw_chunk_315_cast_fp16)[name = tensor("op_2235_cast_fp16")]; tensor var_2236_cast_fp16 = softmax(axis = var_1630, x = aw_chunk_317_cast_fp16)[name = tensor("op_2236_cast_fp16")]; tensor var_2237_cast_fp16 = softmax(axis = var_1630, x = aw_chunk_319_cast_fp16)[name = tensor("op_2237_cast_fp16")]; tensor var_2239_equation_0 = const()[name = tensor("op_2239_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_2239_cast_fp16 = einsum(equation = var_2239_equation_0, values = (var_2127_cast_fp16, var_2222_cast_fp16))[name = tensor("op_2239_cast_fp16")]; tensor var_2241_equation_0 = const()[name = tensor("op_2241_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_2241_cast_fp16 = einsum(equation = var_2241_equation_0, values = (var_2127_cast_fp16, var_2223_cast_fp16))[name = tensor("op_2241_cast_fp16")]; tensor var_2243_equation_0 = const()[name = tensor("op_2243_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_2243_cast_fp16 = einsum(equation = var_2243_equation_0, values = (var_2131_cast_fp16, var_2224_cast_fp16))[name = tensor("op_2243_cast_fp16")]; tensor var_2245_equation_0 = const()[name = tensor("op_2245_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_2245_cast_fp16 = einsum(equation = var_2245_equation_0, values = (var_2131_cast_fp16, var_2225_cast_fp16))[name = tensor("op_2245_cast_fp16")]; tensor var_2247_equation_0 = const()[name = tensor("op_2247_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_2247_cast_fp16 = einsum(equation = var_2247_equation_0, values = (var_2135_cast_fp16, var_2226_cast_fp16))[name = tensor("op_2247_cast_fp16")]; tensor var_2249_equation_0 = const()[name = tensor("op_2249_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_2249_cast_fp16 = einsum(equation = var_2249_equation_0, values = (var_2135_cast_fp16, var_2227_cast_fp16))[name = tensor("op_2249_cast_fp16")]; tensor var_2251_equation_0 = const()[name = tensor("op_2251_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_2251_cast_fp16 = einsum(equation = var_2251_equation_0, values = (var_2139_cast_fp16, var_2228_cast_fp16))[name = tensor("op_2251_cast_fp16")]; tensor var_2253_equation_0 = const()[name = tensor("op_2253_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_2253_cast_fp16 = einsum(equation = var_2253_equation_0, values = (var_2139_cast_fp16, var_2229_cast_fp16))[name = tensor("op_2253_cast_fp16")]; tensor var_2255_equation_0 = const()[name = tensor("op_2255_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_2255_cast_fp16 = einsum(equation = var_2255_equation_0, values = (var_2143_cast_fp16, var_2230_cast_fp16))[name = tensor("op_2255_cast_fp16")]; tensor var_2257_equation_0 = const()[name = tensor("op_2257_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_2257_cast_fp16 = einsum(equation = var_2257_equation_0, values = (var_2143_cast_fp16, var_2231_cast_fp16))[name = tensor("op_2257_cast_fp16")]; tensor var_2259_equation_0 = const()[name = tensor("op_2259_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_2259_cast_fp16 = einsum(equation = var_2259_equation_0, values = (var_2147_cast_fp16, var_2232_cast_fp16))[name = tensor("op_2259_cast_fp16")]; tensor var_2261_equation_0 = const()[name = tensor("op_2261_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_2261_cast_fp16 = einsum(equation = var_2261_equation_0, values = (var_2147_cast_fp16, var_2233_cast_fp16))[name = tensor("op_2261_cast_fp16")]; tensor var_2263_equation_0 = const()[name = tensor("op_2263_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_2263_cast_fp16 = einsum(equation = var_2263_equation_0, values = (var_2151_cast_fp16, var_2234_cast_fp16))[name = tensor("op_2263_cast_fp16")]; tensor var_2265_equation_0 = const()[name = tensor("op_2265_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_2265_cast_fp16 = einsum(equation = var_2265_equation_0, values = (var_2151_cast_fp16, var_2235_cast_fp16))[name = tensor("op_2265_cast_fp16")]; tensor var_2267_equation_0 = const()[name = tensor("op_2267_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_2267_cast_fp16 = einsum(equation = var_2267_equation_0, values = (var_2155_cast_fp16, var_2236_cast_fp16))[name = tensor("op_2267_cast_fp16")]; tensor var_2269_equation_0 = const()[name = tensor("op_2269_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_2269_cast_fp16 = einsum(equation = var_2269_equation_0, values = (var_2155_cast_fp16, var_2237_cast_fp16))[name = tensor("op_2269_cast_fp16")]; tensor var_2271_interleave_0 = const()[name = tensor("op_2271_interleave_0"), val = tensor(false)]; tensor var_2271_cast_fp16 = concat(axis = var_1608, interleave = var_2271_interleave_0, values = (var_2239_cast_fp16, var_2241_cast_fp16))[name = tensor("op_2271_cast_fp16")]; tensor var_2273_interleave_0 = const()[name = tensor("op_2273_interleave_0"), val = tensor(false)]; tensor var_2273_cast_fp16 = concat(axis = var_1608, interleave = var_2273_interleave_0, values = (var_2243_cast_fp16, var_2245_cast_fp16))[name = tensor("op_2273_cast_fp16")]; tensor var_2275_interleave_0 = const()[name = tensor("op_2275_interleave_0"), val = tensor(false)]; tensor var_2275_cast_fp16 = concat(axis = var_1608, interleave = var_2275_interleave_0, values = (var_2247_cast_fp16, var_2249_cast_fp16))[name = tensor("op_2275_cast_fp16")]; tensor var_2277_interleave_0 = const()[name = tensor("op_2277_interleave_0"), val = tensor(false)]; tensor var_2277_cast_fp16 = concat(axis = var_1608, interleave = var_2277_interleave_0, values = (var_2251_cast_fp16, var_2253_cast_fp16))[name = tensor("op_2277_cast_fp16")]; tensor var_2279_interleave_0 = const()[name = tensor("op_2279_interleave_0"), val = tensor(false)]; tensor var_2279_cast_fp16 = concat(axis = var_1608, interleave = var_2279_interleave_0, values = (var_2255_cast_fp16, var_2257_cast_fp16))[name = tensor("op_2279_cast_fp16")]; tensor var_2281_interleave_0 = const()[name = tensor("op_2281_interleave_0"), val = tensor(false)]; tensor var_2281_cast_fp16 = concat(axis = var_1608, interleave = var_2281_interleave_0, values = (var_2259_cast_fp16, var_2261_cast_fp16))[name = tensor("op_2281_cast_fp16")]; tensor var_2283_interleave_0 = const()[name = tensor("op_2283_interleave_0"), val = tensor(false)]; tensor var_2283_cast_fp16 = concat(axis = var_1608, interleave = var_2283_interleave_0, values = (var_2263_cast_fp16, var_2265_cast_fp16))[name = tensor("op_2283_cast_fp16")]; tensor var_2285_interleave_0 = const()[name = tensor("op_2285_interleave_0"), val = tensor(false)]; tensor var_2285_cast_fp16 = concat(axis = var_1608, interleave = var_2285_interleave_0, values = (var_2267_cast_fp16, var_2269_cast_fp16))[name = tensor("op_2285_cast_fp16")]; tensor input_55_interleave_0 = const()[name = tensor("input_55_interleave_0"), val = tensor(false)]; tensor input_55_cast_fp16 = concat(axis = var_1630, interleave = input_55_interleave_0, values = (var_2271_cast_fp16, var_2273_cast_fp16, var_2275_cast_fp16, var_2277_cast_fp16, var_2279_cast_fp16, var_2281_cast_fp16, var_2283_cast_fp16, var_2285_cast_fp16))[name = tensor("input_55_cast_fp16")]; tensor var_2291 = const()[name = tensor("op_2291"), val = tensor([1, 1])]; tensor var_2293 = const()[name = tensor("op_2293"), val = tensor([1, 1])]; tensor var_2295_pad_type_0 = const()[name = tensor("op_2295_pad_type_0"), val = tensor("custom")]; tensor var_2295_pad_0 = const()[name = tensor("op_2295_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(13381376))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(13688640))), 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(13688832)))]; tensor var_2295_cast_fp16 = conv(bias = down_blocks_1_attentions_0_transformer_blocks_0_attn2_to_out_0_bias_to_fp16, dilations = var_2293, groups = var_1630, pad = var_2295_pad_0, pad_type = var_2295_pad_type_0, strides = var_2291, weight = down_blocks_1_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16_palettized, x = input_55_cast_fp16)[name = tensor("op_2295_cast_fp16")]; tensor inputs_11_cast_fp16 = add(x = var_2295_cast_fp16, y = inputs_9_cast_fp16)[name = tensor("inputs_11_cast_fp16")]; tensor var_2299 = const()[name = tensor("op_2299"), val = tensor([1])]; tensor channels_mean_11_cast_fp16 = reduce_mean(axes = var_2299, keep_dims = var_1625, x = inputs_11_cast_fp16)[name = tensor("channels_mean_11_cast_fp16")]; tensor zero_mean_11_cast_fp16 = sub(x = inputs_11_cast_fp16, y = channels_mean_11_cast_fp16)[name = tensor("zero_mean_11_cast_fp16")]; tensor zero_mean_sq_11_cast_fp16 = mul(x = zero_mean_11_cast_fp16, y = zero_mean_11_cast_fp16)[name = tensor("zero_mean_sq_11_cast_fp16")]; tensor var_2303 = const()[name = tensor("op_2303"), val = tensor([1])]; tensor var_2304_cast_fp16 = reduce_mean(axes = var_2303, keep_dims = var_1625, x = zero_mean_sq_11_cast_fp16)[name = tensor("op_2304_cast_fp16")]; tensor var_2305_to_fp16 = const()[name = tensor("op_2305_to_fp16"), val = tensor(0x1.5p-17)]; tensor var_2306_cast_fp16 = add(x = var_2304_cast_fp16, y = var_2305_to_fp16)[name = tensor("op_2306_cast_fp16")]; tensor denom_11_epsilon_0_to_fp16 = const()[name = tensor("denom_11_epsilon_0_to_fp16"), val = tensor(0x1p-24)]; tensor denom_11_cast_fp16 = rsqrt(epsilon = denom_11_epsilon_0_to_fp16, x = var_2306_cast_fp16)[name = tensor("denom_11_cast_fp16")]; tensor out_11_cast_fp16 = mul(x = zero_mean_11_cast_fp16, y = denom_11_cast_fp16)[name = tensor("out_11_cast_fp16")]; tensor var_2310_to_fp16 = const()[name = tensor("op_2310_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(13690176)))]; tensor var_2311_cast_fp16 = add(x = out_11_cast_fp16, y = var_2310_to_fp16)[name = tensor("op_2311_cast_fp16")]; tensor var_2313_to_fp16 = const()[name = tensor("op_2313_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(13691520)))]; tensor input_57_cast_fp16 = mul(x = var_2311_cast_fp16, y = var_2313_to_fp16)[name = tensor("input_57_cast_fp16")]; tensor var_2321 = const()[name = tensor("op_2321"), val = tensor([1, 1])]; tensor var_2323 = const()[name = tensor("op_2323"), val = tensor([1, 1])]; tensor var_2325_pad_type_0 = const()[name = tensor("op_2325_pad_type_0"), val = tensor("custom")]; tensor var_2325_pad_0 = const()[name = tensor("op_2325_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(13692864))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(16150528))), 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(16150720))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(16154624))), name = tensor("down_blocks_1_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16_palettized"), shape = tensor([5120])]; tensor var_2325_cast_fp16 = conv(bias = down_blocks_1_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16_palettized, dilations = var_2323, groups = var_1630, pad = var_2325_pad_0, pad_type = var_2325_pad_type_0, strides = var_2321, weight = down_blocks_1_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16_palettized, x = input_57_cast_fp16)[name = tensor("op_2325_cast_fp16")]; tensor var_2326_split_sizes_0 = const()[name = tensor("op_2326_split_sizes_0"), val = tensor([2560, 2560])]; tensor var_2326_axis_0 = const()[name = tensor("op_2326_axis_0"), val = tensor(1)]; tensor var_2326_cast_fp16_0, tensor var_2326_cast_fp16_1 = split(axis = var_2326_axis_0, split_sizes = var_2326_split_sizes_0, x = var_2325_cast_fp16)[name = tensor("op_2326_cast_fp16")]; tensor var_2328_mode_0 = const()[name = tensor("op_2328_mode_0"), val = tensor("EXACT")]; tensor var_2328_cast_fp16 = gelu(mode = var_2328_mode_0, x = var_2326_cast_fp16_1)[name = tensor("op_2328_cast_fp16")]; tensor input_59_cast_fp16 = mul(x = var_2326_cast_fp16_0, y = var_2328_cast_fp16)[name = tensor("input_59_cast_fp16")]; tensor var_2332 = const()[name = tensor("op_2332"), val = tensor([1, 1])]; tensor var_2334 = const()[name = tensor("op_2334"), val = tensor([1, 1])]; tensor var_2336_pad_type_0 = const()[name = tensor("op_2336_pad_type_0"), val = tensor("custom")]; tensor var_2336_pad_0 = const()[name = tensor("op_2336_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(16154816))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(17383680))), 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(17383872)))]; tensor var_2336_cast_fp16 = conv(bias = down_blocks_1_attentions_0_transformer_blocks_0_ff_net_2_bias_to_fp16, dilations = var_2334, groups = var_1630, pad = var_2336_pad_0, pad_type = var_2336_pad_type_0, strides = var_2332, weight = down_blocks_1_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16_palettized, x = input_59_cast_fp16)[name = tensor("op_2336_cast_fp16")]; tensor hidden_states_33_cast_fp16 = add(x = var_2336_cast_fp16, y = inputs_11_cast_fp16)[name = tensor("hidden_states_33_cast_fp16")]; tensor var_2338 = const()[name = tensor("op_2338"), val = tensor([2, 640, 32, 32])]; tensor input_61_cast_fp16 = reshape(shape = var_2338, x = hidden_states_33_cast_fp16)[name = tensor("input_61_cast_fp16")]; tensor var_2342 = const()[name = tensor("op_2342"), val = tensor([1, 1])]; tensor var_2344 = const()[name = tensor("op_2344"), 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_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(17385216))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(17692480))), 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(17692672)))]; tensor hidden_states_35_cast_fp16 = conv(bias = down_blocks_1_attentions_0_proj_out_bias_to_fp16, dilations = var_2344, groups = var_1630, pad = hidden_states_35_pad_0, pad_type = hidden_states_35_pad_type_0, strides = var_2342, weight = down_blocks_1_attentions_0_proj_out_weight_to_fp16_palettized, x = input_61_cast_fp16)[name = tensor("hidden_states_35_cast_fp16")]; tensor input_63_cast_fp16 = add(x = hidden_states_35_cast_fp16, y = hidden_states_23_cast_fp16)[name = tensor("input_63_cast_fp16")]; tensor var_2351 = const()[name = tensor("op_2351"), val = tensor([2, 2])]; tensor var_2353 = const()[name = tensor("op_2353"), val = tensor([1, 1])]; tensor input_65_pad_type_0 = const()[name = tensor("input_65_pad_type_0"), val = tensor("custom")]; tensor input_65_pad_0 = const()[name = tensor("input_65_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(17694016))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(20458880))), 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(20459072)))]; tensor input_65_cast_fp16 = conv(bias = down_blocks_1_downsamplers_0_conv_bias_to_fp16, dilations = var_2353, groups = var_1630, pad = input_65_pad_0, pad_type = input_65_pad_type_0, strides = var_2351, weight = down_blocks_1_downsamplers_0_conv_weight_to_fp16_palettized, x = input_63_cast_fp16)[name = tensor("input_65_cast_fp16")]; tensor var_2375 = const()[name = tensor("op_2375"), val = tensor(true)]; tensor var_2380 = const()[name = tensor("op_2380"), val = tensor(1)]; tensor reshape_24_shape_0 = const()[name = tensor("reshape_24_shape_0"), val = tensor([2, 32, 20, 16, 16])]; tensor reshape_24_cast_fp16 = reshape(shape = reshape_24_shape_0, x = input_65_cast_fp16)[name = tensor("reshape_24_cast_fp16")]; 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_fp16 = reduce_mean(axes = reduce_mean_18_axes_0, keep_dims = reduce_mean_18_keep_dims_0, x = reshape_24_cast_fp16)[name = tensor("reduce_mean_18_cast_fp16")]; tensor sub_12_cast_fp16 = sub(x = reshape_24_cast_fp16, y = reduce_mean_18_cast_fp16)[name = tensor("sub_12_cast_fp16")]; tensor square_6_cast_fp16 = square(x = sub_12_cast_fp16)[name = tensor("square_6_cast_fp16")]; 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_fp16 = reduce_mean(axes = reduce_mean_20_axes_0, keep_dims = reduce_mean_20_keep_dims_0, x = square_6_cast_fp16)[name = tensor("reduce_mean_20_cast_fp16")]; 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_fp16 = add(x = reduce_mean_20_cast_fp16, y = add_12_y_0_to_fp16)[name = tensor("add_12_cast_fp16")]; tensor sqrt_6_cast_fp16 = sqrt(x = add_12_cast_fp16)[name = tensor("sqrt_6_cast_fp16")]; tensor real_div_6_cast_fp16 = real_div(x = sub_12_cast_fp16, y = sqrt_6_cast_fp16)[name = tensor("real_div_6_cast_fp16")]; tensor reshape_25_shape_0 = const()[name = tensor("reshape_25_shape_0"), val = tensor([2, 640, 16, 16])]; tensor reshape_25_cast_fp16 = reshape(shape = reshape_25_shape_0, x = real_div_6_cast_fp16)[name = tensor("reshape_25_cast_fp16")]; 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(20460416)))]; 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(20461760)))]; 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_fp16 = 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_9_mean_0_to_fp16, variance = add_9_variance_0_to_fp16, x = reshape_25_cast_fp16)[name = tensor("add_13_cast_fp16")]; tensor input_69_cast_fp16 = silu(x = add_13_cast_fp16)[name = tensor("input_69_cast_fp16")]; tensor var_2397 = const()[name = tensor("op_2397"), val = tensor([1, 1])]; tensor var_2399 = const()[name = tensor("op_2399"), 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_2_resnets_0_conv1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(20463104))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(25992768))), 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(25992960)))]; tensor hidden_states_37_cast_fp16 = conv(bias = down_blocks_2_resnets_0_conv1_bias_to_fp16, dilations = var_2399, groups = var_2380, pad = hidden_states_37_pad_0, pad_type = hidden_states_37_pad_type_0, strides = var_2397, weight = down_blocks_2_resnets_0_conv1_weight_to_fp16_palettized, x = input_69_cast_fp16)[name = tensor("hidden_states_37_cast_fp16")]; tensor var_2405 = const()[name = tensor("op_2405"), val = tensor([1, 1])]; tensor var_2407 = const()[name = tensor("op_2407"), 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_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(25995584))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(27224448))), 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(27224640)))]; tensor temb_5_cast_fp16 = conv(bias = down_blocks_2_resnets_0_time_emb_proj_bias_to_fp16, dilations = var_2407, groups = var_2380, pad = temb_5_pad_0, pad_type = temb_5_pad_type_0, strides = var_2405, weight = down_blocks_2_resnets_0_time_emb_proj_weight_to_fp16_palettized, x = input_15_cast_fp16)[name = tensor("temb_5_cast_fp16")]; tensor input_73_cast_fp16 = add(x = hidden_states_37_cast_fp16, y = temb_5_cast_fp16)[name = tensor("input_73_cast_fp16")]; tensor reshape_28_shape_0 = const()[name = tensor("reshape_28_shape_0"), val = tensor([2, 32, 40, 16, 16])]; tensor reshape_28_cast_fp16 = reshape(shape = reshape_28_shape_0, x = input_73_cast_fp16)[name = tensor("reshape_28_cast_fp16")]; 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_fp16 = reduce_mean(axes = reduce_mean_21_axes_0, keep_dims = reduce_mean_21_keep_dims_0, x = reshape_28_cast_fp16)[name = tensor("reduce_mean_21_cast_fp16")]; tensor sub_14_cast_fp16 = sub(x = reshape_28_cast_fp16, y = reduce_mean_21_cast_fp16)[name = tensor("sub_14_cast_fp16")]; tensor square_7_cast_fp16 = square(x = sub_14_cast_fp16)[name = tensor("square_7_cast_fp16")]; 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_fp16 = reduce_mean(axes = reduce_mean_23_axes_0, keep_dims = reduce_mean_23_keep_dims_0, x = square_7_cast_fp16)[name = tensor("reduce_mean_23_cast_fp16")]; 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_fp16 = add(x = reduce_mean_23_cast_fp16, y = add_14_y_0_to_fp16)[name = tensor("add_14_cast_fp16")]; tensor sqrt_7_cast_fp16 = sqrt(x = add_14_cast_fp16)[name = tensor("sqrt_7_cast_fp16")]; tensor real_div_7_cast_fp16 = real_div(x = sub_14_cast_fp16, y = sqrt_7_cast_fp16)[name = tensor("real_div_7_cast_fp16")]; tensor reshape_29_shape_0 = const()[name = tensor("reshape_29_shape_0"), val = tensor([2, 1280, 16, 16])]; tensor reshape_29_cast_fp16 = reshape(shape = reshape_29_shape_0, x = real_div_7_cast_fp16)[name = tensor("reshape_29_cast_fp16")]; 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(27227264)))]; 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(27229888)))]; 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(27232512)))]; 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(27235136)))]; 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_fp16 = 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_fp16)[name = tensor("add_15_cast_fp16")]; tensor input_77_cast_fp16 = silu(x = add_15_cast_fp16)[name = tensor("input_77_cast_fp16")]; tensor var_2417 = const()[name = tensor("op_2417"), val = tensor([1, 1])]; tensor var_2419 = const()[name = tensor("op_2419"), 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_2_resnets_0_conv2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(27237760))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(38297024))), 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(38297216)))]; tensor hidden_states_39_cast_fp16 = conv(bias = down_blocks_2_resnets_0_conv2_bias_to_fp16, dilations = var_2419, groups = var_2380, pad = hidden_states_39_pad_0, pad_type = hidden_states_39_pad_type_0, strides = var_2417, weight = down_blocks_2_resnets_0_conv2_weight_to_fp16_palettized, x = input_77_cast_fp16)[name = tensor("hidden_states_39_cast_fp16")]; tensor var_2424 = const()[name = tensor("op_2424"), val = tensor([1, 1])]; tensor var_2426 = const()[name = tensor("op_2426"), 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(38299840))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(38914304))), 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(38914496)))]; tensor x_3_cast_fp16 = conv(bias = down_blocks_2_resnets_0_conv_shortcut_bias_to_fp16, dilations = var_2426, groups = var_2380, pad = x_3_pad_0, pad_type = x_3_pad_type_0, strides = var_2424, weight = down_blocks_2_resnets_0_conv_shortcut_weight_to_fp16_palettized, x = input_65_cast_fp16)[name = tensor("x_3_cast_fp16")]; tensor hidden_states_41_cast_fp16 = add(x = x_3_cast_fp16, y = hidden_states_39_cast_fp16)[name = tensor("hidden_states_41_cast_fp16")]; tensor reshape_32_shape_0 = const()[name = tensor("reshape_32_shape_0"), val = tensor([2, 32, 40, 16, 16])]; tensor reshape_32_cast_fp16 = reshape(shape = reshape_32_shape_0, x = hidden_states_41_cast_fp16)[name = tensor("reshape_32_cast_fp16")]; 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_fp16 = reduce_mean(axes = reduce_mean_24_axes_0, keep_dims = reduce_mean_24_keep_dims_0, x = reshape_32_cast_fp16)[name = tensor("reduce_mean_24_cast_fp16")]; tensor sub_16_cast_fp16 = sub(x = reshape_32_cast_fp16, y = reduce_mean_24_cast_fp16)[name = tensor("sub_16_cast_fp16")]; tensor square_8_cast_fp16 = square(x = sub_16_cast_fp16)[name = tensor("square_8_cast_fp16")]; 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_fp16 = reduce_mean(axes = reduce_mean_26_axes_0, keep_dims = reduce_mean_26_keep_dims_0, x = square_8_cast_fp16)[name = tensor("reduce_mean_26_cast_fp16")]; 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_fp16 = add(x = reduce_mean_26_cast_fp16, y = add_16_y_0_to_fp16)[name = tensor("add_16_cast_fp16")]; tensor sqrt_8_cast_fp16 = sqrt(x = add_16_cast_fp16)[name = tensor("sqrt_8_cast_fp16")]; tensor real_div_8_cast_fp16 = real_div(x = sub_16_cast_fp16, y = sqrt_8_cast_fp16)[name = tensor("real_div_8_cast_fp16")]; tensor reshape_33_shape_0 = const()[name = tensor("reshape_33_shape_0"), val = tensor([2, 1280, 16, 16])]; tensor reshape_33_cast_fp16 = reshape(shape = reshape_33_shape_0, x = real_div_8_cast_fp16)[name = tensor("reshape_33_cast_fp16")]; 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(38917120)))]; 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(38919744)))]; 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_fp16 = 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_fp16)[name = tensor("add_17_cast_fp16")]; tensor var_2446 = const()[name = tensor("op_2446"), val = tensor([1, 1])]; tensor var_2448 = const()[name = tensor("op_2448"), 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_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(38922368))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(40151232))), 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(40151424)))]; tensor hidden_states_43_cast_fp16 = conv(bias = down_blocks_2_attentions_0_proj_in_bias_to_fp16, dilations = var_2448, groups = var_2380, pad = hidden_states_43_pad_0, pad_type = hidden_states_43_pad_type_0, strides = var_2446, weight = down_blocks_2_attentions_0_proj_in_weight_to_fp16_palettized, x = add_17_cast_fp16)[name = tensor("hidden_states_43_cast_fp16")]; tensor var_2453 = const()[name = tensor("op_2453"), val = tensor([2, 1280, 1, 256])]; tensor inputs_13_cast_fp16 = reshape(shape = var_2453, x = hidden_states_43_cast_fp16)[name = tensor("inputs_13_cast_fp16")]; tensor var_2463 = const()[name = tensor("op_2463"), val = tensor([1])]; tensor channels_mean_13_cast_fp16 = reduce_mean(axes = var_2463, keep_dims = var_2375, x = inputs_13_cast_fp16)[name = tensor("channels_mean_13_cast_fp16")]; tensor zero_mean_13_cast_fp16 = sub(x = inputs_13_cast_fp16, y = channels_mean_13_cast_fp16)[name = tensor("zero_mean_13_cast_fp16")]; tensor zero_mean_sq_13_cast_fp16 = mul(x = zero_mean_13_cast_fp16, y = zero_mean_13_cast_fp16)[name = tensor("zero_mean_sq_13_cast_fp16")]; tensor var_2467 = const()[name = tensor("op_2467"), val = tensor([1])]; tensor var_2468_cast_fp16 = reduce_mean(axes = var_2467, keep_dims = var_2375, x = zero_mean_sq_13_cast_fp16)[name = tensor("op_2468_cast_fp16")]; tensor var_2469_to_fp16 = const()[name = tensor("op_2469_to_fp16"), val = tensor(0x1.5p-17)]; tensor var_2470_cast_fp16 = add(x = var_2468_cast_fp16, y = var_2469_to_fp16)[name = tensor("op_2470_cast_fp16")]; tensor denom_13_epsilon_0_to_fp16 = const()[name = tensor("denom_13_epsilon_0_to_fp16"), val = tensor(0x1p-24)]; tensor denom_13_cast_fp16 = rsqrt(epsilon = denom_13_epsilon_0_to_fp16, x = var_2470_cast_fp16)[name = tensor("denom_13_cast_fp16")]; tensor out_13_cast_fp16 = mul(x = zero_mean_13_cast_fp16, y = denom_13_cast_fp16)[name = tensor("out_13_cast_fp16")]; tensor var_2474_to_fp16 = const()[name = tensor("op_2474_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(40154048)))]; tensor var_2475_cast_fp16 = add(x = out_13_cast_fp16, y = var_2474_to_fp16)[name = tensor("op_2475_cast_fp16")]; tensor var_2477_to_fp16 = const()[name = tensor("op_2477_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(40156672)))]; tensor hidden_states_45_cast_fp16 = mul(x = var_2475_cast_fp16, y = var_2477_to_fp16)[name = tensor("hidden_states_45_cast_fp16")]; tensor var_2484 = const()[name = tensor("op_2484"), val = tensor([1, 1])]; tensor var_2486 = const()[name = tensor("op_2486"), 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_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(40159296))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(41388160))), 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_9_cast_fp16 = conv(dilations = var_2486, groups = var_2380, pad = q_9_pad_0, pad_type = q_9_pad_type_0, strides = var_2484, weight = down_blocks_2_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16_palettized, x = hidden_states_45_cast_fp16)[name = tensor("q_9_cast_fp16")]; tensor var_2490 = const()[name = tensor("op_2490"), val = tensor([1, 1])]; tensor var_2492 = const()[name = tensor("op_2492"), 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_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(41388352))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(42617216))), 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_17_cast_fp16 = conv(dilations = var_2492, groups = var_2380, pad = k_17_pad_0, pad_type = k_17_pad_type_0, strides = var_2490, weight = down_blocks_2_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16_palettized, x = hidden_states_45_cast_fp16)[name = tensor("k_17_cast_fp16")]; tensor var_2496 = const()[name = tensor("op_2496"), val = tensor([1, 1])]; tensor var_2498 = const()[name = tensor("op_2498"), 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_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(42617408))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(43846272))), 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_9_cast_fp16 = conv(dilations = var_2498, groups = var_2380, pad = v_9_pad_0, pad_type = v_9_pad_type_0, strides = var_2496, weight = down_blocks_2_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16_palettized, x = hidden_states_45_cast_fp16)[name = tensor("v_9_cast_fp16")]; tensor var_2502_begin_0 = const()[name = tensor("op_2502_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_2502_end_0 = const()[name = tensor("op_2502_end_0"), val = tensor([2, 160, 1, 256])]; tensor var_2502_end_mask_0 = const()[name = tensor("op_2502_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_2502_cast_fp16 = slice_by_index(begin = var_2502_begin_0, end = var_2502_end_0, end_mask = var_2502_end_mask_0, x = q_9_cast_fp16)[name = tensor("op_2502_cast_fp16")]; tensor var_2506_begin_0 = const()[name = tensor("op_2506_begin_0"), val = tensor([0, 160, 0, 0])]; tensor var_2506_end_0 = const()[name = tensor("op_2506_end_0"), val = tensor([2, 320, 1, 256])]; tensor var_2506_end_mask_0 = const()[name = tensor("op_2506_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_2506_cast_fp16 = slice_by_index(begin = var_2506_begin_0, end = var_2506_end_0, end_mask = var_2506_end_mask_0, x = q_9_cast_fp16)[name = tensor("op_2506_cast_fp16")]; tensor var_2510_begin_0 = const()[name = tensor("op_2510_begin_0"), val = tensor([0, 320, 0, 0])]; tensor var_2510_end_0 = const()[name = tensor("op_2510_end_0"), val = tensor([2, 480, 1, 256])]; tensor var_2510_end_mask_0 = const()[name = tensor("op_2510_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_2510_cast_fp16 = slice_by_index(begin = var_2510_begin_0, end = var_2510_end_0, end_mask = var_2510_end_mask_0, x = q_9_cast_fp16)[name = tensor("op_2510_cast_fp16")]; tensor var_2514_begin_0 = const()[name = tensor("op_2514_begin_0"), val = tensor([0, 480, 0, 0])]; tensor var_2514_end_0 = const()[name = tensor("op_2514_end_0"), val = tensor([2, 640, 1, 256])]; tensor var_2514_end_mask_0 = const()[name = tensor("op_2514_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_2514_cast_fp16 = slice_by_index(begin = var_2514_begin_0, end = var_2514_end_0, end_mask = var_2514_end_mask_0, x = q_9_cast_fp16)[name = tensor("op_2514_cast_fp16")]; tensor var_2518_begin_0 = const()[name = tensor("op_2518_begin_0"), val = tensor([0, 640, 0, 0])]; tensor var_2518_end_0 = const()[name = tensor("op_2518_end_0"), val = tensor([2, 800, 1, 256])]; tensor var_2518_end_mask_0 = const()[name = tensor("op_2518_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_2518_cast_fp16 = slice_by_index(begin = var_2518_begin_0, end = var_2518_end_0, end_mask = var_2518_end_mask_0, x = q_9_cast_fp16)[name = tensor("op_2518_cast_fp16")]; tensor var_2522_begin_0 = const()[name = tensor("op_2522_begin_0"), val = tensor([0, 800, 0, 0])]; tensor var_2522_end_0 = const()[name = tensor("op_2522_end_0"), val = tensor([2, 960, 1, 256])]; tensor var_2522_end_mask_0 = const()[name = tensor("op_2522_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_2522_cast_fp16 = slice_by_index(begin = var_2522_begin_0, end = var_2522_end_0, end_mask = var_2522_end_mask_0, x = q_9_cast_fp16)[name = tensor("op_2522_cast_fp16")]; tensor var_2526_begin_0 = const()[name = tensor("op_2526_begin_0"), val = tensor([0, 960, 0, 0])]; tensor var_2526_end_0 = const()[name = tensor("op_2526_end_0"), val = tensor([2, 1120, 1, 256])]; tensor var_2526_end_mask_0 = const()[name = tensor("op_2526_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_2526_cast_fp16 = slice_by_index(begin = var_2526_begin_0, end = var_2526_end_0, end_mask = var_2526_end_mask_0, x = q_9_cast_fp16)[name = tensor("op_2526_cast_fp16")]; tensor var_2530_begin_0 = const()[name = tensor("op_2530_begin_0"), val = tensor([0, 1120, 0, 0])]; tensor var_2530_end_0 = const()[name = tensor("op_2530_end_0"), val = tensor([2, 1280, 1, 256])]; tensor var_2530_end_mask_0 = const()[name = tensor("op_2530_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_2530_cast_fp16 = slice_by_index(begin = var_2530_begin_0, end = var_2530_end_0, end_mask = var_2530_end_mask_0, x = q_9_cast_fp16)[name = tensor("op_2530_cast_fp16")]; tensor k_19_perm_0 = const()[name = tensor("k_19_perm_0"), val = tensor([0, 3, 2, 1])]; tensor var_2537_begin_0 = const()[name = tensor("op_2537_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_2537_end_0 = const()[name = tensor("op_2537_end_0"), val = tensor([2, 256, 1, 160])]; tensor var_2537_end_mask_0 = const()[name = tensor("op_2537_end_mask_0"), val = tensor([true, true, true, false])]; tensor transpose_13 = transpose(perm = k_19_perm_0, x = k_17_cast_fp16)[name = tensor("transpose_13")]; tensor var_2537_cast_fp16 = slice_by_index(begin = var_2537_begin_0, end = var_2537_end_0, end_mask = var_2537_end_mask_0, x = transpose_13)[name = tensor("op_2537_cast_fp16")]; tensor var_2541_begin_0 = const()[name = tensor("op_2541_begin_0"), val = tensor([0, 0, 0, 160])]; tensor var_2541_end_0 = const()[name = tensor("op_2541_end_0"), val = tensor([2, 256, 1, 320])]; tensor var_2541_end_mask_0 = const()[name = tensor("op_2541_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2541_cast_fp16 = slice_by_index(begin = var_2541_begin_0, end = var_2541_end_0, end_mask = var_2541_end_mask_0, x = transpose_13)[name = tensor("op_2541_cast_fp16")]; tensor var_2545_begin_0 = const()[name = tensor("op_2545_begin_0"), val = tensor([0, 0, 0, 320])]; tensor var_2545_end_0 = const()[name = tensor("op_2545_end_0"), val = tensor([2, 256, 1, 480])]; tensor var_2545_end_mask_0 = const()[name = tensor("op_2545_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2545_cast_fp16 = slice_by_index(begin = var_2545_begin_0, end = var_2545_end_0, end_mask = var_2545_end_mask_0, x = transpose_13)[name = tensor("op_2545_cast_fp16")]; tensor var_2549_begin_0 = const()[name = tensor("op_2549_begin_0"), val = tensor([0, 0, 0, 480])]; tensor var_2549_end_0 = const()[name = tensor("op_2549_end_0"), val = tensor([2, 256, 1, 640])]; tensor var_2549_end_mask_0 = const()[name = tensor("op_2549_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2549_cast_fp16 = slice_by_index(begin = var_2549_begin_0, end = var_2549_end_0, end_mask = var_2549_end_mask_0, x = transpose_13)[name = tensor("op_2549_cast_fp16")]; tensor var_2553_begin_0 = const()[name = tensor("op_2553_begin_0"), val = tensor([0, 0, 0, 640])]; tensor var_2553_end_0 = const()[name = tensor("op_2553_end_0"), val = tensor([2, 256, 1, 800])]; tensor var_2553_end_mask_0 = const()[name = tensor("op_2553_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2553_cast_fp16 = slice_by_index(begin = var_2553_begin_0, end = var_2553_end_0, end_mask = var_2553_end_mask_0, x = transpose_13)[name = tensor("op_2553_cast_fp16")]; tensor var_2557_begin_0 = const()[name = tensor("op_2557_begin_0"), val = tensor([0, 0, 0, 800])]; tensor var_2557_end_0 = const()[name = tensor("op_2557_end_0"), val = tensor([2, 256, 1, 960])]; tensor var_2557_end_mask_0 = const()[name = tensor("op_2557_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2557_cast_fp16 = slice_by_index(begin = var_2557_begin_0, end = var_2557_end_0, end_mask = var_2557_end_mask_0, x = transpose_13)[name = tensor("op_2557_cast_fp16")]; tensor var_2561_begin_0 = const()[name = tensor("op_2561_begin_0"), val = tensor([0, 0, 0, 960])]; tensor var_2561_end_0 = const()[name = tensor("op_2561_end_0"), val = tensor([2, 256, 1, 1120])]; tensor var_2561_end_mask_0 = const()[name = tensor("op_2561_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2561_cast_fp16 = slice_by_index(begin = var_2561_begin_0, end = var_2561_end_0, end_mask = var_2561_end_mask_0, x = transpose_13)[name = tensor("op_2561_cast_fp16")]; tensor var_2565_begin_0 = const()[name = tensor("op_2565_begin_0"), val = tensor([0, 0, 0, 1120])]; tensor var_2565_end_0 = const()[name = tensor("op_2565_end_0"), val = tensor([2, 256, 1, 1280])]; tensor var_2565_end_mask_0 = const()[name = tensor("op_2565_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2565_cast_fp16 = slice_by_index(begin = var_2565_begin_0, end = var_2565_end_0, end_mask = var_2565_end_mask_0, x = transpose_13)[name = tensor("op_2565_cast_fp16")]; tensor var_2567_begin_0 = const()[name = tensor("op_2567_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_2567_end_0 = const()[name = tensor("op_2567_end_0"), val = tensor([2, 160, 1, 256])]; tensor var_2567_end_mask_0 = const()[name = tensor("op_2567_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_2567_cast_fp16 = slice_by_index(begin = var_2567_begin_0, end = var_2567_end_0, end_mask = var_2567_end_mask_0, x = v_9_cast_fp16)[name = tensor("op_2567_cast_fp16")]; tensor var_2571_begin_0 = const()[name = tensor("op_2571_begin_0"), val = tensor([0, 160, 0, 0])]; tensor var_2571_end_0 = const()[name = tensor("op_2571_end_0"), val = tensor([2, 320, 1, 256])]; tensor var_2571_end_mask_0 = const()[name = tensor("op_2571_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_2571_cast_fp16 = slice_by_index(begin = var_2571_begin_0, end = var_2571_end_0, end_mask = var_2571_end_mask_0, x = v_9_cast_fp16)[name = tensor("op_2571_cast_fp16")]; tensor var_2575_begin_0 = const()[name = tensor("op_2575_begin_0"), val = tensor([0, 320, 0, 0])]; tensor var_2575_end_0 = const()[name = tensor("op_2575_end_0"), val = tensor([2, 480, 1, 256])]; tensor var_2575_end_mask_0 = const()[name = tensor("op_2575_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_2575_cast_fp16 = slice_by_index(begin = var_2575_begin_0, end = var_2575_end_0, end_mask = var_2575_end_mask_0, x = v_9_cast_fp16)[name = tensor("op_2575_cast_fp16")]; tensor var_2579_begin_0 = const()[name = tensor("op_2579_begin_0"), val = tensor([0, 480, 0, 0])]; tensor var_2579_end_0 = const()[name = tensor("op_2579_end_0"), val = tensor([2, 640, 1, 256])]; tensor var_2579_end_mask_0 = const()[name = tensor("op_2579_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_2579_cast_fp16 = slice_by_index(begin = var_2579_begin_0, end = var_2579_end_0, end_mask = var_2579_end_mask_0, x = v_9_cast_fp16)[name = tensor("op_2579_cast_fp16")]; tensor var_2583_begin_0 = const()[name = tensor("op_2583_begin_0"), val = tensor([0, 640, 0, 0])]; tensor var_2583_end_0 = const()[name = tensor("op_2583_end_0"), val = tensor([2, 800, 1, 256])]; tensor var_2583_end_mask_0 = const()[name = tensor("op_2583_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_2583_cast_fp16 = slice_by_index(begin = var_2583_begin_0, end = var_2583_end_0, end_mask = var_2583_end_mask_0, x = v_9_cast_fp16)[name = tensor("op_2583_cast_fp16")]; tensor var_2587_begin_0 = const()[name = tensor("op_2587_begin_0"), val = tensor([0, 800, 0, 0])]; tensor var_2587_end_0 = const()[name = tensor("op_2587_end_0"), val = tensor([2, 960, 1, 256])]; tensor var_2587_end_mask_0 = const()[name = tensor("op_2587_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_2587_cast_fp16 = slice_by_index(begin = var_2587_begin_0, end = var_2587_end_0, end_mask = var_2587_end_mask_0, x = v_9_cast_fp16)[name = tensor("op_2587_cast_fp16")]; tensor var_2591_begin_0 = const()[name = tensor("op_2591_begin_0"), val = tensor([0, 960, 0, 0])]; tensor var_2591_end_0 = const()[name = tensor("op_2591_end_0"), val = tensor([2, 1120, 1, 256])]; tensor var_2591_end_mask_0 = const()[name = tensor("op_2591_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_2591_cast_fp16 = slice_by_index(begin = var_2591_begin_0, end = var_2591_end_0, end_mask = var_2591_end_mask_0, x = v_9_cast_fp16)[name = tensor("op_2591_cast_fp16")]; tensor var_2595_begin_0 = const()[name = tensor("op_2595_begin_0"), val = tensor([0, 1120, 0, 0])]; tensor var_2595_end_0 = const()[name = tensor("op_2595_end_0"), val = tensor([2, 1280, 1, 256])]; tensor var_2595_end_mask_0 = const()[name = tensor("op_2595_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_2595_cast_fp16 = slice_by_index(begin = var_2595_begin_0, end = var_2595_end_0, end_mask = var_2595_end_mask_0, x = v_9_cast_fp16)[name = tensor("op_2595_cast_fp16")]; tensor var_2599_equation_0 = const()[name = tensor("op_2599_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_2599_cast_fp16 = einsum(equation = var_2599_equation_0, values = (var_2537_cast_fp16, var_2502_cast_fp16))[name = tensor("op_2599_cast_fp16")]; tensor var_2600_to_fp16 = const()[name = tensor("op_2600_to_fp16"), val = tensor(0x1.43cp-4)]; tensor aw_1_cast_fp16 = mul(x = var_2599_cast_fp16, y = var_2600_to_fp16)[name = tensor("aw_1_cast_fp16")]; tensor var_2603_equation_0 = const()[name = tensor("op_2603_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_2603_cast_fp16 = einsum(equation = var_2603_equation_0, values = (var_2541_cast_fp16, var_2506_cast_fp16))[name = tensor("op_2603_cast_fp16")]; tensor var_2604_to_fp16 = const()[name = tensor("op_2604_to_fp16"), val = tensor(0x1.43cp-4)]; tensor aw_3_cast_fp16 = mul(x = var_2603_cast_fp16, y = var_2604_to_fp16)[name = tensor("aw_3_cast_fp16")]; tensor var_2607_equation_0 = const()[name = tensor("op_2607_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_2607_cast_fp16 = einsum(equation = var_2607_equation_0, values = (var_2545_cast_fp16, var_2510_cast_fp16))[name = tensor("op_2607_cast_fp16")]; tensor var_2608_to_fp16 = const()[name = tensor("op_2608_to_fp16"), val = tensor(0x1.43cp-4)]; tensor aw_5_cast_fp16 = mul(x = var_2607_cast_fp16, y = var_2608_to_fp16)[name = tensor("aw_5_cast_fp16")]; tensor var_2611_equation_0 = const()[name = tensor("op_2611_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_2611_cast_fp16 = einsum(equation = var_2611_equation_0, values = (var_2549_cast_fp16, var_2514_cast_fp16))[name = tensor("op_2611_cast_fp16")]; tensor var_2612_to_fp16 = const()[name = tensor("op_2612_to_fp16"), val = tensor(0x1.43cp-4)]; tensor aw_7_cast_fp16 = mul(x = var_2611_cast_fp16, y = var_2612_to_fp16)[name = tensor("aw_7_cast_fp16")]; tensor var_2615_equation_0 = const()[name = tensor("op_2615_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_2615_cast_fp16 = einsum(equation = var_2615_equation_0, values = (var_2553_cast_fp16, var_2518_cast_fp16))[name = tensor("op_2615_cast_fp16")]; tensor var_2616_to_fp16 = const()[name = tensor("op_2616_to_fp16"), val = tensor(0x1.43cp-4)]; tensor aw_9_cast_fp16 = mul(x = var_2615_cast_fp16, y = var_2616_to_fp16)[name = tensor("aw_9_cast_fp16")]; tensor var_2619_equation_0 = const()[name = tensor("op_2619_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_2619_cast_fp16 = einsum(equation = var_2619_equation_0, values = (var_2557_cast_fp16, var_2522_cast_fp16))[name = tensor("op_2619_cast_fp16")]; tensor var_2620_to_fp16 = const()[name = tensor("op_2620_to_fp16"), val = tensor(0x1.43cp-4)]; tensor aw_11_cast_fp16 = mul(x = var_2619_cast_fp16, y = var_2620_to_fp16)[name = tensor("aw_11_cast_fp16")]; tensor var_2623_equation_0 = const()[name = tensor("op_2623_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_2623_cast_fp16 = einsum(equation = var_2623_equation_0, values = (var_2561_cast_fp16, var_2526_cast_fp16))[name = tensor("op_2623_cast_fp16")]; tensor var_2624_to_fp16 = const()[name = tensor("op_2624_to_fp16"), val = tensor(0x1.43cp-4)]; tensor aw_13_cast_fp16 = mul(x = var_2623_cast_fp16, y = var_2624_to_fp16)[name = tensor("aw_13_cast_fp16")]; tensor var_2627_equation_0 = const()[name = tensor("op_2627_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_2627_cast_fp16 = einsum(equation = var_2627_equation_0, values = (var_2565_cast_fp16, var_2530_cast_fp16))[name = tensor("op_2627_cast_fp16")]; tensor var_2628_to_fp16 = const()[name = tensor("op_2628_to_fp16"), val = tensor(0x1.43cp-4)]; tensor aw_15_cast_fp16 = mul(x = var_2627_cast_fp16, y = var_2628_to_fp16)[name = tensor("aw_15_cast_fp16")]; tensor var_2630_cast_fp16 = softmax(axis = var_2380, x = aw_1_cast_fp16)[name = tensor("op_2630_cast_fp16")]; tensor var_2631_cast_fp16 = softmax(axis = var_2380, x = aw_3_cast_fp16)[name = tensor("op_2631_cast_fp16")]; tensor var_2632_cast_fp16 = softmax(axis = var_2380, x = aw_5_cast_fp16)[name = tensor("op_2632_cast_fp16")]; tensor var_2633_cast_fp16 = softmax(axis = var_2380, x = aw_7_cast_fp16)[name = tensor("op_2633_cast_fp16")]; tensor var_2634_cast_fp16 = softmax(axis = var_2380, x = aw_9_cast_fp16)[name = tensor("op_2634_cast_fp16")]; tensor var_2635_cast_fp16 = softmax(axis = var_2380, x = aw_11_cast_fp16)[name = tensor("op_2635_cast_fp16")]; tensor var_2636_cast_fp16 = softmax(axis = var_2380, x = aw_13_cast_fp16)[name = tensor("op_2636_cast_fp16")]; tensor var_2637_cast_fp16 = softmax(axis = var_2380, x = aw_15_cast_fp16)[name = tensor("op_2637_cast_fp16")]; tensor var_2639_equation_0 = const()[name = tensor("op_2639_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_2639_cast_fp16 = einsum(equation = var_2639_equation_0, values = (var_2567_cast_fp16, var_2630_cast_fp16))[name = tensor("op_2639_cast_fp16")]; tensor var_2641_equation_0 = const()[name = tensor("op_2641_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_2641_cast_fp16 = einsum(equation = var_2641_equation_0, values = (var_2571_cast_fp16, var_2631_cast_fp16))[name = tensor("op_2641_cast_fp16")]; tensor var_2643_equation_0 = const()[name = tensor("op_2643_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_2643_cast_fp16 = einsum(equation = var_2643_equation_0, values = (var_2575_cast_fp16, var_2632_cast_fp16))[name = tensor("op_2643_cast_fp16")]; tensor var_2645_equation_0 = const()[name = tensor("op_2645_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_2645_cast_fp16 = einsum(equation = var_2645_equation_0, values = (var_2579_cast_fp16, var_2633_cast_fp16))[name = tensor("op_2645_cast_fp16")]; tensor var_2647_equation_0 = const()[name = tensor("op_2647_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_2647_cast_fp16 = einsum(equation = var_2647_equation_0, values = (var_2583_cast_fp16, var_2634_cast_fp16))[name = tensor("op_2647_cast_fp16")]; tensor var_2649_equation_0 = const()[name = tensor("op_2649_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_2649_cast_fp16 = einsum(equation = var_2649_equation_0, values = (var_2587_cast_fp16, var_2635_cast_fp16))[name = tensor("op_2649_cast_fp16")]; tensor var_2651_equation_0 = const()[name = tensor("op_2651_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_2651_cast_fp16 = einsum(equation = var_2651_equation_0, values = (var_2591_cast_fp16, var_2636_cast_fp16))[name = tensor("op_2651_cast_fp16")]; tensor var_2653_equation_0 = const()[name = tensor("op_2653_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_2653_cast_fp16 = einsum(equation = var_2653_equation_0, values = (var_2595_cast_fp16, var_2637_cast_fp16))[name = tensor("op_2653_cast_fp16")]; tensor input_81_interleave_0 = const()[name = tensor("input_81_interleave_0"), val = tensor(false)]; tensor input_81_cast_fp16 = concat(axis = var_2380, interleave = input_81_interleave_0, values = (var_2639_cast_fp16, var_2641_cast_fp16, var_2643_cast_fp16, var_2645_cast_fp16, var_2647_cast_fp16, var_2649_cast_fp16, var_2651_cast_fp16, var_2653_cast_fp16))[name = tensor("input_81_cast_fp16")]; tensor var_2659 = const()[name = tensor("op_2659"), val = tensor([1, 1])]; tensor var_2661 = const()[name = tensor("op_2661"), val = tensor([1, 1])]; tensor var_2663_pad_type_0 = const()[name = tensor("op_2663_pad_type_0"), val = tensor("custom")]; tensor var_2663_pad_0 = const()[name = tensor("op_2663_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(43846464))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(45075328))), 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(45075520)))]; tensor var_2663_cast_fp16 = conv(bias = down_blocks_2_attentions_0_transformer_blocks_0_attn1_to_out_0_bias_to_fp16, dilations = var_2661, groups = var_2380, pad = var_2663_pad_0, pad_type = var_2663_pad_type_0, strides = var_2659, weight = down_blocks_2_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16_palettized, x = input_81_cast_fp16)[name = tensor("op_2663_cast_fp16")]; tensor inputs_15_cast_fp16 = add(x = var_2663_cast_fp16, y = inputs_13_cast_fp16)[name = tensor("inputs_15_cast_fp16")]; tensor var_2667 = const()[name = tensor("op_2667"), val = tensor([1])]; tensor channels_mean_15_cast_fp16 = reduce_mean(axes = var_2667, keep_dims = var_2375, x = inputs_15_cast_fp16)[name = tensor("channels_mean_15_cast_fp16")]; tensor zero_mean_15_cast_fp16 = sub(x = inputs_15_cast_fp16, y = channels_mean_15_cast_fp16)[name = tensor("zero_mean_15_cast_fp16")]; tensor zero_mean_sq_15_cast_fp16 = mul(x = zero_mean_15_cast_fp16, y = zero_mean_15_cast_fp16)[name = tensor("zero_mean_sq_15_cast_fp16")]; tensor var_2671 = const()[name = tensor("op_2671"), val = tensor([1])]; tensor var_2672_cast_fp16 = reduce_mean(axes = var_2671, keep_dims = var_2375, x = zero_mean_sq_15_cast_fp16)[name = tensor("op_2672_cast_fp16")]; tensor var_2673_to_fp16 = const()[name = tensor("op_2673_to_fp16"), val = tensor(0x1.5p-17)]; tensor var_2674_cast_fp16 = add(x = var_2672_cast_fp16, y = var_2673_to_fp16)[name = tensor("op_2674_cast_fp16")]; tensor denom_15_epsilon_0_to_fp16 = const()[name = tensor("denom_15_epsilon_0_to_fp16"), val = tensor(0x1p-24)]; tensor denom_15_cast_fp16 = rsqrt(epsilon = denom_15_epsilon_0_to_fp16, x = var_2674_cast_fp16)[name = tensor("denom_15_cast_fp16")]; tensor out_15_cast_fp16 = mul(x = zero_mean_15_cast_fp16, y = denom_15_cast_fp16)[name = tensor("out_15_cast_fp16")]; tensor var_2678_to_fp16 = const()[name = tensor("op_2678_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(45078144)))]; tensor var_2679_cast_fp16 = add(x = out_15_cast_fp16, y = var_2678_to_fp16)[name = tensor("op_2679_cast_fp16")]; tensor var_2681_to_fp16 = const()[name = tensor("op_2681_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(45080768)))]; tensor hidden_states_47_cast_fp16 = mul(x = var_2679_cast_fp16, y = var_2681_to_fp16)[name = tensor("hidden_states_47_cast_fp16")]; tensor var_2688 = const()[name = tensor("op_2688"), val = tensor([1, 1])]; tensor var_2690 = const()[name = tensor("op_2690"), 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_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(45083392))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(46312256))), 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_11_cast_fp16 = conv(dilations = var_2690, groups = var_2380, pad = q_11_pad_0, pad_type = q_11_pad_type_0, strides = var_2688, weight = down_blocks_2_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16_palettized, x = hidden_states_47_cast_fp16)[name = tensor("q_11_cast_fp16")]; tensor var_2694 = const()[name = tensor("op_2694"), val = tensor([1, 1])]; tensor var_2696 = const()[name = tensor("op_2696"), 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_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(46312448))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(47049792))), name = tensor("down_blocks_2_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized"), shape = tensor([1280, 768, 1, 1])]; tensor k_21_cast_fp16 = conv(dilations = var_2696, groups = var_2380, pad = k_21_pad_0, pad_type = k_21_pad_type_0, strides = var_2694, weight = down_blocks_2_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor("k_21_cast_fp16")]; tensor var_2700 = const()[name = tensor("op_2700"), val = tensor([1, 1])]; tensor var_2702 = const()[name = tensor("op_2702"), 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_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(47049984))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(47787328))), name = tensor("down_blocks_2_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized"), shape = tensor([1280, 768, 1, 1])]; tensor v_11_cast_fp16 = conv(dilations = var_2702, groups = var_2380, pad = v_11_pad_0, pad_type = v_11_pad_type_0, strides = var_2700, weight = down_blocks_2_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor("v_11_cast_fp16")]; 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, 160, 1, 256])]; tensor var_2706_end_mask_0 = const()[name = tensor("op_2706_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_2706_cast_fp16 = slice_by_index(begin = var_2706_begin_0, end = var_2706_end_0, end_mask = var_2706_end_mask_0, x = q_11_cast_fp16)[name = tensor("op_2706_cast_fp16")]; tensor var_2710_begin_0 = const()[name = tensor("op_2710_begin_0"), val = tensor([0, 160, 0, 0])]; tensor var_2710_end_0 = const()[name = tensor("op_2710_end_0"), val = tensor([2, 320, 1, 256])]; tensor var_2710_end_mask_0 = const()[name = tensor("op_2710_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_2710_cast_fp16 = slice_by_index(begin = var_2710_begin_0, end = var_2710_end_0, end_mask = var_2710_end_mask_0, x = q_11_cast_fp16)[name = tensor("op_2710_cast_fp16")]; tensor var_2714_begin_0 = const()[name = tensor("op_2714_begin_0"), val = tensor([0, 320, 0, 0])]; tensor var_2714_end_0 = const()[name = tensor("op_2714_end_0"), val = tensor([2, 480, 1, 256])]; tensor var_2714_end_mask_0 = const()[name = tensor("op_2714_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_2714_cast_fp16 = slice_by_index(begin = var_2714_begin_0, end = var_2714_end_0, end_mask = var_2714_end_mask_0, x = q_11_cast_fp16)[name = tensor("op_2714_cast_fp16")]; tensor var_2718_begin_0 = const()[name = tensor("op_2718_begin_0"), val = tensor([0, 480, 0, 0])]; tensor var_2718_end_0 = const()[name = tensor("op_2718_end_0"), val = tensor([2, 640, 1, 256])]; tensor var_2718_end_mask_0 = const()[name = tensor("op_2718_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_2718_cast_fp16 = slice_by_index(begin = var_2718_begin_0, end = var_2718_end_0, end_mask = var_2718_end_mask_0, x = q_11_cast_fp16)[name = tensor("op_2718_cast_fp16")]; tensor var_2722_begin_0 = const()[name = tensor("op_2722_begin_0"), val = tensor([0, 640, 0, 0])]; tensor var_2722_end_0 = const()[name = tensor("op_2722_end_0"), val = tensor([2, 800, 1, 256])]; tensor var_2722_end_mask_0 = const()[name = tensor("op_2722_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_2722_cast_fp16 = slice_by_index(begin = var_2722_begin_0, end = var_2722_end_0, end_mask = var_2722_end_mask_0, x = q_11_cast_fp16)[name = tensor("op_2722_cast_fp16")]; tensor var_2726_begin_0 = const()[name = tensor("op_2726_begin_0"), val = tensor([0, 800, 0, 0])]; tensor var_2726_end_0 = const()[name = tensor("op_2726_end_0"), val = tensor([2, 960, 1, 256])]; tensor var_2726_end_mask_0 = const()[name = tensor("op_2726_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_2726_cast_fp16 = slice_by_index(begin = var_2726_begin_0, end = var_2726_end_0, end_mask = var_2726_end_mask_0, x = q_11_cast_fp16)[name = tensor("op_2726_cast_fp16")]; tensor var_2730_begin_0 = const()[name = tensor("op_2730_begin_0"), val = tensor([0, 960, 0, 0])]; tensor var_2730_end_0 = const()[name = tensor("op_2730_end_0"), val = tensor([2, 1120, 1, 256])]; tensor var_2730_end_mask_0 = const()[name = tensor("op_2730_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_2730_cast_fp16 = slice_by_index(begin = var_2730_begin_0, end = var_2730_end_0, end_mask = var_2730_end_mask_0, x = q_11_cast_fp16)[name = tensor("op_2730_cast_fp16")]; tensor var_2734_begin_0 = const()[name = tensor("op_2734_begin_0"), val = tensor([0, 1120, 0, 0])]; tensor var_2734_end_0 = const()[name = tensor("op_2734_end_0"), val = tensor([2, 1280, 1, 256])]; tensor var_2734_end_mask_0 = const()[name = tensor("op_2734_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_2734_cast_fp16 = slice_by_index(begin = var_2734_begin_0, end = var_2734_end_0, end_mask = var_2734_end_mask_0, x = q_11_cast_fp16)[name = tensor("op_2734_cast_fp16")]; tensor k_23_perm_0 = const()[name = tensor("k_23_perm_0"), val = tensor([0, 3, 2, 1])]; tensor var_2741_begin_0 = const()[name = tensor("op_2741_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_2741_end_0 = const()[name = tensor("op_2741_end_0"), val = tensor([2, 77, 1, 160])]; tensor var_2741_end_mask_0 = const()[name = tensor("op_2741_end_mask_0"), val = tensor([true, true, true, false])]; tensor transpose_12 = transpose(perm = k_23_perm_0, x = k_21_cast_fp16)[name = tensor("transpose_12")]; tensor var_2741_cast_fp16 = slice_by_index(begin = var_2741_begin_0, end = var_2741_end_0, end_mask = var_2741_end_mask_0, x = transpose_12)[name = tensor("op_2741_cast_fp16")]; tensor var_2745_begin_0 = const()[name = tensor("op_2745_begin_0"), val = tensor([0, 0, 0, 160])]; tensor var_2745_end_0 = const()[name = tensor("op_2745_end_0"), val = tensor([2, 77, 1, 320])]; tensor var_2745_end_mask_0 = const()[name = tensor("op_2745_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2745_cast_fp16 = slice_by_index(begin = var_2745_begin_0, end = var_2745_end_0, end_mask = var_2745_end_mask_0, x = transpose_12)[name = tensor("op_2745_cast_fp16")]; tensor var_2749_begin_0 = const()[name = tensor("op_2749_begin_0"), val = tensor([0, 0, 0, 320])]; tensor var_2749_end_0 = const()[name = tensor("op_2749_end_0"), val = tensor([2, 77, 1, 480])]; tensor var_2749_end_mask_0 = const()[name = tensor("op_2749_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2749_cast_fp16 = slice_by_index(begin = var_2749_begin_0, end = var_2749_end_0, end_mask = var_2749_end_mask_0, x = transpose_12)[name = tensor("op_2749_cast_fp16")]; tensor var_2753_begin_0 = const()[name = tensor("op_2753_begin_0"), val = tensor([0, 0, 0, 480])]; tensor var_2753_end_0 = const()[name = tensor("op_2753_end_0"), val = tensor([2, 77, 1, 640])]; tensor var_2753_end_mask_0 = const()[name = tensor("op_2753_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2753_cast_fp16 = slice_by_index(begin = var_2753_begin_0, end = var_2753_end_0, end_mask = var_2753_end_mask_0, x = transpose_12)[name = tensor("op_2753_cast_fp16")]; tensor var_2757_begin_0 = const()[name = tensor("op_2757_begin_0"), val = tensor([0, 0, 0, 640])]; tensor var_2757_end_0 = const()[name = tensor("op_2757_end_0"), val = tensor([2, 77, 1, 800])]; tensor var_2757_end_mask_0 = const()[name = tensor("op_2757_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2757_cast_fp16 = slice_by_index(begin = var_2757_begin_0, end = var_2757_end_0, end_mask = var_2757_end_mask_0, x = transpose_12)[name = tensor("op_2757_cast_fp16")]; tensor var_2761_begin_0 = const()[name = tensor("op_2761_begin_0"), val = tensor([0, 0, 0, 800])]; tensor var_2761_end_0 = const()[name = tensor("op_2761_end_0"), val = tensor([2, 77, 1, 960])]; tensor var_2761_end_mask_0 = const()[name = tensor("op_2761_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2761_cast_fp16 = slice_by_index(begin = var_2761_begin_0, end = var_2761_end_0, end_mask = var_2761_end_mask_0, x = transpose_12)[name = tensor("op_2761_cast_fp16")]; tensor var_2765_begin_0 = const()[name = tensor("op_2765_begin_0"), val = tensor([0, 0, 0, 960])]; tensor var_2765_end_0 = const()[name = tensor("op_2765_end_0"), val = tensor([2, 77, 1, 1120])]; tensor var_2765_end_mask_0 = const()[name = tensor("op_2765_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2765_cast_fp16 = slice_by_index(begin = var_2765_begin_0, end = var_2765_end_0, end_mask = var_2765_end_mask_0, x = transpose_12)[name = tensor("op_2765_cast_fp16")]; tensor var_2769_begin_0 = const()[name = tensor("op_2769_begin_0"), val = tensor([0, 0, 0, 1120])]; tensor var_2769_end_0 = const()[name = tensor("op_2769_end_0"), val = tensor([2, 77, 1, 1280])]; tensor var_2769_end_mask_0 = const()[name = tensor("op_2769_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2769_cast_fp16 = slice_by_index(begin = var_2769_begin_0, end = var_2769_end_0, end_mask = var_2769_end_mask_0, x = transpose_12)[name = tensor("op_2769_cast_fp16")]; tensor var_2771_begin_0 = const()[name = tensor("op_2771_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_2771_end_0 = const()[name = tensor("op_2771_end_0"), val = tensor([2, 160, 1, 77])]; tensor var_2771_end_mask_0 = const()[name = tensor("op_2771_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_2771_cast_fp16 = slice_by_index(begin = var_2771_begin_0, end = var_2771_end_0, end_mask = var_2771_end_mask_0, x = v_11_cast_fp16)[name = tensor("op_2771_cast_fp16")]; tensor var_2775_begin_0 = const()[name = tensor("op_2775_begin_0"), val = tensor([0, 160, 0, 0])]; tensor var_2775_end_0 = const()[name = tensor("op_2775_end_0"), val = tensor([2, 320, 1, 77])]; tensor var_2775_end_mask_0 = const()[name = tensor("op_2775_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_2775_cast_fp16 = slice_by_index(begin = var_2775_begin_0, end = var_2775_end_0, end_mask = var_2775_end_mask_0, x = v_11_cast_fp16)[name = tensor("op_2775_cast_fp16")]; tensor var_2779_begin_0 = const()[name = tensor("op_2779_begin_0"), val = tensor([0, 320, 0, 0])]; tensor var_2779_end_0 = const()[name = tensor("op_2779_end_0"), val = tensor([2, 480, 1, 77])]; tensor var_2779_end_mask_0 = const()[name = tensor("op_2779_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_2779_cast_fp16 = slice_by_index(begin = var_2779_begin_0, end = var_2779_end_0, end_mask = var_2779_end_mask_0, x = v_11_cast_fp16)[name = tensor("op_2779_cast_fp16")]; tensor var_2783_begin_0 = const()[name = tensor("op_2783_begin_0"), val = tensor([0, 480, 0, 0])]; tensor var_2783_end_0 = const()[name = tensor("op_2783_end_0"), val = tensor([2, 640, 1, 77])]; tensor var_2783_end_mask_0 = const()[name = tensor("op_2783_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_2783_cast_fp16 = slice_by_index(begin = var_2783_begin_0, end = var_2783_end_0, end_mask = var_2783_end_mask_0, x = v_11_cast_fp16)[name = tensor("op_2783_cast_fp16")]; tensor var_2787_begin_0 = const()[name = tensor("op_2787_begin_0"), val = tensor([0, 640, 0, 0])]; tensor var_2787_end_0 = const()[name = tensor("op_2787_end_0"), val = tensor([2, 800, 1, 77])]; tensor var_2787_end_mask_0 = const()[name = tensor("op_2787_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_2787_cast_fp16 = slice_by_index(begin = var_2787_begin_0, end = var_2787_end_0, end_mask = var_2787_end_mask_0, x = v_11_cast_fp16)[name = tensor("op_2787_cast_fp16")]; tensor var_2791_begin_0 = const()[name = tensor("op_2791_begin_0"), val = tensor([0, 800, 0, 0])]; tensor var_2791_end_0 = const()[name = tensor("op_2791_end_0"), val = tensor([2, 960, 1, 77])]; tensor var_2791_end_mask_0 = const()[name = tensor("op_2791_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_2791_cast_fp16 = slice_by_index(begin = var_2791_begin_0, end = var_2791_end_0, end_mask = var_2791_end_mask_0, x = v_11_cast_fp16)[name = tensor("op_2791_cast_fp16")]; tensor var_2795_begin_0 = const()[name = tensor("op_2795_begin_0"), val = tensor([0, 960, 0, 0])]; tensor var_2795_end_0 = const()[name = tensor("op_2795_end_0"), val = tensor([2, 1120, 1, 77])]; tensor var_2795_end_mask_0 = const()[name = tensor("op_2795_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_2795_cast_fp16 = slice_by_index(begin = var_2795_begin_0, end = var_2795_end_0, end_mask = var_2795_end_mask_0, x = v_11_cast_fp16)[name = tensor("op_2795_cast_fp16")]; tensor var_2799_begin_0 = const()[name = tensor("op_2799_begin_0"), val = tensor([0, 1120, 0, 0])]; tensor var_2799_end_0 = const()[name = tensor("op_2799_end_0"), val = tensor([2, 1280, 1, 77])]; tensor var_2799_end_mask_0 = const()[name = tensor("op_2799_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_2799_cast_fp16 = slice_by_index(begin = var_2799_begin_0, end = var_2799_end_0, end_mask = var_2799_end_mask_0, x = v_11_cast_fp16)[name = tensor("op_2799_cast_fp16")]; tensor var_2803_equation_0 = const()[name = tensor("op_2803_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_2803_cast_fp16 = einsum(equation = var_2803_equation_0, values = (var_2741_cast_fp16, var_2706_cast_fp16))[name = tensor("op_2803_cast_fp16")]; tensor var_2804_to_fp16 = const()[name = tensor("op_2804_to_fp16"), val = tensor(0x1.43cp-4)]; tensor aw_17_cast_fp16 = mul(x = var_2803_cast_fp16, y = var_2804_to_fp16)[name = tensor("aw_17_cast_fp16")]; tensor var_2807_equation_0 = const()[name = tensor("op_2807_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_2807_cast_fp16 = einsum(equation = var_2807_equation_0, values = (var_2745_cast_fp16, var_2710_cast_fp16))[name = tensor("op_2807_cast_fp16")]; tensor var_2808_to_fp16 = const()[name = tensor("op_2808_to_fp16"), val = tensor(0x1.43cp-4)]; tensor aw_19_cast_fp16 = mul(x = var_2807_cast_fp16, y = var_2808_to_fp16)[name = tensor("aw_19_cast_fp16")]; tensor var_2811_equation_0 = const()[name = tensor("op_2811_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_2811_cast_fp16 = einsum(equation = var_2811_equation_0, values = (var_2749_cast_fp16, var_2714_cast_fp16))[name = tensor("op_2811_cast_fp16")]; tensor var_2812_to_fp16 = const()[name = tensor("op_2812_to_fp16"), val = tensor(0x1.43cp-4)]; tensor aw_21_cast_fp16 = mul(x = var_2811_cast_fp16, y = var_2812_to_fp16)[name = tensor("aw_21_cast_fp16")]; tensor var_2815_equation_0 = const()[name = tensor("op_2815_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_2815_cast_fp16 = einsum(equation = var_2815_equation_0, values = (var_2753_cast_fp16, var_2718_cast_fp16))[name = tensor("op_2815_cast_fp16")]; tensor var_2816_to_fp16 = const()[name = tensor("op_2816_to_fp16"), val = tensor(0x1.43cp-4)]; tensor aw_23_cast_fp16 = mul(x = var_2815_cast_fp16, y = var_2816_to_fp16)[name = tensor("aw_23_cast_fp16")]; tensor var_2819_equation_0 = const()[name = tensor("op_2819_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_2819_cast_fp16 = einsum(equation = var_2819_equation_0, values = (var_2757_cast_fp16, var_2722_cast_fp16))[name = tensor("op_2819_cast_fp16")]; tensor var_2820_to_fp16 = const()[name = tensor("op_2820_to_fp16"), val = tensor(0x1.43cp-4)]; tensor aw_25_cast_fp16 = mul(x = var_2819_cast_fp16, y = var_2820_to_fp16)[name = tensor("aw_25_cast_fp16")]; tensor var_2823_equation_0 = const()[name = tensor("op_2823_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_2823_cast_fp16 = einsum(equation = var_2823_equation_0, values = (var_2761_cast_fp16, var_2726_cast_fp16))[name = tensor("op_2823_cast_fp16")]; tensor var_2824_to_fp16 = const()[name = tensor("op_2824_to_fp16"), val = tensor(0x1.43cp-4)]; tensor aw_27_cast_fp16 = mul(x = var_2823_cast_fp16, y = var_2824_to_fp16)[name = tensor("aw_27_cast_fp16")]; tensor var_2827_equation_0 = const()[name = tensor("op_2827_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_2827_cast_fp16 = einsum(equation = var_2827_equation_0, values = (var_2765_cast_fp16, var_2730_cast_fp16))[name = tensor("op_2827_cast_fp16")]; tensor var_2828_to_fp16 = const()[name = tensor("op_2828_to_fp16"), val = tensor(0x1.43cp-4)]; tensor aw_29_cast_fp16 = mul(x = var_2827_cast_fp16, y = var_2828_to_fp16)[name = tensor("aw_29_cast_fp16")]; tensor var_2831_equation_0 = const()[name = tensor("op_2831_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_2831_cast_fp16 = einsum(equation = var_2831_equation_0, values = (var_2769_cast_fp16, var_2734_cast_fp16))[name = tensor("op_2831_cast_fp16")]; tensor var_2832_to_fp16 = const()[name = tensor("op_2832_to_fp16"), val = tensor(0x1.43cp-4)]; tensor aw_31_cast_fp16 = mul(x = var_2831_cast_fp16, y = var_2832_to_fp16)[name = tensor("aw_31_cast_fp16")]; tensor var_2834_cast_fp16 = softmax(axis = var_2380, x = aw_17_cast_fp16)[name = tensor("op_2834_cast_fp16")]; tensor var_2835_cast_fp16 = softmax(axis = var_2380, x = aw_19_cast_fp16)[name = tensor("op_2835_cast_fp16")]; tensor var_2836_cast_fp16 = softmax(axis = var_2380, x = aw_21_cast_fp16)[name = tensor("op_2836_cast_fp16")]; tensor var_2837_cast_fp16 = softmax(axis = var_2380, x = aw_23_cast_fp16)[name = tensor("op_2837_cast_fp16")]; tensor var_2838_cast_fp16 = softmax(axis = var_2380, x = aw_25_cast_fp16)[name = tensor("op_2838_cast_fp16")]; tensor var_2839_cast_fp16 = softmax(axis = var_2380, x = aw_27_cast_fp16)[name = tensor("op_2839_cast_fp16")]; tensor var_2840_cast_fp16 = softmax(axis = var_2380, x = aw_29_cast_fp16)[name = tensor("op_2840_cast_fp16")]; tensor var_2841_cast_fp16 = softmax(axis = var_2380, x = aw_31_cast_fp16)[name = tensor("op_2841_cast_fp16")]; tensor var_2843_equation_0 = const()[name = tensor("op_2843_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_2843_cast_fp16 = einsum(equation = var_2843_equation_0, values = (var_2771_cast_fp16, var_2834_cast_fp16))[name = tensor("op_2843_cast_fp16")]; tensor var_2845_equation_0 = const()[name = tensor("op_2845_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_2845_cast_fp16 = einsum(equation = var_2845_equation_0, values = (var_2775_cast_fp16, var_2835_cast_fp16))[name = tensor("op_2845_cast_fp16")]; tensor var_2847_equation_0 = const()[name = tensor("op_2847_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_2847_cast_fp16 = einsum(equation = var_2847_equation_0, values = (var_2779_cast_fp16, var_2836_cast_fp16))[name = tensor("op_2847_cast_fp16")]; tensor var_2849_equation_0 = const()[name = tensor("op_2849_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_2849_cast_fp16 = einsum(equation = var_2849_equation_0, values = (var_2783_cast_fp16, var_2837_cast_fp16))[name = tensor("op_2849_cast_fp16")]; tensor var_2851_equation_0 = const()[name = tensor("op_2851_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_2851_cast_fp16 = einsum(equation = var_2851_equation_0, values = (var_2787_cast_fp16, var_2838_cast_fp16))[name = tensor("op_2851_cast_fp16")]; tensor var_2853_equation_0 = const()[name = tensor("op_2853_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_2853_cast_fp16 = einsum(equation = var_2853_equation_0, values = (var_2791_cast_fp16, var_2839_cast_fp16))[name = tensor("op_2853_cast_fp16")]; tensor var_2855_equation_0 = const()[name = tensor("op_2855_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_2855_cast_fp16 = einsum(equation = var_2855_equation_0, values = (var_2795_cast_fp16, var_2840_cast_fp16))[name = tensor("op_2855_cast_fp16")]; tensor var_2857_equation_0 = const()[name = tensor("op_2857_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_2857_cast_fp16 = einsum(equation = var_2857_equation_0, values = (var_2799_cast_fp16, var_2841_cast_fp16))[name = tensor("op_2857_cast_fp16")]; tensor input_83_interleave_0 = const()[name = tensor("input_83_interleave_0"), val = tensor(false)]; tensor input_83_cast_fp16 = concat(axis = var_2380, interleave = input_83_interleave_0, values = (var_2843_cast_fp16, var_2845_cast_fp16, var_2847_cast_fp16, var_2849_cast_fp16, var_2851_cast_fp16, var_2853_cast_fp16, var_2855_cast_fp16, var_2857_cast_fp16))[name = tensor("input_83_cast_fp16")]; tensor var_2863 = const()[name = tensor("op_2863"), val = tensor([1, 1])]; tensor var_2865 = const()[name = tensor("op_2865"), val = tensor([1, 1])]; tensor var_2867_pad_type_0 = const()[name = tensor("op_2867_pad_type_0"), val = tensor("custom")]; tensor var_2867_pad_0 = const()[name = tensor("op_2867_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(47787520))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(49016384))), 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(49016576)))]; tensor var_2867_cast_fp16 = conv(bias = down_blocks_2_attentions_0_transformer_blocks_0_attn2_to_out_0_bias_to_fp16, dilations = var_2865, groups = var_2380, pad = var_2867_pad_0, pad_type = var_2867_pad_type_0, strides = var_2863, weight = down_blocks_2_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16_palettized, x = input_83_cast_fp16)[name = tensor("op_2867_cast_fp16")]; tensor inputs_17_cast_fp16 = add(x = var_2867_cast_fp16, y = inputs_15_cast_fp16)[name = tensor("inputs_17_cast_fp16")]; tensor var_2871 = const()[name = tensor("op_2871"), val = tensor([1])]; tensor channels_mean_17_cast_fp16 = reduce_mean(axes = var_2871, keep_dims = var_2375, x = inputs_17_cast_fp16)[name = tensor("channels_mean_17_cast_fp16")]; tensor zero_mean_17_cast_fp16 = sub(x = inputs_17_cast_fp16, y = channels_mean_17_cast_fp16)[name = tensor("zero_mean_17_cast_fp16")]; tensor zero_mean_sq_17_cast_fp16 = mul(x = zero_mean_17_cast_fp16, y = zero_mean_17_cast_fp16)[name = tensor("zero_mean_sq_17_cast_fp16")]; tensor var_2875 = const()[name = tensor("op_2875"), val = tensor([1])]; tensor var_2876_cast_fp16 = reduce_mean(axes = var_2875, keep_dims = var_2375, x = zero_mean_sq_17_cast_fp16)[name = tensor("op_2876_cast_fp16")]; tensor var_2877_to_fp16 = const()[name = tensor("op_2877_to_fp16"), val = tensor(0x1.5p-17)]; tensor var_2878_cast_fp16 = add(x = var_2876_cast_fp16, y = var_2877_to_fp16)[name = tensor("op_2878_cast_fp16")]; tensor denom_17_epsilon_0_to_fp16 = const()[name = tensor("denom_17_epsilon_0_to_fp16"), val = tensor(0x1p-24)]; tensor denom_17_cast_fp16 = rsqrt(epsilon = denom_17_epsilon_0_to_fp16, x = var_2878_cast_fp16)[name = tensor("denom_17_cast_fp16")]; tensor out_17_cast_fp16 = mul(x = zero_mean_17_cast_fp16, y = denom_17_cast_fp16)[name = tensor("out_17_cast_fp16")]; tensor var_2882_to_fp16 = const()[name = tensor("op_2882_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(49019200)))]; tensor var_2883_cast_fp16 = add(x = out_17_cast_fp16, y = var_2882_to_fp16)[name = tensor("op_2883_cast_fp16")]; tensor var_2885_to_fp16 = const()[name = tensor("op_2885_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(49021824)))]; tensor input_85_cast_fp16 = mul(x = var_2883_cast_fp16, y = var_2885_to_fp16)[name = tensor("input_85_cast_fp16")]; tensor var_2893 = const()[name = tensor("op_2893"), val = tensor([1, 1])]; tensor var_2895 = const()[name = tensor("op_2895"), val = tensor([1, 1])]; tensor var_2897_pad_type_0 = const()[name = tensor("op_2897_pad_type_0"), val = tensor("custom")]; tensor var_2897_pad_0 = const()[name = tensor("op_2897_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(49024448))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(58854912))), 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(58855104))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(58862848))), name = tensor("down_blocks_2_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16_palettized"), shape = tensor([10240])]; tensor var_2897_cast_fp16 = conv(bias = down_blocks_2_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16_palettized, dilations = var_2895, groups = var_2380, pad = var_2897_pad_0, pad_type = var_2897_pad_type_0, strides = var_2893, weight = down_blocks_2_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16_palettized, x = input_85_cast_fp16)[name = tensor("op_2897_cast_fp16")]; tensor var_2898_split_sizes_0 = const()[name = tensor("op_2898_split_sizes_0"), val = tensor([5120, 5120])]; tensor var_2898_axis_0 = const()[name = tensor("op_2898_axis_0"), val = tensor(1)]; tensor var_2898_cast_fp16_0, tensor var_2898_cast_fp16_1 = split(axis = var_2898_axis_0, split_sizes = var_2898_split_sizes_0, x = var_2897_cast_fp16)[name = tensor("op_2898_cast_fp16")]; tensor var_2900_mode_0 = const()[name = tensor("op_2900_mode_0"), val = tensor("EXACT")]; tensor var_2900_cast_fp16 = gelu(mode = var_2900_mode_0, x = var_2898_cast_fp16_1)[name = tensor("op_2900_cast_fp16")]; tensor input_87_cast_fp16 = mul(x = var_2898_cast_fp16_0, y = var_2900_cast_fp16)[name = tensor("input_87_cast_fp16")]; tensor var_2904 = const()[name = tensor("op_2904"), val = tensor([1, 1])]; tensor var_2906 = const()[name = tensor("op_2906"), val = tensor([1, 1])]; tensor var_2908_pad_type_0 = const()[name = tensor("op_2908_pad_type_0"), val = tensor("custom")]; tensor var_2908_pad_0 = const()[name = tensor("op_2908_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(58863040))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(63778304))), 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(63778496)))]; tensor var_2908_cast_fp16 = conv(bias = down_blocks_2_attentions_0_transformer_blocks_0_ff_net_2_bias_to_fp16, dilations = var_2906, groups = var_2380, pad = var_2908_pad_0, pad_type = var_2908_pad_type_0, strides = var_2904, weight = down_blocks_2_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16_palettized, x = input_87_cast_fp16)[name = tensor("op_2908_cast_fp16")]; tensor hidden_states_51_cast_fp16 = add(x = var_2908_cast_fp16, y = inputs_17_cast_fp16)[name = tensor("hidden_states_51_cast_fp16")]; tensor var_2910 = const()[name = tensor("op_2910"), val = tensor([2, 1280, 16, 16])]; tensor input_89_cast_fp16 = reshape(shape = var_2910, x = hidden_states_51_cast_fp16)[name = tensor("input_89_cast_fp16")]; tensor var_2914 = const()[name = tensor("op_2914"), val = tensor([1, 1])]; tensor var_2916 = const()[name = tensor("op_2916"), 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_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(63781120))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(65009984))), 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(65010176)))]; tensor hidden_states_53_cast_fp16 = conv(bias = down_blocks_2_attentions_0_proj_out_bias_to_fp16, dilations = var_2916, groups = var_2380, pad = hidden_states_53_pad_0, pad_type = hidden_states_53_pad_type_0, strides = var_2914, weight = down_blocks_2_attentions_0_proj_out_weight_to_fp16_palettized, x = input_89_cast_fp16)[name = tensor("hidden_states_53_cast_fp16")]; tensor hidden_states_55_cast_fp16 = add(x = hidden_states_53_cast_fp16, y = hidden_states_41_cast_fp16)[name = tensor("hidden_states_55_cast_fp16")]; tensor var_2937 = const()[name = tensor("op_2937"), val = tensor(true)]; tensor var_2942 = const()[name = tensor("op_2942"), val = tensor(1)]; tensor input_91_interleave_0 = const()[name = tensor("input_91_interleave_0"), val = tensor(false)]; tensor input_91_cast_fp16 = concat(axis = var_2942, interleave = input_91_interleave_0, values = (hidden_states_55_cast_fp16, hidden_states_55_cast_fp16))[name = tensor("input_91_cast_fp16")]; tensor reshape_36_shape_0 = const()[name = tensor("reshape_36_shape_0"), val = tensor([2, 32, 80, 16, 16])]; tensor reshape_36_cast_fp16 = reshape(shape = reshape_36_shape_0, x = input_91_cast_fp16)[name = tensor("reshape_36_cast_fp16")]; 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_fp16 = reduce_mean(axes = reduce_mean_27_axes_0, keep_dims = reduce_mean_27_keep_dims_0, x = reshape_36_cast_fp16)[name = tensor("reduce_mean_27_cast_fp16")]; tensor sub_18_cast_fp16 = sub(x = reshape_36_cast_fp16, y = reduce_mean_27_cast_fp16)[name = tensor("sub_18_cast_fp16")]; tensor square_9_cast_fp16 = square(x = sub_18_cast_fp16)[name = tensor("square_9_cast_fp16")]; 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_fp16 = reduce_mean(axes = reduce_mean_29_axes_0, keep_dims = reduce_mean_29_keep_dims_0, x = square_9_cast_fp16)[name = tensor("reduce_mean_29_cast_fp16")]; 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_fp16 = add(x = reduce_mean_29_cast_fp16, y = add_18_y_0_to_fp16)[name = tensor("add_18_cast_fp16")]; tensor sqrt_9_cast_fp16 = sqrt(x = add_18_cast_fp16)[name = tensor("sqrt_9_cast_fp16")]; tensor real_div_9_cast_fp16 = real_div(x = sub_18_cast_fp16, y = sqrt_9_cast_fp16)[name = tensor("real_div_9_cast_fp16")]; tensor reshape_37_shape_0 = const()[name = tensor("reshape_37_shape_0"), val = tensor([2, 2560, 16, 16])]; tensor reshape_37_cast_fp16 = reshape(shape = reshape_37_shape_0, x = real_div_9_cast_fp16)[name = tensor("reshape_37_cast_fp16")]; tensor add_19_mean_0_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(65012800))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(65014784))), name = tensor("add_19_mean_0_to_fp16_palettized"), shape = tensor([2560])]; tensor add_19_variance_0_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(65014976))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(65016960))), name = tensor("add_19_variance_0_to_fp16_palettized"), shape = tensor([2560])]; tensor add_19_gamma_0_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(65017152))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(65019136))), name = tensor("add_19_gamma_0_to_fp16_palettized"), shape = tensor([2560])]; tensor add_19_beta_0_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(65019328))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(65021312))), name = tensor("add_19_beta_0_to_fp16_palettized"), shape = tensor([2560])]; 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_fp16 = batch_norm(beta = add_19_beta_0_to_fp16_palettized, epsilon = add_19_epsilon_0_to_fp16, gamma = add_19_gamma_0_to_fp16_palettized, mean = add_19_mean_0_to_fp16_palettized, variance = add_19_variance_0_to_fp16_palettized, x = reshape_37_cast_fp16)[name = tensor("add_19_cast_fp16")]; tensor input_95_cast_fp16 = silu(x = add_19_cast_fp16)[name = tensor("input_95_cast_fp16")]; tensor var_2967 = const()[name = tensor("op_2967"), val = tensor([1, 1])]; tensor var_2969 = const()[name = tensor("op_2969"), 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 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(65021504))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(87139968))), 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(87140160)))]; tensor hidden_states_57_cast_fp16 = conv(bias = up_blocks_0_resnets_0_conv1_bias_to_fp16, dilations = var_2969, groups = var_2942, pad = hidden_states_57_pad_0, pad_type = hidden_states_57_pad_type_0, strides = var_2967, weight = up_blocks_0_resnets_0_conv1_weight_to_fp16_palettized, x = input_95_cast_fp16)[name = tensor("hidden_states_57_cast_fp16")]; tensor var_2975 = const()[name = tensor("op_2975"), val = tensor([1, 1])]; tensor var_2977 = const()[name = tensor("op_2977"), 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 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(87142784))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(88371648))), 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(88371840)))]; tensor temb_7_cast_fp16 = conv(bias = up_blocks_0_resnets_0_time_emb_proj_bias_to_fp16, dilations = var_2977, groups = var_2942, pad = temb_7_pad_0, pad_type = temb_7_pad_type_0, strides = var_2975, weight = up_blocks_0_resnets_0_time_emb_proj_weight_to_fp16_palettized, x = input_15_cast_fp16)[name = tensor("temb_7_cast_fp16")]; tensor input_99_cast_fp16 = add(x = hidden_states_57_cast_fp16, y = temb_7_cast_fp16)[name = tensor("input_99_cast_fp16")]; tensor reshape_40_shape_0 = const()[name = tensor("reshape_40_shape_0"), val = tensor([2, 32, 40, 16, 16])]; tensor reshape_40_cast_fp16 = reshape(shape = reshape_40_shape_0, x = input_99_cast_fp16)[name = tensor("reshape_40_cast_fp16")]; 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_fp16 = reduce_mean(axes = reduce_mean_30_axes_0, keep_dims = reduce_mean_30_keep_dims_0, x = reshape_40_cast_fp16)[name = tensor("reduce_mean_30_cast_fp16")]; tensor sub_20_cast_fp16 = sub(x = reshape_40_cast_fp16, y = reduce_mean_30_cast_fp16)[name = tensor("sub_20_cast_fp16")]; tensor square_10_cast_fp16 = square(x = sub_20_cast_fp16)[name = tensor("square_10_cast_fp16")]; 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_fp16 = reduce_mean(axes = reduce_mean_32_axes_0, keep_dims = reduce_mean_32_keep_dims_0, x = square_10_cast_fp16)[name = tensor("reduce_mean_32_cast_fp16")]; 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_fp16 = add(x = reduce_mean_32_cast_fp16, y = add_20_y_0_to_fp16)[name = tensor("add_20_cast_fp16")]; tensor sqrt_10_cast_fp16 = sqrt(x = add_20_cast_fp16)[name = tensor("sqrt_10_cast_fp16")]; tensor real_div_10_cast_fp16 = real_div(x = sub_20_cast_fp16, y = sqrt_10_cast_fp16)[name = tensor("real_div_10_cast_fp16")]; tensor reshape_41_shape_0 = const()[name = tensor("reshape_41_shape_0"), val = tensor([2, 1280, 16, 16])]; tensor reshape_41_cast_fp16 = reshape(shape = reshape_41_shape_0, x = real_div_10_cast_fp16)[name = tensor("reshape_41_cast_fp16")]; 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(88374464)))]; 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(88377088)))]; 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_fp16 = 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_fp16)[name = tensor("add_21_cast_fp16")]; tensor input_103_cast_fp16 = silu(x = add_21_cast_fp16)[name = tensor("input_103_cast_fp16")]; tensor var_2987 = const()[name = tensor("op_2987"), val = tensor([1, 1])]; tensor var_2989 = const()[name = tensor("op_2989"), val = tensor([1, 1])]; tensor hidden_states_59_pad_type_0 = const()[name = tensor("hidden_states_59_pad_type_0"), val = tensor("custom")]; tensor hidden_states_59_pad_0 = const()[name = tensor("hidden_states_59_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(88379712))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(99438976))), 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(99439168)))]; tensor hidden_states_59_cast_fp16 = conv(bias = up_blocks_0_resnets_0_conv2_bias_to_fp16, dilations = var_2989, groups = var_2942, pad = hidden_states_59_pad_0, pad_type = hidden_states_59_pad_type_0, strides = var_2987, weight = up_blocks_0_resnets_0_conv2_weight_to_fp16_palettized, x = input_103_cast_fp16)[name = tensor("hidden_states_59_cast_fp16")]; tensor var_2994 = const()[name = tensor("op_2994"), val = tensor([1, 1])]; tensor var_2996 = const()[name = tensor("op_2996"), 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(99441792))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(101899456))), 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(101899648)))]; tensor x_5_cast_fp16 = conv(bias = up_blocks_0_resnets_0_conv_shortcut_bias_to_fp16, dilations = var_2996, groups = var_2942, pad = x_5_pad_0, pad_type = x_5_pad_type_0, strides = var_2994, weight = up_blocks_0_resnets_0_conv_shortcut_weight_to_fp16_palettized, x = input_91_cast_fp16)[name = tensor("x_5_cast_fp16")]; tensor hidden_states_61_cast_fp16 = add(x = x_5_cast_fp16, y = hidden_states_59_cast_fp16)[name = tensor("hidden_states_61_cast_fp16")]; tensor reshape_44_shape_0 = const()[name = tensor("reshape_44_shape_0"), val = tensor([2, 32, 40, 16, 16])]; tensor reshape_44_cast_fp16 = reshape(shape = reshape_44_shape_0, x = hidden_states_61_cast_fp16)[name = tensor("reshape_44_cast_fp16")]; 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_fp16 = reduce_mean(axes = reduce_mean_33_axes_0, keep_dims = reduce_mean_33_keep_dims_0, x = reshape_44_cast_fp16)[name = tensor("reduce_mean_33_cast_fp16")]; tensor sub_22_cast_fp16 = sub(x = reshape_44_cast_fp16, y = reduce_mean_33_cast_fp16)[name = tensor("sub_22_cast_fp16")]; tensor square_11_cast_fp16 = square(x = sub_22_cast_fp16)[name = tensor("square_11_cast_fp16")]; 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_fp16 = reduce_mean(axes = reduce_mean_35_axes_0, keep_dims = reduce_mean_35_keep_dims_0, x = square_11_cast_fp16)[name = tensor("reduce_mean_35_cast_fp16")]; 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_fp16 = add(x = reduce_mean_35_cast_fp16, y = add_22_y_0_to_fp16)[name = tensor("add_22_cast_fp16")]; tensor sqrt_11_cast_fp16 = sqrt(x = add_22_cast_fp16)[name = tensor("sqrt_11_cast_fp16")]; tensor real_div_11_cast_fp16 = real_div(x = sub_22_cast_fp16, y = sqrt_11_cast_fp16)[name = tensor("real_div_11_cast_fp16")]; tensor reshape_45_shape_0 = const()[name = tensor("reshape_45_shape_0"), val = tensor([2, 1280, 16, 16])]; tensor reshape_45_cast_fp16 = reshape(shape = reshape_45_shape_0, x = real_div_11_cast_fp16)[name = tensor("reshape_45_cast_fp16")]; 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(101902272)))]; 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(101904896)))]; 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_fp16 = 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_fp16)[name = tensor("add_23_cast_fp16")]; tensor var_3016 = const()[name = tensor("op_3016"), val = tensor([1, 1])]; tensor var_3018 = const()[name = tensor("op_3018"), val = tensor([1, 1])]; tensor hidden_states_63_pad_type_0 = const()[name = tensor("hidden_states_63_pad_type_0"), val = tensor("custom")]; tensor hidden_states_63_pad_0 = const()[name = tensor("hidden_states_63_pad_0"), val = tensor([0, 0, 0, 0])]; tensor up_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(101907520))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(103136384))), name = tensor("up_blocks_0_attentions_0_proj_in_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; tensor up_blocks_0_attentions_0_proj_in_bias_to_fp16 = const()[name = tensor("up_blocks_0_attentions_0_proj_in_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(103136576)))]; tensor hidden_states_63_cast_fp16 = conv(bias = up_blocks_0_attentions_0_proj_in_bias_to_fp16, dilations = var_3018, groups = var_2942, pad = hidden_states_63_pad_0, pad_type = hidden_states_63_pad_type_0, strides = var_3016, weight = up_blocks_0_attentions_0_proj_in_weight_to_fp16_palettized, x = add_23_cast_fp16)[name = tensor("hidden_states_63_cast_fp16")]; tensor var_3023 = const()[name = tensor("op_3023"), val = tensor([2, 1280, 1, 256])]; tensor inputs_19_cast_fp16 = reshape(shape = var_3023, x = hidden_states_63_cast_fp16)[name = tensor("inputs_19_cast_fp16")]; tensor var_3033 = const()[name = tensor("op_3033"), val = tensor([1])]; tensor channels_mean_19_cast_fp16 = reduce_mean(axes = var_3033, keep_dims = var_2937, x = inputs_19_cast_fp16)[name = tensor("channels_mean_19_cast_fp16")]; tensor zero_mean_19_cast_fp16 = sub(x = inputs_19_cast_fp16, y = channels_mean_19_cast_fp16)[name = tensor("zero_mean_19_cast_fp16")]; tensor zero_mean_sq_19_cast_fp16 = mul(x = zero_mean_19_cast_fp16, y = zero_mean_19_cast_fp16)[name = tensor("zero_mean_sq_19_cast_fp16")]; tensor var_3037 = const()[name = tensor("op_3037"), val = tensor([1])]; tensor var_3038_cast_fp16 = reduce_mean(axes = var_3037, keep_dims = var_2937, x = zero_mean_sq_19_cast_fp16)[name = tensor("op_3038_cast_fp16")]; tensor var_3039_to_fp16 = const()[name = tensor("op_3039_to_fp16"), val = tensor(0x1.5p-17)]; tensor var_3040_cast_fp16 = add(x = var_3038_cast_fp16, y = var_3039_to_fp16)[name = tensor("op_3040_cast_fp16")]; tensor denom_19_epsilon_0_to_fp16 = const()[name = tensor("denom_19_epsilon_0_to_fp16"), val = tensor(0x1p-24)]; tensor denom_19_cast_fp16 = rsqrt(epsilon = denom_19_epsilon_0_to_fp16, x = var_3040_cast_fp16)[name = tensor("denom_19_cast_fp16")]; tensor out_19_cast_fp16 = mul(x = zero_mean_19_cast_fp16, y = denom_19_cast_fp16)[name = tensor("out_19_cast_fp16")]; tensor var_3044_to_fp16 = const()[name = tensor("op_3044_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(103139200)))]; tensor var_3045_cast_fp16 = add(x = out_19_cast_fp16, y = var_3044_to_fp16)[name = tensor("op_3045_cast_fp16")]; tensor var_3047_to_fp16 = const()[name = tensor("op_3047_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(103141824)))]; tensor hidden_states_65_cast_fp16 = mul(x = var_3045_cast_fp16, y = var_3047_to_fp16)[name = tensor("hidden_states_65_cast_fp16")]; tensor var_3054 = const()[name = tensor("op_3054"), val = tensor([1, 1])]; tensor var_3056 = const()[name = tensor("op_3056"), 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 up_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(103144448))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(104373312))), name = tensor("up_blocks_0_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; tensor q_13_cast_fp16 = conv(dilations = var_3056, groups = var_2942, pad = q_13_pad_0, pad_type = q_13_pad_type_0, strides = var_3054, weight = up_blocks_0_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16_palettized, x = hidden_states_65_cast_fp16)[name = tensor("q_13_cast_fp16")]; tensor var_3060 = const()[name = tensor("op_3060"), val = tensor([1, 1])]; tensor var_3062 = const()[name = tensor("op_3062"), 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 up_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(104373504))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(105602368))), name = tensor("up_blocks_0_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; tensor k_25_cast_fp16 = conv(dilations = var_3062, groups = var_2942, pad = k_25_pad_0, pad_type = k_25_pad_type_0, strides = var_3060, weight = up_blocks_0_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16_palettized, x = hidden_states_65_cast_fp16)[name = tensor("k_25_cast_fp16")]; tensor var_3066 = const()[name = tensor("op_3066"), val = tensor([1, 1])]; tensor var_3068 = const()[name = tensor("op_3068"), 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 up_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(105602560))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(106831424))), name = tensor("up_blocks_0_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; tensor v_13_cast_fp16 = conv(dilations = var_3068, groups = var_2942, pad = v_13_pad_0, pad_type = v_13_pad_type_0, strides = var_3066, weight = up_blocks_0_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16_palettized, x = hidden_states_65_cast_fp16)[name = tensor("v_13_cast_fp16")]; tensor var_3072_begin_0 = const()[name = tensor("op_3072_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_3072_end_0 = const()[name = tensor("op_3072_end_0"), val = tensor([2, 160, 1, 256])]; tensor var_3072_end_mask_0 = const()[name = tensor("op_3072_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_3072_cast_fp16 = slice_by_index(begin = var_3072_begin_0, end = var_3072_end_0, end_mask = var_3072_end_mask_0, x = q_13_cast_fp16)[name = tensor("op_3072_cast_fp16")]; tensor var_3076_begin_0 = const()[name = tensor("op_3076_begin_0"), val = tensor([0, 160, 0, 0])]; tensor var_3076_end_0 = const()[name = tensor("op_3076_end_0"), val = tensor([2, 320, 1, 256])]; tensor var_3076_end_mask_0 = const()[name = tensor("op_3076_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_3076_cast_fp16 = slice_by_index(begin = var_3076_begin_0, end = var_3076_end_0, end_mask = var_3076_end_mask_0, x = q_13_cast_fp16)[name = tensor("op_3076_cast_fp16")]; tensor var_3080_begin_0 = const()[name = tensor("op_3080_begin_0"), val = tensor([0, 320, 0, 0])]; tensor var_3080_end_0 = const()[name = tensor("op_3080_end_0"), val = tensor([2, 480, 1, 256])]; tensor var_3080_end_mask_0 = const()[name = tensor("op_3080_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_3080_cast_fp16 = slice_by_index(begin = var_3080_begin_0, end = var_3080_end_0, end_mask = var_3080_end_mask_0, x = q_13_cast_fp16)[name = tensor("op_3080_cast_fp16")]; tensor var_3084_begin_0 = const()[name = tensor("op_3084_begin_0"), val = tensor([0, 480, 0, 0])]; tensor var_3084_end_0 = const()[name = tensor("op_3084_end_0"), val = tensor([2, 640, 1, 256])]; tensor var_3084_end_mask_0 = const()[name = tensor("op_3084_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_3084_cast_fp16 = slice_by_index(begin = var_3084_begin_0, end = var_3084_end_0, end_mask = var_3084_end_mask_0, x = q_13_cast_fp16)[name = tensor("op_3084_cast_fp16")]; tensor var_3088_begin_0 = const()[name = tensor("op_3088_begin_0"), val = tensor([0, 640, 0, 0])]; tensor var_3088_end_0 = const()[name = tensor("op_3088_end_0"), val = tensor([2, 800, 1, 256])]; tensor var_3088_end_mask_0 = const()[name = tensor("op_3088_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_3088_cast_fp16 = slice_by_index(begin = var_3088_begin_0, end = var_3088_end_0, end_mask = var_3088_end_mask_0, x = q_13_cast_fp16)[name = tensor("op_3088_cast_fp16")]; tensor var_3092_begin_0 = const()[name = tensor("op_3092_begin_0"), val = tensor([0, 800, 0, 0])]; tensor var_3092_end_0 = const()[name = tensor("op_3092_end_0"), val = tensor([2, 960, 1, 256])]; tensor var_3092_end_mask_0 = const()[name = tensor("op_3092_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_3092_cast_fp16 = slice_by_index(begin = var_3092_begin_0, end = var_3092_end_0, end_mask = var_3092_end_mask_0, x = q_13_cast_fp16)[name = tensor("op_3092_cast_fp16")]; tensor var_3096_begin_0 = const()[name = tensor("op_3096_begin_0"), val = tensor([0, 960, 0, 0])]; tensor var_3096_end_0 = const()[name = tensor("op_3096_end_0"), val = tensor([2, 1120, 1, 256])]; tensor var_3096_end_mask_0 = const()[name = tensor("op_3096_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_3096_cast_fp16 = slice_by_index(begin = var_3096_begin_0, end = var_3096_end_0, end_mask = var_3096_end_mask_0, x = q_13_cast_fp16)[name = tensor("op_3096_cast_fp16")]; tensor var_3100_begin_0 = const()[name = tensor("op_3100_begin_0"), val = tensor([0, 1120, 0, 0])]; tensor var_3100_end_0 = const()[name = tensor("op_3100_end_0"), val = tensor([2, 1280, 1, 256])]; tensor var_3100_end_mask_0 = const()[name = tensor("op_3100_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_3100_cast_fp16 = slice_by_index(begin = var_3100_begin_0, end = var_3100_end_0, end_mask = var_3100_end_mask_0, x = q_13_cast_fp16)[name = tensor("op_3100_cast_fp16")]; tensor k_27_perm_0 = const()[name = tensor("k_27_perm_0"), val = tensor([0, 3, 2, 1])]; tensor var_3107_begin_0 = const()[name = tensor("op_3107_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_3107_end_0 = const()[name = tensor("op_3107_end_0"), val = tensor([2, 256, 1, 160])]; tensor var_3107_end_mask_0 = const()[name = tensor("op_3107_end_mask_0"), val = tensor([true, true, true, false])]; tensor transpose_11 = transpose(perm = k_27_perm_0, x = k_25_cast_fp16)[name = tensor("transpose_11")]; tensor var_3107_cast_fp16 = slice_by_index(begin = var_3107_begin_0, end = var_3107_end_0, end_mask = var_3107_end_mask_0, x = transpose_11)[name = tensor("op_3107_cast_fp16")]; tensor var_3111_begin_0 = const()[name = tensor("op_3111_begin_0"), val = tensor([0, 0, 0, 160])]; tensor var_3111_end_0 = const()[name = tensor("op_3111_end_0"), val = tensor([2, 256, 1, 320])]; tensor var_3111_end_mask_0 = const()[name = tensor("op_3111_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_3111_cast_fp16 = slice_by_index(begin = var_3111_begin_0, end = var_3111_end_0, end_mask = var_3111_end_mask_0, x = transpose_11)[name = tensor("op_3111_cast_fp16")]; tensor var_3115_begin_0 = const()[name = tensor("op_3115_begin_0"), val = tensor([0, 0, 0, 320])]; tensor var_3115_end_0 = const()[name = tensor("op_3115_end_0"), val = tensor([2, 256, 1, 480])]; tensor var_3115_end_mask_0 = const()[name = tensor("op_3115_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_3115_cast_fp16 = slice_by_index(begin = var_3115_begin_0, end = var_3115_end_0, end_mask = var_3115_end_mask_0, x = transpose_11)[name = tensor("op_3115_cast_fp16")]; tensor var_3119_begin_0 = const()[name = tensor("op_3119_begin_0"), val = tensor([0, 0, 0, 480])]; tensor var_3119_end_0 = const()[name = tensor("op_3119_end_0"), val = tensor([2, 256, 1, 640])]; tensor var_3119_end_mask_0 = const()[name = tensor("op_3119_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_3119_cast_fp16 = slice_by_index(begin = var_3119_begin_0, end = var_3119_end_0, end_mask = var_3119_end_mask_0, x = transpose_11)[name = tensor("op_3119_cast_fp16")]; tensor var_3123_begin_0 = const()[name = tensor("op_3123_begin_0"), val = tensor([0, 0, 0, 640])]; tensor var_3123_end_0 = const()[name = tensor("op_3123_end_0"), val = tensor([2, 256, 1, 800])]; tensor var_3123_end_mask_0 = const()[name = tensor("op_3123_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_3123_cast_fp16 = slice_by_index(begin = var_3123_begin_0, end = var_3123_end_0, end_mask = var_3123_end_mask_0, x = transpose_11)[name = tensor("op_3123_cast_fp16")]; tensor var_3127_begin_0 = const()[name = tensor("op_3127_begin_0"), val = tensor([0, 0, 0, 800])]; tensor var_3127_end_0 = const()[name = tensor("op_3127_end_0"), val = tensor([2, 256, 1, 960])]; tensor var_3127_end_mask_0 = const()[name = tensor("op_3127_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_3127_cast_fp16 = slice_by_index(begin = var_3127_begin_0, end = var_3127_end_0, end_mask = var_3127_end_mask_0, x = transpose_11)[name = tensor("op_3127_cast_fp16")]; tensor var_3131_begin_0 = const()[name = tensor("op_3131_begin_0"), val = tensor([0, 0, 0, 960])]; tensor var_3131_end_0 = const()[name = tensor("op_3131_end_0"), val = tensor([2, 256, 1, 1120])]; tensor var_3131_end_mask_0 = const()[name = tensor("op_3131_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_3131_cast_fp16 = slice_by_index(begin = var_3131_begin_0, end = var_3131_end_0, end_mask = var_3131_end_mask_0, x = transpose_11)[name = tensor("op_3131_cast_fp16")]; tensor var_3135_begin_0 = const()[name = tensor("op_3135_begin_0"), val = tensor([0, 0, 0, 1120])]; tensor var_3135_end_0 = const()[name = tensor("op_3135_end_0"), val = tensor([2, 256, 1, 1280])]; tensor var_3135_end_mask_0 = const()[name = tensor("op_3135_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_3135_cast_fp16 = slice_by_index(begin = var_3135_begin_0, end = var_3135_end_0, end_mask = var_3135_end_mask_0, x = transpose_11)[name = tensor("op_3135_cast_fp16")]; tensor var_3137_begin_0 = const()[name = tensor("op_3137_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_3137_end_0 = const()[name = tensor("op_3137_end_0"), val = tensor([2, 160, 1, 256])]; tensor var_3137_end_mask_0 = const()[name = tensor("op_3137_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_3137_cast_fp16 = slice_by_index(begin = var_3137_begin_0, end = var_3137_end_0, end_mask = var_3137_end_mask_0, x = v_13_cast_fp16)[name = tensor("op_3137_cast_fp16")]; tensor var_3141_begin_0 = const()[name = tensor("op_3141_begin_0"), val = tensor([0, 160, 0, 0])]; tensor var_3141_end_0 = const()[name = tensor("op_3141_end_0"), val = tensor([2, 320, 1, 256])]; tensor var_3141_end_mask_0 = const()[name = tensor("op_3141_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_3141_cast_fp16 = slice_by_index(begin = var_3141_begin_0, end = var_3141_end_0, end_mask = var_3141_end_mask_0, x = v_13_cast_fp16)[name = tensor("op_3141_cast_fp16")]; tensor var_3145_begin_0 = const()[name = tensor("op_3145_begin_0"), val = tensor([0, 320, 0, 0])]; tensor var_3145_end_0 = const()[name = tensor("op_3145_end_0"), val = tensor([2, 480, 1, 256])]; tensor var_3145_end_mask_0 = const()[name = tensor("op_3145_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_3145_cast_fp16 = slice_by_index(begin = var_3145_begin_0, end = var_3145_end_0, end_mask = var_3145_end_mask_0, x = v_13_cast_fp16)[name = tensor("op_3145_cast_fp16")]; tensor var_3149_begin_0 = const()[name = tensor("op_3149_begin_0"), val = tensor([0, 480, 0, 0])]; tensor var_3149_end_0 = const()[name = tensor("op_3149_end_0"), val = tensor([2, 640, 1, 256])]; tensor var_3149_end_mask_0 = const()[name = tensor("op_3149_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_3149_cast_fp16 = slice_by_index(begin = var_3149_begin_0, end = var_3149_end_0, end_mask = var_3149_end_mask_0, x = v_13_cast_fp16)[name = tensor("op_3149_cast_fp16")]; tensor var_3153_begin_0 = const()[name = tensor("op_3153_begin_0"), val = tensor([0, 640, 0, 0])]; tensor var_3153_end_0 = const()[name = tensor("op_3153_end_0"), val = tensor([2, 800, 1, 256])]; tensor var_3153_end_mask_0 = const()[name = tensor("op_3153_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_3153_cast_fp16 = slice_by_index(begin = var_3153_begin_0, end = var_3153_end_0, end_mask = var_3153_end_mask_0, x = v_13_cast_fp16)[name = tensor("op_3153_cast_fp16")]; tensor var_3157_begin_0 = const()[name = tensor("op_3157_begin_0"), val = tensor([0, 800, 0, 0])]; tensor var_3157_end_0 = const()[name = tensor("op_3157_end_0"), val = tensor([2, 960, 1, 256])]; tensor var_3157_end_mask_0 = const()[name = tensor("op_3157_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_3157_cast_fp16 = slice_by_index(begin = var_3157_begin_0, end = var_3157_end_0, end_mask = var_3157_end_mask_0, x = v_13_cast_fp16)[name = tensor("op_3157_cast_fp16")]; tensor var_3161_begin_0 = const()[name = tensor("op_3161_begin_0"), val = tensor([0, 960, 0, 0])]; tensor var_3161_end_0 = const()[name = tensor("op_3161_end_0"), val = tensor([2, 1120, 1, 256])]; tensor var_3161_end_mask_0 = const()[name = tensor("op_3161_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_3161_cast_fp16 = slice_by_index(begin = var_3161_begin_0, end = var_3161_end_0, end_mask = var_3161_end_mask_0, x = v_13_cast_fp16)[name = tensor("op_3161_cast_fp16")]; tensor var_3165_begin_0 = const()[name = tensor("op_3165_begin_0"), val = tensor([0, 1120, 0, 0])]; tensor var_3165_end_0 = const()[name = tensor("op_3165_end_0"), val = tensor([2, 1280, 1, 256])]; tensor var_3165_end_mask_0 = const()[name = tensor("op_3165_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_3165_cast_fp16 = slice_by_index(begin = var_3165_begin_0, end = var_3165_end_0, end_mask = var_3165_end_mask_0, x = v_13_cast_fp16)[name = tensor("op_3165_cast_fp16")]; tensor var_3169_equation_0 = const()[name = tensor("op_3169_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_3169_cast_fp16 = einsum(equation = var_3169_equation_0, values = (var_3107_cast_fp16, var_3072_cast_fp16))[name = tensor("op_3169_cast_fp16")]; tensor var_3170_to_fp16 = const()[name = tensor("op_3170_to_fp16"), val = tensor(0x1.43cp-4)]; tensor aw_33_cast_fp16 = mul(x = var_3169_cast_fp16, y = var_3170_to_fp16)[name = tensor("aw_33_cast_fp16")]; tensor var_3173_equation_0 = const()[name = tensor("op_3173_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_3173_cast_fp16 = einsum(equation = var_3173_equation_0, values = (var_3111_cast_fp16, var_3076_cast_fp16))[name = tensor("op_3173_cast_fp16")]; tensor var_3174_to_fp16 = const()[name = tensor("op_3174_to_fp16"), val = tensor(0x1.43cp-4)]; tensor aw_35_cast_fp16 = mul(x = var_3173_cast_fp16, y = var_3174_to_fp16)[name = tensor("aw_35_cast_fp16")]; tensor var_3177_equation_0 = const()[name = tensor("op_3177_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_3177_cast_fp16 = einsum(equation = var_3177_equation_0, values = (var_3115_cast_fp16, var_3080_cast_fp16))[name = tensor("op_3177_cast_fp16")]; tensor var_3178_to_fp16 = const()[name = tensor("op_3178_to_fp16"), val = tensor(0x1.43cp-4)]; tensor aw_37_cast_fp16 = mul(x = var_3177_cast_fp16, y = var_3178_to_fp16)[name = tensor("aw_37_cast_fp16")]; tensor var_3181_equation_0 = const()[name = tensor("op_3181_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_3181_cast_fp16 = einsum(equation = var_3181_equation_0, values = (var_3119_cast_fp16, var_3084_cast_fp16))[name = tensor("op_3181_cast_fp16")]; tensor var_3182_to_fp16 = const()[name = tensor("op_3182_to_fp16"), val = tensor(0x1.43cp-4)]; tensor aw_39_cast_fp16 = mul(x = var_3181_cast_fp16, y = var_3182_to_fp16)[name = tensor("aw_39_cast_fp16")]; tensor var_3185_equation_0 = const()[name = tensor("op_3185_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_3185_cast_fp16 = einsum(equation = var_3185_equation_0, values = (var_3123_cast_fp16, var_3088_cast_fp16))[name = tensor("op_3185_cast_fp16")]; tensor var_3186_to_fp16 = const()[name = tensor("op_3186_to_fp16"), val = tensor(0x1.43cp-4)]; tensor aw_41_cast_fp16 = mul(x = var_3185_cast_fp16, y = var_3186_to_fp16)[name = tensor("aw_41_cast_fp16")]; tensor var_3189_equation_0 = const()[name = tensor("op_3189_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_3189_cast_fp16 = einsum(equation = var_3189_equation_0, values = (var_3127_cast_fp16, var_3092_cast_fp16))[name = tensor("op_3189_cast_fp16")]; tensor var_3190_to_fp16 = const()[name = tensor("op_3190_to_fp16"), val = tensor(0x1.43cp-4)]; tensor aw_43_cast_fp16 = mul(x = var_3189_cast_fp16, y = var_3190_to_fp16)[name = tensor("aw_43_cast_fp16")]; tensor var_3193_equation_0 = const()[name = tensor("op_3193_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_3193_cast_fp16 = einsum(equation = var_3193_equation_0, values = (var_3131_cast_fp16, var_3096_cast_fp16))[name = tensor("op_3193_cast_fp16")]; tensor var_3194_to_fp16 = const()[name = tensor("op_3194_to_fp16"), val = tensor(0x1.43cp-4)]; tensor aw_45_cast_fp16 = mul(x = var_3193_cast_fp16, y = var_3194_to_fp16)[name = tensor("aw_45_cast_fp16")]; tensor var_3197_equation_0 = const()[name = tensor("op_3197_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_3197_cast_fp16 = einsum(equation = var_3197_equation_0, values = (var_3135_cast_fp16, var_3100_cast_fp16))[name = tensor("op_3197_cast_fp16")]; tensor var_3198_to_fp16 = const()[name = tensor("op_3198_to_fp16"), val = tensor(0x1.43cp-4)]; tensor aw_47_cast_fp16 = mul(x = var_3197_cast_fp16, y = var_3198_to_fp16)[name = tensor("aw_47_cast_fp16")]; tensor var_3200_cast_fp16 = softmax(axis = var_2942, x = aw_33_cast_fp16)[name = tensor("op_3200_cast_fp16")]; tensor var_3201_cast_fp16 = softmax(axis = var_2942, x = aw_35_cast_fp16)[name = tensor("op_3201_cast_fp16")]; tensor var_3202_cast_fp16 = softmax(axis = var_2942, x = aw_37_cast_fp16)[name = tensor("op_3202_cast_fp16")]; tensor var_3203_cast_fp16 = softmax(axis = var_2942, x = aw_39_cast_fp16)[name = tensor("op_3203_cast_fp16")]; tensor var_3204_cast_fp16 = softmax(axis = var_2942, x = aw_41_cast_fp16)[name = tensor("op_3204_cast_fp16")]; tensor var_3205_cast_fp16 = softmax(axis = var_2942, x = aw_43_cast_fp16)[name = tensor("op_3205_cast_fp16")]; tensor var_3206_cast_fp16 = softmax(axis = var_2942, x = aw_45_cast_fp16)[name = tensor("op_3206_cast_fp16")]; tensor var_3207_cast_fp16 = softmax(axis = var_2942, x = aw_47_cast_fp16)[name = tensor("op_3207_cast_fp16")]; tensor var_3209_equation_0 = const()[name = tensor("op_3209_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_3209_cast_fp16 = einsum(equation = var_3209_equation_0, values = (var_3137_cast_fp16, var_3200_cast_fp16))[name = tensor("op_3209_cast_fp16")]; tensor var_3211_equation_0 = const()[name = tensor("op_3211_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_3211_cast_fp16 = einsum(equation = var_3211_equation_0, values = (var_3141_cast_fp16, var_3201_cast_fp16))[name = tensor("op_3211_cast_fp16")]; tensor var_3213_equation_0 = const()[name = tensor("op_3213_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_3213_cast_fp16 = einsum(equation = var_3213_equation_0, values = (var_3145_cast_fp16, var_3202_cast_fp16))[name = tensor("op_3213_cast_fp16")]; tensor var_3215_equation_0 = const()[name = tensor("op_3215_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_3215_cast_fp16 = einsum(equation = var_3215_equation_0, values = (var_3149_cast_fp16, var_3203_cast_fp16))[name = tensor("op_3215_cast_fp16")]; tensor var_3217_equation_0 = const()[name = tensor("op_3217_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_3217_cast_fp16 = einsum(equation = var_3217_equation_0, values = (var_3153_cast_fp16, var_3204_cast_fp16))[name = tensor("op_3217_cast_fp16")]; tensor var_3219_equation_0 = const()[name = tensor("op_3219_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_3219_cast_fp16 = einsum(equation = var_3219_equation_0, values = (var_3157_cast_fp16, var_3205_cast_fp16))[name = tensor("op_3219_cast_fp16")]; tensor var_3221_equation_0 = const()[name = tensor("op_3221_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_3221_cast_fp16 = einsum(equation = var_3221_equation_0, values = (var_3161_cast_fp16, var_3206_cast_fp16))[name = tensor("op_3221_cast_fp16")]; tensor var_3223_equation_0 = const()[name = tensor("op_3223_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_3223_cast_fp16 = einsum(equation = var_3223_equation_0, values = (var_3165_cast_fp16, var_3207_cast_fp16))[name = tensor("op_3223_cast_fp16")]; tensor input_107_interleave_0 = const()[name = tensor("input_107_interleave_0"), val = tensor(false)]; tensor input_107_cast_fp16 = concat(axis = var_2942, interleave = input_107_interleave_0, values = (var_3209_cast_fp16, var_3211_cast_fp16, var_3213_cast_fp16, var_3215_cast_fp16, var_3217_cast_fp16, var_3219_cast_fp16, var_3221_cast_fp16, var_3223_cast_fp16))[name = tensor("input_107_cast_fp16")]; tensor var_3229 = const()[name = tensor("op_3229"), val = tensor([1, 1])]; tensor var_3231 = const()[name = tensor("op_3231"), val = tensor([1, 1])]; tensor var_3233_pad_type_0 = const()[name = tensor("op_3233_pad_type_0"), val = tensor("custom")]; tensor var_3233_pad_0 = const()[name = tensor("op_3233_pad_0"), val = tensor([0, 0, 0, 0])]; tensor up_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(106831616))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(108060480))), name = tensor("up_blocks_0_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; tensor up_blocks_0_attentions_0_transformer_blocks_0_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("up_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(108060672)))]; tensor var_3233_cast_fp16 = conv(bias = up_blocks_0_attentions_0_transformer_blocks_0_attn1_to_out_0_bias_to_fp16, dilations = var_3231, groups = var_2942, pad = var_3233_pad_0, pad_type = var_3233_pad_type_0, strides = var_3229, weight = up_blocks_0_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16_palettized, x = input_107_cast_fp16)[name = tensor("op_3233_cast_fp16")]; tensor inputs_21_cast_fp16 = add(x = var_3233_cast_fp16, y = inputs_19_cast_fp16)[name = tensor("inputs_21_cast_fp16")]; tensor var_3237 = const()[name = tensor("op_3237"), val = tensor([1])]; tensor channels_mean_21_cast_fp16 = reduce_mean(axes = var_3237, keep_dims = var_2937, x = inputs_21_cast_fp16)[name = tensor("channels_mean_21_cast_fp16")]; tensor zero_mean_21_cast_fp16 = sub(x = inputs_21_cast_fp16, y = channels_mean_21_cast_fp16)[name = tensor("zero_mean_21_cast_fp16")]; tensor zero_mean_sq_21_cast_fp16 = mul(x = zero_mean_21_cast_fp16, y = zero_mean_21_cast_fp16)[name = tensor("zero_mean_sq_21_cast_fp16")]; tensor var_3241 = const()[name = tensor("op_3241"), val = tensor([1])]; tensor var_3242_cast_fp16 = reduce_mean(axes = var_3241, keep_dims = var_2937, x = zero_mean_sq_21_cast_fp16)[name = tensor("op_3242_cast_fp16")]; tensor var_3243_to_fp16 = const()[name = tensor("op_3243_to_fp16"), val = tensor(0x1.5p-17)]; tensor var_3244_cast_fp16 = add(x = var_3242_cast_fp16, y = var_3243_to_fp16)[name = tensor("op_3244_cast_fp16")]; tensor denom_21_epsilon_0_to_fp16 = const()[name = tensor("denom_21_epsilon_0_to_fp16"), val = tensor(0x1p-24)]; tensor denom_21_cast_fp16 = rsqrt(epsilon = denom_21_epsilon_0_to_fp16, x = var_3244_cast_fp16)[name = tensor("denom_21_cast_fp16")]; tensor out_21_cast_fp16 = mul(x = zero_mean_21_cast_fp16, y = denom_21_cast_fp16)[name = tensor("out_21_cast_fp16")]; tensor var_3248_to_fp16 = const()[name = tensor("op_3248_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(108063296)))]; tensor var_3249_cast_fp16 = add(x = out_21_cast_fp16, y = var_3248_to_fp16)[name = tensor("op_3249_cast_fp16")]; tensor var_3251_to_fp16 = const()[name = tensor("op_3251_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(108065920)))]; tensor hidden_states_67_cast_fp16 = mul(x = var_3249_cast_fp16, y = var_3251_to_fp16)[name = tensor("hidden_states_67_cast_fp16")]; tensor var_3258 = const()[name = tensor("op_3258"), val = tensor([1, 1])]; tensor var_3260 = const()[name = tensor("op_3260"), 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 up_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(108068544))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(109297408))), name = tensor("up_blocks_0_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; tensor q_15_cast_fp16 = conv(dilations = var_3260, groups = var_2942, pad = q_15_pad_0, pad_type = q_15_pad_type_0, strides = var_3258, weight = up_blocks_0_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16_palettized, x = hidden_states_67_cast_fp16)[name = tensor("q_15_cast_fp16")]; tensor var_3264 = const()[name = tensor("op_3264"), val = tensor([1, 1])]; tensor var_3266 = const()[name = tensor("op_3266"), 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 up_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(109297600))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(110034944))), name = tensor("up_blocks_0_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized"), shape = tensor([1280, 768, 1, 1])]; tensor k_29_cast_fp16 = conv(dilations = var_3266, groups = var_2942, pad = k_29_pad_0, pad_type = k_29_pad_type_0, strides = var_3264, weight = up_blocks_0_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor("k_29_cast_fp16")]; tensor var_3270 = const()[name = tensor("op_3270"), val = tensor([1, 1])]; tensor var_3272 = const()[name = tensor("op_3272"), 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 up_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(110035136))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(110772480))), name = tensor("up_blocks_0_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized"), shape = tensor([1280, 768, 1, 1])]; tensor v_15_cast_fp16 = conv(dilations = var_3272, groups = var_2942, pad = v_15_pad_0, pad_type = v_15_pad_type_0, strides = var_3270, weight = up_blocks_0_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor("v_15_cast_fp16")]; tensor var_3276_begin_0 = const()[name = tensor("op_3276_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_3276_end_0 = const()[name = tensor("op_3276_end_0"), val = tensor([2, 160, 1, 256])]; tensor var_3276_end_mask_0 = const()[name = tensor("op_3276_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_3276_cast_fp16 = slice_by_index(begin = var_3276_begin_0, end = var_3276_end_0, end_mask = var_3276_end_mask_0, x = q_15_cast_fp16)[name = tensor("op_3276_cast_fp16")]; tensor var_3280_begin_0 = const()[name = tensor("op_3280_begin_0"), val = tensor([0, 160, 0, 0])]; tensor var_3280_end_0 = const()[name = tensor("op_3280_end_0"), val = tensor([2, 320, 1, 256])]; tensor var_3280_end_mask_0 = const()[name = tensor("op_3280_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_3280_cast_fp16 = slice_by_index(begin = var_3280_begin_0, end = var_3280_end_0, end_mask = var_3280_end_mask_0, x = q_15_cast_fp16)[name = tensor("op_3280_cast_fp16")]; tensor var_3284_begin_0 = const()[name = tensor("op_3284_begin_0"), val = tensor([0, 320, 0, 0])]; tensor var_3284_end_0 = const()[name = tensor("op_3284_end_0"), val = tensor([2, 480, 1, 256])]; tensor var_3284_end_mask_0 = const()[name = tensor("op_3284_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_3284_cast_fp16 = slice_by_index(begin = var_3284_begin_0, end = var_3284_end_0, end_mask = var_3284_end_mask_0, x = q_15_cast_fp16)[name = tensor("op_3284_cast_fp16")]; tensor var_3288_begin_0 = const()[name = tensor("op_3288_begin_0"), val = tensor([0, 480, 0, 0])]; tensor var_3288_end_0 = const()[name = tensor("op_3288_end_0"), val = tensor([2, 640, 1, 256])]; tensor var_3288_end_mask_0 = const()[name = tensor("op_3288_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_3288_cast_fp16 = slice_by_index(begin = var_3288_begin_0, end = var_3288_end_0, end_mask = var_3288_end_mask_0, x = q_15_cast_fp16)[name = tensor("op_3288_cast_fp16")]; tensor var_3292_begin_0 = const()[name = tensor("op_3292_begin_0"), val = tensor([0, 640, 0, 0])]; tensor var_3292_end_0 = const()[name = tensor("op_3292_end_0"), val = tensor([2, 800, 1, 256])]; tensor var_3292_end_mask_0 = const()[name = tensor("op_3292_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_3292_cast_fp16 = slice_by_index(begin = var_3292_begin_0, end = var_3292_end_0, end_mask = var_3292_end_mask_0, x = q_15_cast_fp16)[name = tensor("op_3292_cast_fp16")]; tensor var_3296_begin_0 = const()[name = tensor("op_3296_begin_0"), val = tensor([0, 800, 0, 0])]; tensor var_3296_end_0 = const()[name = tensor("op_3296_end_0"), val = tensor([2, 960, 1, 256])]; tensor var_3296_end_mask_0 = const()[name = tensor("op_3296_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_3296_cast_fp16 = slice_by_index(begin = var_3296_begin_0, end = var_3296_end_0, end_mask = var_3296_end_mask_0, x = q_15_cast_fp16)[name = tensor("op_3296_cast_fp16")]; tensor var_3300_begin_0 = const()[name = tensor("op_3300_begin_0"), val = tensor([0, 960, 0, 0])]; tensor var_3300_end_0 = const()[name = tensor("op_3300_end_0"), val = tensor([2, 1120, 1, 256])]; tensor var_3300_end_mask_0 = const()[name = tensor("op_3300_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_3300_cast_fp16 = slice_by_index(begin = var_3300_begin_0, end = var_3300_end_0, end_mask = var_3300_end_mask_0, x = q_15_cast_fp16)[name = tensor("op_3300_cast_fp16")]; tensor var_3304_begin_0 = const()[name = tensor("op_3304_begin_0"), val = tensor([0, 1120, 0, 0])]; tensor var_3304_end_0 = const()[name = tensor("op_3304_end_0"), val = tensor([2, 1280, 1, 256])]; tensor var_3304_end_mask_0 = const()[name = tensor("op_3304_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_3304_cast_fp16 = slice_by_index(begin = var_3304_begin_0, end = var_3304_end_0, end_mask = var_3304_end_mask_0, x = q_15_cast_fp16)[name = tensor("op_3304_cast_fp16")]; tensor k_31_perm_0 = const()[name = tensor("k_31_perm_0"), val = tensor([0, 3, 2, 1])]; tensor var_3311_begin_0 = const()[name = tensor("op_3311_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_3311_end_0 = const()[name = tensor("op_3311_end_0"), val = tensor([2, 77, 1, 160])]; tensor var_3311_end_mask_0 = const()[name = tensor("op_3311_end_mask_0"), val = tensor([true, true, true, false])]; tensor transpose_10 = transpose(perm = k_31_perm_0, x = k_29_cast_fp16)[name = tensor("transpose_10")]; tensor var_3311_cast_fp16 = slice_by_index(begin = var_3311_begin_0, end = var_3311_end_0, end_mask = var_3311_end_mask_0, x = transpose_10)[name = tensor("op_3311_cast_fp16")]; tensor var_3315_begin_0 = const()[name = tensor("op_3315_begin_0"), val = tensor([0, 0, 0, 160])]; tensor var_3315_end_0 = const()[name = tensor("op_3315_end_0"), val = tensor([2, 77, 1, 320])]; tensor var_3315_end_mask_0 = const()[name = tensor("op_3315_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_3315_cast_fp16 = slice_by_index(begin = var_3315_begin_0, end = var_3315_end_0, end_mask = var_3315_end_mask_0, x = transpose_10)[name = tensor("op_3315_cast_fp16")]; tensor var_3319_begin_0 = const()[name = tensor("op_3319_begin_0"), val = tensor([0, 0, 0, 320])]; tensor var_3319_end_0 = const()[name = tensor("op_3319_end_0"), val = tensor([2, 77, 1, 480])]; tensor var_3319_end_mask_0 = const()[name = tensor("op_3319_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_3319_cast_fp16 = slice_by_index(begin = var_3319_begin_0, end = var_3319_end_0, end_mask = var_3319_end_mask_0, x = transpose_10)[name = tensor("op_3319_cast_fp16")]; tensor var_3323_begin_0 = const()[name = tensor("op_3323_begin_0"), val = tensor([0, 0, 0, 480])]; tensor var_3323_end_0 = const()[name = tensor("op_3323_end_0"), val = tensor([2, 77, 1, 640])]; tensor var_3323_end_mask_0 = const()[name = tensor("op_3323_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_3323_cast_fp16 = slice_by_index(begin = var_3323_begin_0, end = var_3323_end_0, end_mask = var_3323_end_mask_0, x = transpose_10)[name = tensor("op_3323_cast_fp16")]; tensor var_3327_begin_0 = const()[name = tensor("op_3327_begin_0"), val = tensor([0, 0, 0, 640])]; tensor var_3327_end_0 = const()[name = tensor("op_3327_end_0"), val = tensor([2, 77, 1, 800])]; tensor var_3327_end_mask_0 = const()[name = tensor("op_3327_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_3327_cast_fp16 = slice_by_index(begin = var_3327_begin_0, end = var_3327_end_0, end_mask = var_3327_end_mask_0, x = transpose_10)[name = tensor("op_3327_cast_fp16")]; tensor var_3331_begin_0 = const()[name = tensor("op_3331_begin_0"), val = tensor([0, 0, 0, 800])]; tensor var_3331_end_0 = const()[name = tensor("op_3331_end_0"), val = tensor([2, 77, 1, 960])]; tensor var_3331_end_mask_0 = const()[name = tensor("op_3331_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_3331_cast_fp16 = slice_by_index(begin = var_3331_begin_0, end = var_3331_end_0, end_mask = var_3331_end_mask_0, x = transpose_10)[name = tensor("op_3331_cast_fp16")]; tensor var_3335_begin_0 = const()[name = tensor("op_3335_begin_0"), val = tensor([0, 0, 0, 960])]; tensor var_3335_end_0 = const()[name = tensor("op_3335_end_0"), val = tensor([2, 77, 1, 1120])]; tensor var_3335_end_mask_0 = const()[name = tensor("op_3335_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_3335_cast_fp16 = slice_by_index(begin = var_3335_begin_0, end = var_3335_end_0, end_mask = var_3335_end_mask_0, x = transpose_10)[name = tensor("op_3335_cast_fp16")]; tensor var_3339_begin_0 = const()[name = tensor("op_3339_begin_0"), val = tensor([0, 0, 0, 1120])]; tensor var_3339_end_0 = const()[name = tensor("op_3339_end_0"), val = tensor([2, 77, 1, 1280])]; tensor var_3339_end_mask_0 = const()[name = tensor("op_3339_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_3339_cast_fp16 = slice_by_index(begin = var_3339_begin_0, end = var_3339_end_0, end_mask = var_3339_end_mask_0, x = transpose_10)[name = tensor("op_3339_cast_fp16")]; tensor var_3341_begin_0 = const()[name = tensor("op_3341_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_3341_end_0 = const()[name = tensor("op_3341_end_0"), val = tensor([2, 160, 1, 77])]; tensor var_3341_end_mask_0 = const()[name = tensor("op_3341_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_3341_cast_fp16 = slice_by_index(begin = var_3341_begin_0, end = var_3341_end_0, end_mask = var_3341_end_mask_0, x = v_15_cast_fp16)[name = tensor("op_3341_cast_fp16")]; tensor var_3345_begin_0 = const()[name = tensor("op_3345_begin_0"), val = tensor([0, 160, 0, 0])]; tensor var_3345_end_0 = const()[name = tensor("op_3345_end_0"), val = tensor([2, 320, 1, 77])]; tensor var_3345_end_mask_0 = const()[name = tensor("op_3345_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_3345_cast_fp16 = slice_by_index(begin = var_3345_begin_0, end = var_3345_end_0, end_mask = var_3345_end_mask_0, x = v_15_cast_fp16)[name = tensor("op_3345_cast_fp16")]; tensor var_3349_begin_0 = const()[name = tensor("op_3349_begin_0"), val = tensor([0, 320, 0, 0])]; tensor var_3349_end_0 = const()[name = tensor("op_3349_end_0"), val = tensor([2, 480, 1, 77])]; tensor var_3349_end_mask_0 = const()[name = tensor("op_3349_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_3349_cast_fp16 = slice_by_index(begin = var_3349_begin_0, end = var_3349_end_0, end_mask = var_3349_end_mask_0, x = v_15_cast_fp16)[name = tensor("op_3349_cast_fp16")]; tensor var_3353_begin_0 = const()[name = tensor("op_3353_begin_0"), val = tensor([0, 480, 0, 0])]; tensor var_3353_end_0 = const()[name = tensor("op_3353_end_0"), val = tensor([2, 640, 1, 77])]; tensor var_3353_end_mask_0 = const()[name = tensor("op_3353_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_3353_cast_fp16 = slice_by_index(begin = var_3353_begin_0, end = var_3353_end_0, end_mask = var_3353_end_mask_0, x = v_15_cast_fp16)[name = tensor("op_3353_cast_fp16")]; tensor var_3357_begin_0 = const()[name = tensor("op_3357_begin_0"), val = tensor([0, 640, 0, 0])]; tensor var_3357_end_0 = const()[name = tensor("op_3357_end_0"), val = tensor([2, 800, 1, 77])]; tensor var_3357_end_mask_0 = const()[name = tensor("op_3357_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_3357_cast_fp16 = slice_by_index(begin = var_3357_begin_0, end = var_3357_end_0, end_mask = var_3357_end_mask_0, x = v_15_cast_fp16)[name = tensor("op_3357_cast_fp16")]; tensor var_3361_begin_0 = const()[name = tensor("op_3361_begin_0"), val = tensor([0, 800, 0, 0])]; tensor var_3361_end_0 = const()[name = tensor("op_3361_end_0"), val = tensor([2, 960, 1, 77])]; tensor var_3361_end_mask_0 = const()[name = tensor("op_3361_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_3361_cast_fp16 = slice_by_index(begin = var_3361_begin_0, end = var_3361_end_0, end_mask = var_3361_end_mask_0, x = v_15_cast_fp16)[name = tensor("op_3361_cast_fp16")]; tensor var_3365_begin_0 = const()[name = tensor("op_3365_begin_0"), val = tensor([0, 960, 0, 0])]; tensor var_3365_end_0 = const()[name = tensor("op_3365_end_0"), val = tensor([2, 1120, 1, 77])]; tensor var_3365_end_mask_0 = const()[name = tensor("op_3365_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_3365_cast_fp16 = slice_by_index(begin = var_3365_begin_0, end = var_3365_end_0, end_mask = var_3365_end_mask_0, x = v_15_cast_fp16)[name = tensor("op_3365_cast_fp16")]; tensor var_3369_begin_0 = const()[name = tensor("op_3369_begin_0"), val = tensor([0, 1120, 0, 0])]; tensor var_3369_end_0 = const()[name = tensor("op_3369_end_0"), val = tensor([2, 1280, 1, 77])]; tensor var_3369_end_mask_0 = const()[name = tensor("op_3369_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_3369_cast_fp16 = slice_by_index(begin = var_3369_begin_0, end = var_3369_end_0, end_mask = var_3369_end_mask_0, x = v_15_cast_fp16)[name = tensor("op_3369_cast_fp16")]; tensor var_3373_equation_0 = const()[name = tensor("op_3373_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_3373_cast_fp16 = einsum(equation = var_3373_equation_0, values = (var_3311_cast_fp16, var_3276_cast_fp16))[name = tensor("op_3373_cast_fp16")]; tensor var_3374_to_fp16 = const()[name = tensor("op_3374_to_fp16"), val = tensor(0x1.43cp-4)]; tensor aw_49_cast_fp16 = mul(x = var_3373_cast_fp16, y = var_3374_to_fp16)[name = tensor("aw_49_cast_fp16")]; tensor var_3377_equation_0 = const()[name = tensor("op_3377_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_3377_cast_fp16 = einsum(equation = var_3377_equation_0, values = (var_3315_cast_fp16, var_3280_cast_fp16))[name = tensor("op_3377_cast_fp16")]; tensor var_3378_to_fp16 = const()[name = tensor("op_3378_to_fp16"), val = tensor(0x1.43cp-4)]; tensor aw_51_cast_fp16 = mul(x = var_3377_cast_fp16, y = var_3378_to_fp16)[name = tensor("aw_51_cast_fp16")]; tensor var_3381_equation_0 = const()[name = tensor("op_3381_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_3381_cast_fp16 = einsum(equation = var_3381_equation_0, values = (var_3319_cast_fp16, var_3284_cast_fp16))[name = tensor("op_3381_cast_fp16")]; tensor var_3382_to_fp16 = const()[name = tensor("op_3382_to_fp16"), val = tensor(0x1.43cp-4)]; tensor aw_53_cast_fp16 = mul(x = var_3381_cast_fp16, y = var_3382_to_fp16)[name = tensor("aw_53_cast_fp16")]; tensor var_3385_equation_0 = const()[name = tensor("op_3385_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_3385_cast_fp16 = einsum(equation = var_3385_equation_0, values = (var_3323_cast_fp16, var_3288_cast_fp16))[name = tensor("op_3385_cast_fp16")]; tensor var_3386_to_fp16 = const()[name = tensor("op_3386_to_fp16"), val = tensor(0x1.43cp-4)]; tensor aw_55_cast_fp16 = mul(x = var_3385_cast_fp16, y = var_3386_to_fp16)[name = tensor("aw_55_cast_fp16")]; tensor var_3389_equation_0 = const()[name = tensor("op_3389_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_3389_cast_fp16 = einsum(equation = var_3389_equation_0, values = (var_3327_cast_fp16, var_3292_cast_fp16))[name = tensor("op_3389_cast_fp16")]; tensor var_3390_to_fp16 = const()[name = tensor("op_3390_to_fp16"), val = tensor(0x1.43cp-4)]; tensor aw_57_cast_fp16 = mul(x = var_3389_cast_fp16, y = var_3390_to_fp16)[name = tensor("aw_57_cast_fp16")]; tensor var_3393_equation_0 = const()[name = tensor("op_3393_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_3393_cast_fp16 = einsum(equation = var_3393_equation_0, values = (var_3331_cast_fp16, var_3296_cast_fp16))[name = tensor("op_3393_cast_fp16")]; tensor var_3394_to_fp16 = const()[name = tensor("op_3394_to_fp16"), val = tensor(0x1.43cp-4)]; tensor aw_59_cast_fp16 = mul(x = var_3393_cast_fp16, y = var_3394_to_fp16)[name = tensor("aw_59_cast_fp16")]; tensor var_3397_equation_0 = const()[name = tensor("op_3397_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_3397_cast_fp16 = einsum(equation = var_3397_equation_0, values = (var_3335_cast_fp16, var_3300_cast_fp16))[name = tensor("op_3397_cast_fp16")]; tensor var_3398_to_fp16 = const()[name = tensor("op_3398_to_fp16"), val = tensor(0x1.43cp-4)]; tensor aw_61_cast_fp16 = mul(x = var_3397_cast_fp16, y = var_3398_to_fp16)[name = tensor("aw_61_cast_fp16")]; tensor var_3401_equation_0 = const()[name = tensor("op_3401_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_3401_cast_fp16 = einsum(equation = var_3401_equation_0, values = (var_3339_cast_fp16, var_3304_cast_fp16))[name = tensor("op_3401_cast_fp16")]; tensor var_3402_to_fp16 = const()[name = tensor("op_3402_to_fp16"), val = tensor(0x1.43cp-4)]; tensor aw_63_cast_fp16 = mul(x = var_3401_cast_fp16, y = var_3402_to_fp16)[name = tensor("aw_63_cast_fp16")]; tensor var_3404_cast_fp16 = softmax(axis = var_2942, x = aw_49_cast_fp16)[name = tensor("op_3404_cast_fp16")]; tensor var_3405_cast_fp16 = softmax(axis = var_2942, x = aw_51_cast_fp16)[name = tensor("op_3405_cast_fp16")]; tensor var_3406_cast_fp16 = softmax(axis = var_2942, x = aw_53_cast_fp16)[name = tensor("op_3406_cast_fp16")]; tensor var_3407_cast_fp16 = softmax(axis = var_2942, x = aw_55_cast_fp16)[name = tensor("op_3407_cast_fp16")]; tensor var_3408_cast_fp16 = softmax(axis = var_2942, x = aw_57_cast_fp16)[name = tensor("op_3408_cast_fp16")]; tensor var_3409_cast_fp16 = softmax(axis = var_2942, x = aw_59_cast_fp16)[name = tensor("op_3409_cast_fp16")]; tensor var_3410_cast_fp16 = softmax(axis = var_2942, x = aw_61_cast_fp16)[name = tensor("op_3410_cast_fp16")]; tensor var_3411_cast_fp16 = softmax(axis = var_2942, x = aw_63_cast_fp16)[name = tensor("op_3411_cast_fp16")]; tensor var_3413_equation_0 = const()[name = tensor("op_3413_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_3413_cast_fp16 = einsum(equation = var_3413_equation_0, values = (var_3341_cast_fp16, var_3404_cast_fp16))[name = tensor("op_3413_cast_fp16")]; tensor var_3415_equation_0 = const()[name = tensor("op_3415_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_3415_cast_fp16 = einsum(equation = var_3415_equation_0, values = (var_3345_cast_fp16, var_3405_cast_fp16))[name = tensor("op_3415_cast_fp16")]; tensor var_3417_equation_0 = const()[name = tensor("op_3417_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_3417_cast_fp16 = einsum(equation = var_3417_equation_0, values = (var_3349_cast_fp16, var_3406_cast_fp16))[name = tensor("op_3417_cast_fp16")]; tensor var_3419_equation_0 = const()[name = tensor("op_3419_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_3419_cast_fp16 = einsum(equation = var_3419_equation_0, values = (var_3353_cast_fp16, var_3407_cast_fp16))[name = tensor("op_3419_cast_fp16")]; tensor var_3421_equation_0 = const()[name = tensor("op_3421_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_3421_cast_fp16 = einsum(equation = var_3421_equation_0, values = (var_3357_cast_fp16, var_3408_cast_fp16))[name = tensor("op_3421_cast_fp16")]; tensor var_3423_equation_0 = const()[name = tensor("op_3423_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_3423_cast_fp16 = einsum(equation = var_3423_equation_0, values = (var_3361_cast_fp16, var_3409_cast_fp16))[name = tensor("op_3423_cast_fp16")]; tensor var_3425_equation_0 = const()[name = tensor("op_3425_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_3425_cast_fp16 = einsum(equation = var_3425_equation_0, values = (var_3365_cast_fp16, var_3410_cast_fp16))[name = tensor("op_3425_cast_fp16")]; tensor var_3427_equation_0 = const()[name = tensor("op_3427_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_3427_cast_fp16 = einsum(equation = var_3427_equation_0, values = (var_3369_cast_fp16, var_3411_cast_fp16))[name = tensor("op_3427_cast_fp16")]; tensor input_109_interleave_0 = const()[name = tensor("input_109_interleave_0"), val = tensor(false)]; tensor input_109_cast_fp16 = concat(axis = var_2942, interleave = input_109_interleave_0, values = (var_3413_cast_fp16, var_3415_cast_fp16, var_3417_cast_fp16, var_3419_cast_fp16, var_3421_cast_fp16, var_3423_cast_fp16, var_3425_cast_fp16, var_3427_cast_fp16))[name = tensor("input_109_cast_fp16")]; tensor var_3433 = const()[name = tensor("op_3433"), val = tensor([1, 1])]; tensor var_3435 = const()[name = tensor("op_3435"), val = tensor([1, 1])]; tensor var_3437_pad_type_0 = const()[name = tensor("op_3437_pad_type_0"), val = tensor("custom")]; tensor var_3437_pad_0 = const()[name = tensor("op_3437_pad_0"), val = tensor([0, 0, 0, 0])]; tensor up_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(110772672))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(112001536))), name = tensor("up_blocks_0_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; tensor up_blocks_0_attentions_0_transformer_blocks_0_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("up_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(112001728)))]; tensor var_3437_cast_fp16 = conv(bias = up_blocks_0_attentions_0_transformer_blocks_0_attn2_to_out_0_bias_to_fp16, dilations = var_3435, groups = var_2942, pad = var_3437_pad_0, pad_type = var_3437_pad_type_0, strides = var_3433, weight = up_blocks_0_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16_palettized, x = input_109_cast_fp16)[name = tensor("op_3437_cast_fp16")]; tensor inputs_23_cast_fp16 = add(x = var_3437_cast_fp16, y = inputs_21_cast_fp16)[name = tensor("inputs_23_cast_fp16")]; tensor var_3441 = const()[name = tensor("op_3441"), val = tensor([1])]; tensor channels_mean_23_cast_fp16 = reduce_mean(axes = var_3441, keep_dims = var_2937, x = inputs_23_cast_fp16)[name = tensor("channels_mean_23_cast_fp16")]; tensor zero_mean_23_cast_fp16 = sub(x = inputs_23_cast_fp16, y = channels_mean_23_cast_fp16)[name = tensor("zero_mean_23_cast_fp16")]; tensor zero_mean_sq_23_cast_fp16 = mul(x = zero_mean_23_cast_fp16, y = zero_mean_23_cast_fp16)[name = tensor("zero_mean_sq_23_cast_fp16")]; tensor var_3445 = const()[name = tensor("op_3445"), val = tensor([1])]; tensor var_3446_cast_fp16 = reduce_mean(axes = var_3445, keep_dims = var_2937, x = zero_mean_sq_23_cast_fp16)[name = tensor("op_3446_cast_fp16")]; tensor var_3447_to_fp16 = const()[name = tensor("op_3447_to_fp16"), val = tensor(0x1.5p-17)]; tensor var_3448_cast_fp16 = add(x = var_3446_cast_fp16, y = var_3447_to_fp16)[name = tensor("op_3448_cast_fp16")]; tensor denom_23_epsilon_0_to_fp16 = const()[name = tensor("denom_23_epsilon_0_to_fp16"), val = tensor(0x1p-24)]; tensor denom_23_cast_fp16 = rsqrt(epsilon = denom_23_epsilon_0_to_fp16, x = var_3448_cast_fp16)[name = tensor("denom_23_cast_fp16")]; tensor out_23_cast_fp16 = mul(x = zero_mean_23_cast_fp16, y = denom_23_cast_fp16)[name = tensor("out_23_cast_fp16")]; tensor var_3452_to_fp16 = const()[name = tensor("op_3452_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(112004352)))]; tensor var_3453_cast_fp16 = add(x = out_23_cast_fp16, y = var_3452_to_fp16)[name = tensor("op_3453_cast_fp16")]; tensor var_3455_to_fp16 = const()[name = tensor("op_3455_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(112006976)))]; tensor input_111_cast_fp16 = mul(x = var_3453_cast_fp16, y = var_3455_to_fp16)[name = tensor("input_111_cast_fp16")]; tensor var_3463 = const()[name = tensor("op_3463"), val = tensor([1, 1])]; tensor var_3465 = const()[name = tensor("op_3465"), val = tensor([1, 1])]; tensor var_3467_pad_type_0 = const()[name = tensor("op_3467_pad_type_0"), val = tensor("custom")]; tensor var_3467_pad_0 = const()[name = tensor("op_3467_pad_0"), val = tensor([0, 0, 0, 0])]; tensor up_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(112009600))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(121840064))), name = tensor("up_blocks_0_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16_palettized"), shape = tensor([10240, 1280, 1, 1])]; tensor up_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(121840256))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(121848000))), name = tensor("up_blocks_0_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16_palettized"), shape = tensor([10240])]; tensor var_3467_cast_fp16 = conv(bias = up_blocks_0_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16_palettized, dilations = var_3465, groups = var_2942, pad = var_3467_pad_0, pad_type = var_3467_pad_type_0, strides = var_3463, weight = up_blocks_0_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16_palettized, x = input_111_cast_fp16)[name = tensor("op_3467_cast_fp16")]; tensor var_3468_split_sizes_0 = const()[name = tensor("op_3468_split_sizes_0"), val = tensor([5120, 5120])]; tensor var_3468_axis_0 = const()[name = tensor("op_3468_axis_0"), val = tensor(1)]; tensor var_3468_cast_fp16_0, tensor var_3468_cast_fp16_1 = split(axis = var_3468_axis_0, split_sizes = var_3468_split_sizes_0, x = var_3467_cast_fp16)[name = tensor("op_3468_cast_fp16")]; tensor var_3470_mode_0 = const()[name = tensor("op_3470_mode_0"), val = tensor("EXACT")]; tensor var_3470_cast_fp16 = gelu(mode = var_3470_mode_0, x = var_3468_cast_fp16_1)[name = tensor("op_3470_cast_fp16")]; tensor input_113_cast_fp16 = mul(x = var_3468_cast_fp16_0, y = var_3470_cast_fp16)[name = tensor("input_113_cast_fp16")]; tensor var_3474 = const()[name = tensor("op_3474"), val = tensor([1, 1])]; tensor var_3476 = const()[name = tensor("op_3476"), val = tensor([1, 1])]; tensor var_3478_pad_type_0 = const()[name = tensor("op_3478_pad_type_0"), val = tensor("custom")]; tensor var_3478_pad_0 = const()[name = tensor("op_3478_pad_0"), val = tensor([0, 0, 0, 0])]; tensor up_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(121848192))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(126763456))), name = tensor("up_blocks_0_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16_palettized"), shape = tensor([1280, 5120, 1, 1])]; tensor up_blocks_0_attentions_0_transformer_blocks_0_ff_net_2_bias_to_fp16 = const()[name = tensor("up_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(126763648)))]; tensor var_3478_cast_fp16 = conv(bias = up_blocks_0_attentions_0_transformer_blocks_0_ff_net_2_bias_to_fp16, dilations = var_3476, groups = var_2942, pad = var_3478_pad_0, pad_type = var_3478_pad_type_0, strides = var_3474, weight = up_blocks_0_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16_palettized, x = input_113_cast_fp16)[name = tensor("op_3478_cast_fp16")]; tensor hidden_states_71_cast_fp16 = add(x = var_3478_cast_fp16, y = inputs_23_cast_fp16)[name = tensor("hidden_states_71_cast_fp16")]; tensor var_3480 = const()[name = tensor("op_3480"), val = tensor([2, 1280, 16, 16])]; tensor input_115_cast_fp16 = reshape(shape = var_3480, x = hidden_states_71_cast_fp16)[name = tensor("input_115_cast_fp16")]; tensor var_3484 = const()[name = tensor("op_3484"), val = tensor([1, 1])]; tensor var_3486 = const()[name = tensor("op_3486"), 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([0, 0, 0, 0])]; tensor up_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(126766272))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(127995136))), name = tensor("up_blocks_0_attentions_0_proj_out_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; tensor up_blocks_0_attentions_0_proj_out_bias_to_fp16 = const()[name = tensor("up_blocks_0_attentions_0_proj_out_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(127995328)))]; tensor hidden_states_73_cast_fp16 = conv(bias = up_blocks_0_attentions_0_proj_out_bias_to_fp16, dilations = var_3486, groups = var_2942, pad = hidden_states_73_pad_0, pad_type = hidden_states_73_pad_type_0, strides = var_3484, weight = up_blocks_0_attentions_0_proj_out_weight_to_fp16_palettized, x = input_115_cast_fp16)[name = tensor("hidden_states_73_cast_fp16")]; tensor hidden_states_75_cast_fp16 = add(x = hidden_states_73_cast_fp16, y = hidden_states_61_cast_fp16)[name = tensor("hidden_states_75_cast_fp16")]; tensor input_117_interleave_0 = const()[name = tensor("input_117_interleave_0"), val = tensor(false)]; tensor input_117_cast_fp16 = concat(axis = var_2942, interleave = input_117_interleave_0, values = (hidden_states_75_cast_fp16, input_65_cast_fp16))[name = tensor("input_117_cast_fp16")]; tensor reshape_48_shape_0 = const()[name = tensor("reshape_48_shape_0"), val = tensor([2, 32, 60, 16, 16])]; tensor reshape_48_cast_fp16 = reshape(shape = reshape_48_shape_0, x = input_117_cast_fp16)[name = tensor("reshape_48_cast_fp16")]; 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_fp16 = reduce_mean(axes = reduce_mean_36_axes_0, keep_dims = reduce_mean_36_keep_dims_0, x = reshape_48_cast_fp16)[name = tensor("reduce_mean_36_cast_fp16")]; tensor sub_24_cast_fp16 = sub(x = reshape_48_cast_fp16, y = reduce_mean_36_cast_fp16)[name = tensor("sub_24_cast_fp16")]; tensor square_12_cast_fp16 = square(x = sub_24_cast_fp16)[name = tensor("square_12_cast_fp16")]; 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_fp16 = reduce_mean(axes = reduce_mean_38_axes_0, keep_dims = reduce_mean_38_keep_dims_0, x = square_12_cast_fp16)[name = tensor("reduce_mean_38_cast_fp16")]; 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_fp16 = add(x = reduce_mean_38_cast_fp16, y = add_24_y_0_to_fp16)[name = tensor("add_24_cast_fp16")]; tensor sqrt_12_cast_fp16 = sqrt(x = add_24_cast_fp16)[name = tensor("sqrt_12_cast_fp16")]; tensor real_div_12_cast_fp16 = real_div(x = sub_24_cast_fp16, y = sqrt_12_cast_fp16)[name = tensor("real_div_12_cast_fp16")]; tensor reshape_49_shape_0 = const()[name = tensor("reshape_49_shape_0"), val = tensor([2, 1920, 16, 16])]; tensor reshape_49_cast_fp16 = reshape(shape = reshape_49_shape_0, x = real_div_12_cast_fp16)[name = tensor("reshape_49_cast_fp16")]; tensor add_25_mean_0_to_fp16 = const()[name = tensor("add_25_mean_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(127997952)))]; tensor add_25_variance_0_to_fp16 = const()[name = tensor("add_25_variance_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(128001856)))]; 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(128005760)))]; 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(128009664)))]; 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_fp16 = 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_25_mean_0_to_fp16, variance = add_25_variance_0_to_fp16, x = reshape_49_cast_fp16)[name = tensor("add_25_cast_fp16")]; tensor input_121_cast_fp16 = silu(x = add_25_cast_fp16)[name = tensor("input_121_cast_fp16")]; tensor var_3504 = const()[name = tensor("op_3504"), val = tensor([1, 1])]; tensor var_3506 = const()[name = tensor("op_3506"), val = tensor([1, 1])]; tensor hidden_states_77_pad_type_0 = const()[name = tensor("hidden_states_77_pad_type_0"), val = tensor("custom")]; tensor hidden_states_77_pad_0 = const()[name = tensor("hidden_states_77_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(128013568))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(144602432))), name = tensor("up_blocks_0_resnets_1_conv1_weight_to_fp16_palettized"), shape = tensor([1280, 1920, 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(144602624)))]; tensor hidden_states_77_cast_fp16 = conv(bias = up_blocks_0_resnets_1_conv1_bias_to_fp16, dilations = var_3506, groups = var_2942, pad = hidden_states_77_pad_0, pad_type = hidden_states_77_pad_type_0, strides = var_3504, weight = up_blocks_0_resnets_1_conv1_weight_to_fp16_palettized, x = input_121_cast_fp16)[name = tensor("hidden_states_77_cast_fp16")]; tensor var_3512 = const()[name = tensor("op_3512"), val = tensor([1, 1])]; tensor var_3514 = const()[name = tensor("op_3514"), 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 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(144605248))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(145834112))), 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(145834304)))]; tensor temb_9_cast_fp16 = conv(bias = up_blocks_0_resnets_1_time_emb_proj_bias_to_fp16, dilations = var_3514, groups = var_2942, pad = temb_9_pad_0, pad_type = temb_9_pad_type_0, strides = var_3512, weight = up_blocks_0_resnets_1_time_emb_proj_weight_to_fp16_palettized, x = input_15_cast_fp16)[name = tensor("temb_9_cast_fp16")]; tensor input_125_cast_fp16 = add(x = hidden_states_77_cast_fp16, y = temb_9_cast_fp16)[name = tensor("input_125_cast_fp16")]; tensor reshape_52_shape_0 = const()[name = tensor("reshape_52_shape_0"), val = tensor([2, 32, 40, 16, 16])]; tensor reshape_52_cast_fp16 = reshape(shape = reshape_52_shape_0, x = input_125_cast_fp16)[name = tensor("reshape_52_cast_fp16")]; 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_fp16 = reduce_mean(axes = reduce_mean_39_axes_0, keep_dims = reduce_mean_39_keep_dims_0, x = reshape_52_cast_fp16)[name = tensor("reduce_mean_39_cast_fp16")]; tensor sub_26_cast_fp16 = sub(x = reshape_52_cast_fp16, y = reduce_mean_39_cast_fp16)[name = tensor("sub_26_cast_fp16")]; tensor square_13_cast_fp16 = square(x = sub_26_cast_fp16)[name = tensor("square_13_cast_fp16")]; 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_fp16 = reduce_mean(axes = reduce_mean_41_axes_0, keep_dims = reduce_mean_41_keep_dims_0, x = square_13_cast_fp16)[name = tensor("reduce_mean_41_cast_fp16")]; 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_fp16 = add(x = reduce_mean_41_cast_fp16, y = add_26_y_0_to_fp16)[name = tensor("add_26_cast_fp16")]; tensor sqrt_13_cast_fp16 = sqrt(x = add_26_cast_fp16)[name = tensor("sqrt_13_cast_fp16")]; tensor real_div_13_cast_fp16 = real_div(x = sub_26_cast_fp16, y = sqrt_13_cast_fp16)[name = tensor("real_div_13_cast_fp16")]; tensor reshape_53_shape_0 = const()[name = tensor("reshape_53_shape_0"), val = tensor([2, 1280, 16, 16])]; tensor reshape_53_cast_fp16 = reshape(shape = reshape_53_shape_0, x = real_div_13_cast_fp16)[name = tensor("reshape_53_cast_fp16")]; 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(145836928)))]; 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(145839552)))]; 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_fp16 = 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_15_mean_0_to_fp16, variance = add_15_variance_0_to_fp16, x = reshape_53_cast_fp16)[name = tensor("add_27_cast_fp16")]; tensor input_129_cast_fp16 = silu(x = add_27_cast_fp16)[name = tensor("input_129_cast_fp16")]; tensor var_3524 = const()[name = tensor("op_3524"), val = tensor([1, 1])]; tensor var_3526 = const()[name = tensor("op_3526"), 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([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(145842176))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(156901440))), 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(156901632)))]; tensor hidden_states_79_cast_fp16 = conv(bias = up_blocks_0_resnets_1_conv2_bias_to_fp16, dilations = var_3526, groups = var_2942, pad = hidden_states_79_pad_0, pad_type = hidden_states_79_pad_type_0, strides = var_3524, weight = up_blocks_0_resnets_1_conv2_weight_to_fp16_palettized, x = input_129_cast_fp16)[name = tensor("hidden_states_79_cast_fp16")]; tensor var_3531 = const()[name = tensor("op_3531"), val = tensor([1, 1])]; tensor var_3533 = const()[name = tensor("op_3533"), 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(156904256))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(158747520))), name = tensor("up_blocks_0_resnets_1_conv_shortcut_weight_to_fp16_palettized"), shape = tensor([1280, 1920, 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(158747712)))]; tensor x_7_cast_fp16 = conv(bias = up_blocks_0_resnets_1_conv_shortcut_bias_to_fp16, dilations = var_3533, groups = var_2942, pad = x_7_pad_0, pad_type = x_7_pad_type_0, strides = var_3531, weight = up_blocks_0_resnets_1_conv_shortcut_weight_to_fp16_palettized, x = input_117_cast_fp16)[name = tensor("x_7_cast_fp16")]; tensor hidden_states_81_cast_fp16 = add(x = x_7_cast_fp16, y = hidden_states_79_cast_fp16)[name = tensor("hidden_states_81_cast_fp16")]; tensor reshape_56_shape_0 = const()[name = tensor("reshape_56_shape_0"), val = tensor([2, 32, 40, 16, 16])]; tensor reshape_56_cast_fp16 = reshape(shape = reshape_56_shape_0, x = hidden_states_81_cast_fp16)[name = tensor("reshape_56_cast_fp16")]; 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_fp16 = reduce_mean(axes = reduce_mean_42_axes_0, keep_dims = reduce_mean_42_keep_dims_0, x = reshape_56_cast_fp16)[name = tensor("reduce_mean_42_cast_fp16")]; tensor sub_28_cast_fp16 = sub(x = reshape_56_cast_fp16, y = reduce_mean_42_cast_fp16)[name = tensor("sub_28_cast_fp16")]; tensor square_14_cast_fp16 = square(x = sub_28_cast_fp16)[name = tensor("square_14_cast_fp16")]; 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_fp16 = reduce_mean(axes = reduce_mean_44_axes_0, keep_dims = reduce_mean_44_keep_dims_0, x = square_14_cast_fp16)[name = tensor("reduce_mean_44_cast_fp16")]; 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_fp16 = add(x = reduce_mean_44_cast_fp16, y = add_28_y_0_to_fp16)[name = tensor("add_28_cast_fp16")]; tensor sqrt_14_cast_fp16 = sqrt(x = add_28_cast_fp16)[name = tensor("sqrt_14_cast_fp16")]; tensor real_div_14_cast_fp16 = real_div(x = sub_28_cast_fp16, y = sqrt_14_cast_fp16)[name = tensor("real_div_14_cast_fp16")]; tensor reshape_57_shape_0 = const()[name = tensor("reshape_57_shape_0"), val = tensor([2, 1280, 16, 16])]; tensor reshape_57_cast_fp16 = reshape(shape = reshape_57_shape_0, x = real_div_14_cast_fp16)[name = tensor("reshape_57_cast_fp16")]; 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(158750336)))]; 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(158752960)))]; 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_fp16 = 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_15_mean_0_to_fp16, variance = add_15_variance_0_to_fp16, x = reshape_57_cast_fp16)[name = tensor("add_29_cast_fp16")]; tensor var_3553 = const()[name = tensor("op_3553"), val = tensor([1, 1])]; tensor var_3555 = const()[name = tensor("op_3555"), val = tensor([1, 1])]; tensor hidden_states_83_pad_type_0 = const()[name = tensor("hidden_states_83_pad_type_0"), val = tensor("custom")]; tensor hidden_states_83_pad_0 = const()[name = tensor("hidden_states_83_pad_0"), val = tensor([0, 0, 0, 0])]; tensor up_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(158755584))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(159984448))), name = tensor("up_blocks_0_attentions_1_proj_in_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; tensor up_blocks_0_attentions_1_proj_in_bias_to_fp16 = const()[name = tensor("up_blocks_0_attentions_1_proj_in_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(159984640)))]; tensor hidden_states_83_cast_fp16 = conv(bias = up_blocks_0_attentions_1_proj_in_bias_to_fp16, dilations = var_3555, groups = var_2942, pad = hidden_states_83_pad_0, pad_type = hidden_states_83_pad_type_0, strides = var_3553, weight = up_blocks_0_attentions_1_proj_in_weight_to_fp16_palettized, x = add_29_cast_fp16)[name = tensor("hidden_states_83_cast_fp16")]; tensor var_3560 = const()[name = tensor("op_3560"), val = tensor([2, 1280, 1, 256])]; tensor inputs_25_cast_fp16 = reshape(shape = var_3560, x = hidden_states_83_cast_fp16)[name = tensor("inputs_25_cast_fp16")]; tensor var_3570 = const()[name = tensor("op_3570"), val = tensor([1])]; tensor channels_mean_25_cast_fp16 = reduce_mean(axes = var_3570, keep_dims = var_2937, x = inputs_25_cast_fp16)[name = tensor("channels_mean_25_cast_fp16")]; tensor zero_mean_25_cast_fp16 = sub(x = inputs_25_cast_fp16, y = channels_mean_25_cast_fp16)[name = tensor("zero_mean_25_cast_fp16")]; tensor zero_mean_sq_25_cast_fp16 = mul(x = zero_mean_25_cast_fp16, y = zero_mean_25_cast_fp16)[name = tensor("zero_mean_sq_25_cast_fp16")]; tensor var_3574 = const()[name = tensor("op_3574"), val = tensor([1])]; tensor var_3575_cast_fp16 = reduce_mean(axes = var_3574, keep_dims = var_2937, x = zero_mean_sq_25_cast_fp16)[name = tensor("op_3575_cast_fp16")]; tensor var_3576_to_fp16 = const()[name = tensor("op_3576_to_fp16"), val = tensor(0x1.5p-17)]; tensor var_3577_cast_fp16 = add(x = var_3575_cast_fp16, y = var_3576_to_fp16)[name = tensor("op_3577_cast_fp16")]; tensor denom_25_epsilon_0_to_fp16 = const()[name = tensor("denom_25_epsilon_0_to_fp16"), val = tensor(0x1p-24)]; tensor denom_25_cast_fp16 = rsqrt(epsilon = denom_25_epsilon_0_to_fp16, x = var_3577_cast_fp16)[name = tensor("denom_25_cast_fp16")]; tensor out_25_cast_fp16 = mul(x = zero_mean_25_cast_fp16, y = denom_25_cast_fp16)[name = tensor("out_25_cast_fp16")]; tensor var_3581_to_fp16 = const()[name = tensor("op_3581_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(159987264)))]; tensor var_3582_cast_fp16 = add(x = out_25_cast_fp16, y = var_3581_to_fp16)[name = tensor("op_3582_cast_fp16")]; tensor var_3584_to_fp16 = const()[name = tensor("op_3584_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(159989888)))]; tensor hidden_states_85_cast_fp16 = mul(x = var_3582_cast_fp16, y = var_3584_to_fp16)[name = tensor("hidden_states_85_cast_fp16")]; tensor var_3591 = const()[name = tensor("op_3591"), val = tensor([1, 1])]; tensor var_3593 = const()[name = tensor("op_3593"), 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 up_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(159992512))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(161221376))), name = tensor("up_blocks_0_attentions_1_transformer_blocks_0_attn1_to_q_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; tensor q_17_cast_fp16 = conv(dilations = var_3593, groups = var_2942, pad = q_17_pad_0, pad_type = q_17_pad_type_0, strides = var_3591, weight = up_blocks_0_attentions_1_transformer_blocks_0_attn1_to_q_weight_to_fp16_palettized, x = hidden_states_85_cast_fp16)[name = tensor("q_17_cast_fp16")]; tensor var_3597 = const()[name = tensor("op_3597"), val = tensor([1, 1])]; tensor var_3599 = const()[name = tensor("op_3599"), 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 up_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(161221568))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(162450432))), name = tensor("up_blocks_0_attentions_1_transformer_blocks_0_attn1_to_k_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; tensor k_33_cast_fp16 = conv(dilations = var_3599, groups = var_2942, pad = k_33_pad_0, pad_type = k_33_pad_type_0, strides = var_3597, weight = up_blocks_0_attentions_1_transformer_blocks_0_attn1_to_k_weight_to_fp16_palettized, x = hidden_states_85_cast_fp16)[name = tensor("k_33_cast_fp16")]; tensor var_3603 = const()[name = tensor("op_3603"), val = tensor([1, 1])]; tensor var_3605 = const()[name = tensor("op_3605"), 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 up_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(162450624))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(163679488))), name = tensor("up_blocks_0_attentions_1_transformer_blocks_0_attn1_to_v_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; tensor v_17_cast_fp16 = conv(dilations = var_3605, groups = var_2942, pad = v_17_pad_0, pad_type = v_17_pad_type_0, strides = var_3603, weight = up_blocks_0_attentions_1_transformer_blocks_0_attn1_to_v_weight_to_fp16_palettized, x = hidden_states_85_cast_fp16)[name = tensor("v_17_cast_fp16")]; tensor var_3609_begin_0 = const()[name = tensor("op_3609_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_3609_end_0 = const()[name = tensor("op_3609_end_0"), val = tensor([2, 160, 1, 256])]; tensor var_3609_end_mask_0 = const()[name = tensor("op_3609_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_3609_cast_fp16 = slice_by_index(begin = var_3609_begin_0, end = var_3609_end_0, end_mask = var_3609_end_mask_0, x = q_17_cast_fp16)[name = tensor("op_3609_cast_fp16")]; tensor var_3613_begin_0 = const()[name = tensor("op_3613_begin_0"), val = tensor([0, 160, 0, 0])]; tensor var_3613_end_0 = const()[name = tensor("op_3613_end_0"), val = tensor([2, 320, 1, 256])]; tensor var_3613_end_mask_0 = const()[name = tensor("op_3613_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_3613_cast_fp16 = slice_by_index(begin = var_3613_begin_0, end = var_3613_end_0, end_mask = var_3613_end_mask_0, x = q_17_cast_fp16)[name = tensor("op_3613_cast_fp16")]; tensor var_3617_begin_0 = const()[name = tensor("op_3617_begin_0"), val = tensor([0, 320, 0, 0])]; tensor var_3617_end_0 = const()[name = tensor("op_3617_end_0"), val = tensor([2, 480, 1, 256])]; tensor var_3617_end_mask_0 = const()[name = tensor("op_3617_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_3617_cast_fp16 = slice_by_index(begin = var_3617_begin_0, end = var_3617_end_0, end_mask = var_3617_end_mask_0, x = q_17_cast_fp16)[name = tensor("op_3617_cast_fp16")]; tensor var_3621_begin_0 = const()[name = tensor("op_3621_begin_0"), val = tensor([0, 480, 0, 0])]; tensor var_3621_end_0 = const()[name = tensor("op_3621_end_0"), val = tensor([2, 640, 1, 256])]; tensor var_3621_end_mask_0 = const()[name = tensor("op_3621_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_3621_cast_fp16 = slice_by_index(begin = var_3621_begin_0, end = var_3621_end_0, end_mask = var_3621_end_mask_0, x = q_17_cast_fp16)[name = tensor("op_3621_cast_fp16")]; tensor var_3625_begin_0 = const()[name = tensor("op_3625_begin_0"), val = tensor([0, 640, 0, 0])]; tensor var_3625_end_0 = const()[name = tensor("op_3625_end_0"), val = tensor([2, 800, 1, 256])]; tensor var_3625_end_mask_0 = const()[name = tensor("op_3625_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_3625_cast_fp16 = slice_by_index(begin = var_3625_begin_0, end = var_3625_end_0, end_mask = var_3625_end_mask_0, x = q_17_cast_fp16)[name = tensor("op_3625_cast_fp16")]; tensor var_3629_begin_0 = const()[name = tensor("op_3629_begin_0"), val = tensor([0, 800, 0, 0])]; tensor var_3629_end_0 = const()[name = tensor("op_3629_end_0"), val = tensor([2, 960, 1, 256])]; tensor var_3629_end_mask_0 = const()[name = tensor("op_3629_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_3629_cast_fp16 = slice_by_index(begin = var_3629_begin_0, end = var_3629_end_0, end_mask = var_3629_end_mask_0, x = q_17_cast_fp16)[name = tensor("op_3629_cast_fp16")]; tensor var_3633_begin_0 = const()[name = tensor("op_3633_begin_0"), val = tensor([0, 960, 0, 0])]; tensor var_3633_end_0 = const()[name = tensor("op_3633_end_0"), val = tensor([2, 1120, 1, 256])]; tensor var_3633_end_mask_0 = const()[name = tensor("op_3633_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_3633_cast_fp16 = slice_by_index(begin = var_3633_begin_0, end = var_3633_end_0, end_mask = var_3633_end_mask_0, x = q_17_cast_fp16)[name = tensor("op_3633_cast_fp16")]; tensor var_3637_begin_0 = const()[name = tensor("op_3637_begin_0"), val = tensor([0, 1120, 0, 0])]; tensor var_3637_end_0 = const()[name = tensor("op_3637_end_0"), val = tensor([2, 1280, 1, 256])]; tensor var_3637_end_mask_0 = const()[name = tensor("op_3637_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_3637_cast_fp16 = slice_by_index(begin = var_3637_begin_0, end = var_3637_end_0, end_mask = var_3637_end_mask_0, x = q_17_cast_fp16)[name = tensor("op_3637_cast_fp16")]; tensor k_35_perm_0 = const()[name = tensor("k_35_perm_0"), val = tensor([0, 3, 2, 1])]; tensor var_3644_begin_0 = const()[name = tensor("op_3644_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_3644_end_0 = const()[name = tensor("op_3644_end_0"), val = tensor([2, 256, 1, 160])]; tensor var_3644_end_mask_0 = const()[name = tensor("op_3644_end_mask_0"), val = tensor([true, true, true, false])]; tensor transpose_9 = transpose(perm = k_35_perm_0, x = k_33_cast_fp16)[name = tensor("transpose_9")]; tensor var_3644_cast_fp16 = slice_by_index(begin = var_3644_begin_0, end = var_3644_end_0, end_mask = var_3644_end_mask_0, x = transpose_9)[name = tensor("op_3644_cast_fp16")]; tensor var_3648_begin_0 = const()[name = tensor("op_3648_begin_0"), val = tensor([0, 0, 0, 160])]; tensor var_3648_end_0 = const()[name = tensor("op_3648_end_0"), val = tensor([2, 256, 1, 320])]; tensor var_3648_end_mask_0 = const()[name = tensor("op_3648_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_3648_cast_fp16 = slice_by_index(begin = var_3648_begin_0, end = var_3648_end_0, end_mask = var_3648_end_mask_0, x = transpose_9)[name = tensor("op_3648_cast_fp16")]; tensor var_3652_begin_0 = const()[name = tensor("op_3652_begin_0"), val = tensor([0, 0, 0, 320])]; tensor var_3652_end_0 = const()[name = tensor("op_3652_end_0"), val = tensor([2, 256, 1, 480])]; tensor var_3652_end_mask_0 = const()[name = tensor("op_3652_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_3652_cast_fp16 = slice_by_index(begin = var_3652_begin_0, end = var_3652_end_0, end_mask = var_3652_end_mask_0, x = transpose_9)[name = tensor("op_3652_cast_fp16")]; tensor var_3656_begin_0 = const()[name = tensor("op_3656_begin_0"), val = tensor([0, 0, 0, 480])]; tensor var_3656_end_0 = const()[name = tensor("op_3656_end_0"), val = tensor([2, 256, 1, 640])]; tensor var_3656_end_mask_0 = const()[name = tensor("op_3656_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_3656_cast_fp16 = slice_by_index(begin = var_3656_begin_0, end = var_3656_end_0, end_mask = var_3656_end_mask_0, x = transpose_9)[name = tensor("op_3656_cast_fp16")]; tensor var_3660_begin_0 = const()[name = tensor("op_3660_begin_0"), val = tensor([0, 0, 0, 640])]; tensor var_3660_end_0 = const()[name = tensor("op_3660_end_0"), val = tensor([2, 256, 1, 800])]; tensor var_3660_end_mask_0 = const()[name = tensor("op_3660_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_3660_cast_fp16 = slice_by_index(begin = var_3660_begin_0, end = var_3660_end_0, end_mask = var_3660_end_mask_0, x = transpose_9)[name = tensor("op_3660_cast_fp16")]; tensor var_3664_begin_0 = const()[name = tensor("op_3664_begin_0"), val = tensor([0, 0, 0, 800])]; tensor var_3664_end_0 = const()[name = tensor("op_3664_end_0"), val = tensor([2, 256, 1, 960])]; tensor var_3664_end_mask_0 = const()[name = tensor("op_3664_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_3664_cast_fp16 = slice_by_index(begin = var_3664_begin_0, end = var_3664_end_0, end_mask = var_3664_end_mask_0, x = transpose_9)[name = tensor("op_3664_cast_fp16")]; tensor var_3668_begin_0 = const()[name = tensor("op_3668_begin_0"), val = tensor([0, 0, 0, 960])]; tensor var_3668_end_0 = const()[name = tensor("op_3668_end_0"), val = tensor([2, 256, 1, 1120])]; tensor var_3668_end_mask_0 = const()[name = tensor("op_3668_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_3668_cast_fp16 = slice_by_index(begin = var_3668_begin_0, end = var_3668_end_0, end_mask = var_3668_end_mask_0, x = transpose_9)[name = tensor("op_3668_cast_fp16")]; tensor var_3672_begin_0 = const()[name = tensor("op_3672_begin_0"), val = tensor([0, 0, 0, 1120])]; tensor var_3672_end_0 = const()[name = tensor("op_3672_end_0"), val = tensor([2, 256, 1, 1280])]; tensor var_3672_end_mask_0 = const()[name = tensor("op_3672_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_3672_cast_fp16 = slice_by_index(begin = var_3672_begin_0, end = var_3672_end_0, end_mask = var_3672_end_mask_0, x = transpose_9)[name = tensor("op_3672_cast_fp16")]; tensor var_3674_begin_0 = const()[name = tensor("op_3674_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_3674_end_0 = const()[name = tensor("op_3674_end_0"), val = tensor([2, 160, 1, 256])]; tensor var_3674_end_mask_0 = const()[name = tensor("op_3674_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_3674_cast_fp16 = slice_by_index(begin = var_3674_begin_0, end = var_3674_end_0, end_mask = var_3674_end_mask_0, x = v_17_cast_fp16)[name = tensor("op_3674_cast_fp16")]; tensor var_3678_begin_0 = const()[name = tensor("op_3678_begin_0"), val = tensor([0, 160, 0, 0])]; tensor var_3678_end_0 = const()[name = tensor("op_3678_end_0"), val = tensor([2, 320, 1, 256])]; tensor var_3678_end_mask_0 = const()[name = tensor("op_3678_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_3678_cast_fp16 = slice_by_index(begin = var_3678_begin_0, end = var_3678_end_0, end_mask = var_3678_end_mask_0, x = v_17_cast_fp16)[name = tensor("op_3678_cast_fp16")]; tensor var_3682_begin_0 = const()[name = tensor("op_3682_begin_0"), val = tensor([0, 320, 0, 0])]; tensor var_3682_end_0 = const()[name = tensor("op_3682_end_0"), val = tensor([2, 480, 1, 256])]; tensor var_3682_end_mask_0 = const()[name = tensor("op_3682_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_3682_cast_fp16 = slice_by_index(begin = var_3682_begin_0, end = var_3682_end_0, end_mask = var_3682_end_mask_0, x = v_17_cast_fp16)[name = tensor("op_3682_cast_fp16")]; tensor var_3686_begin_0 = const()[name = tensor("op_3686_begin_0"), val = tensor([0, 480, 0, 0])]; tensor var_3686_end_0 = const()[name = tensor("op_3686_end_0"), val = tensor([2, 640, 1, 256])]; tensor var_3686_end_mask_0 = const()[name = tensor("op_3686_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_3686_cast_fp16 = slice_by_index(begin = var_3686_begin_0, end = var_3686_end_0, end_mask = var_3686_end_mask_0, x = v_17_cast_fp16)[name = tensor("op_3686_cast_fp16")]; tensor var_3690_begin_0 = const()[name = tensor("op_3690_begin_0"), val = tensor([0, 640, 0, 0])]; tensor var_3690_end_0 = const()[name = tensor("op_3690_end_0"), val = tensor([2, 800, 1, 256])]; tensor var_3690_end_mask_0 = const()[name = tensor("op_3690_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_3690_cast_fp16 = slice_by_index(begin = var_3690_begin_0, end = var_3690_end_0, end_mask = var_3690_end_mask_0, x = v_17_cast_fp16)[name = tensor("op_3690_cast_fp16")]; tensor var_3694_begin_0 = const()[name = tensor("op_3694_begin_0"), val = tensor([0, 800, 0, 0])]; tensor var_3694_end_0 = const()[name = tensor("op_3694_end_0"), val = tensor([2, 960, 1, 256])]; tensor var_3694_end_mask_0 = const()[name = tensor("op_3694_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_3694_cast_fp16 = slice_by_index(begin = var_3694_begin_0, end = var_3694_end_0, end_mask = var_3694_end_mask_0, x = v_17_cast_fp16)[name = tensor("op_3694_cast_fp16")]; tensor var_3698_begin_0 = const()[name = tensor("op_3698_begin_0"), val = tensor([0, 960, 0, 0])]; tensor var_3698_end_0 = const()[name = tensor("op_3698_end_0"), val = tensor([2, 1120, 1, 256])]; tensor var_3698_end_mask_0 = const()[name = tensor("op_3698_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_3698_cast_fp16 = slice_by_index(begin = var_3698_begin_0, end = var_3698_end_0, end_mask = var_3698_end_mask_0, x = v_17_cast_fp16)[name = tensor("op_3698_cast_fp16")]; tensor var_3702_begin_0 = const()[name = tensor("op_3702_begin_0"), val = tensor([0, 1120, 0, 0])]; tensor var_3702_end_0 = const()[name = tensor("op_3702_end_0"), val = tensor([2, 1280, 1, 256])]; tensor var_3702_end_mask_0 = const()[name = tensor("op_3702_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_3702_cast_fp16 = slice_by_index(begin = var_3702_begin_0, end = var_3702_end_0, end_mask = var_3702_end_mask_0, x = v_17_cast_fp16)[name = tensor("op_3702_cast_fp16")]; tensor var_3706_equation_0 = const()[name = tensor("op_3706_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_3706_cast_fp16 = einsum(equation = var_3706_equation_0, values = (var_3644_cast_fp16, var_3609_cast_fp16))[name = tensor("op_3706_cast_fp16")]; tensor var_3707_to_fp16 = const()[name = tensor("op_3707_to_fp16"), val = tensor(0x1.43cp-4)]; tensor aw_65_cast_fp16 = mul(x = var_3706_cast_fp16, y = var_3707_to_fp16)[name = tensor("aw_65_cast_fp16")]; tensor var_3710_equation_0 = const()[name = tensor("op_3710_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_3710_cast_fp16 = einsum(equation = var_3710_equation_0, values = (var_3648_cast_fp16, var_3613_cast_fp16))[name = tensor("op_3710_cast_fp16")]; tensor var_3711_to_fp16 = const()[name = tensor("op_3711_to_fp16"), val = tensor(0x1.43cp-4)]; tensor aw_67_cast_fp16 = mul(x = var_3710_cast_fp16, y = var_3711_to_fp16)[name = tensor("aw_67_cast_fp16")]; tensor var_3714_equation_0 = const()[name = tensor("op_3714_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_3714_cast_fp16 = einsum(equation = var_3714_equation_0, values = (var_3652_cast_fp16, var_3617_cast_fp16))[name = tensor("op_3714_cast_fp16")]; tensor var_3715_to_fp16 = const()[name = tensor("op_3715_to_fp16"), val = tensor(0x1.43cp-4)]; tensor aw_69_cast_fp16 = mul(x = var_3714_cast_fp16, y = var_3715_to_fp16)[name = tensor("aw_69_cast_fp16")]; tensor var_3718_equation_0 = const()[name = tensor("op_3718_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_3718_cast_fp16 = einsum(equation = var_3718_equation_0, values = (var_3656_cast_fp16, var_3621_cast_fp16))[name = tensor("op_3718_cast_fp16")]; tensor var_3719_to_fp16 = const()[name = tensor("op_3719_to_fp16"), val = tensor(0x1.43cp-4)]; tensor aw_71_cast_fp16 = mul(x = var_3718_cast_fp16, y = var_3719_to_fp16)[name = tensor("aw_71_cast_fp16")]; tensor var_3722_equation_0 = const()[name = tensor("op_3722_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_3722_cast_fp16 = einsum(equation = var_3722_equation_0, values = (var_3660_cast_fp16, var_3625_cast_fp16))[name = tensor("op_3722_cast_fp16")]; tensor var_3723_to_fp16 = const()[name = tensor("op_3723_to_fp16"), val = tensor(0x1.43cp-4)]; tensor aw_73_cast_fp16 = mul(x = var_3722_cast_fp16, y = var_3723_to_fp16)[name = tensor("aw_73_cast_fp16")]; tensor var_3726_equation_0 = const()[name = tensor("op_3726_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_3726_cast_fp16 = einsum(equation = var_3726_equation_0, values = (var_3664_cast_fp16, var_3629_cast_fp16))[name = tensor("op_3726_cast_fp16")]; tensor var_3727_to_fp16 = const()[name = tensor("op_3727_to_fp16"), val = tensor(0x1.43cp-4)]; tensor aw_75_cast_fp16 = mul(x = var_3726_cast_fp16, y = var_3727_to_fp16)[name = tensor("aw_75_cast_fp16")]; tensor var_3730_equation_0 = const()[name = tensor("op_3730_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_3730_cast_fp16 = einsum(equation = var_3730_equation_0, values = (var_3668_cast_fp16, var_3633_cast_fp16))[name = tensor("op_3730_cast_fp16")]; tensor var_3731_to_fp16 = const()[name = tensor("op_3731_to_fp16"), val = tensor(0x1.43cp-4)]; tensor aw_77_cast_fp16 = mul(x = var_3730_cast_fp16, y = var_3731_to_fp16)[name = tensor("aw_77_cast_fp16")]; tensor var_3734_equation_0 = const()[name = tensor("op_3734_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_3734_cast_fp16 = einsum(equation = var_3734_equation_0, values = (var_3672_cast_fp16, var_3637_cast_fp16))[name = tensor("op_3734_cast_fp16")]; tensor var_3735_to_fp16 = const()[name = tensor("op_3735_to_fp16"), val = tensor(0x1.43cp-4)]; tensor aw_79_cast_fp16 = mul(x = var_3734_cast_fp16, y = var_3735_to_fp16)[name = tensor("aw_79_cast_fp16")]; tensor var_3737_cast_fp16 = softmax(axis = var_2942, x = aw_65_cast_fp16)[name = tensor("op_3737_cast_fp16")]; tensor var_3738_cast_fp16 = softmax(axis = var_2942, x = aw_67_cast_fp16)[name = tensor("op_3738_cast_fp16")]; tensor var_3739_cast_fp16 = softmax(axis = var_2942, x = aw_69_cast_fp16)[name = tensor("op_3739_cast_fp16")]; tensor var_3740_cast_fp16 = softmax(axis = var_2942, x = aw_71_cast_fp16)[name = tensor("op_3740_cast_fp16")]; tensor var_3741_cast_fp16 = softmax(axis = var_2942, x = aw_73_cast_fp16)[name = tensor("op_3741_cast_fp16")]; tensor var_3742_cast_fp16 = softmax(axis = var_2942, x = aw_75_cast_fp16)[name = tensor("op_3742_cast_fp16")]; tensor var_3743_cast_fp16 = softmax(axis = var_2942, x = aw_77_cast_fp16)[name = tensor("op_3743_cast_fp16")]; tensor var_3744_cast_fp16 = softmax(axis = var_2942, x = aw_79_cast_fp16)[name = tensor("op_3744_cast_fp16")]; tensor var_3746_equation_0 = const()[name = tensor("op_3746_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_3746_cast_fp16 = einsum(equation = var_3746_equation_0, values = (var_3674_cast_fp16, var_3737_cast_fp16))[name = tensor("op_3746_cast_fp16")]; tensor var_3748_equation_0 = const()[name = tensor("op_3748_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_3748_cast_fp16 = einsum(equation = var_3748_equation_0, values = (var_3678_cast_fp16, var_3738_cast_fp16))[name = tensor("op_3748_cast_fp16")]; tensor var_3750_equation_0 = const()[name = tensor("op_3750_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_3750_cast_fp16 = einsum(equation = var_3750_equation_0, values = (var_3682_cast_fp16, var_3739_cast_fp16))[name = tensor("op_3750_cast_fp16")]; tensor var_3752_equation_0 = const()[name = tensor("op_3752_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_3752_cast_fp16 = einsum(equation = var_3752_equation_0, values = (var_3686_cast_fp16, var_3740_cast_fp16))[name = tensor("op_3752_cast_fp16")]; tensor var_3754_equation_0 = const()[name = tensor("op_3754_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_3754_cast_fp16 = einsum(equation = var_3754_equation_0, values = (var_3690_cast_fp16, var_3741_cast_fp16))[name = tensor("op_3754_cast_fp16")]; tensor var_3756_equation_0 = const()[name = tensor("op_3756_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_3756_cast_fp16 = einsum(equation = var_3756_equation_0, values = (var_3694_cast_fp16, var_3742_cast_fp16))[name = tensor("op_3756_cast_fp16")]; tensor var_3758_equation_0 = const()[name = tensor("op_3758_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_3758_cast_fp16 = einsum(equation = var_3758_equation_0, values = (var_3698_cast_fp16, var_3743_cast_fp16))[name = tensor("op_3758_cast_fp16")]; tensor var_3760_equation_0 = const()[name = tensor("op_3760_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_3760_cast_fp16 = einsum(equation = var_3760_equation_0, values = (var_3702_cast_fp16, var_3744_cast_fp16))[name = tensor("op_3760_cast_fp16")]; tensor input_133_interleave_0 = const()[name = tensor("input_133_interleave_0"), val = tensor(false)]; tensor input_133_cast_fp16 = concat(axis = var_2942, interleave = input_133_interleave_0, values = (var_3746_cast_fp16, var_3748_cast_fp16, var_3750_cast_fp16, var_3752_cast_fp16, var_3754_cast_fp16, var_3756_cast_fp16, var_3758_cast_fp16, var_3760_cast_fp16))[name = tensor("input_133_cast_fp16")]; tensor var_3766 = const()[name = tensor("op_3766"), val = tensor([1, 1])]; tensor var_3768 = const()[name = tensor("op_3768"), val = tensor([1, 1])]; tensor var_3770_pad_type_0 = const()[name = tensor("op_3770_pad_type_0"), val = tensor("custom")]; tensor var_3770_pad_0 = const()[name = tensor("op_3770_pad_0"), val = tensor([0, 0, 0, 0])]; tensor up_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(163679680))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(164908544))), name = tensor("up_blocks_0_attentions_1_transformer_blocks_0_attn1_to_out_0_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; tensor up_blocks_0_attentions_1_transformer_blocks_0_attn1_to_out_0_bias_to_fp16 = const()[name = tensor("up_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(164908736)))]; tensor var_3770_cast_fp16 = conv(bias = up_blocks_0_attentions_1_transformer_blocks_0_attn1_to_out_0_bias_to_fp16, dilations = var_3768, groups = var_2942, pad = var_3770_pad_0, pad_type = var_3770_pad_type_0, strides = var_3766, weight = up_blocks_0_attentions_1_transformer_blocks_0_attn1_to_out_0_weight_to_fp16_palettized, x = input_133_cast_fp16)[name = tensor("op_3770_cast_fp16")]; tensor inputs_27_cast_fp16 = add(x = var_3770_cast_fp16, y = inputs_25_cast_fp16)[name = tensor("inputs_27_cast_fp16")]; tensor var_3774 = const()[name = tensor("op_3774"), val = tensor([1])]; tensor channels_mean_27_cast_fp16 = reduce_mean(axes = var_3774, keep_dims = var_2937, x = inputs_27_cast_fp16)[name = tensor("channels_mean_27_cast_fp16")]; tensor zero_mean_27_cast_fp16 = sub(x = inputs_27_cast_fp16, y = channels_mean_27_cast_fp16)[name = tensor("zero_mean_27_cast_fp16")]; tensor zero_mean_sq_27_cast_fp16 = mul(x = zero_mean_27_cast_fp16, y = zero_mean_27_cast_fp16)[name = tensor("zero_mean_sq_27_cast_fp16")]; tensor var_3778 = const()[name = tensor("op_3778"), val = tensor([1])]; tensor var_3779_cast_fp16 = reduce_mean(axes = var_3778, keep_dims = var_2937, x = zero_mean_sq_27_cast_fp16)[name = tensor("op_3779_cast_fp16")]; tensor var_3780_to_fp16 = const()[name = tensor("op_3780_to_fp16"), val = tensor(0x1.5p-17)]; tensor var_3781_cast_fp16 = add(x = var_3779_cast_fp16, y = var_3780_to_fp16)[name = tensor("op_3781_cast_fp16")]; tensor denom_27_epsilon_0_to_fp16 = const()[name = tensor("denom_27_epsilon_0_to_fp16"), val = tensor(0x1p-24)]; tensor denom_27_cast_fp16 = rsqrt(epsilon = denom_27_epsilon_0_to_fp16, x = var_3781_cast_fp16)[name = tensor("denom_27_cast_fp16")]; tensor out_27_cast_fp16 = mul(x = zero_mean_27_cast_fp16, y = denom_27_cast_fp16)[name = tensor("out_27_cast_fp16")]; tensor var_3785_to_fp16 = const()[name = tensor("op_3785_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(164911360)))]; tensor var_3786_cast_fp16 = add(x = out_27_cast_fp16, y = var_3785_to_fp16)[name = tensor("op_3786_cast_fp16")]; tensor var_3788_to_fp16 = const()[name = tensor("op_3788_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(164913984)))]; tensor hidden_states_87_cast_fp16 = mul(x = var_3786_cast_fp16, y = var_3788_to_fp16)[name = tensor("hidden_states_87_cast_fp16")]; tensor var_3795 = const()[name = tensor("op_3795"), val = tensor([1, 1])]; tensor var_3797 = const()[name = tensor("op_3797"), 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 up_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(164916608))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(166145472))), name = tensor("up_blocks_0_attentions_1_transformer_blocks_0_attn2_to_q_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; tensor q_19_cast_fp16 = conv(dilations = var_3797, groups = var_2942, pad = q_19_pad_0, pad_type = q_19_pad_type_0, strides = var_3795, weight = up_blocks_0_attentions_1_transformer_blocks_0_attn2_to_q_weight_to_fp16_palettized, x = hidden_states_87_cast_fp16)[name = tensor("q_19_cast_fp16")]; tensor var_3801 = const()[name = tensor("op_3801"), val = tensor([1, 1])]; tensor var_3803 = const()[name = tensor("op_3803"), 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 up_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(166145664))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(166883008))), name = tensor("up_blocks_0_attentions_1_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized"), shape = tensor([1280, 768, 1, 1])]; tensor k_37_cast_fp16 = conv(dilations = var_3803, groups = var_2942, pad = k_37_pad_0, pad_type = k_37_pad_type_0, strides = var_3801, weight = up_blocks_0_attentions_1_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor("k_37_cast_fp16")]; tensor var_3807 = const()[name = tensor("op_3807"), val = tensor([1, 1])]; tensor var_3809 = const()[name = tensor("op_3809"), 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 up_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(166883200))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(167620544))), name = tensor("up_blocks_0_attentions_1_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized"), shape = tensor([1280, 768, 1, 1])]; tensor v_19_cast_fp16 = conv(dilations = var_3809, groups = var_2942, pad = v_19_pad_0, pad_type = v_19_pad_type_0, strides = var_3807, weight = up_blocks_0_attentions_1_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor("v_19_cast_fp16")]; tensor var_3813_begin_0 = const()[name = tensor("op_3813_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_3813_end_0 = const()[name = tensor("op_3813_end_0"), val = tensor([2, 160, 1, 256])]; tensor var_3813_end_mask_0 = const()[name = tensor("op_3813_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_3813_cast_fp16 = slice_by_index(begin = var_3813_begin_0, end = var_3813_end_0, end_mask = var_3813_end_mask_0, x = q_19_cast_fp16)[name = tensor("op_3813_cast_fp16")]; tensor var_3817_begin_0 = const()[name = tensor("op_3817_begin_0"), val = tensor([0, 160, 0, 0])]; tensor var_3817_end_0 = const()[name = tensor("op_3817_end_0"), val = tensor([2, 320, 1, 256])]; tensor var_3817_end_mask_0 = const()[name = tensor("op_3817_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_3817_cast_fp16 = slice_by_index(begin = var_3817_begin_0, end = var_3817_end_0, end_mask = var_3817_end_mask_0, x = q_19_cast_fp16)[name = tensor("op_3817_cast_fp16")]; tensor var_3821_begin_0 = const()[name = tensor("op_3821_begin_0"), val = tensor([0, 320, 0, 0])]; tensor var_3821_end_0 = const()[name = tensor("op_3821_end_0"), val = tensor([2, 480, 1, 256])]; tensor var_3821_end_mask_0 = const()[name = tensor("op_3821_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_3821_cast_fp16 = slice_by_index(begin = var_3821_begin_0, end = var_3821_end_0, end_mask = var_3821_end_mask_0, x = q_19_cast_fp16)[name = tensor("op_3821_cast_fp16")]; tensor var_3825_begin_0 = const()[name = tensor("op_3825_begin_0"), val = tensor([0, 480, 0, 0])]; tensor var_3825_end_0 = const()[name = tensor("op_3825_end_0"), val = tensor([2, 640, 1, 256])]; tensor var_3825_end_mask_0 = const()[name = tensor("op_3825_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_3825_cast_fp16 = slice_by_index(begin = var_3825_begin_0, end = var_3825_end_0, end_mask = var_3825_end_mask_0, x = q_19_cast_fp16)[name = tensor("op_3825_cast_fp16")]; tensor var_3829_begin_0 = const()[name = tensor("op_3829_begin_0"), val = tensor([0, 640, 0, 0])]; tensor var_3829_end_0 = const()[name = tensor("op_3829_end_0"), val = tensor([2, 800, 1, 256])]; tensor var_3829_end_mask_0 = const()[name = tensor("op_3829_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_3829_cast_fp16 = slice_by_index(begin = var_3829_begin_0, end = var_3829_end_0, end_mask = var_3829_end_mask_0, x = q_19_cast_fp16)[name = tensor("op_3829_cast_fp16")]; tensor var_3833_begin_0 = const()[name = tensor("op_3833_begin_0"), val = tensor([0, 800, 0, 0])]; tensor var_3833_end_0 = const()[name = tensor("op_3833_end_0"), val = tensor([2, 960, 1, 256])]; tensor var_3833_end_mask_0 = const()[name = tensor("op_3833_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_3833_cast_fp16 = slice_by_index(begin = var_3833_begin_0, end = var_3833_end_0, end_mask = var_3833_end_mask_0, x = q_19_cast_fp16)[name = tensor("op_3833_cast_fp16")]; tensor var_3837_begin_0 = const()[name = tensor("op_3837_begin_0"), val = tensor([0, 960, 0, 0])]; tensor var_3837_end_0 = const()[name = tensor("op_3837_end_0"), val = tensor([2, 1120, 1, 256])]; tensor var_3837_end_mask_0 = const()[name = tensor("op_3837_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_3837_cast_fp16 = slice_by_index(begin = var_3837_begin_0, end = var_3837_end_0, end_mask = var_3837_end_mask_0, x = q_19_cast_fp16)[name = tensor("op_3837_cast_fp16")]; tensor var_3841_begin_0 = const()[name = tensor("op_3841_begin_0"), val = tensor([0, 1120, 0, 0])]; tensor var_3841_end_0 = const()[name = tensor("op_3841_end_0"), val = tensor([2, 1280, 1, 256])]; tensor var_3841_end_mask_0 = const()[name = tensor("op_3841_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_3841_cast_fp16 = slice_by_index(begin = var_3841_begin_0, end = var_3841_end_0, end_mask = var_3841_end_mask_0, x = q_19_cast_fp16)[name = tensor("op_3841_cast_fp16")]; tensor k_39_perm_0 = const()[name = tensor("k_39_perm_0"), val = tensor([0, 3, 2, 1])]; tensor var_3848_begin_0 = const()[name = tensor("op_3848_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_3848_end_0 = const()[name = tensor("op_3848_end_0"), val = tensor([2, 77, 1, 160])]; tensor var_3848_end_mask_0 = const()[name = tensor("op_3848_end_mask_0"), val = tensor([true, true, true, false])]; tensor transpose_8 = transpose(perm = k_39_perm_0, x = k_37_cast_fp16)[name = tensor("transpose_8")]; tensor var_3848_cast_fp16 = slice_by_index(begin = var_3848_begin_0, end = var_3848_end_0, end_mask = var_3848_end_mask_0, x = transpose_8)[name = tensor("op_3848_cast_fp16")]; tensor var_3852_begin_0 = const()[name = tensor("op_3852_begin_0"), val = tensor([0, 0, 0, 160])]; tensor var_3852_end_0 = const()[name = tensor("op_3852_end_0"), val = tensor([2, 77, 1, 320])]; tensor var_3852_end_mask_0 = const()[name = tensor("op_3852_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_3852_cast_fp16 = slice_by_index(begin = var_3852_begin_0, end = var_3852_end_0, end_mask = var_3852_end_mask_0, x = transpose_8)[name = tensor("op_3852_cast_fp16")]; tensor var_3856_begin_0 = const()[name = tensor("op_3856_begin_0"), val = tensor([0, 0, 0, 320])]; tensor var_3856_end_0 = const()[name = tensor("op_3856_end_0"), val = tensor([2, 77, 1, 480])]; tensor var_3856_end_mask_0 = const()[name = tensor("op_3856_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_3856_cast_fp16 = slice_by_index(begin = var_3856_begin_0, end = var_3856_end_0, end_mask = var_3856_end_mask_0, x = transpose_8)[name = tensor("op_3856_cast_fp16")]; tensor var_3860_begin_0 = const()[name = tensor("op_3860_begin_0"), val = tensor([0, 0, 0, 480])]; tensor var_3860_end_0 = const()[name = tensor("op_3860_end_0"), val = tensor([2, 77, 1, 640])]; tensor var_3860_end_mask_0 = const()[name = tensor("op_3860_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_3860_cast_fp16 = slice_by_index(begin = var_3860_begin_0, end = var_3860_end_0, end_mask = var_3860_end_mask_0, x = transpose_8)[name = tensor("op_3860_cast_fp16")]; tensor var_3864_begin_0 = const()[name = tensor("op_3864_begin_0"), val = tensor([0, 0, 0, 640])]; tensor var_3864_end_0 = const()[name = tensor("op_3864_end_0"), val = tensor([2, 77, 1, 800])]; tensor var_3864_end_mask_0 = const()[name = tensor("op_3864_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_3864_cast_fp16 = slice_by_index(begin = var_3864_begin_0, end = var_3864_end_0, end_mask = var_3864_end_mask_0, x = transpose_8)[name = tensor("op_3864_cast_fp16")]; tensor var_3868_begin_0 = const()[name = tensor("op_3868_begin_0"), val = tensor([0, 0, 0, 800])]; tensor var_3868_end_0 = const()[name = tensor("op_3868_end_0"), val = tensor([2, 77, 1, 960])]; tensor var_3868_end_mask_0 = const()[name = tensor("op_3868_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_3868_cast_fp16 = slice_by_index(begin = var_3868_begin_0, end = var_3868_end_0, end_mask = var_3868_end_mask_0, x = transpose_8)[name = tensor("op_3868_cast_fp16")]; tensor var_3872_begin_0 = const()[name = tensor("op_3872_begin_0"), val = tensor([0, 0, 0, 960])]; tensor var_3872_end_0 = const()[name = tensor("op_3872_end_0"), val = tensor([2, 77, 1, 1120])]; tensor var_3872_end_mask_0 = const()[name = tensor("op_3872_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_3872_cast_fp16 = slice_by_index(begin = var_3872_begin_0, end = var_3872_end_0, end_mask = var_3872_end_mask_0, x = transpose_8)[name = tensor("op_3872_cast_fp16")]; tensor var_3876_begin_0 = const()[name = tensor("op_3876_begin_0"), val = tensor([0, 0, 0, 1120])]; tensor var_3876_end_0 = const()[name = tensor("op_3876_end_0"), val = tensor([2, 77, 1, 1280])]; tensor var_3876_end_mask_0 = const()[name = tensor("op_3876_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_3876_cast_fp16 = slice_by_index(begin = var_3876_begin_0, end = var_3876_end_0, end_mask = var_3876_end_mask_0, x = transpose_8)[name = tensor("op_3876_cast_fp16")]; tensor var_3878_begin_0 = const()[name = tensor("op_3878_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_3878_end_0 = const()[name = tensor("op_3878_end_0"), val = tensor([2, 160, 1, 77])]; tensor var_3878_end_mask_0 = const()[name = tensor("op_3878_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_3878_cast_fp16 = slice_by_index(begin = var_3878_begin_0, end = var_3878_end_0, end_mask = var_3878_end_mask_0, x = v_19_cast_fp16)[name = tensor("op_3878_cast_fp16")]; tensor var_3882_begin_0 = const()[name = tensor("op_3882_begin_0"), val = tensor([0, 160, 0, 0])]; tensor var_3882_end_0 = const()[name = tensor("op_3882_end_0"), val = tensor([2, 320, 1, 77])]; tensor var_3882_end_mask_0 = const()[name = tensor("op_3882_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_3882_cast_fp16 = slice_by_index(begin = var_3882_begin_0, end = var_3882_end_0, end_mask = var_3882_end_mask_0, x = v_19_cast_fp16)[name = tensor("op_3882_cast_fp16")]; tensor var_3886_begin_0 = const()[name = tensor("op_3886_begin_0"), val = tensor([0, 320, 0, 0])]; tensor var_3886_end_0 = const()[name = tensor("op_3886_end_0"), val = tensor([2, 480, 1, 77])]; tensor var_3886_end_mask_0 = const()[name = tensor("op_3886_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_3886_cast_fp16 = slice_by_index(begin = var_3886_begin_0, end = var_3886_end_0, end_mask = var_3886_end_mask_0, x = v_19_cast_fp16)[name = tensor("op_3886_cast_fp16")]; tensor var_3890_begin_0 = const()[name = tensor("op_3890_begin_0"), val = tensor([0, 480, 0, 0])]; tensor var_3890_end_0 = const()[name = tensor("op_3890_end_0"), val = tensor([2, 640, 1, 77])]; tensor var_3890_end_mask_0 = const()[name = tensor("op_3890_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_3890_cast_fp16 = slice_by_index(begin = var_3890_begin_0, end = var_3890_end_0, end_mask = var_3890_end_mask_0, x = v_19_cast_fp16)[name = tensor("op_3890_cast_fp16")]; tensor var_3894_begin_0 = const()[name = tensor("op_3894_begin_0"), val = tensor([0, 640, 0, 0])]; tensor var_3894_end_0 = const()[name = tensor("op_3894_end_0"), val = tensor([2, 800, 1, 77])]; tensor var_3894_end_mask_0 = const()[name = tensor("op_3894_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_3894_cast_fp16 = slice_by_index(begin = var_3894_begin_0, end = var_3894_end_0, end_mask = var_3894_end_mask_0, x = v_19_cast_fp16)[name = tensor("op_3894_cast_fp16")]; tensor var_3898_begin_0 = const()[name = tensor("op_3898_begin_0"), val = tensor([0, 800, 0, 0])]; tensor var_3898_end_0 = const()[name = tensor("op_3898_end_0"), val = tensor([2, 960, 1, 77])]; tensor var_3898_end_mask_0 = const()[name = tensor("op_3898_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_3898_cast_fp16 = slice_by_index(begin = var_3898_begin_0, end = var_3898_end_0, end_mask = var_3898_end_mask_0, x = v_19_cast_fp16)[name = tensor("op_3898_cast_fp16")]; tensor var_3902_begin_0 = const()[name = tensor("op_3902_begin_0"), val = tensor([0, 960, 0, 0])]; tensor var_3902_end_0 = const()[name = tensor("op_3902_end_0"), val = tensor([2, 1120, 1, 77])]; tensor var_3902_end_mask_0 = const()[name = tensor("op_3902_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_3902_cast_fp16 = slice_by_index(begin = var_3902_begin_0, end = var_3902_end_0, end_mask = var_3902_end_mask_0, x = v_19_cast_fp16)[name = tensor("op_3902_cast_fp16")]; tensor var_3906_begin_0 = const()[name = tensor("op_3906_begin_0"), val = tensor([0, 1120, 0, 0])]; tensor var_3906_end_0 = const()[name = tensor("op_3906_end_0"), val = tensor([2, 1280, 1, 77])]; tensor var_3906_end_mask_0 = const()[name = tensor("op_3906_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_3906_cast_fp16 = slice_by_index(begin = var_3906_begin_0, end = var_3906_end_0, end_mask = var_3906_end_mask_0, x = v_19_cast_fp16)[name = tensor("op_3906_cast_fp16")]; tensor var_3910_equation_0 = const()[name = tensor("op_3910_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_3910_cast_fp16 = einsum(equation = var_3910_equation_0, values = (var_3848_cast_fp16, var_3813_cast_fp16))[name = tensor("op_3910_cast_fp16")]; tensor var_3911_to_fp16 = const()[name = tensor("op_3911_to_fp16"), val = tensor(0x1.43cp-4)]; tensor aw_81_cast_fp16 = mul(x = var_3910_cast_fp16, y = var_3911_to_fp16)[name = tensor("aw_81_cast_fp16")]; tensor var_3914_equation_0 = const()[name = tensor("op_3914_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_3914_cast_fp16 = einsum(equation = var_3914_equation_0, values = (var_3852_cast_fp16, var_3817_cast_fp16))[name = tensor("op_3914_cast_fp16")]; tensor var_3915_to_fp16 = const()[name = tensor("op_3915_to_fp16"), val = tensor(0x1.43cp-4)]; tensor aw_83_cast_fp16 = mul(x = var_3914_cast_fp16, y = var_3915_to_fp16)[name = tensor("aw_83_cast_fp16")]; tensor var_3918_equation_0 = const()[name = tensor("op_3918_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_3918_cast_fp16 = einsum(equation = var_3918_equation_0, values = (var_3856_cast_fp16, var_3821_cast_fp16))[name = tensor("op_3918_cast_fp16")]; tensor var_3919_to_fp16 = const()[name = tensor("op_3919_to_fp16"), val = tensor(0x1.43cp-4)]; tensor aw_85_cast_fp16 = mul(x = var_3918_cast_fp16, y = var_3919_to_fp16)[name = tensor("aw_85_cast_fp16")]; tensor var_3922_equation_0 = const()[name = tensor("op_3922_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_3922_cast_fp16 = einsum(equation = var_3922_equation_0, values = (var_3860_cast_fp16, var_3825_cast_fp16))[name = tensor("op_3922_cast_fp16")]; tensor var_3923_to_fp16 = const()[name = tensor("op_3923_to_fp16"), val = tensor(0x1.43cp-4)]; tensor aw_87_cast_fp16 = mul(x = var_3922_cast_fp16, y = var_3923_to_fp16)[name = tensor("aw_87_cast_fp16")]; tensor var_3926_equation_0 = const()[name = tensor("op_3926_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_3926_cast_fp16 = einsum(equation = var_3926_equation_0, values = (var_3864_cast_fp16, var_3829_cast_fp16))[name = tensor("op_3926_cast_fp16")]; tensor var_3927_to_fp16 = const()[name = tensor("op_3927_to_fp16"), val = tensor(0x1.43cp-4)]; tensor aw_89_cast_fp16 = mul(x = var_3926_cast_fp16, y = var_3927_to_fp16)[name = tensor("aw_89_cast_fp16")]; tensor var_3930_equation_0 = const()[name = tensor("op_3930_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_3930_cast_fp16 = einsum(equation = var_3930_equation_0, values = (var_3868_cast_fp16, var_3833_cast_fp16))[name = tensor("op_3930_cast_fp16")]; tensor var_3931_to_fp16 = const()[name = tensor("op_3931_to_fp16"), val = tensor(0x1.43cp-4)]; tensor aw_91_cast_fp16 = mul(x = var_3930_cast_fp16, y = var_3931_to_fp16)[name = tensor("aw_91_cast_fp16")]; tensor var_3934_equation_0 = const()[name = tensor("op_3934_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_3934_cast_fp16 = einsum(equation = var_3934_equation_0, values = (var_3872_cast_fp16, var_3837_cast_fp16))[name = tensor("op_3934_cast_fp16")]; tensor var_3935_to_fp16 = const()[name = tensor("op_3935_to_fp16"), val = tensor(0x1.43cp-4)]; tensor aw_93_cast_fp16 = mul(x = var_3934_cast_fp16, y = var_3935_to_fp16)[name = tensor("aw_93_cast_fp16")]; tensor var_3938_equation_0 = const()[name = tensor("op_3938_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_3938_cast_fp16 = einsum(equation = var_3938_equation_0, values = (var_3876_cast_fp16, var_3841_cast_fp16))[name = tensor("op_3938_cast_fp16")]; tensor var_3939_to_fp16 = const()[name = tensor("op_3939_to_fp16"), val = tensor(0x1.43cp-4)]; tensor aw_cast_fp16 = mul(x = var_3938_cast_fp16, y = var_3939_to_fp16)[name = tensor("aw_cast_fp16")]; tensor var_3941_cast_fp16 = softmax(axis = var_2942, x = aw_81_cast_fp16)[name = tensor("op_3941_cast_fp16")]; tensor var_3942_cast_fp16 = softmax(axis = var_2942, x = aw_83_cast_fp16)[name = tensor("op_3942_cast_fp16")]; tensor var_3943_cast_fp16 = softmax(axis = var_2942, x = aw_85_cast_fp16)[name = tensor("op_3943_cast_fp16")]; tensor var_3944_cast_fp16 = softmax(axis = var_2942, x = aw_87_cast_fp16)[name = tensor("op_3944_cast_fp16")]; tensor var_3945_cast_fp16 = softmax(axis = var_2942, x = aw_89_cast_fp16)[name = tensor("op_3945_cast_fp16")]; tensor var_3946_cast_fp16 = softmax(axis = var_2942, x = aw_91_cast_fp16)[name = tensor("op_3946_cast_fp16")]; tensor var_3947_cast_fp16 = softmax(axis = var_2942, x = aw_93_cast_fp16)[name = tensor("op_3947_cast_fp16")]; tensor var_3948_cast_fp16 = softmax(axis = var_2942, x = aw_cast_fp16)[name = tensor("op_3948_cast_fp16")]; tensor var_3950_equation_0 = const()[name = tensor("op_3950_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_3950_cast_fp16 = einsum(equation = var_3950_equation_0, values = (var_3878_cast_fp16, var_3941_cast_fp16))[name = tensor("op_3950_cast_fp16")]; tensor var_3952_equation_0 = const()[name = tensor("op_3952_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_3952_cast_fp16 = einsum(equation = var_3952_equation_0, values = (var_3882_cast_fp16, var_3942_cast_fp16))[name = tensor("op_3952_cast_fp16")]; tensor var_3954_equation_0 = const()[name = tensor("op_3954_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_3954_cast_fp16 = einsum(equation = var_3954_equation_0, values = (var_3886_cast_fp16, var_3943_cast_fp16))[name = tensor("op_3954_cast_fp16")]; tensor var_3956_equation_0 = const()[name = tensor("op_3956_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_3956_cast_fp16 = einsum(equation = var_3956_equation_0, values = (var_3890_cast_fp16, var_3944_cast_fp16))[name = tensor("op_3956_cast_fp16")]; tensor var_3958_equation_0 = const()[name = tensor("op_3958_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_3958_cast_fp16 = einsum(equation = var_3958_equation_0, values = (var_3894_cast_fp16, var_3945_cast_fp16))[name = tensor("op_3958_cast_fp16")]; tensor var_3960_equation_0 = const()[name = tensor("op_3960_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_3960_cast_fp16 = einsum(equation = var_3960_equation_0, values = (var_3898_cast_fp16, var_3946_cast_fp16))[name = tensor("op_3960_cast_fp16")]; tensor var_3962_equation_0 = const()[name = tensor("op_3962_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_3962_cast_fp16 = einsum(equation = var_3962_equation_0, values = (var_3902_cast_fp16, var_3947_cast_fp16))[name = tensor("op_3962_cast_fp16")]; tensor var_3964_equation_0 = const()[name = tensor("op_3964_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_3964_cast_fp16 = einsum(equation = var_3964_equation_0, values = (var_3906_cast_fp16, var_3948_cast_fp16))[name = tensor("op_3964_cast_fp16")]; tensor input_135_interleave_0 = const()[name = tensor("input_135_interleave_0"), val = tensor(false)]; tensor input_135_cast_fp16 = concat(axis = var_2942, interleave = input_135_interleave_0, values = (var_3950_cast_fp16, var_3952_cast_fp16, var_3954_cast_fp16, var_3956_cast_fp16, var_3958_cast_fp16, var_3960_cast_fp16, var_3962_cast_fp16, var_3964_cast_fp16))[name = tensor("input_135_cast_fp16")]; tensor var_3970 = const()[name = tensor("op_3970"), val = tensor([1, 1])]; tensor var_3972 = const()[name = tensor("op_3972"), val = tensor([1, 1])]; tensor var_3974_pad_type_0 = const()[name = tensor("op_3974_pad_type_0"), val = tensor("custom")]; tensor var_3974_pad_0 = const()[name = tensor("op_3974_pad_0"), val = tensor([0, 0, 0, 0])]; tensor up_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(167620736))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(168849600))), name = tensor("up_blocks_0_attentions_1_transformer_blocks_0_attn2_to_out_0_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; tensor up_blocks_0_attentions_1_transformer_blocks_0_attn2_to_out_0_bias_to_fp16 = const()[name = tensor("up_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(168849792)))]; tensor var_3974_cast_fp16 = conv(bias = up_blocks_0_attentions_1_transformer_blocks_0_attn2_to_out_0_bias_to_fp16, dilations = var_3972, groups = var_2942, pad = var_3974_pad_0, pad_type = var_3974_pad_type_0, strides = var_3970, weight = up_blocks_0_attentions_1_transformer_blocks_0_attn2_to_out_0_weight_to_fp16_palettized, x = input_135_cast_fp16)[name = tensor("op_3974_cast_fp16")]; tensor inputs_29_cast_fp16 = add(x = var_3974_cast_fp16, y = inputs_27_cast_fp16)[name = tensor("inputs_29_cast_fp16")]; tensor var_3978 = const()[name = tensor("op_3978"), val = tensor([1])]; tensor channels_mean_29_cast_fp16 = reduce_mean(axes = var_3978, keep_dims = var_2937, x = inputs_29_cast_fp16)[name = tensor("channels_mean_29_cast_fp16")]; tensor zero_mean_29_cast_fp16 = sub(x = inputs_29_cast_fp16, y = channels_mean_29_cast_fp16)[name = tensor("zero_mean_29_cast_fp16")]; tensor zero_mean_sq_29_cast_fp16 = mul(x = zero_mean_29_cast_fp16, y = zero_mean_29_cast_fp16)[name = tensor("zero_mean_sq_29_cast_fp16")]; tensor var_3982 = const()[name = tensor("op_3982"), val = tensor([1])]; tensor var_3983_cast_fp16 = reduce_mean(axes = var_3982, keep_dims = var_2937, x = zero_mean_sq_29_cast_fp16)[name = tensor("op_3983_cast_fp16")]; tensor var_3984_to_fp16 = const()[name = tensor("op_3984_to_fp16"), val = tensor(0x1.5p-17)]; tensor var_3985_cast_fp16 = add(x = var_3983_cast_fp16, y = var_3984_to_fp16)[name = tensor("op_3985_cast_fp16")]; tensor denom_29_epsilon_0_to_fp16 = const()[name = tensor("denom_29_epsilon_0_to_fp16"), val = tensor(0x1p-24)]; tensor denom_29_cast_fp16 = rsqrt(epsilon = denom_29_epsilon_0_to_fp16, x = var_3985_cast_fp16)[name = tensor("denom_29_cast_fp16")]; tensor out_29_cast_fp16 = mul(x = zero_mean_29_cast_fp16, y = denom_29_cast_fp16)[name = tensor("out_29_cast_fp16")]; tensor var_3989_to_fp16 = const()[name = tensor("op_3989_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(168852416)))]; tensor var_3990_cast_fp16 = add(x = out_29_cast_fp16, y = var_3989_to_fp16)[name = tensor("op_3990_cast_fp16")]; tensor var_3992_to_fp16 = const()[name = tensor("op_3992_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(168855040)))]; tensor input_137_cast_fp16 = mul(x = var_3990_cast_fp16, y = var_3992_to_fp16)[name = tensor("input_137_cast_fp16")]; tensor var_4000 = const()[name = tensor("op_4000"), val = tensor([1, 1])]; tensor var_4002 = const()[name = tensor("op_4002"), val = tensor([1, 1])]; tensor var_4004_pad_type_0 = const()[name = tensor("op_4004_pad_type_0"), val = tensor("custom")]; tensor var_4004_pad_0 = const()[name = tensor("op_4004_pad_0"), val = tensor([0, 0, 0, 0])]; tensor up_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(168857664))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(178688128))), name = tensor("up_blocks_0_attentions_1_transformer_blocks_0_ff_net_0_proj_weight_to_fp16_palettized"), shape = tensor([10240, 1280, 1, 1])]; tensor up_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(178688320))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(178696064))), name = tensor("up_blocks_0_attentions_1_transformer_blocks_0_ff_net_0_proj_bias_to_fp16_palettized"), shape = tensor([10240])]; tensor var_4004_cast_fp16 = conv(bias = up_blocks_0_attentions_1_transformer_blocks_0_ff_net_0_proj_bias_to_fp16_palettized, dilations = var_4002, groups = var_2942, pad = var_4004_pad_0, pad_type = var_4004_pad_type_0, strides = var_4000, weight = up_blocks_0_attentions_1_transformer_blocks_0_ff_net_0_proj_weight_to_fp16_palettized, x = input_137_cast_fp16)[name = tensor("op_4004_cast_fp16")]; tensor var_4005_split_sizes_0 = const()[name = tensor("op_4005_split_sizes_0"), val = tensor([5120, 5120])]; tensor var_4005_axis_0 = const()[name = tensor("op_4005_axis_0"), val = tensor(1)]; tensor var_4005_cast_fp16_0, tensor var_4005_cast_fp16_1 = split(axis = var_4005_axis_0, split_sizes = var_4005_split_sizes_0, x = var_4004_cast_fp16)[name = tensor("op_4005_cast_fp16")]; tensor var_4007_mode_0 = const()[name = tensor("op_4007_mode_0"), val = tensor("EXACT")]; tensor var_4007_cast_fp16 = gelu(mode = var_4007_mode_0, x = var_4005_cast_fp16_1)[name = tensor("op_4007_cast_fp16")]; tensor input_139_cast_fp16 = mul(x = var_4005_cast_fp16_0, y = var_4007_cast_fp16)[name = tensor("input_139_cast_fp16")]; tensor var_4011 = const()[name = tensor("op_4011"), val = tensor([1, 1])]; tensor var_4013 = const()[name = tensor("op_4013"), val = tensor([1, 1])]; tensor var_4015_pad_type_0 = const()[name = tensor("op_4015_pad_type_0"), val = tensor("custom")]; tensor var_4015_pad_0 = const()[name = tensor("op_4015_pad_0"), val = tensor([0, 0, 0, 0])]; tensor up_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(178696256))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(183611520))), name = tensor("up_blocks_0_attentions_1_transformer_blocks_0_ff_net_2_weight_to_fp16_palettized"), shape = tensor([1280, 5120, 1, 1])]; tensor up_blocks_0_attentions_1_transformer_blocks_0_ff_net_2_bias_to_fp16 = const()[name = tensor("up_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(183611712)))]; tensor var_4015_cast_fp16 = conv(bias = up_blocks_0_attentions_1_transformer_blocks_0_ff_net_2_bias_to_fp16, dilations = var_4013, groups = var_2942, pad = var_4015_pad_0, pad_type = var_4015_pad_type_0, strides = var_4011, weight = up_blocks_0_attentions_1_transformer_blocks_0_ff_net_2_weight_to_fp16_palettized, x = input_139_cast_fp16)[name = tensor("op_4015_cast_fp16")]; tensor hidden_states_91_cast_fp16 = add(x = var_4015_cast_fp16, y = inputs_29_cast_fp16)[name = tensor("hidden_states_91_cast_fp16")]; tensor var_4017 = const()[name = tensor("op_4017"), val = tensor([2, 1280, 16, 16])]; tensor input_141_cast_fp16 = reshape(shape = var_4017, x = hidden_states_91_cast_fp16)[name = tensor("input_141_cast_fp16")]; tensor var_4021 = const()[name = tensor("op_4021"), val = tensor([1, 1])]; tensor var_4023 = const()[name = tensor("op_4023"), 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([0, 0, 0, 0])]; tensor up_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(183614336))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(184843200))), name = tensor("up_blocks_0_attentions_1_proj_out_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; tensor up_blocks_0_attentions_1_proj_out_bias_to_fp16 = const()[name = tensor("up_blocks_0_attentions_1_proj_out_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(184843392)))]; tensor hidden_states_93_cast_fp16 = conv(bias = up_blocks_0_attentions_1_proj_out_bias_to_fp16, dilations = var_4023, groups = var_2942, pad = hidden_states_93_pad_0, pad_type = hidden_states_93_pad_type_0, strides = var_4021, weight = up_blocks_0_attentions_1_proj_out_weight_to_fp16_palettized, x = input_141_cast_fp16)[name = tensor("hidden_states_93_cast_fp16")]; tensor input_143_cast_fp16 = add(x = hidden_states_93_cast_fp16, y = hidden_states_81_cast_fp16)[name = tensor("input_143_cast_fp16")]; tensor input_145_scale_factor_height_0 = const()[name = tensor("input_145_scale_factor_height_0"), val = tensor(0x1p+1)]; tensor input_145_scale_factor_width_0 = const()[name = tensor("input_145_scale_factor_width_0"), val = tensor(0x1p+1)]; tensor input_145_cast_fp16 = upsample_nearest_neighbor(scale_factor_height = input_145_scale_factor_height_0, scale_factor_width = input_145_scale_factor_width_0, x = input_143_cast_fp16)[name = tensor("input_145_cast_fp16")]; tensor var_4032 = const()[name = tensor("op_4032"), val = tensor([1, 1])]; tensor var_4034 = const()[name = tensor("op_4034"), val = tensor([1, 1])]; tensor hidden_states_95_pad_type_0 = const()[name = tensor("hidden_states_95_pad_type_0"), val = tensor("custom")]; tensor hidden_states_95_pad_0 = const()[name = tensor("hidden_states_95_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(184846016))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(195905280))), 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(195905472)))]; tensor hidden_states_95_cast_fp16 = conv(bias = up_blocks_0_upsamplers_0_conv_bias_to_fp16, dilations = var_4034, groups = var_2942, pad = hidden_states_95_pad_0, pad_type = hidden_states_95_pad_type_0, strides = var_4032, weight = up_blocks_0_upsamplers_0_conv_weight_to_fp16_palettized, x = input_145_cast_fp16)[name = tensor("hidden_states_95_cast_fp16")]; tensor var_4039 = const()[name = tensor("op_4039"), val = tensor(3)]; tensor var_4056 = const()[name = tensor("op_4056"), val = tensor(true)]; tensor var_4061 = const()[name = tensor("op_4061"), val = tensor(1)]; tensor input_147_interleave_0 = const()[name = tensor("input_147_interleave_0"), val = tensor(false)]; tensor input_147_cast_fp16 = concat(axis = var_4061, interleave = input_147_interleave_0, values = (hidden_states_95_cast_fp16, input_63_cast_fp16))[name = tensor("input_147_cast_fp16")]; tensor reshape_60_shape_0 = const()[name = tensor("reshape_60_shape_0"), val = tensor([2, 32, 60, 32, 32])]; tensor reshape_60_cast_fp16 = reshape(shape = reshape_60_shape_0, x = input_147_cast_fp16)[name = tensor("reshape_60_cast_fp16")]; 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_fp16 = reduce_mean(axes = reduce_mean_45_axes_0, keep_dims = reduce_mean_45_keep_dims_0, x = reshape_60_cast_fp16)[name = tensor("reduce_mean_45_cast_fp16")]; tensor sub_30_cast_fp16 = sub(x = reshape_60_cast_fp16, y = reduce_mean_45_cast_fp16)[name = tensor("sub_30_cast_fp16")]; tensor square_15_cast_fp16 = square(x = sub_30_cast_fp16)[name = tensor("square_15_cast_fp16")]; 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_fp16 = reduce_mean(axes = reduce_mean_47_axes_0, keep_dims = reduce_mean_47_keep_dims_0, x = square_15_cast_fp16)[name = tensor("reduce_mean_47_cast_fp16")]; 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_fp16 = add(x = reduce_mean_47_cast_fp16, y = add_30_y_0_to_fp16)[name = tensor("add_30_cast_fp16")]; tensor sqrt_15_cast_fp16 = sqrt(x = add_30_cast_fp16)[name = tensor("sqrt_15_cast_fp16")]; tensor real_div_15_cast_fp16 = real_div(x = sub_30_cast_fp16, y = sqrt_15_cast_fp16)[name = tensor("real_div_15_cast_fp16")]; tensor reshape_61_shape_0 = const()[name = tensor("reshape_61_shape_0"), val = tensor([2, 1920, 32, 32])]; tensor reshape_61_cast_fp16 = reshape(shape = reshape_61_shape_0, x = real_div_15_cast_fp16)[name = tensor("reshape_61_cast_fp16")]; 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(195908096)))]; 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(195912000)))]; 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_fp16 = 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_25_mean_0_to_fp16, variance = add_25_variance_0_to_fp16, x = reshape_61_cast_fp16)[name = tensor("add_31_cast_fp16")]; tensor input_151_cast_fp16 = silu(x = add_31_cast_fp16)[name = tensor("input_151_cast_fp16")]; tensor var_4086 = const()[name = tensor("op_4086"), val = tensor([1, 1])]; tensor var_4088 = const()[name = tensor("op_4088"), 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([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(195915904))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(204210368))), name = tensor("up_blocks_1_resnets_0_conv1_weight_to_fp16_palettized"), shape = tensor([640, 1920, 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(204210560)))]; tensor hidden_states_97_cast_fp16 = conv(bias = up_blocks_1_resnets_0_conv1_bias_to_fp16, dilations = var_4088, groups = var_4061, pad = hidden_states_97_pad_0, pad_type = hidden_states_97_pad_type_0, strides = var_4086, weight = up_blocks_1_resnets_0_conv1_weight_to_fp16_palettized, x = input_151_cast_fp16)[name = tensor("hidden_states_97_cast_fp16")]; tensor var_4094 = const()[name = tensor("op_4094"), val = tensor([1, 1])]; tensor var_4096 = const()[name = tensor("op_4096"), 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 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(204211904))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(204826368))), name = tensor("up_blocks_1_resnets_0_time_emb_proj_weight_to_fp16_palettized"), shape = tensor([640, 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(204826560)))]; tensor temb_11_cast_fp16 = conv(bias = up_blocks_1_resnets_0_time_emb_proj_bias_to_fp16, dilations = var_4096, groups = var_4061, pad = temb_11_pad_0, pad_type = temb_11_pad_type_0, strides = var_4094, weight = up_blocks_1_resnets_0_time_emb_proj_weight_to_fp16_palettized, x = input_15_cast_fp16)[name = tensor("temb_11_cast_fp16")]; tensor input_155_cast_fp16 = add(x = hidden_states_97_cast_fp16, y = temb_11_cast_fp16)[name = tensor("input_155_cast_fp16")]; tensor reshape_64_shape_0 = const()[name = tensor("reshape_64_shape_0"), val = tensor([2, 32, 20, 32, 32])]; tensor reshape_64_cast_fp16 = reshape(shape = reshape_64_shape_0, x = input_155_cast_fp16)[name = tensor("reshape_64_cast_fp16")]; 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_fp16 = reduce_mean(axes = reduce_mean_48_axes_0, keep_dims = reduce_mean_48_keep_dims_0, x = reshape_64_cast_fp16)[name = tensor("reduce_mean_48_cast_fp16")]; tensor sub_32_cast_fp16 = sub(x = reshape_64_cast_fp16, y = reduce_mean_48_cast_fp16)[name = tensor("sub_32_cast_fp16")]; tensor square_16_cast_fp16 = square(x = sub_32_cast_fp16)[name = tensor("square_16_cast_fp16")]; 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_fp16 = reduce_mean(axes = reduce_mean_50_axes_0, keep_dims = reduce_mean_50_keep_dims_0, x = square_16_cast_fp16)[name = tensor("reduce_mean_50_cast_fp16")]; 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_fp16 = add(x = reduce_mean_50_cast_fp16, y = add_32_y_0_to_fp16)[name = tensor("add_32_cast_fp16")]; tensor sqrt_16_cast_fp16 = sqrt(x = add_32_cast_fp16)[name = tensor("sqrt_16_cast_fp16")]; tensor real_div_16_cast_fp16 = real_div(x = sub_32_cast_fp16, y = sqrt_16_cast_fp16)[name = tensor("real_div_16_cast_fp16")]; tensor reshape_65_shape_0 = const()[name = tensor("reshape_65_shape_0"), val = tensor([2, 640, 32, 32])]; tensor reshape_65_cast_fp16 = reshape(shape = reshape_65_shape_0, x = real_div_16_cast_fp16)[name = tensor("reshape_65_cast_fp16")]; 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(204827904)))]; 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(204829248)))]; 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_fp16 = 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_9_mean_0_to_fp16, variance = add_9_variance_0_to_fp16, x = reshape_65_cast_fp16)[name = tensor("add_33_cast_fp16")]; tensor input_159_cast_fp16 = silu(x = add_33_cast_fp16)[name = tensor("input_159_cast_fp16")]; tensor var_4106 = const()[name = tensor("op_4106"), val = tensor([1, 1])]; tensor var_4108 = const()[name = tensor("op_4108"), val = tensor([1, 1])]; tensor hidden_states_99_pad_type_0 = const()[name = tensor("hidden_states_99_pad_type_0"), val = tensor("custom")]; tensor hidden_states_99_pad_0 = const()[name = tensor("hidden_states_99_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(204830592))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(207595456))), name = tensor("up_blocks_1_resnets_0_conv2_weight_to_fp16_palettized"), shape = tensor([640, 640, 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(207595648)))]; tensor hidden_states_99_cast_fp16 = conv(bias = up_blocks_1_resnets_0_conv2_bias_to_fp16, dilations = var_4108, groups = var_4061, pad = hidden_states_99_pad_0, pad_type = hidden_states_99_pad_type_0, strides = var_4106, weight = up_blocks_1_resnets_0_conv2_weight_to_fp16_palettized, x = input_159_cast_fp16)[name = tensor("hidden_states_99_cast_fp16")]; tensor var_4113 = const()[name = tensor("op_4113"), val = tensor([1, 1])]; tensor var_4115 = const()[name = tensor("op_4115"), 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_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(207596992))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(208518656))), name = tensor("up_blocks_1_resnets_0_conv_shortcut_weight_to_fp16_palettized"), shape = tensor([640, 1920, 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(208518848)))]; tensor x_9_cast_fp16 = conv(bias = up_blocks_1_resnets_0_conv_shortcut_bias_to_fp16, dilations = var_4115, groups = var_4061, pad = x_9_pad_0, pad_type = x_9_pad_type_0, strides = var_4113, weight = up_blocks_1_resnets_0_conv_shortcut_weight_to_fp16_palettized, x = input_147_cast_fp16)[name = tensor("x_9_cast_fp16")]; tensor hidden_states_101_cast_fp16 = add(x = x_9_cast_fp16, y = hidden_states_99_cast_fp16)[name = tensor("hidden_states_101_cast_fp16")]; tensor reshape_68_shape_0 = const()[name = tensor("reshape_68_shape_0"), val = tensor([2, 32, 20, 32, 32])]; tensor reshape_68_cast_fp16 = reshape(shape = reshape_68_shape_0, x = hidden_states_101_cast_fp16)[name = tensor("reshape_68_cast_fp16")]; 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_fp16 = reduce_mean(axes = reduce_mean_51_axes_0, keep_dims = reduce_mean_51_keep_dims_0, x = reshape_68_cast_fp16)[name = tensor("reduce_mean_51_cast_fp16")]; tensor sub_34_cast_fp16 = sub(x = reshape_68_cast_fp16, y = reduce_mean_51_cast_fp16)[name = tensor("sub_34_cast_fp16")]; tensor square_17_cast_fp16 = square(x = sub_34_cast_fp16)[name = tensor("square_17_cast_fp16")]; 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_fp16 = reduce_mean(axes = reduce_mean_53_axes_0, keep_dims = reduce_mean_53_keep_dims_0, x = square_17_cast_fp16)[name = tensor("reduce_mean_53_cast_fp16")]; 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_fp16 = add(x = reduce_mean_53_cast_fp16, y = add_34_y_0_to_fp16)[name = tensor("add_34_cast_fp16")]; tensor sqrt_17_cast_fp16 = sqrt(x = add_34_cast_fp16)[name = tensor("sqrt_17_cast_fp16")]; tensor real_div_17_cast_fp16 = real_div(x = sub_34_cast_fp16, y = sqrt_17_cast_fp16)[name = tensor("real_div_17_cast_fp16")]; tensor reshape_69_shape_0 = const()[name = tensor("reshape_69_shape_0"), val = tensor([2, 640, 32, 32])]; tensor reshape_69_cast_fp16 = reshape(shape = reshape_69_shape_0, x = real_div_17_cast_fp16)[name = tensor("reshape_69_cast_fp16")]; 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(208520192)))]; 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(208521536)))]; 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_fp16 = 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_9_mean_0_to_fp16, variance = add_9_variance_0_to_fp16, x = reshape_69_cast_fp16)[name = tensor("add_35_cast_fp16")]; tensor var_4135 = const()[name = tensor("op_4135"), val = tensor([1, 1])]; tensor var_4137 = const()[name = tensor("op_4137"), val = tensor([1, 1])]; tensor hidden_states_103_pad_type_0 = const()[name = tensor("hidden_states_103_pad_type_0"), val = tensor("custom")]; tensor hidden_states_103_pad_0 = const()[name = tensor("hidden_states_103_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(208522880))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(208830144))), name = tensor("up_blocks_1_attentions_0_proj_in_weight_to_fp16_palettized"), shape = tensor([640, 640, 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(208830336)))]; tensor hidden_states_103_cast_fp16 = conv(bias = up_blocks_1_attentions_0_proj_in_bias_to_fp16, dilations = var_4137, groups = var_4061, pad = hidden_states_103_pad_0, pad_type = hidden_states_103_pad_type_0, strides = var_4135, weight = up_blocks_1_attentions_0_proj_in_weight_to_fp16_palettized, x = add_35_cast_fp16)[name = tensor("hidden_states_103_cast_fp16")]; tensor var_4142 = const()[name = tensor("op_4142"), val = tensor([2, 640, 1, 1024])]; tensor inputs_31_cast_fp16 = reshape(shape = var_4142, x = hidden_states_103_cast_fp16)[name = tensor("inputs_31_cast_fp16")]; tensor var_4152 = const()[name = tensor("op_4152"), val = tensor([1])]; tensor channels_mean_31_cast_fp16 = reduce_mean(axes = var_4152, keep_dims = var_4056, x = inputs_31_cast_fp16)[name = tensor("channels_mean_31_cast_fp16")]; tensor zero_mean_31_cast_fp16 = sub(x = inputs_31_cast_fp16, y = channels_mean_31_cast_fp16)[name = tensor("zero_mean_31_cast_fp16")]; tensor zero_mean_sq_31_cast_fp16 = mul(x = zero_mean_31_cast_fp16, y = zero_mean_31_cast_fp16)[name = tensor("zero_mean_sq_31_cast_fp16")]; tensor var_4156 = const()[name = tensor("op_4156"), val = tensor([1])]; tensor var_4157_cast_fp16 = reduce_mean(axes = var_4156, keep_dims = var_4056, x = zero_mean_sq_31_cast_fp16)[name = tensor("op_4157_cast_fp16")]; tensor var_4158_to_fp16 = const()[name = tensor("op_4158_to_fp16"), val = tensor(0x1.5p-17)]; tensor var_4159_cast_fp16 = add(x = var_4157_cast_fp16, y = var_4158_to_fp16)[name = tensor("op_4159_cast_fp16")]; tensor denom_31_epsilon_0_to_fp16 = const()[name = tensor("denom_31_epsilon_0_to_fp16"), val = tensor(0x1p-24)]; tensor denom_31_cast_fp16 = rsqrt(epsilon = denom_31_epsilon_0_to_fp16, x = var_4159_cast_fp16)[name = tensor("denom_31_cast_fp16")]; tensor out_31_cast_fp16 = mul(x = zero_mean_31_cast_fp16, y = denom_31_cast_fp16)[name = tensor("out_31_cast_fp16")]; tensor var_4163_to_fp16 = const()[name = tensor("op_4163_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(208831680)))]; tensor var_4164_cast_fp16 = add(x = out_31_cast_fp16, y = var_4163_to_fp16)[name = tensor("op_4164_cast_fp16")]; tensor var_4166_to_fp16 = const()[name = tensor("op_4166_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(208833024)))]; tensor hidden_states_105_cast_fp16 = mul(x = var_4164_cast_fp16, y = var_4166_to_fp16)[name = tensor("hidden_states_105_cast_fp16")]; tensor var_4173 = const()[name = tensor("op_4173"), val = tensor([1, 1])]; tensor var_4175 = const()[name = tensor("op_4175"), 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 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(208834368))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(209141632))), name = tensor("up_blocks_1_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16_palettized"), shape = tensor([640, 640, 1, 1])]; tensor q_21_cast_fp16 = conv(dilations = var_4175, groups = var_4061, pad = q_21_pad_0, pad_type = q_21_pad_type_0, strides = var_4173, weight = up_blocks_1_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16_palettized, x = hidden_states_105_cast_fp16)[name = tensor("q_21_cast_fp16")]; tensor var_4179 = const()[name = tensor("op_4179"), val = tensor([1, 1])]; tensor var_4181 = const()[name = tensor("op_4181"), 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 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(209141824))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(209449088))), name = tensor("up_blocks_1_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16_palettized"), shape = tensor([640, 640, 1, 1])]; tensor k_41_cast_fp16 = conv(dilations = var_4181, groups = var_4061, pad = k_41_pad_0, pad_type = k_41_pad_type_0, strides = var_4179, weight = up_blocks_1_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16_palettized, x = hidden_states_105_cast_fp16)[name = tensor("k_41_cast_fp16")]; tensor var_4185 = const()[name = tensor("op_4185"), val = tensor([1, 1])]; tensor var_4187 = const()[name = tensor("op_4187"), 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 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(209449280))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(209756544))), name = tensor("up_blocks_1_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16_palettized"), shape = tensor([640, 640, 1, 1])]; tensor v_21_cast_fp16 = conv(dilations = var_4187, groups = var_4061, pad = v_21_pad_0, pad_type = v_21_pad_type_0, strides = var_4185, weight = up_blocks_1_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16_palettized, x = hidden_states_105_cast_fp16)[name = tensor("v_21_cast_fp16")]; tensor var_4191_begin_0 = const()[name = tensor("op_4191_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_4191_end_0 = const()[name = tensor("op_4191_end_0"), val = tensor([2, 80, 1, 1024])]; tensor var_4191_end_mask_0 = const()[name = tensor("op_4191_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_4191_cast_fp16 = slice_by_index(begin = var_4191_begin_0, end = var_4191_end_0, end_mask = var_4191_end_mask_0, x = q_21_cast_fp16)[name = tensor("op_4191_cast_fp16")]; tensor var_4195_begin_0 = const()[name = tensor("op_4195_begin_0"), val = tensor([0, 80, 0, 0])]; tensor var_4195_end_0 = const()[name = tensor("op_4195_end_0"), val = tensor([2, 160, 1, 1024])]; tensor var_4195_end_mask_0 = const()[name = tensor("op_4195_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_4195_cast_fp16 = slice_by_index(begin = var_4195_begin_0, end = var_4195_end_0, end_mask = var_4195_end_mask_0, x = q_21_cast_fp16)[name = tensor("op_4195_cast_fp16")]; tensor var_4199_begin_0 = const()[name = tensor("op_4199_begin_0"), val = tensor([0, 160, 0, 0])]; tensor var_4199_end_0 = const()[name = tensor("op_4199_end_0"), val = tensor([2, 240, 1, 1024])]; tensor var_4199_end_mask_0 = const()[name = tensor("op_4199_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_4199_cast_fp16 = slice_by_index(begin = var_4199_begin_0, end = var_4199_end_0, end_mask = var_4199_end_mask_0, x = q_21_cast_fp16)[name = tensor("op_4199_cast_fp16")]; tensor var_4203_begin_0 = const()[name = tensor("op_4203_begin_0"), val = tensor([0, 240, 0, 0])]; tensor var_4203_end_0 = const()[name = tensor("op_4203_end_0"), val = tensor([2, 320, 1, 1024])]; tensor var_4203_end_mask_0 = const()[name = tensor("op_4203_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_4203_cast_fp16 = slice_by_index(begin = var_4203_begin_0, end = var_4203_end_0, end_mask = var_4203_end_mask_0, x = q_21_cast_fp16)[name = tensor("op_4203_cast_fp16")]; tensor var_4207_begin_0 = const()[name = tensor("op_4207_begin_0"), val = tensor([0, 320, 0, 0])]; tensor var_4207_end_0 = const()[name = tensor("op_4207_end_0"), val = tensor([2, 400, 1, 1024])]; tensor var_4207_end_mask_0 = const()[name = tensor("op_4207_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_4207_cast_fp16 = slice_by_index(begin = var_4207_begin_0, end = var_4207_end_0, end_mask = var_4207_end_mask_0, x = q_21_cast_fp16)[name = tensor("op_4207_cast_fp16")]; tensor var_4211_begin_0 = const()[name = tensor("op_4211_begin_0"), val = tensor([0, 400, 0, 0])]; tensor var_4211_end_0 = const()[name = tensor("op_4211_end_0"), val = tensor([2, 480, 1, 1024])]; tensor var_4211_end_mask_0 = const()[name = tensor("op_4211_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_4211_cast_fp16 = slice_by_index(begin = var_4211_begin_0, end = var_4211_end_0, end_mask = var_4211_end_mask_0, x = q_21_cast_fp16)[name = tensor("op_4211_cast_fp16")]; tensor var_4215_begin_0 = const()[name = tensor("op_4215_begin_0"), val = tensor([0, 480, 0, 0])]; tensor var_4215_end_0 = const()[name = tensor("op_4215_end_0"), val = tensor([2, 560, 1, 1024])]; tensor var_4215_end_mask_0 = const()[name = tensor("op_4215_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_4215_cast_fp16 = slice_by_index(begin = var_4215_begin_0, end = var_4215_end_0, end_mask = var_4215_end_mask_0, x = q_21_cast_fp16)[name = tensor("op_4215_cast_fp16")]; tensor var_4219_begin_0 = const()[name = tensor("op_4219_begin_0"), val = tensor([0, 560, 0, 0])]; tensor var_4219_end_0 = const()[name = tensor("op_4219_end_0"), val = tensor([2, 640, 1, 1024])]; tensor var_4219_end_mask_0 = const()[name = tensor("op_4219_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_4219_cast_fp16 = slice_by_index(begin = var_4219_begin_0, end = var_4219_end_0, end_mask = var_4219_end_mask_0, x = q_21_cast_fp16)[name = tensor("op_4219_cast_fp16")]; tensor var_4222_begin_0 = const()[name = tensor("op_4222_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_4222_end_0 = const()[name = tensor("op_4222_end_0"), val = tensor([2, 80, 1, 512])]; tensor var_4222_end_mask_0 = const()[name = tensor("op_4222_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4222_cast_fp16 = slice_by_index(begin = var_4222_begin_0, end = var_4222_end_0, end_mask = var_4222_end_mask_0, x = var_4191_cast_fp16)[name = tensor("op_4222_cast_fp16")]; tensor var_4223_begin_0 = const()[name = tensor("op_4223_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_4223_end_0 = const()[name = tensor("op_4223_end_0"), val = tensor([2, 80, 1, 1024])]; tensor var_4223_end_mask_0 = const()[name = tensor("op_4223_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4223_cast_fp16 = slice_by_index(begin = var_4223_begin_0, end = var_4223_end_0, end_mask = var_4223_end_mask_0, x = var_4191_cast_fp16)[name = tensor("op_4223_cast_fp16")]; tensor var_4224_begin_0 = const()[name = tensor("op_4224_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_4224_end_0 = const()[name = tensor("op_4224_end_0"), val = tensor([2, 80, 1, 512])]; tensor var_4224_end_mask_0 = const()[name = tensor("op_4224_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4224_cast_fp16 = slice_by_index(begin = var_4224_begin_0, end = var_4224_end_0, end_mask = var_4224_end_mask_0, x = var_4195_cast_fp16)[name = tensor("op_4224_cast_fp16")]; tensor var_4225_begin_0 = const()[name = tensor("op_4225_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_4225_end_0 = const()[name = tensor("op_4225_end_0"), val = tensor([2, 80, 1, 1024])]; tensor var_4225_end_mask_0 = const()[name = tensor("op_4225_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4225_cast_fp16 = slice_by_index(begin = var_4225_begin_0, end = var_4225_end_0, end_mask = var_4225_end_mask_0, x = var_4195_cast_fp16)[name = tensor("op_4225_cast_fp16")]; tensor var_4226_begin_0 = const()[name = tensor("op_4226_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_4226_end_0 = const()[name = tensor("op_4226_end_0"), val = tensor([2, 80, 1, 512])]; tensor var_4226_end_mask_0 = const()[name = tensor("op_4226_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4226_cast_fp16 = slice_by_index(begin = var_4226_begin_0, end = var_4226_end_0, end_mask = var_4226_end_mask_0, x = var_4199_cast_fp16)[name = tensor("op_4226_cast_fp16")]; tensor var_4227_begin_0 = const()[name = tensor("op_4227_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_4227_end_0 = const()[name = tensor("op_4227_end_0"), val = tensor([2, 80, 1, 1024])]; tensor var_4227_end_mask_0 = const()[name = tensor("op_4227_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4227_cast_fp16 = slice_by_index(begin = var_4227_begin_0, end = var_4227_end_0, end_mask = var_4227_end_mask_0, x = var_4199_cast_fp16)[name = tensor("op_4227_cast_fp16")]; tensor var_4228_begin_0 = const()[name = tensor("op_4228_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_4228_end_0 = const()[name = tensor("op_4228_end_0"), val = tensor([2, 80, 1, 512])]; tensor var_4228_end_mask_0 = const()[name = tensor("op_4228_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4228_cast_fp16 = slice_by_index(begin = var_4228_begin_0, end = var_4228_end_0, end_mask = var_4228_end_mask_0, x = var_4203_cast_fp16)[name = tensor("op_4228_cast_fp16")]; tensor var_4229_begin_0 = const()[name = tensor("op_4229_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_4229_end_0 = const()[name = tensor("op_4229_end_0"), val = tensor([2, 80, 1, 1024])]; tensor var_4229_end_mask_0 = const()[name = tensor("op_4229_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4229_cast_fp16 = slice_by_index(begin = var_4229_begin_0, end = var_4229_end_0, end_mask = var_4229_end_mask_0, x = var_4203_cast_fp16)[name = tensor("op_4229_cast_fp16")]; tensor var_4230_begin_0 = const()[name = tensor("op_4230_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_4230_end_0 = const()[name = tensor("op_4230_end_0"), val = tensor([2, 80, 1, 512])]; tensor var_4230_end_mask_0 = const()[name = tensor("op_4230_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4230_cast_fp16 = slice_by_index(begin = var_4230_begin_0, end = var_4230_end_0, end_mask = var_4230_end_mask_0, x = var_4207_cast_fp16)[name = tensor("op_4230_cast_fp16")]; tensor var_4231_begin_0 = const()[name = tensor("op_4231_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_4231_end_0 = const()[name = tensor("op_4231_end_0"), val = tensor([2, 80, 1, 1024])]; tensor var_4231_end_mask_0 = const()[name = tensor("op_4231_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4231_cast_fp16 = slice_by_index(begin = var_4231_begin_0, end = var_4231_end_0, end_mask = var_4231_end_mask_0, x = var_4207_cast_fp16)[name = tensor("op_4231_cast_fp16")]; tensor var_4232_begin_0 = const()[name = tensor("op_4232_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_4232_end_0 = const()[name = tensor("op_4232_end_0"), val = tensor([2, 80, 1, 512])]; tensor var_4232_end_mask_0 = const()[name = tensor("op_4232_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4232_cast_fp16 = slice_by_index(begin = var_4232_begin_0, end = var_4232_end_0, end_mask = var_4232_end_mask_0, x = var_4211_cast_fp16)[name = tensor("op_4232_cast_fp16")]; tensor var_4233_begin_0 = const()[name = tensor("op_4233_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_4233_end_0 = const()[name = tensor("op_4233_end_0"), val = tensor([2, 80, 1, 1024])]; tensor var_4233_end_mask_0 = const()[name = tensor("op_4233_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4233_cast_fp16 = slice_by_index(begin = var_4233_begin_0, end = var_4233_end_0, end_mask = var_4233_end_mask_0, x = var_4211_cast_fp16)[name = tensor("op_4233_cast_fp16")]; tensor var_4234_begin_0 = const()[name = tensor("op_4234_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_4234_end_0 = const()[name = tensor("op_4234_end_0"), val = tensor([2, 80, 1, 512])]; tensor var_4234_end_mask_0 = const()[name = tensor("op_4234_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4234_cast_fp16 = slice_by_index(begin = var_4234_begin_0, end = var_4234_end_0, end_mask = var_4234_end_mask_0, x = var_4215_cast_fp16)[name = tensor("op_4234_cast_fp16")]; tensor var_4235_begin_0 = const()[name = tensor("op_4235_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_4235_end_0 = const()[name = tensor("op_4235_end_0"), val = tensor([2, 80, 1, 1024])]; tensor var_4235_end_mask_0 = const()[name = tensor("op_4235_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4235_cast_fp16 = slice_by_index(begin = var_4235_begin_0, end = var_4235_end_0, end_mask = var_4235_end_mask_0, x = var_4215_cast_fp16)[name = tensor("op_4235_cast_fp16")]; tensor var_4236_begin_0 = const()[name = tensor("op_4236_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_4236_end_0 = const()[name = tensor("op_4236_end_0"), val = tensor([2, 80, 1, 512])]; tensor var_4236_end_mask_0 = const()[name = tensor("op_4236_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4236_cast_fp16 = slice_by_index(begin = var_4236_begin_0, end = var_4236_end_0, end_mask = var_4236_end_mask_0, x = var_4219_cast_fp16)[name = tensor("op_4236_cast_fp16")]; tensor var_4237_begin_0 = const()[name = tensor("op_4237_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_4237_end_0 = const()[name = tensor("op_4237_end_0"), val = tensor([2, 80, 1, 1024])]; tensor var_4237_end_mask_0 = const()[name = tensor("op_4237_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4237_cast_fp16 = slice_by_index(begin = var_4237_begin_0, end = var_4237_end_0, end_mask = var_4237_end_mask_0, x = var_4219_cast_fp16)[name = tensor("op_4237_cast_fp16")]; tensor k_43_perm_0 = const()[name = tensor("k_43_perm_0"), val = tensor([0, 3, 2, 1])]; tensor var_4242_begin_0 = const()[name = tensor("op_4242_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_4242_end_0 = const()[name = tensor("op_4242_end_0"), val = tensor([2, 1024, 1, 80])]; tensor var_4242_end_mask_0 = const()[name = tensor("op_4242_end_mask_0"), val = tensor([true, true, true, false])]; tensor transpose_7 = transpose(perm = k_43_perm_0, x = k_41_cast_fp16)[name = tensor("transpose_7")]; tensor var_4242_cast_fp16 = slice_by_index(begin = var_4242_begin_0, end = var_4242_end_0, end_mask = var_4242_end_mask_0, x = transpose_7)[name = tensor("op_4242_cast_fp16")]; tensor var_4246_begin_0 = const()[name = tensor("op_4246_begin_0"), val = tensor([0, 0, 0, 80])]; tensor var_4246_end_0 = const()[name = tensor("op_4246_end_0"), val = tensor([2, 1024, 1, 160])]; tensor var_4246_end_mask_0 = const()[name = tensor("op_4246_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4246_cast_fp16 = slice_by_index(begin = var_4246_begin_0, end = var_4246_end_0, end_mask = var_4246_end_mask_0, x = transpose_7)[name = tensor("op_4246_cast_fp16")]; tensor var_4250_begin_0 = const()[name = tensor("op_4250_begin_0"), val = tensor([0, 0, 0, 160])]; tensor var_4250_end_0 = const()[name = tensor("op_4250_end_0"), val = tensor([2, 1024, 1, 240])]; tensor var_4250_end_mask_0 = const()[name = tensor("op_4250_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4250_cast_fp16 = slice_by_index(begin = var_4250_begin_0, end = var_4250_end_0, end_mask = var_4250_end_mask_0, x = transpose_7)[name = tensor("op_4250_cast_fp16")]; tensor var_4254_begin_0 = const()[name = tensor("op_4254_begin_0"), val = tensor([0, 0, 0, 240])]; tensor var_4254_end_0 = const()[name = tensor("op_4254_end_0"), val = tensor([2, 1024, 1, 320])]; tensor var_4254_end_mask_0 = const()[name = tensor("op_4254_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4254_cast_fp16 = slice_by_index(begin = var_4254_begin_0, end = var_4254_end_0, end_mask = var_4254_end_mask_0, x = transpose_7)[name = tensor("op_4254_cast_fp16")]; tensor var_4258_begin_0 = const()[name = tensor("op_4258_begin_0"), val = tensor([0, 0, 0, 320])]; tensor var_4258_end_0 = const()[name = tensor("op_4258_end_0"), val = tensor([2, 1024, 1, 400])]; tensor var_4258_end_mask_0 = const()[name = tensor("op_4258_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4258_cast_fp16 = slice_by_index(begin = var_4258_begin_0, end = var_4258_end_0, end_mask = var_4258_end_mask_0, x = transpose_7)[name = tensor("op_4258_cast_fp16")]; tensor var_4262_begin_0 = const()[name = tensor("op_4262_begin_0"), val = tensor([0, 0, 0, 400])]; tensor var_4262_end_0 = const()[name = tensor("op_4262_end_0"), val = tensor([2, 1024, 1, 480])]; tensor var_4262_end_mask_0 = const()[name = tensor("op_4262_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4262_cast_fp16 = slice_by_index(begin = var_4262_begin_0, end = var_4262_end_0, end_mask = var_4262_end_mask_0, x = transpose_7)[name = tensor("op_4262_cast_fp16")]; tensor var_4266_begin_0 = const()[name = tensor("op_4266_begin_0"), val = tensor([0, 0, 0, 480])]; tensor var_4266_end_0 = const()[name = tensor("op_4266_end_0"), val = tensor([2, 1024, 1, 560])]; tensor var_4266_end_mask_0 = const()[name = tensor("op_4266_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4266_cast_fp16 = slice_by_index(begin = var_4266_begin_0, end = var_4266_end_0, end_mask = var_4266_end_mask_0, x = transpose_7)[name = tensor("op_4266_cast_fp16")]; tensor var_4270_begin_0 = const()[name = tensor("op_4270_begin_0"), val = tensor([0, 0, 0, 560])]; tensor var_4270_end_0 = const()[name = tensor("op_4270_end_0"), val = tensor([2, 1024, 1, 640])]; tensor var_4270_end_mask_0 = const()[name = tensor("op_4270_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4270_cast_fp16 = slice_by_index(begin = var_4270_begin_0, end = var_4270_end_0, end_mask = var_4270_end_mask_0, x = transpose_7)[name = tensor("op_4270_cast_fp16")]; tensor var_4272_begin_0 = const()[name = tensor("op_4272_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_4272_end_0 = const()[name = tensor("op_4272_end_0"), val = tensor([2, 80, 1, 1024])]; tensor var_4272_end_mask_0 = const()[name = tensor("op_4272_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_4272_cast_fp16 = slice_by_index(begin = var_4272_begin_0, end = var_4272_end_0, end_mask = var_4272_end_mask_0, x = v_21_cast_fp16)[name = tensor("op_4272_cast_fp16")]; tensor var_4276_begin_0 = const()[name = tensor("op_4276_begin_0"), val = tensor([0, 80, 0, 0])]; tensor var_4276_end_0 = const()[name = tensor("op_4276_end_0"), val = tensor([2, 160, 1, 1024])]; tensor var_4276_end_mask_0 = const()[name = tensor("op_4276_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_4276_cast_fp16 = slice_by_index(begin = var_4276_begin_0, end = var_4276_end_0, end_mask = var_4276_end_mask_0, x = v_21_cast_fp16)[name = tensor("op_4276_cast_fp16")]; tensor var_4280_begin_0 = const()[name = tensor("op_4280_begin_0"), val = tensor([0, 160, 0, 0])]; tensor var_4280_end_0 = const()[name = tensor("op_4280_end_0"), val = tensor([2, 240, 1, 1024])]; tensor var_4280_end_mask_0 = const()[name = tensor("op_4280_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_4280_cast_fp16 = slice_by_index(begin = var_4280_begin_0, end = var_4280_end_0, end_mask = var_4280_end_mask_0, x = v_21_cast_fp16)[name = tensor("op_4280_cast_fp16")]; tensor var_4284_begin_0 = const()[name = tensor("op_4284_begin_0"), val = tensor([0, 240, 0, 0])]; tensor var_4284_end_0 = const()[name = tensor("op_4284_end_0"), val = tensor([2, 320, 1, 1024])]; tensor var_4284_end_mask_0 = const()[name = tensor("op_4284_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_4284_cast_fp16 = slice_by_index(begin = var_4284_begin_0, end = var_4284_end_0, end_mask = var_4284_end_mask_0, x = v_21_cast_fp16)[name = tensor("op_4284_cast_fp16")]; tensor var_4288_begin_0 = const()[name = tensor("op_4288_begin_0"), val = tensor([0, 320, 0, 0])]; tensor var_4288_end_0 = const()[name = tensor("op_4288_end_0"), val = tensor([2, 400, 1, 1024])]; tensor var_4288_end_mask_0 = const()[name = tensor("op_4288_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_4288_cast_fp16 = slice_by_index(begin = var_4288_begin_0, end = var_4288_end_0, end_mask = var_4288_end_mask_0, x = v_21_cast_fp16)[name = tensor("op_4288_cast_fp16")]; tensor var_4292_begin_0 = const()[name = tensor("op_4292_begin_0"), val = tensor([0, 400, 0, 0])]; tensor var_4292_end_0 = const()[name = tensor("op_4292_end_0"), val = tensor([2, 480, 1, 1024])]; tensor var_4292_end_mask_0 = const()[name = tensor("op_4292_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_4292_cast_fp16 = slice_by_index(begin = var_4292_begin_0, end = var_4292_end_0, end_mask = var_4292_end_mask_0, x = v_21_cast_fp16)[name = tensor("op_4292_cast_fp16")]; tensor var_4296_begin_0 = const()[name = tensor("op_4296_begin_0"), val = tensor([0, 480, 0, 0])]; tensor var_4296_end_0 = const()[name = tensor("op_4296_end_0"), val = tensor([2, 560, 1, 1024])]; tensor var_4296_end_mask_0 = const()[name = tensor("op_4296_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_4296_cast_fp16 = slice_by_index(begin = var_4296_begin_0, end = var_4296_end_0, end_mask = var_4296_end_mask_0, x = v_21_cast_fp16)[name = tensor("op_4296_cast_fp16")]; tensor var_4300_begin_0 = const()[name = tensor("op_4300_begin_0"), val = tensor([0, 560, 0, 0])]; tensor var_4300_end_0 = const()[name = tensor("op_4300_end_0"), val = tensor([2, 640, 1, 1024])]; tensor var_4300_end_mask_0 = const()[name = tensor("op_4300_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_4300_cast_fp16 = slice_by_index(begin = var_4300_begin_0, end = var_4300_end_0, end_mask = var_4300_end_mask_0, x = v_21_cast_fp16)[name = tensor("op_4300_cast_fp16")]; tensor var_4304_equation_0 = const()[name = tensor("op_4304_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_4304_cast_fp16 = einsum(equation = var_4304_equation_0, values = (var_4242_cast_fp16, var_4222_cast_fp16))[name = tensor("op_4304_cast_fp16")]; tensor var_4305_to_fp16 = const()[name = tensor("op_4305_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_321_cast_fp16 = mul(x = var_4304_cast_fp16, y = var_4305_to_fp16)[name = tensor("aw_chunk_321_cast_fp16")]; tensor var_4308_equation_0 = const()[name = tensor("op_4308_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_4308_cast_fp16 = einsum(equation = var_4308_equation_0, values = (var_4242_cast_fp16, var_4223_cast_fp16))[name = tensor("op_4308_cast_fp16")]; tensor var_4309_to_fp16 = const()[name = tensor("op_4309_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_323_cast_fp16 = mul(x = var_4308_cast_fp16, y = var_4309_to_fp16)[name = tensor("aw_chunk_323_cast_fp16")]; tensor var_4312_equation_0 = const()[name = tensor("op_4312_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_4312_cast_fp16 = einsum(equation = var_4312_equation_0, values = (var_4246_cast_fp16, var_4224_cast_fp16))[name = tensor("op_4312_cast_fp16")]; tensor var_4313_to_fp16 = const()[name = tensor("op_4313_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_325_cast_fp16 = mul(x = var_4312_cast_fp16, y = var_4313_to_fp16)[name = tensor("aw_chunk_325_cast_fp16")]; tensor var_4316_equation_0 = const()[name = tensor("op_4316_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_4316_cast_fp16 = einsum(equation = var_4316_equation_0, values = (var_4246_cast_fp16, var_4225_cast_fp16))[name = tensor("op_4316_cast_fp16")]; tensor var_4317_to_fp16 = const()[name = tensor("op_4317_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_327_cast_fp16 = mul(x = var_4316_cast_fp16, y = var_4317_to_fp16)[name = tensor("aw_chunk_327_cast_fp16")]; tensor var_4320_equation_0 = const()[name = tensor("op_4320_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_4320_cast_fp16 = einsum(equation = var_4320_equation_0, values = (var_4250_cast_fp16, var_4226_cast_fp16))[name = tensor("op_4320_cast_fp16")]; tensor var_4321_to_fp16 = const()[name = tensor("op_4321_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_329_cast_fp16 = mul(x = var_4320_cast_fp16, y = var_4321_to_fp16)[name = tensor("aw_chunk_329_cast_fp16")]; tensor var_4324_equation_0 = const()[name = tensor("op_4324_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_4324_cast_fp16 = einsum(equation = var_4324_equation_0, values = (var_4250_cast_fp16, var_4227_cast_fp16))[name = tensor("op_4324_cast_fp16")]; tensor var_4325_to_fp16 = const()[name = tensor("op_4325_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_331_cast_fp16 = mul(x = var_4324_cast_fp16, y = var_4325_to_fp16)[name = tensor("aw_chunk_331_cast_fp16")]; tensor var_4328_equation_0 = const()[name = tensor("op_4328_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_4328_cast_fp16 = einsum(equation = var_4328_equation_0, values = (var_4254_cast_fp16, var_4228_cast_fp16))[name = tensor("op_4328_cast_fp16")]; tensor var_4329_to_fp16 = const()[name = tensor("op_4329_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_333_cast_fp16 = mul(x = var_4328_cast_fp16, y = var_4329_to_fp16)[name = tensor("aw_chunk_333_cast_fp16")]; tensor var_4332_equation_0 = const()[name = tensor("op_4332_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_4332_cast_fp16 = einsum(equation = var_4332_equation_0, values = (var_4254_cast_fp16, var_4229_cast_fp16))[name = tensor("op_4332_cast_fp16")]; tensor var_4333_to_fp16 = const()[name = tensor("op_4333_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_335_cast_fp16 = mul(x = var_4332_cast_fp16, y = var_4333_to_fp16)[name = tensor("aw_chunk_335_cast_fp16")]; tensor var_4336_equation_0 = const()[name = tensor("op_4336_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_4336_cast_fp16 = einsum(equation = var_4336_equation_0, values = (var_4258_cast_fp16, var_4230_cast_fp16))[name = tensor("op_4336_cast_fp16")]; tensor var_4337_to_fp16 = const()[name = tensor("op_4337_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_337_cast_fp16 = mul(x = var_4336_cast_fp16, y = var_4337_to_fp16)[name = tensor("aw_chunk_337_cast_fp16")]; tensor var_4340_equation_0 = const()[name = tensor("op_4340_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_4340_cast_fp16 = einsum(equation = var_4340_equation_0, values = (var_4258_cast_fp16, var_4231_cast_fp16))[name = tensor("op_4340_cast_fp16")]; tensor var_4341_to_fp16 = const()[name = tensor("op_4341_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_339_cast_fp16 = mul(x = var_4340_cast_fp16, y = var_4341_to_fp16)[name = tensor("aw_chunk_339_cast_fp16")]; tensor var_4344_equation_0 = const()[name = tensor("op_4344_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_4344_cast_fp16 = einsum(equation = var_4344_equation_0, values = (var_4262_cast_fp16, var_4232_cast_fp16))[name = tensor("op_4344_cast_fp16")]; tensor var_4345_to_fp16 = const()[name = tensor("op_4345_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_341_cast_fp16 = mul(x = var_4344_cast_fp16, y = var_4345_to_fp16)[name = tensor("aw_chunk_341_cast_fp16")]; tensor var_4348_equation_0 = const()[name = tensor("op_4348_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_4348_cast_fp16 = einsum(equation = var_4348_equation_0, values = (var_4262_cast_fp16, var_4233_cast_fp16))[name = tensor("op_4348_cast_fp16")]; tensor var_4349_to_fp16 = const()[name = tensor("op_4349_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_343_cast_fp16 = mul(x = var_4348_cast_fp16, y = var_4349_to_fp16)[name = tensor("aw_chunk_343_cast_fp16")]; tensor var_4352_equation_0 = const()[name = tensor("op_4352_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_4352_cast_fp16 = einsum(equation = var_4352_equation_0, values = (var_4266_cast_fp16, var_4234_cast_fp16))[name = tensor("op_4352_cast_fp16")]; tensor var_4353_to_fp16 = const()[name = tensor("op_4353_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_345_cast_fp16 = mul(x = var_4352_cast_fp16, y = var_4353_to_fp16)[name = tensor("aw_chunk_345_cast_fp16")]; tensor var_4356_equation_0 = const()[name = tensor("op_4356_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_4356_cast_fp16 = einsum(equation = var_4356_equation_0, values = (var_4266_cast_fp16, var_4235_cast_fp16))[name = tensor("op_4356_cast_fp16")]; tensor var_4357_to_fp16 = const()[name = tensor("op_4357_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_347_cast_fp16 = mul(x = var_4356_cast_fp16, y = var_4357_to_fp16)[name = tensor("aw_chunk_347_cast_fp16")]; tensor var_4360_equation_0 = const()[name = tensor("op_4360_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_4360_cast_fp16 = einsum(equation = var_4360_equation_0, values = (var_4270_cast_fp16, var_4236_cast_fp16))[name = tensor("op_4360_cast_fp16")]; tensor var_4361_to_fp16 = const()[name = tensor("op_4361_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_349_cast_fp16 = mul(x = var_4360_cast_fp16, y = var_4361_to_fp16)[name = tensor("aw_chunk_349_cast_fp16")]; tensor var_4364_equation_0 = const()[name = tensor("op_4364_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_4364_cast_fp16 = einsum(equation = var_4364_equation_0, values = (var_4270_cast_fp16, var_4237_cast_fp16))[name = tensor("op_4364_cast_fp16")]; tensor var_4365_to_fp16 = const()[name = tensor("op_4365_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_351_cast_fp16 = mul(x = var_4364_cast_fp16, y = var_4365_to_fp16)[name = tensor("aw_chunk_351_cast_fp16")]; tensor var_4367_cast_fp16 = softmax(axis = var_4061, x = aw_chunk_321_cast_fp16)[name = tensor("op_4367_cast_fp16")]; tensor var_4368_cast_fp16 = softmax(axis = var_4061, x = aw_chunk_323_cast_fp16)[name = tensor("op_4368_cast_fp16")]; tensor var_4369_cast_fp16 = softmax(axis = var_4061, x = aw_chunk_325_cast_fp16)[name = tensor("op_4369_cast_fp16")]; tensor var_4370_cast_fp16 = softmax(axis = var_4061, x = aw_chunk_327_cast_fp16)[name = tensor("op_4370_cast_fp16")]; tensor var_4371_cast_fp16 = softmax(axis = var_4061, x = aw_chunk_329_cast_fp16)[name = tensor("op_4371_cast_fp16")]; tensor var_4372_cast_fp16 = softmax(axis = var_4061, x = aw_chunk_331_cast_fp16)[name = tensor("op_4372_cast_fp16")]; tensor var_4373_cast_fp16 = softmax(axis = var_4061, x = aw_chunk_333_cast_fp16)[name = tensor("op_4373_cast_fp16")]; tensor var_4374_cast_fp16 = softmax(axis = var_4061, x = aw_chunk_335_cast_fp16)[name = tensor("op_4374_cast_fp16")]; tensor var_4375_cast_fp16 = softmax(axis = var_4061, x = aw_chunk_337_cast_fp16)[name = tensor("op_4375_cast_fp16")]; tensor var_4376_cast_fp16 = softmax(axis = var_4061, x = aw_chunk_339_cast_fp16)[name = tensor("op_4376_cast_fp16")]; tensor var_4377_cast_fp16 = softmax(axis = var_4061, x = aw_chunk_341_cast_fp16)[name = tensor("op_4377_cast_fp16")]; tensor var_4378_cast_fp16 = softmax(axis = var_4061, x = aw_chunk_343_cast_fp16)[name = tensor("op_4378_cast_fp16")]; tensor var_4379_cast_fp16 = softmax(axis = var_4061, x = aw_chunk_345_cast_fp16)[name = tensor("op_4379_cast_fp16")]; tensor var_4380_cast_fp16 = softmax(axis = var_4061, x = aw_chunk_347_cast_fp16)[name = tensor("op_4380_cast_fp16")]; tensor var_4381_cast_fp16 = softmax(axis = var_4061, x = aw_chunk_349_cast_fp16)[name = tensor("op_4381_cast_fp16")]; tensor var_4382_cast_fp16 = softmax(axis = var_4061, x = aw_chunk_351_cast_fp16)[name = tensor("op_4382_cast_fp16")]; tensor var_4384_equation_0 = const()[name = tensor("op_4384_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_4384_cast_fp16 = einsum(equation = var_4384_equation_0, values = (var_4272_cast_fp16, var_4367_cast_fp16))[name = tensor("op_4384_cast_fp16")]; tensor var_4386_equation_0 = const()[name = tensor("op_4386_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_4386_cast_fp16 = einsum(equation = var_4386_equation_0, values = (var_4272_cast_fp16, var_4368_cast_fp16))[name = tensor("op_4386_cast_fp16")]; tensor var_4388_equation_0 = const()[name = tensor("op_4388_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_4388_cast_fp16 = einsum(equation = var_4388_equation_0, values = (var_4276_cast_fp16, var_4369_cast_fp16))[name = tensor("op_4388_cast_fp16")]; tensor var_4390_equation_0 = const()[name = tensor("op_4390_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_4390_cast_fp16 = einsum(equation = var_4390_equation_0, values = (var_4276_cast_fp16, var_4370_cast_fp16))[name = tensor("op_4390_cast_fp16")]; tensor var_4392_equation_0 = const()[name = tensor("op_4392_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_4392_cast_fp16 = einsum(equation = var_4392_equation_0, values = (var_4280_cast_fp16, var_4371_cast_fp16))[name = tensor("op_4392_cast_fp16")]; tensor var_4394_equation_0 = const()[name = tensor("op_4394_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_4394_cast_fp16 = einsum(equation = var_4394_equation_0, values = (var_4280_cast_fp16, var_4372_cast_fp16))[name = tensor("op_4394_cast_fp16")]; tensor var_4396_equation_0 = const()[name = tensor("op_4396_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_4396_cast_fp16 = einsum(equation = var_4396_equation_0, values = (var_4284_cast_fp16, var_4373_cast_fp16))[name = tensor("op_4396_cast_fp16")]; tensor var_4398_equation_0 = const()[name = tensor("op_4398_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_4398_cast_fp16 = einsum(equation = var_4398_equation_0, values = (var_4284_cast_fp16, var_4374_cast_fp16))[name = tensor("op_4398_cast_fp16")]; tensor var_4400_equation_0 = const()[name = tensor("op_4400_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_4400_cast_fp16 = einsum(equation = var_4400_equation_0, values = (var_4288_cast_fp16, var_4375_cast_fp16))[name = tensor("op_4400_cast_fp16")]; tensor var_4402_equation_0 = const()[name = tensor("op_4402_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_4402_cast_fp16 = einsum(equation = var_4402_equation_0, values = (var_4288_cast_fp16, var_4376_cast_fp16))[name = tensor("op_4402_cast_fp16")]; tensor var_4404_equation_0 = const()[name = tensor("op_4404_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_4404_cast_fp16 = einsum(equation = var_4404_equation_0, values = (var_4292_cast_fp16, var_4377_cast_fp16))[name = tensor("op_4404_cast_fp16")]; tensor var_4406_equation_0 = const()[name = tensor("op_4406_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_4406_cast_fp16 = einsum(equation = var_4406_equation_0, values = (var_4292_cast_fp16, var_4378_cast_fp16))[name = tensor("op_4406_cast_fp16")]; tensor var_4408_equation_0 = const()[name = tensor("op_4408_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_4408_cast_fp16 = einsum(equation = var_4408_equation_0, values = (var_4296_cast_fp16, var_4379_cast_fp16))[name = tensor("op_4408_cast_fp16")]; tensor var_4410_equation_0 = const()[name = tensor("op_4410_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_4410_cast_fp16 = einsum(equation = var_4410_equation_0, values = (var_4296_cast_fp16, var_4380_cast_fp16))[name = tensor("op_4410_cast_fp16")]; tensor var_4412_equation_0 = const()[name = tensor("op_4412_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_4412_cast_fp16 = einsum(equation = var_4412_equation_0, values = (var_4300_cast_fp16, var_4381_cast_fp16))[name = tensor("op_4412_cast_fp16")]; tensor var_4414_equation_0 = const()[name = tensor("op_4414_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_4414_cast_fp16 = einsum(equation = var_4414_equation_0, values = (var_4300_cast_fp16, var_4382_cast_fp16))[name = tensor("op_4414_cast_fp16")]; tensor var_4416_interleave_0 = const()[name = tensor("op_4416_interleave_0"), val = tensor(false)]; tensor var_4416_cast_fp16 = concat(axis = var_4039, interleave = var_4416_interleave_0, values = (var_4384_cast_fp16, var_4386_cast_fp16))[name = tensor("op_4416_cast_fp16")]; tensor var_4418_interleave_0 = const()[name = tensor("op_4418_interleave_0"), val = tensor(false)]; tensor var_4418_cast_fp16 = concat(axis = var_4039, interleave = var_4418_interleave_0, values = (var_4388_cast_fp16, var_4390_cast_fp16))[name = tensor("op_4418_cast_fp16")]; tensor var_4420_interleave_0 = const()[name = tensor("op_4420_interleave_0"), val = tensor(false)]; tensor var_4420_cast_fp16 = concat(axis = var_4039, interleave = var_4420_interleave_0, values = (var_4392_cast_fp16, var_4394_cast_fp16))[name = tensor("op_4420_cast_fp16")]; tensor var_4422_interleave_0 = const()[name = tensor("op_4422_interleave_0"), val = tensor(false)]; tensor var_4422_cast_fp16 = concat(axis = var_4039, interleave = var_4422_interleave_0, values = (var_4396_cast_fp16, var_4398_cast_fp16))[name = tensor("op_4422_cast_fp16")]; tensor var_4424_interleave_0 = const()[name = tensor("op_4424_interleave_0"), val = tensor(false)]; tensor var_4424_cast_fp16 = concat(axis = var_4039, interleave = var_4424_interleave_0, values = (var_4400_cast_fp16, var_4402_cast_fp16))[name = tensor("op_4424_cast_fp16")]; tensor var_4426_interleave_0 = const()[name = tensor("op_4426_interleave_0"), val = tensor(false)]; tensor var_4426_cast_fp16 = concat(axis = var_4039, interleave = var_4426_interleave_0, values = (var_4404_cast_fp16, var_4406_cast_fp16))[name = tensor("op_4426_cast_fp16")]; tensor var_4428_interleave_0 = const()[name = tensor("op_4428_interleave_0"), val = tensor(false)]; tensor var_4428_cast_fp16 = concat(axis = var_4039, interleave = var_4428_interleave_0, values = (var_4408_cast_fp16, var_4410_cast_fp16))[name = tensor("op_4428_cast_fp16")]; tensor var_4430_interleave_0 = const()[name = tensor("op_4430_interleave_0"), val = tensor(false)]; tensor var_4430_cast_fp16 = concat(axis = var_4039, interleave = var_4430_interleave_0, values = (var_4412_cast_fp16, var_4414_cast_fp16))[name = tensor("op_4430_cast_fp16")]; tensor input_163_interleave_0 = const()[name = tensor("input_163_interleave_0"), val = tensor(false)]; tensor input_163_cast_fp16 = concat(axis = var_4061, interleave = input_163_interleave_0, values = (var_4416_cast_fp16, var_4418_cast_fp16, var_4420_cast_fp16, var_4422_cast_fp16, var_4424_cast_fp16, var_4426_cast_fp16, var_4428_cast_fp16, var_4430_cast_fp16))[name = tensor("input_163_cast_fp16")]; tensor var_4436 = const()[name = tensor("op_4436"), val = tensor([1, 1])]; tensor var_4438 = const()[name = tensor("op_4438"), val = tensor([1, 1])]; tensor var_4440_pad_type_0 = const()[name = tensor("op_4440_pad_type_0"), val = tensor("custom")]; tensor var_4440_pad_0 = const()[name = tensor("op_4440_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(209756736))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(210064000))), name = tensor("up_blocks_1_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16_palettized"), shape = tensor([640, 640, 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(210064192)))]; tensor var_4440_cast_fp16 = conv(bias = up_blocks_1_attentions_0_transformer_blocks_0_attn1_to_out_0_bias_to_fp16, dilations = var_4438, groups = var_4061, pad = var_4440_pad_0, pad_type = var_4440_pad_type_0, strides = var_4436, weight = up_blocks_1_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16_palettized, x = input_163_cast_fp16)[name = tensor("op_4440_cast_fp16")]; tensor inputs_33_cast_fp16 = add(x = var_4440_cast_fp16, y = inputs_31_cast_fp16)[name = tensor("inputs_33_cast_fp16")]; tensor var_4444 = const()[name = tensor("op_4444"), val = tensor([1])]; tensor channels_mean_33_cast_fp16 = reduce_mean(axes = var_4444, keep_dims = var_4056, x = inputs_33_cast_fp16)[name = tensor("channels_mean_33_cast_fp16")]; tensor zero_mean_33_cast_fp16 = sub(x = inputs_33_cast_fp16, y = channels_mean_33_cast_fp16)[name = tensor("zero_mean_33_cast_fp16")]; tensor zero_mean_sq_33_cast_fp16 = mul(x = zero_mean_33_cast_fp16, y = zero_mean_33_cast_fp16)[name = tensor("zero_mean_sq_33_cast_fp16")]; tensor var_4448 = const()[name = tensor("op_4448"), val = tensor([1])]; tensor var_4449_cast_fp16 = reduce_mean(axes = var_4448, keep_dims = var_4056, x = zero_mean_sq_33_cast_fp16)[name = tensor("op_4449_cast_fp16")]; tensor var_4450_to_fp16 = const()[name = tensor("op_4450_to_fp16"), val = tensor(0x1.5p-17)]; tensor var_4451_cast_fp16 = add(x = var_4449_cast_fp16, y = var_4450_to_fp16)[name = tensor("op_4451_cast_fp16")]; tensor denom_33_epsilon_0_to_fp16 = const()[name = tensor("denom_33_epsilon_0_to_fp16"), val = tensor(0x1p-24)]; tensor denom_33_cast_fp16 = rsqrt(epsilon = denom_33_epsilon_0_to_fp16, x = var_4451_cast_fp16)[name = tensor("denom_33_cast_fp16")]; tensor out_33_cast_fp16 = mul(x = zero_mean_33_cast_fp16, y = denom_33_cast_fp16)[name = tensor("out_33_cast_fp16")]; tensor var_4455_to_fp16 = const()[name = tensor("op_4455_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(210065536)))]; tensor var_4456_cast_fp16 = add(x = out_33_cast_fp16, y = var_4455_to_fp16)[name = tensor("op_4456_cast_fp16")]; tensor var_4458_to_fp16 = const()[name = tensor("op_4458_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(210066880)))]; tensor hidden_states_107_cast_fp16 = mul(x = var_4456_cast_fp16, y = var_4458_to_fp16)[name = tensor("hidden_states_107_cast_fp16")]; tensor var_4465 = const()[name = tensor("op_4465"), val = tensor([1, 1])]; tensor var_4467 = const()[name = tensor("op_4467"), 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 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(210068224))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(210375488))), name = tensor("up_blocks_1_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16_palettized"), shape = tensor([640, 640, 1, 1])]; tensor q_23_cast_fp16 = conv(dilations = var_4467, groups = var_4061, pad = q_23_pad_0, pad_type = q_23_pad_type_0, strides = var_4465, weight = up_blocks_1_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16_palettized, x = hidden_states_107_cast_fp16)[name = tensor("q_23_cast_fp16")]; tensor var_4471 = const()[name = tensor("op_4471"), val = tensor([1, 1])]; tensor var_4473 = const()[name = tensor("op_4473"), 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 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(210375680))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(210744384))), name = tensor("up_blocks_1_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized"), shape = tensor([640, 768, 1, 1])]; tensor k_45_cast_fp16 = conv(dilations = var_4473, groups = var_4061, pad = k_45_pad_0, pad_type = k_45_pad_type_0, strides = var_4471, weight = up_blocks_1_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor("k_45_cast_fp16")]; tensor var_4477 = const()[name = tensor("op_4477"), val = tensor([1, 1])]; tensor var_4479 = const()[name = tensor("op_4479"), 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 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(210744576))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(211113280))), name = tensor("up_blocks_1_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized"), shape = tensor([640, 768, 1, 1])]; tensor v_23_cast_fp16 = conv(dilations = var_4479, groups = var_4061, pad = v_23_pad_0, pad_type = v_23_pad_type_0, strides = var_4477, weight = up_blocks_1_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor("v_23_cast_fp16")]; tensor var_4483_begin_0 = const()[name = tensor("op_4483_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_4483_end_0 = const()[name = tensor("op_4483_end_0"), val = tensor([2, 80, 1, 1024])]; tensor var_4483_end_mask_0 = const()[name = tensor("op_4483_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_4483_cast_fp16 = slice_by_index(begin = var_4483_begin_0, end = var_4483_end_0, end_mask = var_4483_end_mask_0, x = q_23_cast_fp16)[name = tensor("op_4483_cast_fp16")]; tensor var_4487_begin_0 = const()[name = tensor("op_4487_begin_0"), val = tensor([0, 80, 0, 0])]; tensor var_4487_end_0 = const()[name = tensor("op_4487_end_0"), val = tensor([2, 160, 1, 1024])]; tensor var_4487_end_mask_0 = const()[name = tensor("op_4487_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_4487_cast_fp16 = slice_by_index(begin = var_4487_begin_0, end = var_4487_end_0, end_mask = var_4487_end_mask_0, x = q_23_cast_fp16)[name = tensor("op_4487_cast_fp16")]; tensor var_4491_begin_0 = const()[name = tensor("op_4491_begin_0"), val = tensor([0, 160, 0, 0])]; tensor var_4491_end_0 = const()[name = tensor("op_4491_end_0"), val = tensor([2, 240, 1, 1024])]; tensor var_4491_end_mask_0 = const()[name = tensor("op_4491_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_4491_cast_fp16 = slice_by_index(begin = var_4491_begin_0, end = var_4491_end_0, end_mask = var_4491_end_mask_0, x = q_23_cast_fp16)[name = tensor("op_4491_cast_fp16")]; tensor var_4495_begin_0 = const()[name = tensor("op_4495_begin_0"), val = tensor([0, 240, 0, 0])]; tensor var_4495_end_0 = const()[name = tensor("op_4495_end_0"), val = tensor([2, 320, 1, 1024])]; tensor var_4495_end_mask_0 = const()[name = tensor("op_4495_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_4495_cast_fp16 = slice_by_index(begin = var_4495_begin_0, end = var_4495_end_0, end_mask = var_4495_end_mask_0, x = q_23_cast_fp16)[name = tensor("op_4495_cast_fp16")]; tensor var_4499_begin_0 = const()[name = tensor("op_4499_begin_0"), val = tensor([0, 320, 0, 0])]; tensor var_4499_end_0 = const()[name = tensor("op_4499_end_0"), val = tensor([2, 400, 1, 1024])]; tensor var_4499_end_mask_0 = const()[name = tensor("op_4499_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_4499_cast_fp16 = slice_by_index(begin = var_4499_begin_0, end = var_4499_end_0, end_mask = var_4499_end_mask_0, x = q_23_cast_fp16)[name = tensor("op_4499_cast_fp16")]; tensor var_4503_begin_0 = const()[name = tensor("op_4503_begin_0"), val = tensor([0, 400, 0, 0])]; tensor var_4503_end_0 = const()[name = tensor("op_4503_end_0"), val = tensor([2, 480, 1, 1024])]; tensor var_4503_end_mask_0 = const()[name = tensor("op_4503_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_4503_cast_fp16 = slice_by_index(begin = var_4503_begin_0, end = var_4503_end_0, end_mask = var_4503_end_mask_0, x = q_23_cast_fp16)[name = tensor("op_4503_cast_fp16")]; tensor var_4507_begin_0 = const()[name = tensor("op_4507_begin_0"), val = tensor([0, 480, 0, 0])]; tensor var_4507_end_0 = const()[name = tensor("op_4507_end_0"), val = tensor([2, 560, 1, 1024])]; tensor var_4507_end_mask_0 = const()[name = tensor("op_4507_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_4507_cast_fp16 = slice_by_index(begin = var_4507_begin_0, end = var_4507_end_0, end_mask = var_4507_end_mask_0, x = q_23_cast_fp16)[name = tensor("op_4507_cast_fp16")]; tensor var_4511_begin_0 = const()[name = tensor("op_4511_begin_0"), val = tensor([0, 560, 0, 0])]; tensor var_4511_end_0 = const()[name = tensor("op_4511_end_0"), val = tensor([2, 640, 1, 1024])]; tensor var_4511_end_mask_0 = const()[name = tensor("op_4511_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_4511_cast_fp16 = slice_by_index(begin = var_4511_begin_0, end = var_4511_end_0, end_mask = var_4511_end_mask_0, x = q_23_cast_fp16)[name = tensor("op_4511_cast_fp16")]; tensor var_4514_begin_0 = const()[name = tensor("op_4514_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_4514_end_0 = const()[name = tensor("op_4514_end_0"), val = tensor([2, 80, 1, 512])]; tensor var_4514_end_mask_0 = const()[name = tensor("op_4514_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4514_cast_fp16 = slice_by_index(begin = var_4514_begin_0, end = var_4514_end_0, end_mask = var_4514_end_mask_0, x = var_4483_cast_fp16)[name = tensor("op_4514_cast_fp16")]; tensor var_4515_begin_0 = const()[name = tensor("op_4515_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_4515_end_0 = const()[name = tensor("op_4515_end_0"), val = tensor([2, 80, 1, 1024])]; tensor var_4515_end_mask_0 = const()[name = tensor("op_4515_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4515_cast_fp16 = slice_by_index(begin = var_4515_begin_0, end = var_4515_end_0, end_mask = var_4515_end_mask_0, x = var_4483_cast_fp16)[name = tensor("op_4515_cast_fp16")]; tensor var_4516_begin_0 = const()[name = tensor("op_4516_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_4516_end_0 = const()[name = tensor("op_4516_end_0"), val = tensor([2, 80, 1, 512])]; tensor var_4516_end_mask_0 = const()[name = tensor("op_4516_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4516_cast_fp16 = slice_by_index(begin = var_4516_begin_0, end = var_4516_end_0, end_mask = var_4516_end_mask_0, x = var_4487_cast_fp16)[name = tensor("op_4516_cast_fp16")]; tensor var_4517_begin_0 = const()[name = tensor("op_4517_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_4517_end_0 = const()[name = tensor("op_4517_end_0"), val = tensor([2, 80, 1, 1024])]; tensor var_4517_end_mask_0 = const()[name = tensor("op_4517_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4517_cast_fp16 = slice_by_index(begin = var_4517_begin_0, end = var_4517_end_0, end_mask = var_4517_end_mask_0, x = var_4487_cast_fp16)[name = tensor("op_4517_cast_fp16")]; tensor var_4518_begin_0 = const()[name = tensor("op_4518_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_4518_end_0 = const()[name = tensor("op_4518_end_0"), val = tensor([2, 80, 1, 512])]; tensor var_4518_end_mask_0 = const()[name = tensor("op_4518_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4518_cast_fp16 = slice_by_index(begin = var_4518_begin_0, end = var_4518_end_0, end_mask = var_4518_end_mask_0, x = var_4491_cast_fp16)[name = tensor("op_4518_cast_fp16")]; tensor var_4519_begin_0 = const()[name = tensor("op_4519_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_4519_end_0 = const()[name = tensor("op_4519_end_0"), val = tensor([2, 80, 1, 1024])]; tensor var_4519_end_mask_0 = const()[name = tensor("op_4519_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4519_cast_fp16 = slice_by_index(begin = var_4519_begin_0, end = var_4519_end_0, end_mask = var_4519_end_mask_0, x = var_4491_cast_fp16)[name = tensor("op_4519_cast_fp16")]; tensor var_4520_begin_0 = const()[name = tensor("op_4520_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_4520_end_0 = const()[name = tensor("op_4520_end_0"), val = tensor([2, 80, 1, 512])]; tensor var_4520_end_mask_0 = const()[name = tensor("op_4520_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4520_cast_fp16 = slice_by_index(begin = var_4520_begin_0, end = var_4520_end_0, end_mask = var_4520_end_mask_0, x = var_4495_cast_fp16)[name = tensor("op_4520_cast_fp16")]; tensor var_4521_begin_0 = const()[name = tensor("op_4521_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_4521_end_0 = const()[name = tensor("op_4521_end_0"), val = tensor([2, 80, 1, 1024])]; tensor var_4521_end_mask_0 = const()[name = tensor("op_4521_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4521_cast_fp16 = slice_by_index(begin = var_4521_begin_0, end = var_4521_end_0, end_mask = var_4521_end_mask_0, x = var_4495_cast_fp16)[name = tensor("op_4521_cast_fp16")]; tensor var_4522_begin_0 = const()[name = tensor("op_4522_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_4522_end_0 = const()[name = tensor("op_4522_end_0"), val = tensor([2, 80, 1, 512])]; tensor var_4522_end_mask_0 = const()[name = tensor("op_4522_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4522_cast_fp16 = slice_by_index(begin = var_4522_begin_0, end = var_4522_end_0, end_mask = var_4522_end_mask_0, x = var_4499_cast_fp16)[name = tensor("op_4522_cast_fp16")]; tensor var_4523_begin_0 = const()[name = tensor("op_4523_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_4523_end_0 = const()[name = tensor("op_4523_end_0"), val = tensor([2, 80, 1, 1024])]; tensor var_4523_end_mask_0 = const()[name = tensor("op_4523_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4523_cast_fp16 = slice_by_index(begin = var_4523_begin_0, end = var_4523_end_0, end_mask = var_4523_end_mask_0, x = var_4499_cast_fp16)[name = tensor("op_4523_cast_fp16")]; tensor var_4524_begin_0 = const()[name = tensor("op_4524_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_4524_end_0 = const()[name = tensor("op_4524_end_0"), val = tensor([2, 80, 1, 512])]; tensor var_4524_end_mask_0 = const()[name = tensor("op_4524_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4524_cast_fp16 = slice_by_index(begin = var_4524_begin_0, end = var_4524_end_0, end_mask = var_4524_end_mask_0, x = var_4503_cast_fp16)[name = tensor("op_4524_cast_fp16")]; tensor var_4525_begin_0 = const()[name = tensor("op_4525_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_4525_end_0 = const()[name = tensor("op_4525_end_0"), val = tensor([2, 80, 1, 1024])]; tensor var_4525_end_mask_0 = const()[name = tensor("op_4525_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4525_cast_fp16 = slice_by_index(begin = var_4525_begin_0, end = var_4525_end_0, end_mask = var_4525_end_mask_0, x = var_4503_cast_fp16)[name = tensor("op_4525_cast_fp16")]; 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, 80, 1, 512])]; tensor var_4526_end_mask_0 = const()[name = tensor("op_4526_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4526_cast_fp16 = slice_by_index(begin = var_4526_begin_0, end = var_4526_end_0, end_mask = var_4526_end_mask_0, x = var_4507_cast_fp16)[name = tensor("op_4526_cast_fp16")]; tensor var_4527_begin_0 = const()[name = tensor("op_4527_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_4527_end_0 = const()[name = tensor("op_4527_end_0"), val = tensor([2, 80, 1, 1024])]; tensor var_4527_end_mask_0 = const()[name = tensor("op_4527_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4527_cast_fp16 = slice_by_index(begin = var_4527_begin_0, end = var_4527_end_0, end_mask = var_4527_end_mask_0, x = var_4507_cast_fp16)[name = tensor("op_4527_cast_fp16")]; tensor var_4528_begin_0 = const()[name = tensor("op_4528_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_4528_end_0 = const()[name = tensor("op_4528_end_0"), val = tensor([2, 80, 1, 512])]; tensor var_4528_end_mask_0 = const()[name = tensor("op_4528_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4528_cast_fp16 = slice_by_index(begin = var_4528_begin_0, end = var_4528_end_0, end_mask = var_4528_end_mask_0, x = var_4511_cast_fp16)[name = tensor("op_4528_cast_fp16")]; tensor var_4529_begin_0 = const()[name = tensor("op_4529_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_4529_end_0 = const()[name = tensor("op_4529_end_0"), val = tensor([2, 80, 1, 1024])]; tensor var_4529_end_mask_0 = const()[name = tensor("op_4529_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4529_cast_fp16 = slice_by_index(begin = var_4529_begin_0, end = var_4529_end_0, end_mask = var_4529_end_mask_0, x = var_4511_cast_fp16)[name = tensor("op_4529_cast_fp16")]; tensor k_47_perm_0 = const()[name = tensor("k_47_perm_0"), val = tensor([0, 3, 2, 1])]; tensor var_4534_begin_0 = const()[name = tensor("op_4534_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_4534_end_0 = const()[name = tensor("op_4534_end_0"), val = tensor([2, 77, 1, 80])]; tensor var_4534_end_mask_0 = const()[name = tensor("op_4534_end_mask_0"), val = tensor([true, true, true, false])]; tensor transpose_6 = transpose(perm = k_47_perm_0, x = k_45_cast_fp16)[name = tensor("transpose_6")]; tensor var_4534_cast_fp16 = slice_by_index(begin = var_4534_begin_0, end = var_4534_end_0, end_mask = var_4534_end_mask_0, x = transpose_6)[name = tensor("op_4534_cast_fp16")]; tensor var_4538_begin_0 = const()[name = tensor("op_4538_begin_0"), val = tensor([0, 0, 0, 80])]; tensor var_4538_end_0 = const()[name = tensor("op_4538_end_0"), val = tensor([2, 77, 1, 160])]; tensor var_4538_end_mask_0 = const()[name = tensor("op_4538_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4538_cast_fp16 = slice_by_index(begin = var_4538_begin_0, end = var_4538_end_0, end_mask = var_4538_end_mask_0, x = transpose_6)[name = tensor("op_4538_cast_fp16")]; tensor var_4542_begin_0 = const()[name = tensor("op_4542_begin_0"), val = tensor([0, 0, 0, 160])]; tensor var_4542_end_0 = const()[name = tensor("op_4542_end_0"), val = tensor([2, 77, 1, 240])]; tensor var_4542_end_mask_0 = const()[name = tensor("op_4542_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4542_cast_fp16 = slice_by_index(begin = var_4542_begin_0, end = var_4542_end_0, end_mask = var_4542_end_mask_0, x = transpose_6)[name = tensor("op_4542_cast_fp16")]; tensor var_4546_begin_0 = const()[name = tensor("op_4546_begin_0"), val = tensor([0, 0, 0, 240])]; tensor var_4546_end_0 = const()[name = tensor("op_4546_end_0"), val = tensor([2, 77, 1, 320])]; tensor var_4546_end_mask_0 = const()[name = tensor("op_4546_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4546_cast_fp16 = slice_by_index(begin = var_4546_begin_0, end = var_4546_end_0, end_mask = var_4546_end_mask_0, x = transpose_6)[name = tensor("op_4546_cast_fp16")]; tensor var_4550_begin_0 = const()[name = tensor("op_4550_begin_0"), val = tensor([0, 0, 0, 320])]; tensor var_4550_end_0 = const()[name = tensor("op_4550_end_0"), val = tensor([2, 77, 1, 400])]; tensor var_4550_end_mask_0 = const()[name = tensor("op_4550_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4550_cast_fp16 = slice_by_index(begin = var_4550_begin_0, end = var_4550_end_0, end_mask = var_4550_end_mask_0, x = transpose_6)[name = tensor("op_4550_cast_fp16")]; tensor var_4554_begin_0 = const()[name = tensor("op_4554_begin_0"), val = tensor([0, 0, 0, 400])]; tensor var_4554_end_0 = const()[name = tensor("op_4554_end_0"), val = tensor([2, 77, 1, 480])]; tensor var_4554_end_mask_0 = const()[name = tensor("op_4554_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4554_cast_fp16 = slice_by_index(begin = var_4554_begin_0, end = var_4554_end_0, end_mask = var_4554_end_mask_0, x = transpose_6)[name = tensor("op_4554_cast_fp16")]; tensor var_4558_begin_0 = const()[name = tensor("op_4558_begin_0"), val = tensor([0, 0, 0, 480])]; tensor var_4558_end_0 = const()[name = tensor("op_4558_end_0"), val = tensor([2, 77, 1, 560])]; tensor var_4558_end_mask_0 = const()[name = tensor("op_4558_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4558_cast_fp16 = slice_by_index(begin = var_4558_begin_0, end = var_4558_end_0, end_mask = var_4558_end_mask_0, x = transpose_6)[name = tensor("op_4558_cast_fp16")]; tensor var_4562_begin_0 = const()[name = tensor("op_4562_begin_0"), val = tensor([0, 0, 0, 560])]; 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_fp16 = slice_by_index(begin = var_4562_begin_0, end = var_4562_end_0, end_mask = var_4562_end_mask_0, x = transpose_6)[name = tensor("op_4562_cast_fp16")]; tensor var_4564_begin_0 = const()[name = tensor("op_4564_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_4564_end_0 = const()[name = tensor("op_4564_end_0"), val = tensor([2, 80, 1, 77])]; tensor var_4564_end_mask_0 = const()[name = tensor("op_4564_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_4564_cast_fp16 = slice_by_index(begin = var_4564_begin_0, end = var_4564_end_0, end_mask = var_4564_end_mask_0, x = v_23_cast_fp16)[name = tensor("op_4564_cast_fp16")]; tensor var_4568_begin_0 = const()[name = tensor("op_4568_begin_0"), val = tensor([0, 80, 0, 0])]; tensor var_4568_end_0 = const()[name = tensor("op_4568_end_0"), val = tensor([2, 160, 1, 77])]; tensor var_4568_end_mask_0 = const()[name = tensor("op_4568_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_4568_cast_fp16 = slice_by_index(begin = var_4568_begin_0, end = var_4568_end_0, end_mask = var_4568_end_mask_0, x = v_23_cast_fp16)[name = tensor("op_4568_cast_fp16")]; tensor var_4572_begin_0 = const()[name = tensor("op_4572_begin_0"), val = tensor([0, 160, 0, 0])]; tensor var_4572_end_0 = const()[name = tensor("op_4572_end_0"), val = tensor([2, 240, 1, 77])]; tensor var_4572_end_mask_0 = const()[name = tensor("op_4572_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_4572_cast_fp16 = slice_by_index(begin = var_4572_begin_0, end = var_4572_end_0, end_mask = var_4572_end_mask_0, x = v_23_cast_fp16)[name = tensor("op_4572_cast_fp16")]; tensor var_4576_begin_0 = const()[name = tensor("op_4576_begin_0"), val = tensor([0, 240, 0, 0])]; tensor var_4576_end_0 = const()[name = tensor("op_4576_end_0"), val = tensor([2, 320, 1, 77])]; tensor var_4576_end_mask_0 = const()[name = tensor("op_4576_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_4576_cast_fp16 = slice_by_index(begin = var_4576_begin_0, end = var_4576_end_0, end_mask = var_4576_end_mask_0, x = v_23_cast_fp16)[name = tensor("op_4576_cast_fp16")]; tensor var_4580_begin_0 = const()[name = tensor("op_4580_begin_0"), val = tensor([0, 320, 0, 0])]; tensor var_4580_end_0 = const()[name = tensor("op_4580_end_0"), val = tensor([2, 400, 1, 77])]; tensor var_4580_end_mask_0 = const()[name = tensor("op_4580_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_4580_cast_fp16 = slice_by_index(begin = var_4580_begin_0, end = var_4580_end_0, end_mask = var_4580_end_mask_0, x = v_23_cast_fp16)[name = tensor("op_4580_cast_fp16")]; tensor var_4584_begin_0 = const()[name = tensor("op_4584_begin_0"), val = tensor([0, 400, 0, 0])]; tensor var_4584_end_0 = const()[name = tensor("op_4584_end_0"), val = tensor([2, 480, 1, 77])]; tensor var_4584_end_mask_0 = const()[name = tensor("op_4584_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_4584_cast_fp16 = slice_by_index(begin = var_4584_begin_0, end = var_4584_end_0, end_mask = var_4584_end_mask_0, x = v_23_cast_fp16)[name = tensor("op_4584_cast_fp16")]; tensor var_4588_begin_0 = const()[name = tensor("op_4588_begin_0"), val = tensor([0, 480, 0, 0])]; tensor var_4588_end_0 = const()[name = tensor("op_4588_end_0"), val = tensor([2, 560, 1, 77])]; tensor var_4588_end_mask_0 = const()[name = tensor("op_4588_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_4588_cast_fp16 = slice_by_index(begin = var_4588_begin_0, end = var_4588_end_0, end_mask = var_4588_end_mask_0, x = v_23_cast_fp16)[name = tensor("op_4588_cast_fp16")]; tensor var_4592_begin_0 = const()[name = tensor("op_4592_begin_0"), val = tensor([0, 560, 0, 0])]; tensor var_4592_end_0 = const()[name = tensor("op_4592_end_0"), val = tensor([2, 640, 1, 77])]; tensor var_4592_end_mask_0 = const()[name = tensor("op_4592_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_4592_cast_fp16 = slice_by_index(begin = var_4592_begin_0, end = var_4592_end_0, end_mask = var_4592_end_mask_0, x = v_23_cast_fp16)[name = tensor("op_4592_cast_fp16")]; tensor var_4596_equation_0 = const()[name = tensor("op_4596_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_4596_cast_fp16 = einsum(equation = var_4596_equation_0, values = (var_4534_cast_fp16, var_4514_cast_fp16))[name = tensor("op_4596_cast_fp16")]; tensor var_4597_to_fp16 = const()[name = tensor("op_4597_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_353_cast_fp16 = mul(x = var_4596_cast_fp16, y = var_4597_to_fp16)[name = tensor("aw_chunk_353_cast_fp16")]; tensor var_4600_equation_0 = const()[name = tensor("op_4600_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_4600_cast_fp16 = einsum(equation = var_4600_equation_0, values = (var_4534_cast_fp16, var_4515_cast_fp16))[name = tensor("op_4600_cast_fp16")]; tensor var_4601_to_fp16 = const()[name = tensor("op_4601_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_355_cast_fp16 = mul(x = var_4600_cast_fp16, y = var_4601_to_fp16)[name = tensor("aw_chunk_355_cast_fp16")]; tensor var_4604_equation_0 = const()[name = tensor("op_4604_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_4604_cast_fp16 = einsum(equation = var_4604_equation_0, values = (var_4538_cast_fp16, var_4516_cast_fp16))[name = tensor("op_4604_cast_fp16")]; tensor var_4605_to_fp16 = const()[name = tensor("op_4605_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_357_cast_fp16 = mul(x = var_4604_cast_fp16, y = var_4605_to_fp16)[name = tensor("aw_chunk_357_cast_fp16")]; tensor var_4608_equation_0 = const()[name = tensor("op_4608_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_4608_cast_fp16 = einsum(equation = var_4608_equation_0, values = (var_4538_cast_fp16, var_4517_cast_fp16))[name = tensor("op_4608_cast_fp16")]; tensor var_4609_to_fp16 = const()[name = tensor("op_4609_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_359_cast_fp16 = mul(x = var_4608_cast_fp16, y = var_4609_to_fp16)[name = tensor("aw_chunk_359_cast_fp16")]; tensor var_4612_equation_0 = const()[name = tensor("op_4612_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_4612_cast_fp16 = einsum(equation = var_4612_equation_0, values = (var_4542_cast_fp16, var_4518_cast_fp16))[name = tensor("op_4612_cast_fp16")]; tensor var_4613_to_fp16 = const()[name = tensor("op_4613_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_361_cast_fp16 = mul(x = var_4612_cast_fp16, y = var_4613_to_fp16)[name = tensor("aw_chunk_361_cast_fp16")]; tensor var_4616_equation_0 = const()[name = tensor("op_4616_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_4616_cast_fp16 = einsum(equation = var_4616_equation_0, values = (var_4542_cast_fp16, var_4519_cast_fp16))[name = tensor("op_4616_cast_fp16")]; tensor var_4617_to_fp16 = const()[name = tensor("op_4617_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_363_cast_fp16 = mul(x = var_4616_cast_fp16, y = var_4617_to_fp16)[name = tensor("aw_chunk_363_cast_fp16")]; tensor var_4620_equation_0 = const()[name = tensor("op_4620_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_4620_cast_fp16 = einsum(equation = var_4620_equation_0, values = (var_4546_cast_fp16, var_4520_cast_fp16))[name = tensor("op_4620_cast_fp16")]; tensor var_4621_to_fp16 = const()[name = tensor("op_4621_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_365_cast_fp16 = mul(x = var_4620_cast_fp16, y = var_4621_to_fp16)[name = tensor("aw_chunk_365_cast_fp16")]; tensor var_4624_equation_0 = const()[name = tensor("op_4624_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_4624_cast_fp16 = einsum(equation = var_4624_equation_0, values = (var_4546_cast_fp16, var_4521_cast_fp16))[name = tensor("op_4624_cast_fp16")]; tensor var_4625_to_fp16 = const()[name = tensor("op_4625_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_367_cast_fp16 = mul(x = var_4624_cast_fp16, y = var_4625_to_fp16)[name = tensor("aw_chunk_367_cast_fp16")]; tensor var_4628_equation_0 = const()[name = tensor("op_4628_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_4628_cast_fp16 = einsum(equation = var_4628_equation_0, values = (var_4550_cast_fp16, var_4522_cast_fp16))[name = tensor("op_4628_cast_fp16")]; tensor var_4629_to_fp16 = const()[name = tensor("op_4629_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_369_cast_fp16 = mul(x = var_4628_cast_fp16, y = var_4629_to_fp16)[name = tensor("aw_chunk_369_cast_fp16")]; tensor var_4632_equation_0 = const()[name = tensor("op_4632_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_4632_cast_fp16 = einsum(equation = var_4632_equation_0, values = (var_4550_cast_fp16, var_4523_cast_fp16))[name = tensor("op_4632_cast_fp16")]; tensor var_4633_to_fp16 = const()[name = tensor("op_4633_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_371_cast_fp16 = mul(x = var_4632_cast_fp16, y = var_4633_to_fp16)[name = tensor("aw_chunk_371_cast_fp16")]; tensor var_4636_equation_0 = const()[name = tensor("op_4636_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_4636_cast_fp16 = einsum(equation = var_4636_equation_0, values = (var_4554_cast_fp16, var_4524_cast_fp16))[name = tensor("op_4636_cast_fp16")]; tensor var_4637_to_fp16 = const()[name = tensor("op_4637_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_373_cast_fp16 = mul(x = var_4636_cast_fp16, y = var_4637_to_fp16)[name = tensor("aw_chunk_373_cast_fp16")]; tensor var_4640_equation_0 = const()[name = tensor("op_4640_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_4640_cast_fp16 = einsum(equation = var_4640_equation_0, values = (var_4554_cast_fp16, var_4525_cast_fp16))[name = tensor("op_4640_cast_fp16")]; tensor var_4641_to_fp16 = const()[name = tensor("op_4641_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_375_cast_fp16 = mul(x = var_4640_cast_fp16, y = var_4641_to_fp16)[name = tensor("aw_chunk_375_cast_fp16")]; tensor var_4644_equation_0 = const()[name = tensor("op_4644_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_4644_cast_fp16 = einsum(equation = var_4644_equation_0, values = (var_4558_cast_fp16, var_4526_cast_fp16))[name = tensor("op_4644_cast_fp16")]; tensor var_4645_to_fp16 = const()[name = tensor("op_4645_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_377_cast_fp16 = mul(x = var_4644_cast_fp16, y = var_4645_to_fp16)[name = tensor("aw_chunk_377_cast_fp16")]; tensor var_4648_equation_0 = const()[name = tensor("op_4648_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_4648_cast_fp16 = einsum(equation = var_4648_equation_0, values = (var_4558_cast_fp16, var_4527_cast_fp16))[name = tensor("op_4648_cast_fp16")]; tensor var_4649_to_fp16 = const()[name = tensor("op_4649_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_379_cast_fp16 = mul(x = var_4648_cast_fp16, y = var_4649_to_fp16)[name = tensor("aw_chunk_379_cast_fp16")]; tensor var_4652_equation_0 = const()[name = tensor("op_4652_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_4652_cast_fp16 = einsum(equation = var_4652_equation_0, values = (var_4562_cast_fp16, var_4528_cast_fp16))[name = tensor("op_4652_cast_fp16")]; tensor var_4653_to_fp16 = const()[name = tensor("op_4653_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_381_cast_fp16 = mul(x = var_4652_cast_fp16, y = var_4653_to_fp16)[name = tensor("aw_chunk_381_cast_fp16")]; tensor var_4656_equation_0 = const()[name = tensor("op_4656_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_4656_cast_fp16 = einsum(equation = var_4656_equation_0, values = (var_4562_cast_fp16, var_4529_cast_fp16))[name = tensor("op_4656_cast_fp16")]; tensor var_4657_to_fp16 = const()[name = tensor("op_4657_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_383_cast_fp16 = mul(x = var_4656_cast_fp16, y = var_4657_to_fp16)[name = tensor("aw_chunk_383_cast_fp16")]; tensor var_4659_cast_fp16 = softmax(axis = var_4061, x = aw_chunk_353_cast_fp16)[name = tensor("op_4659_cast_fp16")]; tensor var_4660_cast_fp16 = softmax(axis = var_4061, x = aw_chunk_355_cast_fp16)[name = tensor("op_4660_cast_fp16")]; tensor var_4661_cast_fp16 = softmax(axis = var_4061, x = aw_chunk_357_cast_fp16)[name = tensor("op_4661_cast_fp16")]; tensor var_4662_cast_fp16 = softmax(axis = var_4061, x = aw_chunk_359_cast_fp16)[name = tensor("op_4662_cast_fp16")]; tensor var_4663_cast_fp16 = softmax(axis = var_4061, x = aw_chunk_361_cast_fp16)[name = tensor("op_4663_cast_fp16")]; tensor var_4664_cast_fp16 = softmax(axis = var_4061, x = aw_chunk_363_cast_fp16)[name = tensor("op_4664_cast_fp16")]; tensor var_4665_cast_fp16 = softmax(axis = var_4061, x = aw_chunk_365_cast_fp16)[name = tensor("op_4665_cast_fp16")]; tensor var_4666_cast_fp16 = softmax(axis = var_4061, x = aw_chunk_367_cast_fp16)[name = tensor("op_4666_cast_fp16")]; tensor var_4667_cast_fp16 = softmax(axis = var_4061, x = aw_chunk_369_cast_fp16)[name = tensor("op_4667_cast_fp16")]; tensor var_4668_cast_fp16 = softmax(axis = var_4061, x = aw_chunk_371_cast_fp16)[name = tensor("op_4668_cast_fp16")]; tensor var_4669_cast_fp16 = softmax(axis = var_4061, x = aw_chunk_373_cast_fp16)[name = tensor("op_4669_cast_fp16")]; tensor var_4670_cast_fp16 = softmax(axis = var_4061, x = aw_chunk_375_cast_fp16)[name = tensor("op_4670_cast_fp16")]; tensor var_4671_cast_fp16 = softmax(axis = var_4061, x = aw_chunk_377_cast_fp16)[name = tensor("op_4671_cast_fp16")]; tensor var_4672_cast_fp16 = softmax(axis = var_4061, x = aw_chunk_379_cast_fp16)[name = tensor("op_4672_cast_fp16")]; tensor var_4673_cast_fp16 = softmax(axis = var_4061, x = aw_chunk_381_cast_fp16)[name = tensor("op_4673_cast_fp16")]; tensor var_4674_cast_fp16 = softmax(axis = var_4061, x = aw_chunk_383_cast_fp16)[name = tensor("op_4674_cast_fp16")]; tensor var_4676_equation_0 = const()[name = tensor("op_4676_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_4676_cast_fp16 = einsum(equation = var_4676_equation_0, values = (var_4564_cast_fp16, var_4659_cast_fp16))[name = tensor("op_4676_cast_fp16")]; tensor var_4678_equation_0 = const()[name = tensor("op_4678_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_4678_cast_fp16 = einsum(equation = var_4678_equation_0, values = (var_4564_cast_fp16, var_4660_cast_fp16))[name = tensor("op_4678_cast_fp16")]; tensor var_4680_equation_0 = const()[name = tensor("op_4680_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_4680_cast_fp16 = einsum(equation = var_4680_equation_0, values = (var_4568_cast_fp16, var_4661_cast_fp16))[name = tensor("op_4680_cast_fp16")]; tensor var_4682_equation_0 = const()[name = tensor("op_4682_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_4682_cast_fp16 = einsum(equation = var_4682_equation_0, values = (var_4568_cast_fp16, var_4662_cast_fp16))[name = tensor("op_4682_cast_fp16")]; tensor var_4684_equation_0 = const()[name = tensor("op_4684_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_4684_cast_fp16 = einsum(equation = var_4684_equation_0, values = (var_4572_cast_fp16, var_4663_cast_fp16))[name = tensor("op_4684_cast_fp16")]; tensor var_4686_equation_0 = const()[name = tensor("op_4686_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_4686_cast_fp16 = einsum(equation = var_4686_equation_0, values = (var_4572_cast_fp16, var_4664_cast_fp16))[name = tensor("op_4686_cast_fp16")]; tensor var_4688_equation_0 = const()[name = tensor("op_4688_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_4688_cast_fp16 = einsum(equation = var_4688_equation_0, values = (var_4576_cast_fp16, var_4665_cast_fp16))[name = tensor("op_4688_cast_fp16")]; tensor var_4690_equation_0 = const()[name = tensor("op_4690_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_4690_cast_fp16 = einsum(equation = var_4690_equation_0, values = (var_4576_cast_fp16, var_4666_cast_fp16))[name = tensor("op_4690_cast_fp16")]; tensor var_4692_equation_0 = const()[name = tensor("op_4692_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_4692_cast_fp16 = einsum(equation = var_4692_equation_0, values = (var_4580_cast_fp16, var_4667_cast_fp16))[name = tensor("op_4692_cast_fp16")]; tensor var_4694_equation_0 = const()[name = tensor("op_4694_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_4694_cast_fp16 = einsum(equation = var_4694_equation_0, values = (var_4580_cast_fp16, var_4668_cast_fp16))[name = tensor("op_4694_cast_fp16")]; tensor var_4696_equation_0 = const()[name = tensor("op_4696_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_4696_cast_fp16 = einsum(equation = var_4696_equation_0, values = (var_4584_cast_fp16, var_4669_cast_fp16))[name = tensor("op_4696_cast_fp16")]; tensor var_4698_equation_0 = const()[name = tensor("op_4698_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_4698_cast_fp16 = einsum(equation = var_4698_equation_0, values = (var_4584_cast_fp16, var_4670_cast_fp16))[name = tensor("op_4698_cast_fp16")]; tensor var_4700_equation_0 = const()[name = tensor("op_4700_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_4700_cast_fp16 = einsum(equation = var_4700_equation_0, values = (var_4588_cast_fp16, var_4671_cast_fp16))[name = tensor("op_4700_cast_fp16")]; tensor var_4702_equation_0 = const()[name = tensor("op_4702_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_4702_cast_fp16 = einsum(equation = var_4702_equation_0, values = (var_4588_cast_fp16, var_4672_cast_fp16))[name = tensor("op_4702_cast_fp16")]; tensor var_4704_equation_0 = const()[name = tensor("op_4704_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_4704_cast_fp16 = einsum(equation = var_4704_equation_0, values = (var_4592_cast_fp16, var_4673_cast_fp16))[name = tensor("op_4704_cast_fp16")]; tensor var_4706_equation_0 = const()[name = tensor("op_4706_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_4706_cast_fp16 = einsum(equation = var_4706_equation_0, values = (var_4592_cast_fp16, var_4674_cast_fp16))[name = tensor("op_4706_cast_fp16")]; tensor var_4708_interleave_0 = const()[name = tensor("op_4708_interleave_0"), val = tensor(false)]; tensor var_4708_cast_fp16 = concat(axis = var_4039, interleave = var_4708_interleave_0, values = (var_4676_cast_fp16, var_4678_cast_fp16))[name = tensor("op_4708_cast_fp16")]; tensor var_4710_interleave_0 = const()[name = tensor("op_4710_interleave_0"), val = tensor(false)]; tensor var_4710_cast_fp16 = concat(axis = var_4039, interleave = var_4710_interleave_0, values = (var_4680_cast_fp16, var_4682_cast_fp16))[name = tensor("op_4710_cast_fp16")]; tensor var_4712_interleave_0 = const()[name = tensor("op_4712_interleave_0"), val = tensor(false)]; tensor var_4712_cast_fp16 = concat(axis = var_4039, interleave = var_4712_interleave_0, values = (var_4684_cast_fp16, var_4686_cast_fp16))[name = tensor("op_4712_cast_fp16")]; tensor var_4714_interleave_0 = const()[name = tensor("op_4714_interleave_0"), val = tensor(false)]; tensor var_4714_cast_fp16 = concat(axis = var_4039, interleave = var_4714_interleave_0, values = (var_4688_cast_fp16, var_4690_cast_fp16))[name = tensor("op_4714_cast_fp16")]; tensor var_4716_interleave_0 = const()[name = tensor("op_4716_interleave_0"), val = tensor(false)]; tensor var_4716_cast_fp16 = concat(axis = var_4039, interleave = var_4716_interleave_0, values = (var_4692_cast_fp16, var_4694_cast_fp16))[name = tensor("op_4716_cast_fp16")]; tensor var_4718_interleave_0 = const()[name = tensor("op_4718_interleave_0"), val = tensor(false)]; tensor var_4718_cast_fp16 = concat(axis = var_4039, interleave = var_4718_interleave_0, values = (var_4696_cast_fp16, var_4698_cast_fp16))[name = tensor("op_4718_cast_fp16")]; tensor var_4720_interleave_0 = const()[name = tensor("op_4720_interleave_0"), val = tensor(false)]; tensor var_4720_cast_fp16 = concat(axis = var_4039, interleave = var_4720_interleave_0, values = (var_4700_cast_fp16, var_4702_cast_fp16))[name = tensor("op_4720_cast_fp16")]; tensor var_4722_interleave_0 = const()[name = tensor("op_4722_interleave_0"), val = tensor(false)]; tensor var_4722_cast_fp16 = concat(axis = var_4039, interleave = var_4722_interleave_0, values = (var_4704_cast_fp16, var_4706_cast_fp16))[name = tensor("op_4722_cast_fp16")]; tensor input_165_interleave_0 = const()[name = tensor("input_165_interleave_0"), val = tensor(false)]; tensor input_165_cast_fp16 = concat(axis = var_4061, interleave = input_165_interleave_0, values = (var_4708_cast_fp16, var_4710_cast_fp16, var_4712_cast_fp16, var_4714_cast_fp16, var_4716_cast_fp16, var_4718_cast_fp16, var_4720_cast_fp16, var_4722_cast_fp16))[name = tensor("input_165_cast_fp16")]; tensor var_4728 = const()[name = tensor("op_4728"), val = tensor([1, 1])]; tensor var_4730 = const()[name = tensor("op_4730"), val = tensor([1, 1])]; tensor var_4732_pad_type_0 = const()[name = tensor("op_4732_pad_type_0"), val = tensor("custom")]; tensor var_4732_pad_0 = const()[name = tensor("op_4732_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(211113472))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(211420736))), name = tensor("up_blocks_1_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16_palettized"), shape = tensor([640, 640, 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(211420928)))]; tensor var_4732_cast_fp16 = conv(bias = up_blocks_1_attentions_0_transformer_blocks_0_attn2_to_out_0_bias_to_fp16, dilations = var_4730, groups = var_4061, pad = var_4732_pad_0, pad_type = var_4732_pad_type_0, strides = var_4728, weight = up_blocks_1_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16_palettized, x = input_165_cast_fp16)[name = tensor("op_4732_cast_fp16")]; tensor inputs_35_cast_fp16 = add(x = var_4732_cast_fp16, y = inputs_33_cast_fp16)[name = tensor("inputs_35_cast_fp16")]; tensor var_4736 = const()[name = tensor("op_4736"), val = tensor([1])]; tensor channels_mean_35_cast_fp16 = reduce_mean(axes = var_4736, keep_dims = var_4056, x = inputs_35_cast_fp16)[name = tensor("channels_mean_35_cast_fp16")]; tensor zero_mean_35_cast_fp16 = sub(x = inputs_35_cast_fp16, y = channels_mean_35_cast_fp16)[name = tensor("zero_mean_35_cast_fp16")]; tensor zero_mean_sq_35_cast_fp16 = mul(x = zero_mean_35_cast_fp16, y = zero_mean_35_cast_fp16)[name = tensor("zero_mean_sq_35_cast_fp16")]; tensor var_4740 = const()[name = tensor("op_4740"), val = tensor([1])]; tensor var_4741_cast_fp16 = reduce_mean(axes = var_4740, keep_dims = var_4056, x = zero_mean_sq_35_cast_fp16)[name = tensor("op_4741_cast_fp16")]; tensor var_4742_to_fp16 = const()[name = tensor("op_4742_to_fp16"), val = tensor(0x1.5p-17)]; tensor var_4743_cast_fp16 = add(x = var_4741_cast_fp16, y = var_4742_to_fp16)[name = tensor("op_4743_cast_fp16")]; tensor denom_35_epsilon_0_to_fp16 = const()[name = tensor("denom_35_epsilon_0_to_fp16"), val = tensor(0x1p-24)]; tensor denom_35_cast_fp16 = rsqrt(epsilon = denom_35_epsilon_0_to_fp16, x = var_4743_cast_fp16)[name = tensor("denom_35_cast_fp16")]; tensor out_35_cast_fp16 = mul(x = zero_mean_35_cast_fp16, y = denom_35_cast_fp16)[name = tensor("out_35_cast_fp16")]; tensor var_4747_to_fp16 = const()[name = tensor("op_4747_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(211422272)))]; tensor var_4748_cast_fp16 = add(x = out_35_cast_fp16, y = var_4747_to_fp16)[name = tensor("op_4748_cast_fp16")]; tensor var_4750_to_fp16 = const()[name = tensor("op_4750_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(211423616)))]; tensor input_167_cast_fp16 = mul(x = var_4748_cast_fp16, y = var_4750_to_fp16)[name = tensor("input_167_cast_fp16")]; tensor var_4758 = const()[name = tensor("op_4758"), val = tensor([1, 1])]; tensor var_4760 = const()[name = tensor("op_4760"), val = tensor([1, 1])]; tensor var_4762_pad_type_0 = const()[name = tensor("op_4762_pad_type_0"), val = tensor("custom")]; tensor var_4762_pad_0 = const()[name = tensor("op_4762_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(211424960))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(213882624))), name = tensor("up_blocks_1_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16_palettized"), shape = tensor([5120, 640, 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(213882816))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(213886720))), name = tensor("up_blocks_1_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16_palettized"), shape = tensor([5120])]; tensor var_4762_cast_fp16 = conv(bias = up_blocks_1_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16_palettized, dilations = var_4760, groups = var_4061, pad = var_4762_pad_0, pad_type = var_4762_pad_type_0, strides = var_4758, weight = up_blocks_1_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16_palettized, x = input_167_cast_fp16)[name = tensor("op_4762_cast_fp16")]; tensor var_4763_split_sizes_0 = const()[name = tensor("op_4763_split_sizes_0"), val = tensor([2560, 2560])]; tensor var_4763_axis_0 = const()[name = tensor("op_4763_axis_0"), val = tensor(1)]; tensor var_4763_cast_fp16_0, tensor var_4763_cast_fp16_1 = split(axis = var_4763_axis_0, split_sizes = var_4763_split_sizes_0, x = var_4762_cast_fp16)[name = tensor("op_4763_cast_fp16")]; tensor var_4765_mode_0 = const()[name = tensor("op_4765_mode_0"), val = tensor("EXACT")]; tensor var_4765_cast_fp16 = gelu(mode = var_4765_mode_0, x = var_4763_cast_fp16_1)[name = tensor("op_4765_cast_fp16")]; tensor input_169_cast_fp16 = mul(x = var_4763_cast_fp16_0, y = var_4765_cast_fp16)[name = tensor("input_169_cast_fp16")]; tensor var_4769 = const()[name = tensor("op_4769"), val = tensor([1, 1])]; tensor var_4771 = const()[name = tensor("op_4771"), val = tensor([1, 1])]; tensor var_4773_pad_type_0 = const()[name = tensor("op_4773_pad_type_0"), val = tensor("custom")]; tensor var_4773_pad_0 = const()[name = tensor("op_4773_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(213886912))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(215115776))), name = tensor("up_blocks_1_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16_palettized"), shape = tensor([640, 2560, 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(215115968)))]; tensor var_4773_cast_fp16 = conv(bias = up_blocks_1_attentions_0_transformer_blocks_0_ff_net_2_bias_to_fp16, dilations = var_4771, groups = var_4061, pad = var_4773_pad_0, pad_type = var_4773_pad_type_0, strides = var_4769, weight = up_blocks_1_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16_palettized, x = input_169_cast_fp16)[name = tensor("op_4773_cast_fp16")]; tensor hidden_states_111_cast_fp16 = add(x = var_4773_cast_fp16, y = inputs_35_cast_fp16)[name = tensor("hidden_states_111_cast_fp16")]; tensor var_4775 = const()[name = tensor("op_4775"), val = tensor([2, 640, 32, 32])]; tensor input_171_cast_fp16 = reshape(shape = var_4775, x = hidden_states_111_cast_fp16)[name = tensor("input_171_cast_fp16")]; tensor var_4779 = const()[name = tensor("op_4779"), val = tensor([1, 1])]; tensor var_4781 = const()[name = tensor("op_4781"), 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([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(215117312))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(215424576))), name = tensor("up_blocks_1_attentions_0_proj_out_weight_to_fp16_palettized"), shape = tensor([640, 640, 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(215424768)))]; tensor hidden_states_113_cast_fp16 = conv(bias = up_blocks_1_attentions_0_proj_out_bias_to_fp16, dilations = var_4781, groups = var_4061, pad = hidden_states_113_pad_0, pad_type = hidden_states_113_pad_type_0, strides = var_4779, weight = up_blocks_1_attentions_0_proj_out_weight_to_fp16_palettized, x = input_171_cast_fp16)[name = tensor("hidden_states_113_cast_fp16")]; tensor hidden_states_115_cast_fp16 = add(x = hidden_states_113_cast_fp16, y = hidden_states_101_cast_fp16)[name = tensor("hidden_states_115_cast_fp16")]; tensor input_173_interleave_0 = const()[name = tensor("input_173_interleave_0"), val = tensor(false)]; tensor input_173_cast_fp16 = concat(axis = var_4061, interleave = input_173_interleave_0, values = (hidden_states_115_cast_fp16, input_37_cast_fp16))[name = tensor("input_173_cast_fp16")]; tensor reshape_72_shape_0 = const()[name = tensor("reshape_72_shape_0"), val = tensor([2, 32, 30, 32, 32])]; tensor reshape_72_cast_fp16 = reshape(shape = reshape_72_shape_0, x = input_173_cast_fp16)[name = tensor("reshape_72_cast_fp16")]; 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_fp16 = reduce_mean(axes = reduce_mean_54_axes_0, keep_dims = reduce_mean_54_keep_dims_0, x = reshape_72_cast_fp16)[name = tensor("reduce_mean_54_cast_fp16")]; tensor sub_36_cast_fp16 = sub(x = reshape_72_cast_fp16, y = reduce_mean_54_cast_fp16)[name = tensor("sub_36_cast_fp16")]; tensor square_18_cast_fp16 = square(x = sub_36_cast_fp16)[name = tensor("square_18_cast_fp16")]; 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_fp16 = reduce_mean(axes = reduce_mean_56_axes_0, keep_dims = reduce_mean_56_keep_dims_0, x = square_18_cast_fp16)[name = tensor("reduce_mean_56_cast_fp16")]; 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_fp16 = add(x = reduce_mean_56_cast_fp16, y = add_36_y_0_to_fp16)[name = tensor("add_36_cast_fp16")]; tensor sqrt_18_cast_fp16 = sqrt(x = add_36_cast_fp16)[name = tensor("sqrt_18_cast_fp16")]; tensor real_div_18_cast_fp16 = real_div(x = sub_36_cast_fp16, y = sqrt_18_cast_fp16)[name = tensor("real_div_18_cast_fp16")]; tensor reshape_73_shape_0 = const()[name = tensor("reshape_73_shape_0"), val = tensor([2, 960, 32, 32])]; tensor reshape_73_cast_fp16 = reshape(shape = reshape_73_shape_0, x = real_div_18_cast_fp16)[name = tensor("reshape_73_cast_fp16")]; tensor add_37_mean_0_to_fp16 = const()[name = tensor("add_37_mean_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(215426112)))]; tensor add_37_variance_0_to_fp16 = const()[name = tensor("add_37_variance_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(215428096)))]; 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(215430080)))]; 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(215432064)))]; 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_fp16 = 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_37_mean_0_to_fp16, variance = add_37_variance_0_to_fp16, x = reshape_73_cast_fp16)[name = tensor("add_37_cast_fp16")]; tensor input_177_cast_fp16 = silu(x = add_37_cast_fp16)[name = tensor("input_177_cast_fp16")]; tensor var_4799 = const()[name = tensor("op_4799"), val = tensor([1, 1])]; tensor var_4801 = const()[name = tensor("op_4801"), 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 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(215434048))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(219581312))), name = tensor("up_blocks_1_resnets_1_conv1_weight_to_fp16_palettized"), shape = tensor([640, 960, 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(219581504)))]; tensor hidden_states_117_cast_fp16 = conv(bias = up_blocks_1_resnets_1_conv1_bias_to_fp16, dilations = var_4801, groups = var_4061, pad = hidden_states_117_pad_0, pad_type = hidden_states_117_pad_type_0, strides = var_4799, weight = up_blocks_1_resnets_1_conv1_weight_to_fp16_palettized, x = input_177_cast_fp16)[name = tensor("hidden_states_117_cast_fp16")]; tensor var_4807 = const()[name = tensor("op_4807"), val = tensor([1, 1])]; tensor var_4809 = const()[name = tensor("op_4809"), 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 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(219582848))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(220197312))), name = tensor("up_blocks_1_resnets_1_time_emb_proj_weight_to_fp16_palettized"), shape = tensor([640, 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(220197504)))]; tensor temb_13_cast_fp16 = conv(bias = up_blocks_1_resnets_1_time_emb_proj_bias_to_fp16, dilations = var_4809, groups = var_4061, pad = temb_13_pad_0, pad_type = temb_13_pad_type_0, strides = var_4807, weight = up_blocks_1_resnets_1_time_emb_proj_weight_to_fp16_palettized, x = input_15_cast_fp16)[name = tensor("temb_13_cast_fp16")]; tensor input_181_cast_fp16 = add(x = hidden_states_117_cast_fp16, y = temb_13_cast_fp16)[name = tensor("input_181_cast_fp16")]; tensor reshape_76_shape_0 = const()[name = tensor("reshape_76_shape_0"), val = tensor([2, 32, 20, 32, 32])]; tensor reshape_76_cast_fp16 = reshape(shape = reshape_76_shape_0, x = input_181_cast_fp16)[name = tensor("reshape_76_cast_fp16")]; 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_fp16 = reduce_mean(axes = reduce_mean_57_axes_0, keep_dims = reduce_mean_57_keep_dims_0, x = reshape_76_cast_fp16)[name = tensor("reduce_mean_57_cast_fp16")]; tensor sub_38_cast_fp16 = sub(x = reshape_76_cast_fp16, y = reduce_mean_57_cast_fp16)[name = tensor("sub_38_cast_fp16")]; tensor square_19_cast_fp16 = square(x = sub_38_cast_fp16)[name = tensor("square_19_cast_fp16")]; 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_fp16 = reduce_mean(axes = reduce_mean_59_axes_0, keep_dims = reduce_mean_59_keep_dims_0, x = square_19_cast_fp16)[name = tensor("reduce_mean_59_cast_fp16")]; 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_fp16 = add(x = reduce_mean_59_cast_fp16, y = add_38_y_0_to_fp16)[name = tensor("add_38_cast_fp16")]; tensor sqrt_19_cast_fp16 = sqrt(x = add_38_cast_fp16)[name = tensor("sqrt_19_cast_fp16")]; tensor real_div_19_cast_fp16 = real_div(x = sub_38_cast_fp16, y = sqrt_19_cast_fp16)[name = tensor("real_div_19_cast_fp16")]; tensor reshape_77_shape_0 = const()[name = tensor("reshape_77_shape_0"), val = tensor([2, 640, 32, 32])]; tensor reshape_77_cast_fp16 = reshape(shape = reshape_77_shape_0, x = real_div_19_cast_fp16)[name = tensor("reshape_77_cast_fp16")]; 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(220198848)))]; 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(220200192)))]; 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_fp16 = 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_9_mean_0_to_fp16, variance = add_9_variance_0_to_fp16, x = reshape_77_cast_fp16)[name = tensor("add_39_cast_fp16")]; tensor input_185_cast_fp16 = silu(x = add_39_cast_fp16)[name = tensor("input_185_cast_fp16")]; tensor var_4819 = const()[name = tensor("op_4819"), val = tensor([1, 1])]; tensor var_4821 = const()[name = tensor("op_4821"), 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 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(220201536))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(222966400))), name = tensor("up_blocks_1_resnets_1_conv2_weight_to_fp16_palettized"), shape = tensor([640, 640, 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(222966592)))]; tensor hidden_states_119_cast_fp16 = conv(bias = up_blocks_1_resnets_1_conv2_bias_to_fp16, dilations = var_4821, groups = var_4061, pad = hidden_states_119_pad_0, pad_type = hidden_states_119_pad_type_0, strides = var_4819, weight = up_blocks_1_resnets_1_conv2_weight_to_fp16_palettized, x = input_185_cast_fp16)[name = tensor("hidden_states_119_cast_fp16")]; tensor var_4826 = const()[name = tensor("op_4826"), val = tensor([1, 1])]; tensor var_4828 = const()[name = tensor("op_4828"), 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_1_conv_shortcut_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(222967936))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(223428800))), name = tensor("up_blocks_1_resnets_1_conv_shortcut_weight_to_fp16_palettized"), shape = tensor([640, 960, 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(223428992)))]; tensor x_11_cast_fp16 = conv(bias = up_blocks_1_resnets_1_conv_shortcut_bias_to_fp16, dilations = var_4828, groups = var_4061, pad = x_11_pad_0, pad_type = x_11_pad_type_0, strides = var_4826, weight = up_blocks_1_resnets_1_conv_shortcut_weight_to_fp16_palettized, x = input_173_cast_fp16)[name = tensor("x_11_cast_fp16")]; tensor hidden_states_121_cast_fp16 = add(x = x_11_cast_fp16, y = hidden_states_119_cast_fp16)[name = tensor("hidden_states_121_cast_fp16")]; tensor reshape_80_shape_0 = const()[name = tensor("reshape_80_shape_0"), val = tensor([2, 32, 20, 32, 32])]; tensor reshape_80_cast_fp16 = reshape(shape = reshape_80_shape_0, x = hidden_states_121_cast_fp16)[name = tensor("reshape_80_cast_fp16")]; 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_fp16 = reduce_mean(axes = reduce_mean_60_axes_0, keep_dims = reduce_mean_60_keep_dims_0, x = reshape_80_cast_fp16)[name = tensor("reduce_mean_60_cast_fp16")]; tensor sub_40_cast_fp16 = sub(x = reshape_80_cast_fp16, y = reduce_mean_60_cast_fp16)[name = tensor("sub_40_cast_fp16")]; tensor square_20_cast_fp16 = square(x = sub_40_cast_fp16)[name = tensor("square_20_cast_fp16")]; 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_fp16 = reduce_mean(axes = reduce_mean_62_axes_0, keep_dims = reduce_mean_62_keep_dims_0, x = square_20_cast_fp16)[name = tensor("reduce_mean_62_cast_fp16")]; tensor add_40_y_0_to_fp16 = const()[name = tensor("add_40_y_0_to_fp16"), val = tensor(0x1.1p-20)]; tensor add_40_cast_fp16 = add(x = reduce_mean_62_cast_fp16, y = add_40_y_0_to_fp16)[name = tensor("add_40_cast_fp16")]; tensor sqrt_20_cast_fp16 = sqrt(x = add_40_cast_fp16)[name = tensor("sqrt_20_cast_fp16")]; tensor real_div_20_cast_fp16 = real_div(x = sub_40_cast_fp16, y = sqrt_20_cast_fp16)[name = tensor("real_div_20_cast_fp16")]; tensor reshape_81_shape_0 = const()[name = tensor("reshape_81_shape_0"), val = tensor([2, 640, 32, 32])]; tensor reshape_81_cast_fp16 = reshape(shape = reshape_81_shape_0, x = real_div_20_cast_fp16)[name = tensor("reshape_81_cast_fp16")]; 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(223430336)))]; 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(223431680)))]; 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_fp16 = 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_9_mean_0_to_fp16, variance = add_9_variance_0_to_fp16, x = reshape_81_cast_fp16)[name = tensor("add_41_cast_fp16")]; tensor var_4848 = const()[name = tensor("op_4848"), val = tensor([1, 1])]; tensor var_4850 = const()[name = tensor("op_4850"), 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 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(223433024))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(223740288))), name = tensor("up_blocks_1_attentions_1_proj_in_weight_to_fp16_palettized"), shape = tensor([640, 640, 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(223740480)))]; tensor hidden_states_123_cast_fp16 = conv(bias = up_blocks_1_attentions_1_proj_in_bias_to_fp16, dilations = var_4850, groups = var_4061, pad = hidden_states_123_pad_0, pad_type = hidden_states_123_pad_type_0, strides = var_4848, weight = up_blocks_1_attentions_1_proj_in_weight_to_fp16_palettized, x = add_41_cast_fp16)[name = tensor("hidden_states_123_cast_fp16")]; tensor var_4855 = const()[name = tensor("op_4855"), val = tensor([2, 640, 1, 1024])]; tensor inputs_37_cast_fp16 = reshape(shape = var_4855, x = hidden_states_123_cast_fp16)[name = tensor("inputs_37_cast_fp16")]; tensor var_4865 = const()[name = tensor("op_4865"), val = tensor([1])]; tensor channels_mean_37_cast_fp16 = reduce_mean(axes = var_4865, keep_dims = var_4056, x = inputs_37_cast_fp16)[name = tensor("channels_mean_37_cast_fp16")]; tensor zero_mean_37_cast_fp16 = sub(x = inputs_37_cast_fp16, y = channels_mean_37_cast_fp16)[name = tensor("zero_mean_37_cast_fp16")]; tensor zero_mean_sq_37_cast_fp16 = mul(x = zero_mean_37_cast_fp16, y = zero_mean_37_cast_fp16)[name = tensor("zero_mean_sq_37_cast_fp16")]; tensor var_4869 = const()[name = tensor("op_4869"), val = tensor([1])]; tensor var_4870_cast_fp16 = reduce_mean(axes = var_4869, keep_dims = var_4056, x = zero_mean_sq_37_cast_fp16)[name = tensor("op_4870_cast_fp16")]; tensor var_4871_to_fp16 = const()[name = tensor("op_4871_to_fp16"), val = tensor(0x1.5p-17)]; tensor var_4872_cast_fp16 = add(x = var_4870_cast_fp16, y = var_4871_to_fp16)[name = tensor("op_4872_cast_fp16")]; tensor denom_37_epsilon_0_to_fp16 = const()[name = tensor("denom_37_epsilon_0_to_fp16"), val = tensor(0x1p-24)]; tensor denom_37_cast_fp16 = rsqrt(epsilon = denom_37_epsilon_0_to_fp16, x = var_4872_cast_fp16)[name = tensor("denom_37_cast_fp16")]; tensor out_37_cast_fp16 = mul(x = zero_mean_37_cast_fp16, y = denom_37_cast_fp16)[name = tensor("out_37_cast_fp16")]; tensor var_4876_to_fp16 = const()[name = tensor("op_4876_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(223741824)))]; tensor var_4877_cast_fp16 = add(x = out_37_cast_fp16, y = var_4876_to_fp16)[name = tensor("op_4877_cast_fp16")]; tensor var_4879_to_fp16 = const()[name = tensor("op_4879_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(223743168)))]; tensor hidden_states_125_cast_fp16 = mul(x = var_4877_cast_fp16, y = var_4879_to_fp16)[name = tensor("hidden_states_125_cast_fp16")]; tensor var_4886 = const()[name = tensor("op_4886"), val = tensor([1, 1])]; tensor var_4888 = const()[name = tensor("op_4888"), 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 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(223744512))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(224051776))), name = tensor("up_blocks_1_attentions_1_transformer_blocks_0_attn1_to_q_weight_to_fp16_palettized"), shape = tensor([640, 640, 1, 1])]; tensor q_25_cast_fp16 = conv(dilations = var_4888, groups = var_4061, pad = q_25_pad_0, pad_type = q_25_pad_type_0, strides = var_4886, weight = up_blocks_1_attentions_1_transformer_blocks_0_attn1_to_q_weight_to_fp16_palettized, x = hidden_states_125_cast_fp16)[name = tensor("q_25_cast_fp16")]; tensor var_4892 = const()[name = tensor("op_4892"), val = tensor([1, 1])]; tensor var_4894 = const()[name = tensor("op_4894"), 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 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(224051968))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(224359232))), name = tensor("up_blocks_1_attentions_1_transformer_blocks_0_attn1_to_k_weight_to_fp16_palettized"), shape = tensor([640, 640, 1, 1])]; tensor k_49_cast_fp16 = conv(dilations = var_4894, groups = var_4061, pad = k_49_pad_0, pad_type = k_49_pad_type_0, strides = var_4892, weight = up_blocks_1_attentions_1_transformer_blocks_0_attn1_to_k_weight_to_fp16_palettized, x = hidden_states_125_cast_fp16)[name = tensor("k_49_cast_fp16")]; tensor var_4898 = const()[name = tensor("op_4898"), val = tensor([1, 1])]; tensor var_4900 = const()[name = tensor("op_4900"), 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 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(224359424))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(224666688))), name = tensor("up_blocks_1_attentions_1_transformer_blocks_0_attn1_to_v_weight_to_fp16_palettized"), shape = tensor([640, 640, 1, 1])]; tensor v_25_cast_fp16 = conv(dilations = var_4900, groups = var_4061, pad = v_25_pad_0, pad_type = v_25_pad_type_0, strides = var_4898, weight = up_blocks_1_attentions_1_transformer_blocks_0_attn1_to_v_weight_to_fp16_palettized, x = hidden_states_125_cast_fp16)[name = tensor("v_25_cast_fp16")]; tensor var_4904_begin_0 = const()[name = tensor("op_4904_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_4904_end_0 = const()[name = tensor("op_4904_end_0"), val = tensor([2, 80, 1, 1024])]; tensor var_4904_end_mask_0 = const()[name = tensor("op_4904_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_4904_cast_fp16 = slice_by_index(begin = var_4904_begin_0, end = var_4904_end_0, end_mask = var_4904_end_mask_0, x = q_25_cast_fp16)[name = tensor("op_4904_cast_fp16")]; tensor var_4908_begin_0 = const()[name = tensor("op_4908_begin_0"), val = tensor([0, 80, 0, 0])]; tensor var_4908_end_0 = const()[name = tensor("op_4908_end_0"), val = tensor([2, 160, 1, 1024])]; tensor var_4908_end_mask_0 = const()[name = tensor("op_4908_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_4908_cast_fp16 = slice_by_index(begin = var_4908_begin_0, end = var_4908_end_0, end_mask = var_4908_end_mask_0, x = q_25_cast_fp16)[name = tensor("op_4908_cast_fp16")]; tensor var_4912_begin_0 = const()[name = tensor("op_4912_begin_0"), val = tensor([0, 160, 0, 0])]; tensor var_4912_end_0 = const()[name = tensor("op_4912_end_0"), val = tensor([2, 240, 1, 1024])]; tensor var_4912_end_mask_0 = const()[name = tensor("op_4912_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_4912_cast_fp16 = slice_by_index(begin = var_4912_begin_0, end = var_4912_end_0, end_mask = var_4912_end_mask_0, x = q_25_cast_fp16)[name = tensor("op_4912_cast_fp16")]; tensor var_4916_begin_0 = const()[name = tensor("op_4916_begin_0"), val = tensor([0, 240, 0, 0])]; tensor var_4916_end_0 = const()[name = tensor("op_4916_end_0"), val = tensor([2, 320, 1, 1024])]; tensor var_4916_end_mask_0 = const()[name = tensor("op_4916_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_4916_cast_fp16 = slice_by_index(begin = var_4916_begin_0, end = var_4916_end_0, end_mask = var_4916_end_mask_0, x = q_25_cast_fp16)[name = tensor("op_4916_cast_fp16")]; tensor var_4920_begin_0 = const()[name = tensor("op_4920_begin_0"), val = tensor([0, 320, 0, 0])]; tensor var_4920_end_0 = const()[name = tensor("op_4920_end_0"), val = tensor([2, 400, 1, 1024])]; tensor var_4920_end_mask_0 = const()[name = tensor("op_4920_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_4920_cast_fp16 = slice_by_index(begin = var_4920_begin_0, end = var_4920_end_0, end_mask = var_4920_end_mask_0, x = q_25_cast_fp16)[name = tensor("op_4920_cast_fp16")]; tensor var_4924_begin_0 = const()[name = tensor("op_4924_begin_0"), val = tensor([0, 400, 0, 0])]; tensor var_4924_end_0 = const()[name = tensor("op_4924_end_0"), val = tensor([2, 480, 1, 1024])]; tensor var_4924_end_mask_0 = const()[name = tensor("op_4924_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_4924_cast_fp16 = slice_by_index(begin = var_4924_begin_0, end = var_4924_end_0, end_mask = var_4924_end_mask_0, x = q_25_cast_fp16)[name = tensor("op_4924_cast_fp16")]; tensor var_4928_begin_0 = const()[name = tensor("op_4928_begin_0"), val = tensor([0, 480, 0, 0])]; tensor var_4928_end_0 = const()[name = tensor("op_4928_end_0"), val = tensor([2, 560, 1, 1024])]; tensor var_4928_end_mask_0 = const()[name = tensor("op_4928_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_4928_cast_fp16 = slice_by_index(begin = var_4928_begin_0, end = var_4928_end_0, end_mask = var_4928_end_mask_0, x = q_25_cast_fp16)[name = tensor("op_4928_cast_fp16")]; tensor var_4932_begin_0 = const()[name = tensor("op_4932_begin_0"), val = tensor([0, 560, 0, 0])]; tensor var_4932_end_0 = const()[name = tensor("op_4932_end_0"), val = tensor([2, 640, 1, 1024])]; tensor var_4932_end_mask_0 = const()[name = tensor("op_4932_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_4932_cast_fp16 = slice_by_index(begin = var_4932_begin_0, end = var_4932_end_0, end_mask = var_4932_end_mask_0, x = q_25_cast_fp16)[name = tensor("op_4932_cast_fp16")]; tensor var_4935_begin_0 = const()[name = tensor("op_4935_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_4935_end_0 = const()[name = tensor("op_4935_end_0"), val = tensor([2, 80, 1, 512])]; tensor var_4935_end_mask_0 = const()[name = tensor("op_4935_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4935_cast_fp16 = slice_by_index(begin = var_4935_begin_0, end = var_4935_end_0, end_mask = var_4935_end_mask_0, x = var_4904_cast_fp16)[name = tensor("op_4935_cast_fp16")]; tensor var_4936_begin_0 = const()[name = tensor("op_4936_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_4936_end_0 = const()[name = tensor("op_4936_end_0"), val = tensor([2, 80, 1, 1024])]; tensor var_4936_end_mask_0 = const()[name = tensor("op_4936_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4936_cast_fp16 = slice_by_index(begin = var_4936_begin_0, end = var_4936_end_0, end_mask = var_4936_end_mask_0, x = var_4904_cast_fp16)[name = tensor("op_4936_cast_fp16")]; tensor var_4937_begin_0 = const()[name = tensor("op_4937_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_4937_end_0 = const()[name = tensor("op_4937_end_0"), val = tensor([2, 80, 1, 512])]; tensor var_4937_end_mask_0 = const()[name = tensor("op_4937_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4937_cast_fp16 = slice_by_index(begin = var_4937_begin_0, end = var_4937_end_0, end_mask = var_4937_end_mask_0, x = var_4908_cast_fp16)[name = tensor("op_4937_cast_fp16")]; tensor var_4938_begin_0 = const()[name = tensor("op_4938_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_4938_end_0 = const()[name = tensor("op_4938_end_0"), val = tensor([2, 80, 1, 1024])]; tensor var_4938_end_mask_0 = const()[name = tensor("op_4938_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4938_cast_fp16 = slice_by_index(begin = var_4938_begin_0, end = var_4938_end_0, end_mask = var_4938_end_mask_0, x = var_4908_cast_fp16)[name = tensor("op_4938_cast_fp16")]; tensor var_4939_begin_0 = const()[name = tensor("op_4939_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_4939_end_0 = const()[name = tensor("op_4939_end_0"), val = tensor([2, 80, 1, 512])]; tensor var_4939_end_mask_0 = const()[name = tensor("op_4939_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4939_cast_fp16 = slice_by_index(begin = var_4939_begin_0, end = var_4939_end_0, end_mask = var_4939_end_mask_0, x = var_4912_cast_fp16)[name = tensor("op_4939_cast_fp16")]; tensor var_4940_begin_0 = const()[name = tensor("op_4940_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_4940_end_0 = const()[name = tensor("op_4940_end_0"), val = tensor([2, 80, 1, 1024])]; tensor var_4940_end_mask_0 = const()[name = tensor("op_4940_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4940_cast_fp16 = slice_by_index(begin = var_4940_begin_0, end = var_4940_end_0, end_mask = var_4940_end_mask_0, x = var_4912_cast_fp16)[name = tensor("op_4940_cast_fp16")]; tensor var_4941_begin_0 = const()[name = tensor("op_4941_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_4941_end_0 = const()[name = tensor("op_4941_end_0"), val = tensor([2, 80, 1, 512])]; tensor var_4941_end_mask_0 = const()[name = tensor("op_4941_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4941_cast_fp16 = slice_by_index(begin = var_4941_begin_0, end = var_4941_end_0, end_mask = var_4941_end_mask_0, x = var_4916_cast_fp16)[name = tensor("op_4941_cast_fp16")]; tensor var_4942_begin_0 = const()[name = tensor("op_4942_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_4942_end_0 = const()[name = tensor("op_4942_end_0"), val = tensor([2, 80, 1, 1024])]; tensor var_4942_end_mask_0 = const()[name = tensor("op_4942_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4942_cast_fp16 = slice_by_index(begin = var_4942_begin_0, end = var_4942_end_0, end_mask = var_4942_end_mask_0, x = var_4916_cast_fp16)[name = tensor("op_4942_cast_fp16")]; tensor var_4943_begin_0 = const()[name = tensor("op_4943_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_4943_end_0 = const()[name = tensor("op_4943_end_0"), val = tensor([2, 80, 1, 512])]; tensor var_4943_end_mask_0 = const()[name = tensor("op_4943_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4943_cast_fp16 = slice_by_index(begin = var_4943_begin_0, end = var_4943_end_0, end_mask = var_4943_end_mask_0, x = var_4920_cast_fp16)[name = tensor("op_4943_cast_fp16")]; tensor var_4944_begin_0 = const()[name = tensor("op_4944_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_4944_end_0 = const()[name = tensor("op_4944_end_0"), val = tensor([2, 80, 1, 1024])]; tensor var_4944_end_mask_0 = const()[name = tensor("op_4944_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4944_cast_fp16 = slice_by_index(begin = var_4944_begin_0, end = var_4944_end_0, end_mask = var_4944_end_mask_0, x = var_4920_cast_fp16)[name = tensor("op_4944_cast_fp16")]; tensor var_4945_begin_0 = const()[name = tensor("op_4945_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_4945_end_0 = const()[name = tensor("op_4945_end_0"), val = tensor([2, 80, 1, 512])]; tensor var_4945_end_mask_0 = const()[name = tensor("op_4945_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4945_cast_fp16 = slice_by_index(begin = var_4945_begin_0, end = var_4945_end_0, end_mask = var_4945_end_mask_0, x = var_4924_cast_fp16)[name = tensor("op_4945_cast_fp16")]; tensor var_4946_begin_0 = const()[name = tensor("op_4946_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_4946_end_0 = const()[name = tensor("op_4946_end_0"), val = tensor([2, 80, 1, 1024])]; tensor var_4946_end_mask_0 = const()[name = tensor("op_4946_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4946_cast_fp16 = slice_by_index(begin = var_4946_begin_0, end = var_4946_end_0, end_mask = var_4946_end_mask_0, x = var_4924_cast_fp16)[name = tensor("op_4946_cast_fp16")]; tensor var_4947_begin_0 = const()[name = tensor("op_4947_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_4947_end_0 = const()[name = tensor("op_4947_end_0"), val = tensor([2, 80, 1, 512])]; tensor var_4947_end_mask_0 = const()[name = tensor("op_4947_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4947_cast_fp16 = slice_by_index(begin = var_4947_begin_0, end = var_4947_end_0, end_mask = var_4947_end_mask_0, x = var_4928_cast_fp16)[name = tensor("op_4947_cast_fp16")]; tensor var_4948_begin_0 = const()[name = tensor("op_4948_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_4948_end_0 = const()[name = tensor("op_4948_end_0"), val = tensor([2, 80, 1, 1024])]; tensor var_4948_end_mask_0 = const()[name = tensor("op_4948_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4948_cast_fp16 = slice_by_index(begin = var_4948_begin_0, end = var_4948_end_0, end_mask = var_4948_end_mask_0, x = var_4928_cast_fp16)[name = tensor("op_4948_cast_fp16")]; tensor var_4949_begin_0 = const()[name = tensor("op_4949_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_4949_end_0 = const()[name = tensor("op_4949_end_0"), val = tensor([2, 80, 1, 512])]; tensor var_4949_end_mask_0 = const()[name = tensor("op_4949_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4949_cast_fp16 = slice_by_index(begin = var_4949_begin_0, end = var_4949_end_0, end_mask = var_4949_end_mask_0, x = var_4932_cast_fp16)[name = tensor("op_4949_cast_fp16")]; tensor var_4950_begin_0 = const()[name = tensor("op_4950_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_4950_end_0 = const()[name = tensor("op_4950_end_0"), val = tensor([2, 80, 1, 1024])]; tensor var_4950_end_mask_0 = const()[name = tensor("op_4950_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4950_cast_fp16 = slice_by_index(begin = var_4950_begin_0, end = var_4950_end_0, end_mask = var_4950_end_mask_0, x = var_4932_cast_fp16)[name = tensor("op_4950_cast_fp16")]; tensor k_51_perm_0 = const()[name = tensor("k_51_perm_0"), val = tensor([0, 3, 2, 1])]; tensor var_4955_begin_0 = const()[name = tensor("op_4955_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_4955_end_0 = const()[name = tensor("op_4955_end_0"), val = tensor([2, 1024, 1, 80])]; tensor var_4955_end_mask_0 = const()[name = tensor("op_4955_end_mask_0"), val = tensor([true, true, true, false])]; tensor transpose_5 = transpose(perm = k_51_perm_0, x = k_49_cast_fp16)[name = tensor("transpose_5")]; tensor var_4955_cast_fp16 = slice_by_index(begin = var_4955_begin_0, end = var_4955_end_0, end_mask = var_4955_end_mask_0, x = transpose_5)[name = tensor("op_4955_cast_fp16")]; tensor var_4959_begin_0 = const()[name = tensor("op_4959_begin_0"), val = tensor([0, 0, 0, 80])]; tensor var_4959_end_0 = const()[name = tensor("op_4959_end_0"), val = tensor([2, 1024, 1, 160])]; tensor var_4959_end_mask_0 = const()[name = tensor("op_4959_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4959_cast_fp16 = slice_by_index(begin = var_4959_begin_0, end = var_4959_end_0, end_mask = var_4959_end_mask_0, x = transpose_5)[name = tensor("op_4959_cast_fp16")]; tensor var_4963_begin_0 = const()[name = tensor("op_4963_begin_0"), val = tensor([0, 0, 0, 160])]; tensor var_4963_end_0 = const()[name = tensor("op_4963_end_0"), val = tensor([2, 1024, 1, 240])]; tensor var_4963_end_mask_0 = const()[name = tensor("op_4963_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4963_cast_fp16 = slice_by_index(begin = var_4963_begin_0, end = var_4963_end_0, end_mask = var_4963_end_mask_0, x = transpose_5)[name = tensor("op_4963_cast_fp16")]; tensor var_4967_begin_0 = const()[name = tensor("op_4967_begin_0"), val = tensor([0, 0, 0, 240])]; tensor var_4967_end_0 = const()[name = tensor("op_4967_end_0"), val = tensor([2, 1024, 1, 320])]; tensor var_4967_end_mask_0 = const()[name = tensor("op_4967_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4967_cast_fp16 = slice_by_index(begin = var_4967_begin_0, end = var_4967_end_0, end_mask = var_4967_end_mask_0, x = transpose_5)[name = tensor("op_4967_cast_fp16")]; tensor var_4971_begin_0 = const()[name = tensor("op_4971_begin_0"), val = tensor([0, 0, 0, 320])]; tensor var_4971_end_0 = const()[name = tensor("op_4971_end_0"), val = tensor([2, 1024, 1, 400])]; tensor var_4971_end_mask_0 = const()[name = tensor("op_4971_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4971_cast_fp16 = slice_by_index(begin = var_4971_begin_0, end = var_4971_end_0, end_mask = var_4971_end_mask_0, x = transpose_5)[name = tensor("op_4971_cast_fp16")]; tensor var_4975_begin_0 = const()[name = tensor("op_4975_begin_0"), val = tensor([0, 0, 0, 400])]; tensor var_4975_end_0 = const()[name = tensor("op_4975_end_0"), val = tensor([2, 1024, 1, 480])]; tensor var_4975_end_mask_0 = const()[name = tensor("op_4975_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4975_cast_fp16 = slice_by_index(begin = var_4975_begin_0, end = var_4975_end_0, end_mask = var_4975_end_mask_0, x = transpose_5)[name = tensor("op_4975_cast_fp16")]; tensor var_4979_begin_0 = const()[name = tensor("op_4979_begin_0"), val = tensor([0, 0, 0, 480])]; tensor var_4979_end_0 = const()[name = tensor("op_4979_end_0"), val = tensor([2, 1024, 1, 560])]; tensor var_4979_end_mask_0 = const()[name = tensor("op_4979_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4979_cast_fp16 = slice_by_index(begin = var_4979_begin_0, end = var_4979_end_0, end_mask = var_4979_end_mask_0, x = transpose_5)[name = tensor("op_4979_cast_fp16")]; tensor var_4983_begin_0 = const()[name = tensor("op_4983_begin_0"), val = tensor([0, 0, 0, 560])]; tensor var_4983_end_0 = const()[name = tensor("op_4983_end_0"), val = tensor([2, 1024, 1, 640])]; tensor var_4983_end_mask_0 = const()[name = tensor("op_4983_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4983_cast_fp16 = slice_by_index(begin = var_4983_begin_0, end = var_4983_end_0, end_mask = var_4983_end_mask_0, x = transpose_5)[name = tensor("op_4983_cast_fp16")]; tensor var_4985_begin_0 = const()[name = tensor("op_4985_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_4985_end_0 = const()[name = tensor("op_4985_end_0"), val = tensor([2, 80, 1, 1024])]; tensor var_4985_end_mask_0 = const()[name = tensor("op_4985_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_4985_cast_fp16 = slice_by_index(begin = var_4985_begin_0, end = var_4985_end_0, end_mask = var_4985_end_mask_0, x = v_25_cast_fp16)[name = tensor("op_4985_cast_fp16")]; tensor var_4989_begin_0 = const()[name = tensor("op_4989_begin_0"), val = tensor([0, 80, 0, 0])]; tensor var_4989_end_0 = const()[name = tensor("op_4989_end_0"), val = tensor([2, 160, 1, 1024])]; tensor var_4989_end_mask_0 = const()[name = tensor("op_4989_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_4989_cast_fp16 = slice_by_index(begin = var_4989_begin_0, end = var_4989_end_0, end_mask = var_4989_end_mask_0, x = v_25_cast_fp16)[name = tensor("op_4989_cast_fp16")]; tensor var_4993_begin_0 = const()[name = tensor("op_4993_begin_0"), val = tensor([0, 160, 0, 0])]; tensor var_4993_end_0 = const()[name = tensor("op_4993_end_0"), val = tensor([2, 240, 1, 1024])]; tensor var_4993_end_mask_0 = const()[name = tensor("op_4993_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_4993_cast_fp16 = slice_by_index(begin = var_4993_begin_0, end = var_4993_end_0, end_mask = var_4993_end_mask_0, x = v_25_cast_fp16)[name = tensor("op_4993_cast_fp16")]; tensor var_4997_begin_0 = const()[name = tensor("op_4997_begin_0"), val = tensor([0, 240, 0, 0])]; tensor var_4997_end_0 = const()[name = tensor("op_4997_end_0"), val = tensor([2, 320, 1, 1024])]; tensor var_4997_end_mask_0 = const()[name = tensor("op_4997_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_4997_cast_fp16 = slice_by_index(begin = var_4997_begin_0, end = var_4997_end_0, end_mask = var_4997_end_mask_0, x = v_25_cast_fp16)[name = tensor("op_4997_cast_fp16")]; tensor var_5001_begin_0 = const()[name = tensor("op_5001_begin_0"), val = tensor([0, 320, 0, 0])]; tensor var_5001_end_0 = const()[name = tensor("op_5001_end_0"), val = tensor([2, 400, 1, 1024])]; tensor var_5001_end_mask_0 = const()[name = tensor("op_5001_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_5001_cast_fp16 = slice_by_index(begin = var_5001_begin_0, end = var_5001_end_0, end_mask = var_5001_end_mask_0, x = v_25_cast_fp16)[name = tensor("op_5001_cast_fp16")]; tensor var_5005_begin_0 = const()[name = tensor("op_5005_begin_0"), val = tensor([0, 400, 0, 0])]; tensor var_5005_end_0 = const()[name = tensor("op_5005_end_0"), val = tensor([2, 480, 1, 1024])]; tensor var_5005_end_mask_0 = const()[name = tensor("op_5005_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_5005_cast_fp16 = slice_by_index(begin = var_5005_begin_0, end = var_5005_end_0, end_mask = var_5005_end_mask_0, x = v_25_cast_fp16)[name = tensor("op_5005_cast_fp16")]; tensor var_5009_begin_0 = const()[name = tensor("op_5009_begin_0"), val = tensor([0, 480, 0, 0])]; tensor var_5009_end_0 = const()[name = tensor("op_5009_end_0"), val = tensor([2, 560, 1, 1024])]; tensor var_5009_end_mask_0 = const()[name = tensor("op_5009_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_5009_cast_fp16 = slice_by_index(begin = var_5009_begin_0, end = var_5009_end_0, end_mask = var_5009_end_mask_0, x = v_25_cast_fp16)[name = tensor("op_5009_cast_fp16")]; tensor var_5013_begin_0 = const()[name = tensor("op_5013_begin_0"), val = tensor([0, 560, 0, 0])]; tensor var_5013_end_0 = const()[name = tensor("op_5013_end_0"), val = tensor([2, 640, 1, 1024])]; tensor var_5013_end_mask_0 = const()[name = tensor("op_5013_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_5013_cast_fp16 = slice_by_index(begin = var_5013_begin_0, end = var_5013_end_0, end_mask = var_5013_end_mask_0, x = v_25_cast_fp16)[name = tensor("op_5013_cast_fp16")]; tensor var_5017_equation_0 = const()[name = tensor("op_5017_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_5017_cast_fp16 = einsum(equation = var_5017_equation_0, values = (var_4955_cast_fp16, var_4935_cast_fp16))[name = tensor("op_5017_cast_fp16")]; tensor var_5018_to_fp16 = const()[name = tensor("op_5018_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_385_cast_fp16 = mul(x = var_5017_cast_fp16, y = var_5018_to_fp16)[name = tensor("aw_chunk_385_cast_fp16")]; tensor var_5021_equation_0 = const()[name = tensor("op_5021_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_5021_cast_fp16 = einsum(equation = var_5021_equation_0, values = (var_4955_cast_fp16, var_4936_cast_fp16))[name = tensor("op_5021_cast_fp16")]; tensor var_5022_to_fp16 = const()[name = tensor("op_5022_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_387_cast_fp16 = mul(x = var_5021_cast_fp16, y = var_5022_to_fp16)[name = tensor("aw_chunk_387_cast_fp16")]; tensor var_5025_equation_0 = const()[name = tensor("op_5025_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_5025_cast_fp16 = einsum(equation = var_5025_equation_0, values = (var_4959_cast_fp16, var_4937_cast_fp16))[name = tensor("op_5025_cast_fp16")]; tensor var_5026_to_fp16 = const()[name = tensor("op_5026_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_389_cast_fp16 = mul(x = var_5025_cast_fp16, y = var_5026_to_fp16)[name = tensor("aw_chunk_389_cast_fp16")]; tensor var_5029_equation_0 = const()[name = tensor("op_5029_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_5029_cast_fp16 = einsum(equation = var_5029_equation_0, values = (var_4959_cast_fp16, var_4938_cast_fp16))[name = tensor("op_5029_cast_fp16")]; tensor var_5030_to_fp16 = const()[name = tensor("op_5030_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_391_cast_fp16 = mul(x = var_5029_cast_fp16, y = var_5030_to_fp16)[name = tensor("aw_chunk_391_cast_fp16")]; tensor var_5033_equation_0 = const()[name = tensor("op_5033_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_5033_cast_fp16 = einsum(equation = var_5033_equation_0, values = (var_4963_cast_fp16, var_4939_cast_fp16))[name = tensor("op_5033_cast_fp16")]; tensor var_5034_to_fp16 = const()[name = tensor("op_5034_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_393_cast_fp16 = mul(x = var_5033_cast_fp16, y = var_5034_to_fp16)[name = tensor("aw_chunk_393_cast_fp16")]; tensor var_5037_equation_0 = const()[name = tensor("op_5037_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_5037_cast_fp16 = einsum(equation = var_5037_equation_0, values = (var_4963_cast_fp16, var_4940_cast_fp16))[name = tensor("op_5037_cast_fp16")]; tensor var_5038_to_fp16 = const()[name = tensor("op_5038_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_395_cast_fp16 = mul(x = var_5037_cast_fp16, y = var_5038_to_fp16)[name = tensor("aw_chunk_395_cast_fp16")]; tensor var_5041_equation_0 = const()[name = tensor("op_5041_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_5041_cast_fp16 = einsum(equation = var_5041_equation_0, values = (var_4967_cast_fp16, var_4941_cast_fp16))[name = tensor("op_5041_cast_fp16")]; tensor var_5042_to_fp16 = const()[name = tensor("op_5042_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_397_cast_fp16 = mul(x = var_5041_cast_fp16, y = var_5042_to_fp16)[name = tensor("aw_chunk_397_cast_fp16")]; tensor var_5045_equation_0 = const()[name = tensor("op_5045_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_5045_cast_fp16 = einsum(equation = var_5045_equation_0, values = (var_4967_cast_fp16, var_4942_cast_fp16))[name = tensor("op_5045_cast_fp16")]; tensor var_5046_to_fp16 = const()[name = tensor("op_5046_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_399_cast_fp16 = mul(x = var_5045_cast_fp16, y = var_5046_to_fp16)[name = tensor("aw_chunk_399_cast_fp16")]; tensor var_5049_equation_0 = const()[name = tensor("op_5049_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_5049_cast_fp16 = einsum(equation = var_5049_equation_0, values = (var_4971_cast_fp16, var_4943_cast_fp16))[name = tensor("op_5049_cast_fp16")]; tensor var_5050_to_fp16 = const()[name = tensor("op_5050_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_401_cast_fp16 = mul(x = var_5049_cast_fp16, y = var_5050_to_fp16)[name = tensor("aw_chunk_401_cast_fp16")]; tensor var_5053_equation_0 = const()[name = tensor("op_5053_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_5053_cast_fp16 = einsum(equation = var_5053_equation_0, values = (var_4971_cast_fp16, var_4944_cast_fp16))[name = tensor("op_5053_cast_fp16")]; tensor var_5054_to_fp16 = const()[name = tensor("op_5054_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_403_cast_fp16 = mul(x = var_5053_cast_fp16, y = var_5054_to_fp16)[name = tensor("aw_chunk_403_cast_fp16")]; tensor var_5057_equation_0 = const()[name = tensor("op_5057_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_5057_cast_fp16 = einsum(equation = var_5057_equation_0, values = (var_4975_cast_fp16, var_4945_cast_fp16))[name = tensor("op_5057_cast_fp16")]; tensor var_5058_to_fp16 = const()[name = tensor("op_5058_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_405_cast_fp16 = mul(x = var_5057_cast_fp16, y = var_5058_to_fp16)[name = tensor("aw_chunk_405_cast_fp16")]; tensor var_5061_equation_0 = const()[name = tensor("op_5061_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_5061_cast_fp16 = einsum(equation = var_5061_equation_0, values = (var_4975_cast_fp16, var_4946_cast_fp16))[name = tensor("op_5061_cast_fp16")]; tensor var_5062_to_fp16 = const()[name = tensor("op_5062_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_407_cast_fp16 = mul(x = var_5061_cast_fp16, y = var_5062_to_fp16)[name = tensor("aw_chunk_407_cast_fp16")]; tensor var_5065_equation_0 = const()[name = tensor("op_5065_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_5065_cast_fp16 = einsum(equation = var_5065_equation_0, values = (var_4979_cast_fp16, var_4947_cast_fp16))[name = tensor("op_5065_cast_fp16")]; tensor var_5066_to_fp16 = const()[name = tensor("op_5066_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_409_cast_fp16 = mul(x = var_5065_cast_fp16, y = var_5066_to_fp16)[name = tensor("aw_chunk_409_cast_fp16")]; tensor var_5069_equation_0 = const()[name = tensor("op_5069_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_5069_cast_fp16 = einsum(equation = var_5069_equation_0, values = (var_4979_cast_fp16, var_4948_cast_fp16))[name = tensor("op_5069_cast_fp16")]; tensor var_5070_to_fp16 = const()[name = tensor("op_5070_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_411_cast_fp16 = mul(x = var_5069_cast_fp16, y = var_5070_to_fp16)[name = tensor("aw_chunk_411_cast_fp16")]; tensor var_5073_equation_0 = const()[name = tensor("op_5073_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_5073_cast_fp16 = einsum(equation = var_5073_equation_0, values = (var_4983_cast_fp16, var_4949_cast_fp16))[name = tensor("op_5073_cast_fp16")]; tensor var_5074_to_fp16 = const()[name = tensor("op_5074_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_413_cast_fp16 = mul(x = var_5073_cast_fp16, y = var_5074_to_fp16)[name = tensor("aw_chunk_413_cast_fp16")]; tensor var_5077_equation_0 = const()[name = tensor("op_5077_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_5077_cast_fp16 = einsum(equation = var_5077_equation_0, values = (var_4983_cast_fp16, var_4950_cast_fp16))[name = tensor("op_5077_cast_fp16")]; tensor var_5078_to_fp16 = const()[name = tensor("op_5078_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_415_cast_fp16 = mul(x = var_5077_cast_fp16, y = var_5078_to_fp16)[name = tensor("aw_chunk_415_cast_fp16")]; tensor var_5080_cast_fp16 = softmax(axis = var_4061, x = aw_chunk_385_cast_fp16)[name = tensor("op_5080_cast_fp16")]; tensor var_5081_cast_fp16 = softmax(axis = var_4061, x = aw_chunk_387_cast_fp16)[name = tensor("op_5081_cast_fp16")]; tensor var_5082_cast_fp16 = softmax(axis = var_4061, x = aw_chunk_389_cast_fp16)[name = tensor("op_5082_cast_fp16")]; tensor var_5083_cast_fp16 = softmax(axis = var_4061, x = aw_chunk_391_cast_fp16)[name = tensor("op_5083_cast_fp16")]; tensor var_5084_cast_fp16 = softmax(axis = var_4061, x = aw_chunk_393_cast_fp16)[name = tensor("op_5084_cast_fp16")]; tensor var_5085_cast_fp16 = softmax(axis = var_4061, x = aw_chunk_395_cast_fp16)[name = tensor("op_5085_cast_fp16")]; tensor var_5086_cast_fp16 = softmax(axis = var_4061, x = aw_chunk_397_cast_fp16)[name = tensor("op_5086_cast_fp16")]; tensor var_5087_cast_fp16 = softmax(axis = var_4061, x = aw_chunk_399_cast_fp16)[name = tensor("op_5087_cast_fp16")]; tensor var_5088_cast_fp16 = softmax(axis = var_4061, x = aw_chunk_401_cast_fp16)[name = tensor("op_5088_cast_fp16")]; tensor var_5089_cast_fp16 = softmax(axis = var_4061, x = aw_chunk_403_cast_fp16)[name = tensor("op_5089_cast_fp16")]; tensor var_5090_cast_fp16 = softmax(axis = var_4061, x = aw_chunk_405_cast_fp16)[name = tensor("op_5090_cast_fp16")]; tensor var_5091_cast_fp16 = softmax(axis = var_4061, x = aw_chunk_407_cast_fp16)[name = tensor("op_5091_cast_fp16")]; tensor var_5092_cast_fp16 = softmax(axis = var_4061, x = aw_chunk_409_cast_fp16)[name = tensor("op_5092_cast_fp16")]; tensor var_5093_cast_fp16 = softmax(axis = var_4061, x = aw_chunk_411_cast_fp16)[name = tensor("op_5093_cast_fp16")]; tensor var_5094_cast_fp16 = softmax(axis = var_4061, x = aw_chunk_413_cast_fp16)[name = tensor("op_5094_cast_fp16")]; tensor var_5095_cast_fp16 = softmax(axis = var_4061, x = aw_chunk_415_cast_fp16)[name = tensor("op_5095_cast_fp16")]; tensor var_5097_equation_0 = const()[name = tensor("op_5097_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_5097_cast_fp16 = einsum(equation = var_5097_equation_0, values = (var_4985_cast_fp16, var_5080_cast_fp16))[name = tensor("op_5097_cast_fp16")]; tensor var_5099_equation_0 = const()[name = tensor("op_5099_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_5099_cast_fp16 = einsum(equation = var_5099_equation_0, values = (var_4985_cast_fp16, var_5081_cast_fp16))[name = tensor("op_5099_cast_fp16")]; tensor var_5101_equation_0 = const()[name = tensor("op_5101_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_5101_cast_fp16 = einsum(equation = var_5101_equation_0, values = (var_4989_cast_fp16, var_5082_cast_fp16))[name = tensor("op_5101_cast_fp16")]; tensor var_5103_equation_0 = const()[name = tensor("op_5103_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_5103_cast_fp16 = einsum(equation = var_5103_equation_0, values = (var_4989_cast_fp16, var_5083_cast_fp16))[name = tensor("op_5103_cast_fp16")]; tensor var_5105_equation_0 = const()[name = tensor("op_5105_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_5105_cast_fp16 = einsum(equation = var_5105_equation_0, values = (var_4993_cast_fp16, var_5084_cast_fp16))[name = tensor("op_5105_cast_fp16")]; tensor var_5107_equation_0 = const()[name = tensor("op_5107_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_5107_cast_fp16 = einsum(equation = var_5107_equation_0, values = (var_4993_cast_fp16, var_5085_cast_fp16))[name = tensor("op_5107_cast_fp16")]; tensor var_5109_equation_0 = const()[name = tensor("op_5109_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_5109_cast_fp16 = einsum(equation = var_5109_equation_0, values = (var_4997_cast_fp16, var_5086_cast_fp16))[name = tensor("op_5109_cast_fp16")]; tensor var_5111_equation_0 = const()[name = tensor("op_5111_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_5111_cast_fp16 = einsum(equation = var_5111_equation_0, values = (var_4997_cast_fp16, var_5087_cast_fp16))[name = tensor("op_5111_cast_fp16")]; tensor var_5113_equation_0 = const()[name = tensor("op_5113_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_5113_cast_fp16 = einsum(equation = var_5113_equation_0, values = (var_5001_cast_fp16, var_5088_cast_fp16))[name = tensor("op_5113_cast_fp16")]; tensor var_5115_equation_0 = const()[name = tensor("op_5115_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_5115_cast_fp16 = einsum(equation = var_5115_equation_0, values = (var_5001_cast_fp16, var_5089_cast_fp16))[name = tensor("op_5115_cast_fp16")]; tensor var_5117_equation_0 = const()[name = tensor("op_5117_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_5117_cast_fp16 = einsum(equation = var_5117_equation_0, values = (var_5005_cast_fp16, var_5090_cast_fp16))[name = tensor("op_5117_cast_fp16")]; tensor var_5119_equation_0 = const()[name = tensor("op_5119_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_5119_cast_fp16 = einsum(equation = var_5119_equation_0, values = (var_5005_cast_fp16, var_5091_cast_fp16))[name = tensor("op_5119_cast_fp16")]; tensor var_5121_equation_0 = const()[name = tensor("op_5121_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_5121_cast_fp16 = einsum(equation = var_5121_equation_0, values = (var_5009_cast_fp16, var_5092_cast_fp16))[name = tensor("op_5121_cast_fp16")]; tensor var_5123_equation_0 = const()[name = tensor("op_5123_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_5123_cast_fp16 = einsum(equation = var_5123_equation_0, values = (var_5009_cast_fp16, var_5093_cast_fp16))[name = tensor("op_5123_cast_fp16")]; tensor var_5125_equation_0 = const()[name = tensor("op_5125_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_5125_cast_fp16 = einsum(equation = var_5125_equation_0, values = (var_5013_cast_fp16, var_5094_cast_fp16))[name = tensor("op_5125_cast_fp16")]; tensor var_5127_equation_0 = const()[name = tensor("op_5127_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_5127_cast_fp16 = einsum(equation = var_5127_equation_0, values = (var_5013_cast_fp16, var_5095_cast_fp16))[name = tensor("op_5127_cast_fp16")]; tensor var_5129_interleave_0 = const()[name = tensor("op_5129_interleave_0"), val = tensor(false)]; tensor var_5129_cast_fp16 = concat(axis = var_4039, interleave = var_5129_interleave_0, values = (var_5097_cast_fp16, var_5099_cast_fp16))[name = tensor("op_5129_cast_fp16")]; tensor var_5131_interleave_0 = const()[name = tensor("op_5131_interleave_0"), val = tensor(false)]; tensor var_5131_cast_fp16 = concat(axis = var_4039, interleave = var_5131_interleave_0, values = (var_5101_cast_fp16, var_5103_cast_fp16))[name = tensor("op_5131_cast_fp16")]; tensor var_5133_interleave_0 = const()[name = tensor("op_5133_interleave_0"), val = tensor(false)]; tensor var_5133_cast_fp16 = concat(axis = var_4039, interleave = var_5133_interleave_0, values = (var_5105_cast_fp16, var_5107_cast_fp16))[name = tensor("op_5133_cast_fp16")]; tensor var_5135_interleave_0 = const()[name = tensor("op_5135_interleave_0"), val = tensor(false)]; tensor var_5135_cast_fp16 = concat(axis = var_4039, interleave = var_5135_interleave_0, values = (var_5109_cast_fp16, var_5111_cast_fp16))[name = tensor("op_5135_cast_fp16")]; tensor var_5137_interleave_0 = const()[name = tensor("op_5137_interleave_0"), val = tensor(false)]; tensor var_5137_cast_fp16 = concat(axis = var_4039, interleave = var_5137_interleave_0, values = (var_5113_cast_fp16, var_5115_cast_fp16))[name = tensor("op_5137_cast_fp16")]; tensor var_5139_interleave_0 = const()[name = tensor("op_5139_interleave_0"), val = tensor(false)]; tensor var_5139_cast_fp16 = concat(axis = var_4039, interleave = var_5139_interleave_0, values = (var_5117_cast_fp16, var_5119_cast_fp16))[name = tensor("op_5139_cast_fp16")]; tensor var_5141_interleave_0 = const()[name = tensor("op_5141_interleave_0"), val = tensor(false)]; tensor var_5141_cast_fp16 = concat(axis = var_4039, interleave = var_5141_interleave_0, values = (var_5121_cast_fp16, var_5123_cast_fp16))[name = tensor("op_5141_cast_fp16")]; tensor var_5143_interleave_0 = const()[name = tensor("op_5143_interleave_0"), val = tensor(false)]; tensor var_5143_cast_fp16 = concat(axis = var_4039, interleave = var_5143_interleave_0, values = (var_5125_cast_fp16, var_5127_cast_fp16))[name = tensor("op_5143_cast_fp16")]; tensor input_189_interleave_0 = const()[name = tensor("input_189_interleave_0"), val = tensor(false)]; tensor input_189_cast_fp16 = concat(axis = var_4061, interleave = input_189_interleave_0, values = (var_5129_cast_fp16, var_5131_cast_fp16, var_5133_cast_fp16, var_5135_cast_fp16, var_5137_cast_fp16, var_5139_cast_fp16, var_5141_cast_fp16, var_5143_cast_fp16))[name = tensor("input_189_cast_fp16")]; tensor var_5149 = const()[name = tensor("op_5149"), val = tensor([1, 1])]; tensor var_5151 = const()[name = tensor("op_5151"), val = tensor([1, 1])]; tensor var_5153_pad_type_0 = const()[name = tensor("op_5153_pad_type_0"), val = tensor("custom")]; tensor var_5153_pad_0 = const()[name = tensor("op_5153_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(224666880))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(224974144))), name = tensor("up_blocks_1_attentions_1_transformer_blocks_0_attn1_to_out_0_weight_to_fp16_palettized"), shape = tensor([640, 640, 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(224974336)))]; tensor var_5153_cast_fp16 = conv(bias = up_blocks_1_attentions_1_transformer_blocks_0_attn1_to_out_0_bias_to_fp16, dilations = var_5151, groups = var_4061, pad = var_5153_pad_0, pad_type = var_5153_pad_type_0, strides = var_5149, weight = up_blocks_1_attentions_1_transformer_blocks_0_attn1_to_out_0_weight_to_fp16_palettized, x = input_189_cast_fp16)[name = tensor("op_5153_cast_fp16")]; tensor inputs_39_cast_fp16 = add(x = var_5153_cast_fp16, y = inputs_37_cast_fp16)[name = tensor("inputs_39_cast_fp16")]; tensor var_5157 = const()[name = tensor("op_5157"), val = tensor([1])]; tensor channels_mean_39_cast_fp16 = reduce_mean(axes = var_5157, keep_dims = var_4056, x = inputs_39_cast_fp16)[name = tensor("channels_mean_39_cast_fp16")]; tensor zero_mean_39_cast_fp16 = sub(x = inputs_39_cast_fp16, y = channels_mean_39_cast_fp16)[name = tensor("zero_mean_39_cast_fp16")]; tensor zero_mean_sq_39_cast_fp16 = mul(x = zero_mean_39_cast_fp16, y = zero_mean_39_cast_fp16)[name = tensor("zero_mean_sq_39_cast_fp16")]; tensor var_5161 = const()[name = tensor("op_5161"), val = tensor([1])]; tensor var_5162_cast_fp16 = reduce_mean(axes = var_5161, keep_dims = var_4056, x = zero_mean_sq_39_cast_fp16)[name = tensor("op_5162_cast_fp16")]; tensor var_5163_to_fp16 = const()[name = tensor("op_5163_to_fp16"), val = tensor(0x1.5p-17)]; tensor var_5164_cast_fp16 = add(x = var_5162_cast_fp16, y = var_5163_to_fp16)[name = tensor("op_5164_cast_fp16")]; tensor denom_39_epsilon_0_to_fp16 = const()[name = tensor("denom_39_epsilon_0_to_fp16"), val = tensor(0x1p-24)]; tensor denom_39_cast_fp16 = rsqrt(epsilon = denom_39_epsilon_0_to_fp16, x = var_5164_cast_fp16)[name = tensor("denom_39_cast_fp16")]; tensor out_39_cast_fp16 = mul(x = zero_mean_39_cast_fp16, y = denom_39_cast_fp16)[name = tensor("out_39_cast_fp16")]; tensor var_5168_to_fp16 = const()[name = tensor("op_5168_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(224975680)))]; tensor var_5169_cast_fp16 = add(x = out_39_cast_fp16, y = var_5168_to_fp16)[name = tensor("op_5169_cast_fp16")]; tensor var_5171_to_fp16 = const()[name = tensor("op_5171_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(224977024)))]; tensor hidden_states_127_cast_fp16 = mul(x = var_5169_cast_fp16, y = var_5171_to_fp16)[name = tensor("hidden_states_127_cast_fp16")]; tensor var_5178 = const()[name = tensor("op_5178"), val = tensor([1, 1])]; tensor var_5180 = const()[name = tensor("op_5180"), 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 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(224978368))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(225285632))), name = tensor("up_blocks_1_attentions_1_transformer_blocks_0_attn2_to_q_weight_to_fp16_palettized"), shape = tensor([640, 640, 1, 1])]; tensor q_27_cast_fp16 = conv(dilations = var_5180, groups = var_4061, pad = q_27_pad_0, pad_type = q_27_pad_type_0, strides = var_5178, weight = up_blocks_1_attentions_1_transformer_blocks_0_attn2_to_q_weight_to_fp16_palettized, x = hidden_states_127_cast_fp16)[name = tensor("q_27_cast_fp16")]; tensor var_5184 = const()[name = tensor("op_5184"), val = tensor([1, 1])]; tensor var_5186 = const()[name = tensor("op_5186"), 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 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(225285824))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(225654528))), name = tensor("up_blocks_1_attentions_1_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized"), shape = tensor([640, 768, 1, 1])]; tensor k_53_cast_fp16 = conv(dilations = var_5186, groups = var_4061, pad = k_53_pad_0, pad_type = k_53_pad_type_0, strides = var_5184, weight = up_blocks_1_attentions_1_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor("k_53_cast_fp16")]; tensor var_5190 = const()[name = tensor("op_5190"), val = tensor([1, 1])]; tensor var_5192 = const()[name = tensor("op_5192"), 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 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(225654720))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(226023424))), name = tensor("up_blocks_1_attentions_1_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized"), shape = tensor([640, 768, 1, 1])]; tensor v_27_cast_fp16 = conv(dilations = var_5192, groups = var_4061, pad = v_27_pad_0, pad_type = v_27_pad_type_0, strides = var_5190, weight = up_blocks_1_attentions_1_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor("v_27_cast_fp16")]; tensor var_5196_begin_0 = const()[name = tensor("op_5196_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_5196_end_0 = const()[name = tensor("op_5196_end_0"), val = tensor([2, 80, 1, 1024])]; tensor var_5196_end_mask_0 = const()[name = tensor("op_5196_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_5196_cast_fp16 = slice_by_index(begin = var_5196_begin_0, end = var_5196_end_0, end_mask = var_5196_end_mask_0, x = q_27_cast_fp16)[name = tensor("op_5196_cast_fp16")]; tensor var_5200_begin_0 = const()[name = tensor("op_5200_begin_0"), val = tensor([0, 80, 0, 0])]; tensor var_5200_end_0 = const()[name = tensor("op_5200_end_0"), val = tensor([2, 160, 1, 1024])]; tensor var_5200_end_mask_0 = const()[name = tensor("op_5200_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_5200_cast_fp16 = slice_by_index(begin = var_5200_begin_0, end = var_5200_end_0, end_mask = var_5200_end_mask_0, x = q_27_cast_fp16)[name = tensor("op_5200_cast_fp16")]; tensor var_5204_begin_0 = const()[name = tensor("op_5204_begin_0"), val = tensor([0, 160, 0, 0])]; tensor var_5204_end_0 = const()[name = tensor("op_5204_end_0"), val = tensor([2, 240, 1, 1024])]; tensor var_5204_end_mask_0 = const()[name = tensor("op_5204_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_5204_cast_fp16 = slice_by_index(begin = var_5204_begin_0, end = var_5204_end_0, end_mask = var_5204_end_mask_0, x = q_27_cast_fp16)[name = tensor("op_5204_cast_fp16")]; tensor var_5208_begin_0 = const()[name = tensor("op_5208_begin_0"), val = tensor([0, 240, 0, 0])]; tensor var_5208_end_0 = const()[name = tensor("op_5208_end_0"), val = tensor([2, 320, 1, 1024])]; tensor var_5208_end_mask_0 = const()[name = tensor("op_5208_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_5208_cast_fp16 = slice_by_index(begin = var_5208_begin_0, end = var_5208_end_0, end_mask = var_5208_end_mask_0, x = q_27_cast_fp16)[name = tensor("op_5208_cast_fp16")]; tensor var_5212_begin_0 = const()[name = tensor("op_5212_begin_0"), val = tensor([0, 320, 0, 0])]; tensor var_5212_end_0 = const()[name = tensor("op_5212_end_0"), val = tensor([2, 400, 1, 1024])]; tensor var_5212_end_mask_0 = const()[name = tensor("op_5212_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_5212_cast_fp16 = slice_by_index(begin = var_5212_begin_0, end = var_5212_end_0, end_mask = var_5212_end_mask_0, x = q_27_cast_fp16)[name = tensor("op_5212_cast_fp16")]; tensor var_5216_begin_0 = const()[name = tensor("op_5216_begin_0"), val = tensor([0, 400, 0, 0])]; tensor var_5216_end_0 = const()[name = tensor("op_5216_end_0"), val = tensor([2, 480, 1, 1024])]; tensor var_5216_end_mask_0 = const()[name = tensor("op_5216_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_5216_cast_fp16 = slice_by_index(begin = var_5216_begin_0, end = var_5216_end_0, end_mask = var_5216_end_mask_0, x = q_27_cast_fp16)[name = tensor("op_5216_cast_fp16")]; tensor var_5220_begin_0 = const()[name = tensor("op_5220_begin_0"), val = tensor([0, 480, 0, 0])]; tensor var_5220_end_0 = const()[name = tensor("op_5220_end_0"), val = tensor([2, 560, 1, 1024])]; tensor var_5220_end_mask_0 = const()[name = tensor("op_5220_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_5220_cast_fp16 = slice_by_index(begin = var_5220_begin_0, end = var_5220_end_0, end_mask = var_5220_end_mask_0, x = q_27_cast_fp16)[name = tensor("op_5220_cast_fp16")]; tensor var_5224_begin_0 = const()[name = tensor("op_5224_begin_0"), val = tensor([0, 560, 0, 0])]; tensor var_5224_end_0 = const()[name = tensor("op_5224_end_0"), val = tensor([2, 640, 1, 1024])]; tensor var_5224_end_mask_0 = const()[name = tensor("op_5224_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_5224_cast_fp16 = slice_by_index(begin = var_5224_begin_0, end = var_5224_end_0, end_mask = var_5224_end_mask_0, x = q_27_cast_fp16)[name = tensor("op_5224_cast_fp16")]; tensor var_5227_begin_0 = const()[name = tensor("op_5227_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_5227_end_0 = const()[name = tensor("op_5227_end_0"), val = tensor([2, 80, 1, 512])]; tensor var_5227_end_mask_0 = const()[name = tensor("op_5227_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5227_cast_fp16 = slice_by_index(begin = var_5227_begin_0, end = var_5227_end_0, end_mask = var_5227_end_mask_0, x = var_5196_cast_fp16)[name = tensor("op_5227_cast_fp16")]; tensor var_5228_begin_0 = const()[name = tensor("op_5228_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_5228_end_0 = const()[name = tensor("op_5228_end_0"), val = tensor([2, 80, 1, 1024])]; tensor var_5228_end_mask_0 = const()[name = tensor("op_5228_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5228_cast_fp16 = slice_by_index(begin = var_5228_begin_0, end = var_5228_end_0, end_mask = var_5228_end_mask_0, x = var_5196_cast_fp16)[name = tensor("op_5228_cast_fp16")]; tensor var_5229_begin_0 = const()[name = tensor("op_5229_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_5229_end_0 = const()[name = tensor("op_5229_end_0"), val = tensor([2, 80, 1, 512])]; tensor var_5229_end_mask_0 = const()[name = tensor("op_5229_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5229_cast_fp16 = slice_by_index(begin = var_5229_begin_0, end = var_5229_end_0, end_mask = var_5229_end_mask_0, x = var_5200_cast_fp16)[name = tensor("op_5229_cast_fp16")]; tensor var_5230_begin_0 = const()[name = tensor("op_5230_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_5230_end_0 = const()[name = tensor("op_5230_end_0"), val = tensor([2, 80, 1, 1024])]; tensor var_5230_end_mask_0 = const()[name = tensor("op_5230_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5230_cast_fp16 = slice_by_index(begin = var_5230_begin_0, end = var_5230_end_0, end_mask = var_5230_end_mask_0, x = var_5200_cast_fp16)[name = tensor("op_5230_cast_fp16")]; tensor var_5231_begin_0 = const()[name = tensor("op_5231_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_5231_end_0 = const()[name = tensor("op_5231_end_0"), val = tensor([2, 80, 1, 512])]; tensor var_5231_end_mask_0 = const()[name = tensor("op_5231_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5231_cast_fp16 = slice_by_index(begin = var_5231_begin_0, end = var_5231_end_0, end_mask = var_5231_end_mask_0, x = var_5204_cast_fp16)[name = tensor("op_5231_cast_fp16")]; tensor var_5232_begin_0 = const()[name = tensor("op_5232_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_5232_end_0 = const()[name = tensor("op_5232_end_0"), val = tensor([2, 80, 1, 1024])]; tensor var_5232_end_mask_0 = const()[name = tensor("op_5232_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5232_cast_fp16 = slice_by_index(begin = var_5232_begin_0, end = var_5232_end_0, end_mask = var_5232_end_mask_0, x = var_5204_cast_fp16)[name = tensor("op_5232_cast_fp16")]; tensor var_5233_begin_0 = const()[name = tensor("op_5233_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_5233_end_0 = const()[name = tensor("op_5233_end_0"), val = tensor([2, 80, 1, 512])]; tensor var_5233_end_mask_0 = const()[name = tensor("op_5233_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5233_cast_fp16 = slice_by_index(begin = var_5233_begin_0, end = var_5233_end_0, end_mask = var_5233_end_mask_0, x = var_5208_cast_fp16)[name = tensor("op_5233_cast_fp16")]; tensor var_5234_begin_0 = const()[name = tensor("op_5234_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_5234_end_0 = const()[name = tensor("op_5234_end_0"), val = tensor([2, 80, 1, 1024])]; tensor var_5234_end_mask_0 = const()[name = tensor("op_5234_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5234_cast_fp16 = slice_by_index(begin = var_5234_begin_0, end = var_5234_end_0, end_mask = var_5234_end_mask_0, x = var_5208_cast_fp16)[name = tensor("op_5234_cast_fp16")]; tensor var_5235_begin_0 = const()[name = tensor("op_5235_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_5235_end_0 = const()[name = tensor("op_5235_end_0"), val = tensor([2, 80, 1, 512])]; tensor var_5235_end_mask_0 = const()[name = tensor("op_5235_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5235_cast_fp16 = slice_by_index(begin = var_5235_begin_0, end = var_5235_end_0, end_mask = var_5235_end_mask_0, x = var_5212_cast_fp16)[name = tensor("op_5235_cast_fp16")]; tensor var_5236_begin_0 = const()[name = tensor("op_5236_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_5236_end_0 = const()[name = tensor("op_5236_end_0"), val = tensor([2, 80, 1, 1024])]; tensor var_5236_end_mask_0 = const()[name = tensor("op_5236_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5236_cast_fp16 = slice_by_index(begin = var_5236_begin_0, end = var_5236_end_0, end_mask = var_5236_end_mask_0, x = var_5212_cast_fp16)[name = tensor("op_5236_cast_fp16")]; tensor var_5237_begin_0 = const()[name = tensor("op_5237_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_5237_end_0 = const()[name = tensor("op_5237_end_0"), val = tensor([2, 80, 1, 512])]; tensor var_5237_end_mask_0 = const()[name = tensor("op_5237_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5237_cast_fp16 = slice_by_index(begin = var_5237_begin_0, end = var_5237_end_0, end_mask = var_5237_end_mask_0, x = var_5216_cast_fp16)[name = tensor("op_5237_cast_fp16")]; tensor var_5238_begin_0 = const()[name = tensor("op_5238_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_5238_end_0 = const()[name = tensor("op_5238_end_0"), val = tensor([2, 80, 1, 1024])]; tensor var_5238_end_mask_0 = const()[name = tensor("op_5238_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5238_cast_fp16 = slice_by_index(begin = var_5238_begin_0, end = var_5238_end_0, end_mask = var_5238_end_mask_0, x = var_5216_cast_fp16)[name = tensor("op_5238_cast_fp16")]; tensor var_5239_begin_0 = const()[name = tensor("op_5239_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_5239_end_0 = const()[name = tensor("op_5239_end_0"), val = tensor([2, 80, 1, 512])]; tensor var_5239_end_mask_0 = const()[name = tensor("op_5239_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5239_cast_fp16 = slice_by_index(begin = var_5239_begin_0, end = var_5239_end_0, end_mask = var_5239_end_mask_0, x = var_5220_cast_fp16)[name = tensor("op_5239_cast_fp16")]; tensor var_5240_begin_0 = const()[name = tensor("op_5240_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_5240_end_0 = const()[name = tensor("op_5240_end_0"), val = tensor([2, 80, 1, 1024])]; tensor var_5240_end_mask_0 = const()[name = tensor("op_5240_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5240_cast_fp16 = slice_by_index(begin = var_5240_begin_0, end = var_5240_end_0, end_mask = var_5240_end_mask_0, x = var_5220_cast_fp16)[name = tensor("op_5240_cast_fp16")]; tensor var_5241_begin_0 = const()[name = tensor("op_5241_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_5241_end_0 = const()[name = tensor("op_5241_end_0"), val = tensor([2, 80, 1, 512])]; tensor var_5241_end_mask_0 = const()[name = tensor("op_5241_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5241_cast_fp16 = slice_by_index(begin = var_5241_begin_0, end = var_5241_end_0, end_mask = var_5241_end_mask_0, x = var_5224_cast_fp16)[name = tensor("op_5241_cast_fp16")]; tensor var_5242_begin_0 = const()[name = tensor("op_5242_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_5242_end_0 = const()[name = tensor("op_5242_end_0"), val = tensor([2, 80, 1, 1024])]; tensor var_5242_end_mask_0 = const()[name = tensor("op_5242_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5242_cast_fp16 = slice_by_index(begin = var_5242_begin_0, end = var_5242_end_0, end_mask = var_5242_end_mask_0, x = var_5224_cast_fp16)[name = tensor("op_5242_cast_fp16")]; tensor k_55_perm_0 = const()[name = tensor("k_55_perm_0"), val = tensor([0, 3, 2, 1])]; tensor var_5247_begin_0 = const()[name = tensor("op_5247_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_5247_end_0 = const()[name = tensor("op_5247_end_0"), val = tensor([2, 77, 1, 80])]; tensor var_5247_end_mask_0 = const()[name = tensor("op_5247_end_mask_0"), val = tensor([true, true, true, false])]; tensor transpose_4 = transpose(perm = k_55_perm_0, x = k_53_cast_fp16)[name = tensor("transpose_4")]; tensor var_5247_cast_fp16 = slice_by_index(begin = var_5247_begin_0, end = var_5247_end_0, end_mask = var_5247_end_mask_0, x = transpose_4)[name = tensor("op_5247_cast_fp16")]; tensor var_5251_begin_0 = const()[name = tensor("op_5251_begin_0"), val = tensor([0, 0, 0, 80])]; tensor var_5251_end_0 = const()[name = tensor("op_5251_end_0"), val = tensor([2, 77, 1, 160])]; tensor var_5251_end_mask_0 = const()[name = tensor("op_5251_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5251_cast_fp16 = slice_by_index(begin = var_5251_begin_0, end = var_5251_end_0, end_mask = var_5251_end_mask_0, x = transpose_4)[name = tensor("op_5251_cast_fp16")]; tensor var_5255_begin_0 = const()[name = tensor("op_5255_begin_0"), val = tensor([0, 0, 0, 160])]; tensor var_5255_end_0 = const()[name = tensor("op_5255_end_0"), val = tensor([2, 77, 1, 240])]; tensor var_5255_end_mask_0 = const()[name = tensor("op_5255_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5255_cast_fp16 = slice_by_index(begin = var_5255_begin_0, end = var_5255_end_0, end_mask = var_5255_end_mask_0, x = transpose_4)[name = tensor("op_5255_cast_fp16")]; tensor var_5259_begin_0 = const()[name = tensor("op_5259_begin_0"), val = tensor([0, 0, 0, 240])]; tensor var_5259_end_0 = const()[name = tensor("op_5259_end_0"), val = tensor([2, 77, 1, 320])]; tensor var_5259_end_mask_0 = const()[name = tensor("op_5259_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5259_cast_fp16 = slice_by_index(begin = var_5259_begin_0, end = var_5259_end_0, end_mask = var_5259_end_mask_0, x = transpose_4)[name = tensor("op_5259_cast_fp16")]; tensor var_5263_begin_0 = const()[name = tensor("op_5263_begin_0"), val = tensor([0, 0, 0, 320])]; tensor var_5263_end_0 = const()[name = tensor("op_5263_end_0"), val = tensor([2, 77, 1, 400])]; tensor var_5263_end_mask_0 = const()[name = tensor("op_5263_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5263_cast_fp16 = slice_by_index(begin = var_5263_begin_0, end = var_5263_end_0, end_mask = var_5263_end_mask_0, x = transpose_4)[name = tensor("op_5263_cast_fp16")]; tensor var_5267_begin_0 = const()[name = tensor("op_5267_begin_0"), val = tensor([0, 0, 0, 400])]; tensor var_5267_end_0 = const()[name = tensor("op_5267_end_0"), val = tensor([2, 77, 1, 480])]; tensor var_5267_end_mask_0 = const()[name = tensor("op_5267_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5267_cast_fp16 = slice_by_index(begin = var_5267_begin_0, end = var_5267_end_0, end_mask = var_5267_end_mask_0, x = transpose_4)[name = tensor("op_5267_cast_fp16")]; tensor var_5271_begin_0 = const()[name = tensor("op_5271_begin_0"), val = tensor([0, 0, 0, 480])]; tensor var_5271_end_0 = const()[name = tensor("op_5271_end_0"), val = tensor([2, 77, 1, 560])]; tensor var_5271_end_mask_0 = const()[name = tensor("op_5271_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5271_cast_fp16 = slice_by_index(begin = var_5271_begin_0, end = var_5271_end_0, end_mask = var_5271_end_mask_0, x = transpose_4)[name = tensor("op_5271_cast_fp16")]; tensor var_5275_begin_0 = const()[name = tensor("op_5275_begin_0"), val = tensor([0, 0, 0, 560])]; tensor var_5275_end_0 = const()[name = tensor("op_5275_end_0"), val = tensor([2, 77, 1, 640])]; tensor var_5275_end_mask_0 = const()[name = tensor("op_5275_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5275_cast_fp16 = slice_by_index(begin = var_5275_begin_0, end = var_5275_end_0, end_mask = var_5275_end_mask_0, x = transpose_4)[name = tensor("op_5275_cast_fp16")]; tensor var_5277_begin_0 = const()[name = tensor("op_5277_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_5277_end_0 = const()[name = tensor("op_5277_end_0"), val = tensor([2, 80, 1, 77])]; tensor var_5277_end_mask_0 = const()[name = tensor("op_5277_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_5277_cast_fp16 = slice_by_index(begin = var_5277_begin_0, end = var_5277_end_0, end_mask = var_5277_end_mask_0, x = v_27_cast_fp16)[name = tensor("op_5277_cast_fp16")]; tensor var_5281_begin_0 = const()[name = tensor("op_5281_begin_0"), val = tensor([0, 80, 0, 0])]; tensor var_5281_end_0 = const()[name = tensor("op_5281_end_0"), val = tensor([2, 160, 1, 77])]; tensor var_5281_end_mask_0 = const()[name = tensor("op_5281_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_5281_cast_fp16 = slice_by_index(begin = var_5281_begin_0, end = var_5281_end_0, end_mask = var_5281_end_mask_0, x = v_27_cast_fp16)[name = tensor("op_5281_cast_fp16")]; tensor var_5285_begin_0 = const()[name = tensor("op_5285_begin_0"), val = tensor([0, 160, 0, 0])]; tensor var_5285_end_0 = const()[name = tensor("op_5285_end_0"), val = tensor([2, 240, 1, 77])]; tensor var_5285_end_mask_0 = const()[name = tensor("op_5285_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_5285_cast_fp16 = slice_by_index(begin = var_5285_begin_0, end = var_5285_end_0, end_mask = var_5285_end_mask_0, x = v_27_cast_fp16)[name = tensor("op_5285_cast_fp16")]; tensor var_5289_begin_0 = const()[name = tensor("op_5289_begin_0"), val = tensor([0, 240, 0, 0])]; tensor var_5289_end_0 = const()[name = tensor("op_5289_end_0"), val = tensor([2, 320, 1, 77])]; tensor var_5289_end_mask_0 = const()[name = tensor("op_5289_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_5289_cast_fp16 = slice_by_index(begin = var_5289_begin_0, end = var_5289_end_0, end_mask = var_5289_end_mask_0, x = v_27_cast_fp16)[name = tensor("op_5289_cast_fp16")]; tensor var_5293_begin_0 = const()[name = tensor("op_5293_begin_0"), val = tensor([0, 320, 0, 0])]; tensor var_5293_end_0 = const()[name = tensor("op_5293_end_0"), val = tensor([2, 400, 1, 77])]; tensor var_5293_end_mask_0 = const()[name = tensor("op_5293_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_5293_cast_fp16 = slice_by_index(begin = var_5293_begin_0, end = var_5293_end_0, end_mask = var_5293_end_mask_0, x = v_27_cast_fp16)[name = tensor("op_5293_cast_fp16")]; tensor var_5297_begin_0 = const()[name = tensor("op_5297_begin_0"), val = tensor([0, 400, 0, 0])]; tensor var_5297_end_0 = const()[name = tensor("op_5297_end_0"), val = tensor([2, 480, 1, 77])]; tensor var_5297_end_mask_0 = const()[name = tensor("op_5297_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_5297_cast_fp16 = slice_by_index(begin = var_5297_begin_0, end = var_5297_end_0, end_mask = var_5297_end_mask_0, x = v_27_cast_fp16)[name = tensor("op_5297_cast_fp16")]; tensor var_5301_begin_0 = const()[name = tensor("op_5301_begin_0"), val = tensor([0, 480, 0, 0])]; tensor var_5301_end_0 = const()[name = tensor("op_5301_end_0"), val = tensor([2, 560, 1, 77])]; tensor var_5301_end_mask_0 = const()[name = tensor("op_5301_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_5301_cast_fp16 = slice_by_index(begin = var_5301_begin_0, end = var_5301_end_0, end_mask = var_5301_end_mask_0, x = v_27_cast_fp16)[name = tensor("op_5301_cast_fp16")]; tensor var_5305_begin_0 = const()[name = tensor("op_5305_begin_0"), val = tensor([0, 560, 0, 0])]; tensor var_5305_end_0 = const()[name = tensor("op_5305_end_0"), val = tensor([2, 640, 1, 77])]; tensor var_5305_end_mask_0 = const()[name = tensor("op_5305_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_5305_cast_fp16 = slice_by_index(begin = var_5305_begin_0, end = var_5305_end_0, end_mask = var_5305_end_mask_0, x = v_27_cast_fp16)[name = tensor("op_5305_cast_fp16")]; tensor var_5309_equation_0 = const()[name = tensor("op_5309_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_5309_cast_fp16 = einsum(equation = var_5309_equation_0, values = (var_5247_cast_fp16, var_5227_cast_fp16))[name = tensor("op_5309_cast_fp16")]; tensor var_5310_to_fp16 = const()[name = tensor("op_5310_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_417_cast_fp16 = mul(x = var_5309_cast_fp16, y = var_5310_to_fp16)[name = tensor("aw_chunk_417_cast_fp16")]; tensor var_5313_equation_0 = const()[name = tensor("op_5313_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_5313_cast_fp16 = einsum(equation = var_5313_equation_0, values = (var_5247_cast_fp16, var_5228_cast_fp16))[name = tensor("op_5313_cast_fp16")]; tensor var_5314_to_fp16 = const()[name = tensor("op_5314_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_419_cast_fp16 = mul(x = var_5313_cast_fp16, y = var_5314_to_fp16)[name = tensor("aw_chunk_419_cast_fp16")]; tensor var_5317_equation_0 = const()[name = tensor("op_5317_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_5317_cast_fp16 = einsum(equation = var_5317_equation_0, values = (var_5251_cast_fp16, var_5229_cast_fp16))[name = tensor("op_5317_cast_fp16")]; tensor var_5318_to_fp16 = const()[name = tensor("op_5318_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_421_cast_fp16 = mul(x = var_5317_cast_fp16, y = var_5318_to_fp16)[name = tensor("aw_chunk_421_cast_fp16")]; tensor var_5321_equation_0 = const()[name = tensor("op_5321_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_5321_cast_fp16 = einsum(equation = var_5321_equation_0, values = (var_5251_cast_fp16, var_5230_cast_fp16))[name = tensor("op_5321_cast_fp16")]; tensor var_5322_to_fp16 = const()[name = tensor("op_5322_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_423_cast_fp16 = mul(x = var_5321_cast_fp16, y = var_5322_to_fp16)[name = tensor("aw_chunk_423_cast_fp16")]; tensor var_5325_equation_0 = const()[name = tensor("op_5325_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_5325_cast_fp16 = einsum(equation = var_5325_equation_0, values = (var_5255_cast_fp16, var_5231_cast_fp16))[name = tensor("op_5325_cast_fp16")]; tensor var_5326_to_fp16 = const()[name = tensor("op_5326_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_425_cast_fp16 = mul(x = var_5325_cast_fp16, y = var_5326_to_fp16)[name = tensor("aw_chunk_425_cast_fp16")]; tensor var_5329_equation_0 = const()[name = tensor("op_5329_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_5329_cast_fp16 = einsum(equation = var_5329_equation_0, values = (var_5255_cast_fp16, var_5232_cast_fp16))[name = tensor("op_5329_cast_fp16")]; tensor var_5330_to_fp16 = const()[name = tensor("op_5330_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_427_cast_fp16 = mul(x = var_5329_cast_fp16, y = var_5330_to_fp16)[name = tensor("aw_chunk_427_cast_fp16")]; tensor var_5333_equation_0 = const()[name = tensor("op_5333_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_5333_cast_fp16 = einsum(equation = var_5333_equation_0, values = (var_5259_cast_fp16, var_5233_cast_fp16))[name = tensor("op_5333_cast_fp16")]; tensor var_5334_to_fp16 = const()[name = tensor("op_5334_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_429_cast_fp16 = mul(x = var_5333_cast_fp16, y = var_5334_to_fp16)[name = tensor("aw_chunk_429_cast_fp16")]; tensor var_5337_equation_0 = const()[name = tensor("op_5337_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_5337_cast_fp16 = einsum(equation = var_5337_equation_0, values = (var_5259_cast_fp16, var_5234_cast_fp16))[name = tensor("op_5337_cast_fp16")]; tensor var_5338_to_fp16 = const()[name = tensor("op_5338_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_431_cast_fp16 = mul(x = var_5337_cast_fp16, y = var_5338_to_fp16)[name = tensor("aw_chunk_431_cast_fp16")]; tensor var_5341_equation_0 = const()[name = tensor("op_5341_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_5341_cast_fp16 = einsum(equation = var_5341_equation_0, values = (var_5263_cast_fp16, var_5235_cast_fp16))[name = tensor("op_5341_cast_fp16")]; tensor var_5342_to_fp16 = const()[name = tensor("op_5342_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_433_cast_fp16 = mul(x = var_5341_cast_fp16, y = var_5342_to_fp16)[name = tensor("aw_chunk_433_cast_fp16")]; tensor var_5345_equation_0 = const()[name = tensor("op_5345_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_5345_cast_fp16 = einsum(equation = var_5345_equation_0, values = (var_5263_cast_fp16, var_5236_cast_fp16))[name = tensor("op_5345_cast_fp16")]; tensor var_5346_to_fp16 = const()[name = tensor("op_5346_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_435_cast_fp16 = mul(x = var_5345_cast_fp16, y = var_5346_to_fp16)[name = tensor("aw_chunk_435_cast_fp16")]; tensor var_5349_equation_0 = const()[name = tensor("op_5349_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_5349_cast_fp16 = einsum(equation = var_5349_equation_0, values = (var_5267_cast_fp16, var_5237_cast_fp16))[name = tensor("op_5349_cast_fp16")]; tensor var_5350_to_fp16 = const()[name = tensor("op_5350_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_437_cast_fp16 = mul(x = var_5349_cast_fp16, y = var_5350_to_fp16)[name = tensor("aw_chunk_437_cast_fp16")]; tensor var_5353_equation_0 = const()[name = tensor("op_5353_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_5353_cast_fp16 = einsum(equation = var_5353_equation_0, values = (var_5267_cast_fp16, var_5238_cast_fp16))[name = tensor("op_5353_cast_fp16")]; tensor var_5354_to_fp16 = const()[name = tensor("op_5354_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_439_cast_fp16 = mul(x = var_5353_cast_fp16, y = var_5354_to_fp16)[name = tensor("aw_chunk_439_cast_fp16")]; tensor var_5357_equation_0 = const()[name = tensor("op_5357_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_5357_cast_fp16 = einsum(equation = var_5357_equation_0, values = (var_5271_cast_fp16, var_5239_cast_fp16))[name = tensor("op_5357_cast_fp16")]; tensor var_5358_to_fp16 = const()[name = tensor("op_5358_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_441_cast_fp16 = mul(x = var_5357_cast_fp16, y = var_5358_to_fp16)[name = tensor("aw_chunk_441_cast_fp16")]; tensor var_5361_equation_0 = const()[name = tensor("op_5361_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_5361_cast_fp16 = einsum(equation = var_5361_equation_0, values = (var_5271_cast_fp16, var_5240_cast_fp16))[name = tensor("op_5361_cast_fp16")]; tensor var_5362_to_fp16 = const()[name = tensor("op_5362_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_443_cast_fp16 = mul(x = var_5361_cast_fp16, y = var_5362_to_fp16)[name = tensor("aw_chunk_443_cast_fp16")]; tensor var_5365_equation_0 = const()[name = tensor("op_5365_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_5365_cast_fp16 = einsum(equation = var_5365_equation_0, values = (var_5275_cast_fp16, var_5241_cast_fp16))[name = tensor("op_5365_cast_fp16")]; tensor var_5366_to_fp16 = const()[name = tensor("op_5366_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_445_cast_fp16 = mul(x = var_5365_cast_fp16, y = var_5366_to_fp16)[name = tensor("aw_chunk_445_cast_fp16")]; tensor var_5369_equation_0 = const()[name = tensor("op_5369_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_5369_cast_fp16 = einsum(equation = var_5369_equation_0, values = (var_5275_cast_fp16, var_5242_cast_fp16))[name = tensor("op_5369_cast_fp16")]; tensor var_5370_to_fp16 = const()[name = tensor("op_5370_to_fp16"), val = tensor(0x1.cap-4)]; tensor aw_chunk_447_cast_fp16 = mul(x = var_5369_cast_fp16, y = var_5370_to_fp16)[name = tensor("aw_chunk_447_cast_fp16")]; tensor var_5372_cast_fp16 = softmax(axis = var_4061, x = aw_chunk_417_cast_fp16)[name = tensor("op_5372_cast_fp16")]; tensor var_5373_cast_fp16 = softmax(axis = var_4061, x = aw_chunk_419_cast_fp16)[name = tensor("op_5373_cast_fp16")]; tensor var_5374_cast_fp16 = softmax(axis = var_4061, x = aw_chunk_421_cast_fp16)[name = tensor("op_5374_cast_fp16")]; tensor var_5375_cast_fp16 = softmax(axis = var_4061, x = aw_chunk_423_cast_fp16)[name = tensor("op_5375_cast_fp16")]; tensor var_5376_cast_fp16 = softmax(axis = var_4061, x = aw_chunk_425_cast_fp16)[name = tensor("op_5376_cast_fp16")]; tensor var_5377_cast_fp16 = softmax(axis = var_4061, x = aw_chunk_427_cast_fp16)[name = tensor("op_5377_cast_fp16")]; tensor var_5378_cast_fp16 = softmax(axis = var_4061, x = aw_chunk_429_cast_fp16)[name = tensor("op_5378_cast_fp16")]; tensor var_5379_cast_fp16 = softmax(axis = var_4061, x = aw_chunk_431_cast_fp16)[name = tensor("op_5379_cast_fp16")]; tensor var_5380_cast_fp16 = softmax(axis = var_4061, x = aw_chunk_433_cast_fp16)[name = tensor("op_5380_cast_fp16")]; tensor var_5381_cast_fp16 = softmax(axis = var_4061, x = aw_chunk_435_cast_fp16)[name = tensor("op_5381_cast_fp16")]; tensor var_5382_cast_fp16 = softmax(axis = var_4061, x = aw_chunk_437_cast_fp16)[name = tensor("op_5382_cast_fp16")]; tensor var_5383_cast_fp16 = softmax(axis = var_4061, x = aw_chunk_439_cast_fp16)[name = tensor("op_5383_cast_fp16")]; tensor var_5384_cast_fp16 = softmax(axis = var_4061, x = aw_chunk_441_cast_fp16)[name = tensor("op_5384_cast_fp16")]; tensor var_5385_cast_fp16 = softmax(axis = var_4061, x = aw_chunk_443_cast_fp16)[name = tensor("op_5385_cast_fp16")]; tensor var_5386_cast_fp16 = softmax(axis = var_4061, x = aw_chunk_445_cast_fp16)[name = tensor("op_5386_cast_fp16")]; tensor var_5387_cast_fp16 = softmax(axis = var_4061, x = aw_chunk_447_cast_fp16)[name = tensor("op_5387_cast_fp16")]; tensor var_5389_equation_0 = const()[name = tensor("op_5389_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_5389_cast_fp16 = einsum(equation = var_5389_equation_0, values = (var_5277_cast_fp16, var_5372_cast_fp16))[name = tensor("op_5389_cast_fp16")]; tensor var_5391_equation_0 = const()[name = tensor("op_5391_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_5391_cast_fp16 = einsum(equation = var_5391_equation_0, values = (var_5277_cast_fp16, var_5373_cast_fp16))[name = tensor("op_5391_cast_fp16")]; tensor var_5393_equation_0 = const()[name = tensor("op_5393_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_5393_cast_fp16 = einsum(equation = var_5393_equation_0, values = (var_5281_cast_fp16, var_5374_cast_fp16))[name = tensor("op_5393_cast_fp16")]; tensor var_5395_equation_0 = const()[name = tensor("op_5395_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_5395_cast_fp16 = einsum(equation = var_5395_equation_0, values = (var_5281_cast_fp16, var_5375_cast_fp16))[name = tensor("op_5395_cast_fp16")]; tensor var_5397_equation_0 = const()[name = tensor("op_5397_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_5397_cast_fp16 = einsum(equation = var_5397_equation_0, values = (var_5285_cast_fp16, var_5376_cast_fp16))[name = tensor("op_5397_cast_fp16")]; tensor var_5399_equation_0 = const()[name = tensor("op_5399_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_5399_cast_fp16 = einsum(equation = var_5399_equation_0, values = (var_5285_cast_fp16, var_5377_cast_fp16))[name = tensor("op_5399_cast_fp16")]; tensor var_5401_equation_0 = const()[name = tensor("op_5401_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_5401_cast_fp16 = einsum(equation = var_5401_equation_0, values = (var_5289_cast_fp16, var_5378_cast_fp16))[name = tensor("op_5401_cast_fp16")]; tensor var_5403_equation_0 = const()[name = tensor("op_5403_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_5403_cast_fp16 = einsum(equation = var_5403_equation_0, values = (var_5289_cast_fp16, var_5379_cast_fp16))[name = tensor("op_5403_cast_fp16")]; tensor var_5405_equation_0 = const()[name = tensor("op_5405_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_5405_cast_fp16 = einsum(equation = var_5405_equation_0, values = (var_5293_cast_fp16, var_5380_cast_fp16))[name = tensor("op_5405_cast_fp16")]; tensor var_5407_equation_0 = const()[name = tensor("op_5407_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_5407_cast_fp16 = einsum(equation = var_5407_equation_0, values = (var_5293_cast_fp16, var_5381_cast_fp16))[name = tensor("op_5407_cast_fp16")]; tensor var_5409_equation_0 = const()[name = tensor("op_5409_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_5409_cast_fp16 = einsum(equation = var_5409_equation_0, values = (var_5297_cast_fp16, var_5382_cast_fp16))[name = tensor("op_5409_cast_fp16")]; tensor var_5411_equation_0 = const()[name = tensor("op_5411_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_5411_cast_fp16 = einsum(equation = var_5411_equation_0, values = (var_5297_cast_fp16, var_5383_cast_fp16))[name = tensor("op_5411_cast_fp16")]; tensor var_5413_equation_0 = const()[name = tensor("op_5413_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_5413_cast_fp16 = einsum(equation = var_5413_equation_0, values = (var_5301_cast_fp16, var_5384_cast_fp16))[name = tensor("op_5413_cast_fp16")]; tensor var_5415_equation_0 = const()[name = tensor("op_5415_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_5415_cast_fp16 = einsum(equation = var_5415_equation_0, values = (var_5301_cast_fp16, var_5385_cast_fp16))[name = tensor("op_5415_cast_fp16")]; tensor var_5417_equation_0 = const()[name = tensor("op_5417_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_5417_cast_fp16 = einsum(equation = var_5417_equation_0, values = (var_5305_cast_fp16, var_5386_cast_fp16))[name = tensor("op_5417_cast_fp16")]; tensor var_5419_equation_0 = const()[name = tensor("op_5419_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_5419_cast_fp16 = einsum(equation = var_5419_equation_0, values = (var_5305_cast_fp16, var_5387_cast_fp16))[name = tensor("op_5419_cast_fp16")]; tensor var_5421_interleave_0 = const()[name = tensor("op_5421_interleave_0"), val = tensor(false)]; tensor var_5421_cast_fp16 = concat(axis = var_4039, interleave = var_5421_interleave_0, values = (var_5389_cast_fp16, var_5391_cast_fp16))[name = tensor("op_5421_cast_fp16")]; tensor var_5423_interleave_0 = const()[name = tensor("op_5423_interleave_0"), val = tensor(false)]; tensor var_5423_cast_fp16 = concat(axis = var_4039, interleave = var_5423_interleave_0, values = (var_5393_cast_fp16, var_5395_cast_fp16))[name = tensor("op_5423_cast_fp16")]; tensor var_5425_interleave_0 = const()[name = tensor("op_5425_interleave_0"), val = tensor(false)]; tensor var_5425_cast_fp16 = concat(axis = var_4039, interleave = var_5425_interleave_0, values = (var_5397_cast_fp16, var_5399_cast_fp16))[name = tensor("op_5425_cast_fp16")]; tensor var_5427_interleave_0 = const()[name = tensor("op_5427_interleave_0"), val = tensor(false)]; tensor var_5427_cast_fp16 = concat(axis = var_4039, interleave = var_5427_interleave_0, values = (var_5401_cast_fp16, var_5403_cast_fp16))[name = tensor("op_5427_cast_fp16")]; tensor var_5429_interleave_0 = const()[name = tensor("op_5429_interleave_0"), val = tensor(false)]; tensor var_5429_cast_fp16 = concat(axis = var_4039, interleave = var_5429_interleave_0, values = (var_5405_cast_fp16, var_5407_cast_fp16))[name = tensor("op_5429_cast_fp16")]; tensor var_5431_interleave_0 = const()[name = tensor("op_5431_interleave_0"), val = tensor(false)]; tensor var_5431_cast_fp16 = concat(axis = var_4039, interleave = var_5431_interleave_0, values = (var_5409_cast_fp16, var_5411_cast_fp16))[name = tensor("op_5431_cast_fp16")]; tensor var_5433_interleave_0 = const()[name = tensor("op_5433_interleave_0"), val = tensor(false)]; tensor var_5433_cast_fp16 = concat(axis = var_4039, interleave = var_5433_interleave_0, values = (var_5413_cast_fp16, var_5415_cast_fp16))[name = tensor("op_5433_cast_fp16")]; tensor var_5435_interleave_0 = const()[name = tensor("op_5435_interleave_0"), val = tensor(false)]; tensor var_5435_cast_fp16 = concat(axis = var_4039, interleave = var_5435_interleave_0, values = (var_5417_cast_fp16, var_5419_cast_fp16))[name = tensor("op_5435_cast_fp16")]; tensor input_191_interleave_0 = const()[name = tensor("input_191_interleave_0"), val = tensor(false)]; tensor input_191_cast_fp16 = concat(axis = var_4061, interleave = input_191_interleave_0, values = (var_5421_cast_fp16, var_5423_cast_fp16, var_5425_cast_fp16, var_5427_cast_fp16, var_5429_cast_fp16, var_5431_cast_fp16, var_5433_cast_fp16, var_5435_cast_fp16))[name = tensor("input_191_cast_fp16")]; tensor var_5441 = const()[name = tensor("op_5441"), val = tensor([1, 1])]; tensor var_5443 = const()[name = tensor("op_5443"), val = tensor([1, 1])]; tensor var_5445_pad_type_0 = const()[name = tensor("op_5445_pad_type_0"), val = tensor("custom")]; tensor var_5445_pad_0 = const()[name = tensor("op_5445_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(226023616))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(226330880))), name = tensor("up_blocks_1_attentions_1_transformer_blocks_0_attn2_to_out_0_weight_to_fp16_palettized"), shape = tensor([640, 640, 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(226331072)))]; tensor var_5445_cast_fp16 = conv(bias = up_blocks_1_attentions_1_transformer_blocks_0_attn2_to_out_0_bias_to_fp16, dilations = var_5443, groups = var_4061, pad = var_5445_pad_0, pad_type = var_5445_pad_type_0, strides = var_5441, weight = up_blocks_1_attentions_1_transformer_blocks_0_attn2_to_out_0_weight_to_fp16_palettized, x = input_191_cast_fp16)[name = tensor("op_5445_cast_fp16")]; tensor inputs_41_cast_fp16 = add(x = var_5445_cast_fp16, y = inputs_39_cast_fp16)[name = tensor("inputs_41_cast_fp16")]; tensor var_5449 = const()[name = tensor("op_5449"), val = tensor([1])]; tensor channels_mean_41_cast_fp16 = reduce_mean(axes = var_5449, keep_dims = var_4056, x = inputs_41_cast_fp16)[name = tensor("channels_mean_41_cast_fp16")]; tensor zero_mean_41_cast_fp16 = sub(x = inputs_41_cast_fp16, y = channels_mean_41_cast_fp16)[name = tensor("zero_mean_41_cast_fp16")]; tensor zero_mean_sq_41_cast_fp16 = mul(x = zero_mean_41_cast_fp16, y = zero_mean_41_cast_fp16)[name = tensor("zero_mean_sq_41_cast_fp16")]; tensor var_5453 = const()[name = tensor("op_5453"), val = tensor([1])]; tensor var_5454_cast_fp16 = reduce_mean(axes = var_5453, keep_dims = var_4056, x = zero_mean_sq_41_cast_fp16)[name = tensor("op_5454_cast_fp16")]; tensor var_5455_to_fp16 = const()[name = tensor("op_5455_to_fp16"), val = tensor(0x1.5p-17)]; tensor var_5456_cast_fp16 = add(x = var_5454_cast_fp16, y = var_5455_to_fp16)[name = tensor("op_5456_cast_fp16")]; tensor denom_41_epsilon_0_to_fp16 = const()[name = tensor("denom_41_epsilon_0_to_fp16"), val = tensor(0x1p-24)]; tensor denom_41_cast_fp16 = rsqrt(epsilon = denom_41_epsilon_0_to_fp16, x = var_5456_cast_fp16)[name = tensor("denom_41_cast_fp16")]; tensor out_41_cast_fp16 = mul(x = zero_mean_41_cast_fp16, y = denom_41_cast_fp16)[name = tensor("out_41_cast_fp16")]; tensor var_5460_to_fp16 = const()[name = tensor("op_5460_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(226332416)))]; tensor var_5461_cast_fp16 = add(x = out_41_cast_fp16, y = var_5460_to_fp16)[name = tensor("op_5461_cast_fp16")]; tensor var_5463_to_fp16 = const()[name = tensor("op_5463_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(226333760)))]; tensor input_193_cast_fp16 = mul(x = var_5461_cast_fp16, y = var_5463_to_fp16)[name = tensor("input_193_cast_fp16")]; tensor var_5471 = const()[name = tensor("op_5471"), val = tensor([1, 1])]; tensor var_5473 = const()[name = tensor("op_5473"), val = tensor([1, 1])]; tensor var_5475_pad_type_0 = const()[name = tensor("op_5475_pad_type_0"), val = tensor("custom")]; tensor var_5475_pad_0 = const()[name = tensor("op_5475_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(226335104))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(228792768))), name = tensor("up_blocks_1_attentions_1_transformer_blocks_0_ff_net_0_proj_weight_to_fp16_palettized"), shape = tensor([5120, 640, 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(228792960))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(228796864))), name = tensor("up_blocks_1_attentions_1_transformer_blocks_0_ff_net_0_proj_bias_to_fp16_palettized"), shape = tensor([5120])]; tensor var_5475_cast_fp16 = conv(bias = up_blocks_1_attentions_1_transformer_blocks_0_ff_net_0_proj_bias_to_fp16_palettized, dilations = var_5473, groups = var_4061, pad = var_5475_pad_0, pad_type = var_5475_pad_type_0, strides = var_5471, weight = up_blocks_1_attentions_1_transformer_blocks_0_ff_net_0_proj_weight_to_fp16_palettized, x = input_193_cast_fp16)[name = tensor("op_5475_cast_fp16")]; tensor var_5476_split_sizes_0 = const()[name = tensor("op_5476_split_sizes_0"), val = tensor([2560, 2560])]; tensor var_5476_axis_0 = const()[name = tensor("op_5476_axis_0"), val = tensor(1)]; tensor var_5476_cast_fp16_0, tensor var_5476_cast_fp16_1 = split(axis = var_5476_axis_0, split_sizes = var_5476_split_sizes_0, x = var_5475_cast_fp16)[name = tensor("op_5476_cast_fp16")]; tensor var_5478_mode_0 = const()[name = tensor("op_5478_mode_0"), val = tensor("EXACT")]; tensor var_5478_cast_fp16 = gelu(mode = var_5478_mode_0, x = var_5476_cast_fp16_1)[name = tensor("op_5478_cast_fp16")]; tensor input_195_cast_fp16 = mul(x = var_5476_cast_fp16_0, y = var_5478_cast_fp16)[name = tensor("input_195_cast_fp16")]; tensor var_5482 = const()[name = tensor("op_5482"), val = tensor([1, 1])]; tensor var_5484 = const()[name = tensor("op_5484"), val = tensor([1, 1])]; tensor var_5486_pad_type_0 = const()[name = tensor("op_5486_pad_type_0"), val = tensor("custom")]; tensor var_5486_pad_0 = const()[name = tensor("op_5486_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(228797056))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(230025920))), name = tensor("up_blocks_1_attentions_1_transformer_blocks_0_ff_net_2_weight_to_fp16_palettized"), shape = tensor([640, 2560, 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(230026112)))]; tensor var_5486_cast_fp16 = conv(bias = up_blocks_1_attentions_1_transformer_blocks_0_ff_net_2_bias_to_fp16, dilations = var_5484, groups = var_4061, pad = var_5486_pad_0, pad_type = var_5486_pad_type_0, strides = var_5482, weight = up_blocks_1_attentions_1_transformer_blocks_0_ff_net_2_weight_to_fp16_palettized, x = input_195_cast_fp16)[name = tensor("op_5486_cast_fp16")]; tensor hidden_states_131_cast_fp16 = add(x = var_5486_cast_fp16, y = inputs_41_cast_fp16)[name = tensor("hidden_states_131_cast_fp16")]; tensor var_5488 = const()[name = tensor("op_5488"), val = tensor([2, 640, 32, 32])]; tensor input_197_cast_fp16 = reshape(shape = var_5488, x = hidden_states_131_cast_fp16)[name = tensor("input_197_cast_fp16")]; tensor var_5492 = const()[name = tensor("op_5492"), val = tensor([1, 1])]; tensor var_5494 = const()[name = tensor("op_5494"), 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 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(230027456))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(230334720))), name = tensor("up_blocks_1_attentions_1_proj_out_weight_to_fp16_palettized"), shape = tensor([640, 640, 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(230334912)))]; tensor hidden_states_133_cast_fp16 = conv(bias = up_blocks_1_attentions_1_proj_out_bias_to_fp16, dilations = var_5494, groups = var_4061, pad = hidden_states_133_pad_0, pad_type = hidden_states_133_pad_type_0, strides = var_5492, weight = up_blocks_1_attentions_1_proj_out_weight_to_fp16_palettized, x = input_197_cast_fp16)[name = tensor("hidden_states_133_cast_fp16")]; tensor input_199_cast_fp16 = add(x = hidden_states_133_cast_fp16, y = hidden_states_121_cast_fp16)[name = tensor("input_199_cast_fp16")]; tensor input_201_scale_factor_height_0 = const()[name = tensor("input_201_scale_factor_height_0"), val = tensor(0x1p+1)]; tensor input_201_scale_factor_width_0 = const()[name = tensor("input_201_scale_factor_width_0"), val = tensor(0x1p+1)]; tensor input_201_cast_fp16 = upsample_nearest_neighbor(scale_factor_height = input_201_scale_factor_height_0, scale_factor_width = input_201_scale_factor_width_0, x = input_199_cast_fp16)[name = tensor("input_201_cast_fp16")]; tensor var_5503 = const()[name = tensor("op_5503"), val = tensor([1, 1])]; tensor var_5505 = const()[name = tensor("op_5505"), 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 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(230336256))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(233101120))), name = tensor("up_blocks_1_upsamplers_0_conv_weight_to_fp16_palettized"), shape = tensor([640, 640, 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(233101312)))]; tensor hidden_states_135_cast_fp16 = conv(bias = up_blocks_1_upsamplers_0_conv_bias_to_fp16, dilations = var_5505, groups = var_4061, pad = hidden_states_135_pad_0, pad_type = hidden_states_135_pad_type_0, strides = var_5503, weight = up_blocks_1_upsamplers_0_conv_weight_to_fp16_palettized, x = input_201_cast_fp16)[name = tensor("hidden_states_135_cast_fp16")]; tensor var_5509 = const()[name = tensor("op_5509"), val = tensor(3)]; tensor var_5532 = const()[name = tensor("op_5532"), val = tensor(true)]; tensor var_5537 = const()[name = tensor("op_5537"), val = tensor(1)]; tensor input_203_interleave_0 = const()[name = tensor("input_203_interleave_0"), val = tensor(false)]; tensor input_203_cast_fp16 = concat(axis = var_5537, interleave = input_203_interleave_0, values = (hidden_states_135_cast_fp16, input_35_cast_fp16))[name = tensor("input_203_cast_fp16")]; tensor reshape_84_shape_0 = const()[name = tensor("reshape_84_shape_0"), val = tensor([2, 32, 30, 64, 64])]; tensor reshape_84_cast_fp16 = reshape(shape = reshape_84_shape_0, x = input_203_cast_fp16)[name = tensor("reshape_84_cast_fp16")]; 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_fp16 = reduce_mean(axes = reduce_mean_63_axes_0, keep_dims = reduce_mean_63_keep_dims_0, x = reshape_84_cast_fp16)[name = tensor("reduce_mean_63_cast_fp16")]; tensor sub_42_cast_fp16 = sub(x = reshape_84_cast_fp16, y = reduce_mean_63_cast_fp16)[name = tensor("sub_42_cast_fp16")]; tensor square_21_cast_fp16 = square(x = sub_42_cast_fp16)[name = tensor("square_21_cast_fp16")]; 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_fp16 = reduce_mean(axes = reduce_mean_65_axes_0, keep_dims = reduce_mean_65_keep_dims_0, x = square_21_cast_fp16)[name = tensor("reduce_mean_65_cast_fp16")]; 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_fp16 = add(x = reduce_mean_65_cast_fp16, y = add_42_y_0_to_fp16)[name = tensor("add_42_cast_fp16")]; tensor sqrt_21_cast_fp16 = sqrt(x = add_42_cast_fp16)[name = tensor("sqrt_21_cast_fp16")]; tensor real_div_21_cast_fp16 = real_div(x = sub_42_cast_fp16, y = sqrt_21_cast_fp16)[name = tensor("real_div_21_cast_fp16")]; tensor reshape_85_shape_0 = const()[name = tensor("reshape_85_shape_0"), val = tensor([2, 960, 64, 64])]; tensor reshape_85_cast_fp16 = reshape(shape = reshape_85_shape_0, x = real_div_21_cast_fp16)[name = tensor("reshape_85_cast_fp16")]; 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(233102656)))]; 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(233104640)))]; 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_fp16 = 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_37_mean_0_to_fp16, variance = add_37_variance_0_to_fp16, x = reshape_85_cast_fp16)[name = tensor("add_43_cast_fp16")]; tensor input_207_cast_fp16 = silu(x = add_43_cast_fp16)[name = tensor("input_207_cast_fp16")]; tensor var_5560 = const()[name = tensor("op_5560"), val = tensor([1, 1])]; tensor var_5562 = const()[name = tensor("op_5562"), 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 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(233106624))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(235180288))), name = tensor("up_blocks_2_resnets_0_conv1_weight_to_fp16_palettized"), shape = tensor([320, 960, 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(235180480)))]; tensor hidden_states_137_cast_fp16 = conv(bias = up_blocks_2_resnets_0_conv1_bias_to_fp16, dilations = var_5562, groups = var_5537, pad = hidden_states_137_pad_0, pad_type = hidden_states_137_pad_type_0, strides = var_5560, weight = up_blocks_2_resnets_0_conv1_weight_to_fp16_palettized, x = input_207_cast_fp16)[name = tensor("hidden_states_137_cast_fp16")]; tensor var_5568 = const()[name = tensor("op_5568"), val = tensor([1, 1])]; tensor var_5570 = const()[name = tensor("op_5570"), 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 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(235181184))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(235488448))), name = tensor("up_blocks_2_resnets_0_time_emb_proj_weight_to_fp16_palettized"), shape = tensor([320, 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(235488640)))]; tensor temb_15_cast_fp16 = conv(bias = up_blocks_2_resnets_0_time_emb_proj_bias_to_fp16, dilations = var_5570, groups = var_5537, pad = temb_15_pad_0, pad_type = temb_15_pad_type_0, strides = var_5568, weight = up_blocks_2_resnets_0_time_emb_proj_weight_to_fp16_palettized, x = input_15_cast_fp16)[name = tensor("temb_15_cast_fp16")]; tensor input_211_cast_fp16 = add(x = hidden_states_137_cast_fp16, y = temb_15_cast_fp16)[name = tensor("input_211_cast_fp16")]; tensor reshape_88_shape_0 = const()[name = tensor("reshape_88_shape_0"), val = tensor([2, 32, 10, 64, 64])]; tensor reshape_88_cast_fp16 = reshape(shape = reshape_88_shape_0, x = input_211_cast_fp16)[name = tensor("reshape_88_cast_fp16")]; 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_fp16 = reduce_mean(axes = reduce_mean_66_axes_0, keep_dims = reduce_mean_66_keep_dims_0, x = reshape_88_cast_fp16)[name = tensor("reduce_mean_66_cast_fp16")]; tensor sub_44_cast_fp16 = sub(x = reshape_88_cast_fp16, y = reduce_mean_66_cast_fp16)[name = tensor("sub_44_cast_fp16")]; tensor square_22_cast_fp16 = square(x = sub_44_cast_fp16)[name = tensor("square_22_cast_fp16")]; 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_fp16 = reduce_mean(axes = reduce_mean_68_axes_0, keep_dims = reduce_mean_68_keep_dims_0, x = square_22_cast_fp16)[name = tensor("reduce_mean_68_cast_fp16")]; 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_fp16 = add(x = reduce_mean_68_cast_fp16, y = add_44_y_0_to_fp16)[name = tensor("add_44_cast_fp16")]; tensor sqrt_22_cast_fp16 = sqrt(x = add_44_cast_fp16)[name = tensor("sqrt_22_cast_fp16")]; tensor real_div_22_cast_fp16 = real_div(x = sub_44_cast_fp16, y = sqrt_22_cast_fp16)[name = tensor("real_div_22_cast_fp16")]; tensor reshape_89_shape_0 = const()[name = tensor("reshape_89_shape_0"), val = tensor([2, 320, 64, 64])]; tensor reshape_89_cast_fp16 = reshape(shape = reshape_89_shape_0, x = real_div_22_cast_fp16)[name = tensor("reshape_89_cast_fp16")]; 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(235489344)))]; 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(235490048)))]; 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_fp16 = 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_1_mean_0_to_fp16, variance = add_1_variance_0_to_fp16, x = reshape_89_cast_fp16)[name = tensor("add_45_cast_fp16")]; tensor input_215_cast_fp16 = silu(x = add_45_cast_fp16)[name = tensor("input_215_cast_fp16")]; tensor var_5580 = const()[name = tensor("op_5580"), val = tensor([1, 1])]; tensor var_5582 = const()[name = tensor("op_5582"), val = tensor([1, 1])]; tensor hidden_states_139_pad_type_0 = const()[name = tensor("hidden_states_139_pad_type_0"), val = tensor("custom")]; tensor hidden_states_139_pad_0 = const()[name = tensor("hidden_states_139_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(235490752))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(236182016))), name = tensor("up_blocks_2_resnets_0_conv2_weight_to_fp16_palettized"), shape = tensor([320, 320, 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(236182208)))]; tensor hidden_states_139_cast_fp16 = conv(bias = up_blocks_2_resnets_0_conv2_bias_to_fp16, dilations = var_5582, groups = var_5537, pad = hidden_states_139_pad_0, pad_type = hidden_states_139_pad_type_0, strides = var_5580, weight = up_blocks_2_resnets_0_conv2_weight_to_fp16_palettized, x = input_215_cast_fp16)[name = tensor("hidden_states_139_cast_fp16")]; tensor var_5587 = const()[name = tensor("op_5587"), val = tensor([1, 1])]; tensor var_5589 = const()[name = tensor("op_5589"), 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_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(236182912))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(236413376))), name = tensor("up_blocks_2_resnets_0_conv_shortcut_weight_to_fp16_palettized"), shape = tensor([320, 960, 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(236413568)))]; tensor x_13_cast_fp16 = conv(bias = up_blocks_2_resnets_0_conv_shortcut_bias_to_fp16, dilations = var_5589, groups = var_5537, pad = x_13_pad_0, pad_type = x_13_pad_type_0, strides = var_5587, weight = up_blocks_2_resnets_0_conv_shortcut_weight_to_fp16_palettized, x = input_203_cast_fp16)[name = tensor("x_13_cast_fp16")]; tensor hidden_states_141_cast_fp16 = add(x = x_13_cast_fp16, y = hidden_states_139_cast_fp16)[name = tensor("hidden_states_141_cast_fp16")]; tensor reshape_92_shape_0 = const()[name = tensor("reshape_92_shape_0"), val = tensor([2, 32, 10, 64, 64])]; tensor reshape_92_cast_fp16 = reshape(shape = reshape_92_shape_0, x = hidden_states_141_cast_fp16)[name = tensor("reshape_92_cast_fp16")]; 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_fp16 = reduce_mean(axes = reduce_mean_69_axes_0, keep_dims = reduce_mean_69_keep_dims_0, x = reshape_92_cast_fp16)[name = tensor("reduce_mean_69_cast_fp16")]; tensor sub_46_cast_fp16 = sub(x = reshape_92_cast_fp16, y = reduce_mean_69_cast_fp16)[name = tensor("sub_46_cast_fp16")]; tensor square_23_cast_fp16 = square(x = sub_46_cast_fp16)[name = tensor("square_23_cast_fp16")]; 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_fp16 = reduce_mean(axes = reduce_mean_71_axes_0, keep_dims = reduce_mean_71_keep_dims_0, x = square_23_cast_fp16)[name = tensor("reduce_mean_71_cast_fp16")]; tensor add_46_y_0_to_fp16 = const()[name = tensor("add_46_y_0_to_fp16"), val = tensor(0x1.1p-20)]; tensor add_46_cast_fp16 = add(x = reduce_mean_71_cast_fp16, y = add_46_y_0_to_fp16)[name = tensor("add_46_cast_fp16")]; tensor sqrt_23_cast_fp16 = sqrt(x = add_46_cast_fp16)[name = tensor("sqrt_23_cast_fp16")]; tensor real_div_23_cast_fp16 = real_div(x = sub_46_cast_fp16, y = sqrt_23_cast_fp16)[name = tensor("real_div_23_cast_fp16")]; tensor reshape_93_shape_0 = const()[name = tensor("reshape_93_shape_0"), val = tensor([2, 320, 64, 64])]; tensor reshape_93_cast_fp16 = reshape(shape = reshape_93_shape_0, x = real_div_23_cast_fp16)[name = tensor("reshape_93_cast_fp16")]; 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(236414272)))]; 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(236414976)))]; 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_fp16 = 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_1_mean_0_to_fp16, variance = add_1_variance_0_to_fp16, x = reshape_93_cast_fp16)[name = tensor("add_47_cast_fp16")]; tensor var_5609 = const()[name = tensor("op_5609"), val = tensor([1, 1])]; tensor var_5611 = const()[name = tensor("op_5611"), 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([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(236415680))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(236492544))), name = tensor("up_blocks_2_attentions_0_proj_in_weight_to_fp16_palettized"), shape = tensor([320, 320, 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(236492736)))]; tensor hidden_states_143_cast_fp16 = conv(bias = up_blocks_2_attentions_0_proj_in_bias_to_fp16, dilations = var_5611, groups = var_5537, pad = hidden_states_143_pad_0, pad_type = hidden_states_143_pad_type_0, strides = var_5609, weight = up_blocks_2_attentions_0_proj_in_weight_to_fp16_palettized, x = add_47_cast_fp16)[name = tensor("hidden_states_143_cast_fp16")]; tensor var_5616 = const()[name = tensor("op_5616"), val = tensor([2, 320, 1, 4096])]; tensor inputs_43_cast_fp16 = reshape(shape = var_5616, x = hidden_states_143_cast_fp16)[name = tensor("inputs_43_cast_fp16")]; tensor var_5626 = const()[name = tensor("op_5626"), val = tensor([1])]; tensor channels_mean_43_cast_fp16 = reduce_mean(axes = var_5626, keep_dims = var_5532, x = inputs_43_cast_fp16)[name = tensor("channels_mean_43_cast_fp16")]; tensor zero_mean_43_cast_fp16 = sub(x = inputs_43_cast_fp16, y = channels_mean_43_cast_fp16)[name = tensor("zero_mean_43_cast_fp16")]; tensor zero_mean_sq_43_cast_fp16 = mul(x = zero_mean_43_cast_fp16, y = zero_mean_43_cast_fp16)[name = tensor("zero_mean_sq_43_cast_fp16")]; tensor var_5630 = const()[name = tensor("op_5630"), val = tensor([1])]; tensor var_5631_cast_fp16 = reduce_mean(axes = var_5630, keep_dims = var_5532, x = zero_mean_sq_43_cast_fp16)[name = tensor("op_5631_cast_fp16")]; tensor var_5632_to_fp16 = const()[name = tensor("op_5632_to_fp16"), val = tensor(0x1.5p-17)]; tensor var_5633_cast_fp16 = add(x = var_5631_cast_fp16, y = var_5632_to_fp16)[name = tensor("op_5633_cast_fp16")]; tensor denom_43_epsilon_0_to_fp16 = const()[name = tensor("denom_43_epsilon_0_to_fp16"), val = tensor(0x1p-24)]; tensor denom_43_cast_fp16 = rsqrt(epsilon = denom_43_epsilon_0_to_fp16, x = var_5633_cast_fp16)[name = tensor("denom_43_cast_fp16")]; tensor out_43_cast_fp16 = mul(x = zero_mean_43_cast_fp16, y = denom_43_cast_fp16)[name = tensor("out_43_cast_fp16")]; tensor var_5637_to_fp16 = const()[name = tensor("op_5637_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(236493440)))]; tensor var_5638_cast_fp16 = add(x = out_43_cast_fp16, y = var_5637_to_fp16)[name = tensor("op_5638_cast_fp16")]; tensor var_5640_to_fp16 = const()[name = tensor("op_5640_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(236494144)))]; tensor hidden_states_145_cast_fp16 = mul(x = var_5638_cast_fp16, y = var_5640_to_fp16)[name = tensor("hidden_states_145_cast_fp16")]; tensor var_5647 = const()[name = tensor("op_5647"), val = tensor([1, 1])]; tensor var_5649 = const()[name = tensor("op_5649"), 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_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(236494848))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(236571712))), name = tensor("up_blocks_2_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16_palettized"), shape = tensor([320, 320, 1, 1])]; tensor q_29_cast_fp16 = conv(dilations = var_5649, groups = var_5537, pad = q_29_pad_0, pad_type = q_29_pad_type_0, strides = var_5647, weight = up_blocks_2_attentions_0_transformer_blocks_0_attn1_to_q_weight_to_fp16_palettized, x = hidden_states_145_cast_fp16)[name = tensor("q_29_cast_fp16")]; tensor var_5653 = const()[name = tensor("op_5653"), val = tensor([1, 1])]; tensor var_5655 = const()[name = tensor("op_5655"), 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_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(236571904))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(236648768))), name = tensor("up_blocks_2_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16_palettized"), shape = tensor([320, 320, 1, 1])]; tensor k_57_cast_fp16 = conv(dilations = var_5655, groups = var_5537, pad = k_57_pad_0, pad_type = k_57_pad_type_0, strides = var_5653, weight = up_blocks_2_attentions_0_transformer_blocks_0_attn1_to_k_weight_to_fp16_palettized, x = hidden_states_145_cast_fp16)[name = tensor("k_57_cast_fp16")]; tensor var_5659 = const()[name = tensor("op_5659"), val = tensor([1, 1])]; tensor var_5661 = const()[name = tensor("op_5661"), 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_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(236648960))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(236725824))), name = tensor("up_blocks_2_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16_palettized"), shape = tensor([320, 320, 1, 1])]; tensor v_29_cast_fp16 = conv(dilations = var_5661, groups = var_5537, pad = v_29_pad_0, pad_type = v_29_pad_type_0, strides = var_5659, weight = up_blocks_2_attentions_0_transformer_blocks_0_attn1_to_v_weight_to_fp16_palettized, x = hidden_states_145_cast_fp16)[name = tensor("v_29_cast_fp16")]; tensor var_5665_begin_0 = const()[name = tensor("op_5665_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_5665_end_0 = const()[name = tensor("op_5665_end_0"), val = tensor([2, 40, 1, 4096])]; tensor var_5665_end_mask_0 = const()[name = tensor("op_5665_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_5665_cast_fp16 = slice_by_index(begin = var_5665_begin_0, end = var_5665_end_0, end_mask = var_5665_end_mask_0, x = q_29_cast_fp16)[name = tensor("op_5665_cast_fp16")]; tensor var_5669_begin_0 = const()[name = tensor("op_5669_begin_0"), val = tensor([0, 40, 0, 0])]; tensor var_5669_end_0 = const()[name = tensor("op_5669_end_0"), val = tensor([2, 80, 1, 4096])]; tensor var_5669_end_mask_0 = const()[name = tensor("op_5669_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_5669_cast_fp16 = slice_by_index(begin = var_5669_begin_0, end = var_5669_end_0, end_mask = var_5669_end_mask_0, x = q_29_cast_fp16)[name = tensor("op_5669_cast_fp16")]; tensor var_5673_begin_0 = const()[name = tensor("op_5673_begin_0"), val = tensor([0, 80, 0, 0])]; tensor var_5673_end_0 = const()[name = tensor("op_5673_end_0"), val = tensor([2, 120, 1, 4096])]; tensor var_5673_end_mask_0 = const()[name = tensor("op_5673_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_5673_cast_fp16 = slice_by_index(begin = var_5673_begin_0, end = var_5673_end_0, end_mask = var_5673_end_mask_0, x = q_29_cast_fp16)[name = tensor("op_5673_cast_fp16")]; tensor var_5677_begin_0 = const()[name = tensor("op_5677_begin_0"), val = tensor([0, 120, 0, 0])]; tensor var_5677_end_0 = const()[name = tensor("op_5677_end_0"), val = tensor([2, 160, 1, 4096])]; tensor var_5677_end_mask_0 = const()[name = tensor("op_5677_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_5677_cast_fp16 = slice_by_index(begin = var_5677_begin_0, end = var_5677_end_0, end_mask = var_5677_end_mask_0, x = q_29_cast_fp16)[name = tensor("op_5677_cast_fp16")]; tensor var_5681_begin_0 = const()[name = tensor("op_5681_begin_0"), val = tensor([0, 160, 0, 0])]; tensor var_5681_end_0 = const()[name = tensor("op_5681_end_0"), val = tensor([2, 200, 1, 4096])]; tensor var_5681_end_mask_0 = const()[name = tensor("op_5681_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_5681_cast_fp16 = slice_by_index(begin = var_5681_begin_0, end = var_5681_end_0, end_mask = var_5681_end_mask_0, x = q_29_cast_fp16)[name = tensor("op_5681_cast_fp16")]; tensor var_5685_begin_0 = const()[name = tensor("op_5685_begin_0"), val = tensor([0, 200, 0, 0])]; tensor var_5685_end_0 = const()[name = tensor("op_5685_end_0"), val = tensor([2, 240, 1, 4096])]; tensor var_5685_end_mask_0 = const()[name = tensor("op_5685_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_5685_cast_fp16 = slice_by_index(begin = var_5685_begin_0, end = var_5685_end_0, end_mask = var_5685_end_mask_0, x = q_29_cast_fp16)[name = tensor("op_5685_cast_fp16")]; tensor var_5689_begin_0 = const()[name = tensor("op_5689_begin_0"), val = tensor([0, 240, 0, 0])]; tensor var_5689_end_0 = const()[name = tensor("op_5689_end_0"), val = tensor([2, 280, 1, 4096])]; tensor var_5689_end_mask_0 = const()[name = tensor("op_5689_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_5689_cast_fp16 = slice_by_index(begin = var_5689_begin_0, end = var_5689_end_0, end_mask = var_5689_end_mask_0, x = q_29_cast_fp16)[name = tensor("op_5689_cast_fp16")]; tensor var_5693_begin_0 = const()[name = tensor("op_5693_begin_0"), val = tensor([0, 280, 0, 0])]; tensor var_5693_end_0 = const()[name = tensor("op_5693_end_0"), val = tensor([2, 320, 1, 4096])]; tensor var_5693_end_mask_0 = const()[name = tensor("op_5693_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_5693_cast_fp16 = slice_by_index(begin = var_5693_begin_0, end = var_5693_end_0, end_mask = var_5693_end_mask_0, x = q_29_cast_fp16)[name = tensor("op_5693_cast_fp16")]; tensor var_5696_begin_0 = const()[name = tensor("op_5696_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_5696_end_0 = const()[name = tensor("op_5696_end_0"), val = tensor([2, 40, 1, 512])]; tensor var_5696_end_mask_0 = const()[name = tensor("op_5696_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5696_cast_fp16 = slice_by_index(begin = var_5696_begin_0, end = var_5696_end_0, end_mask = var_5696_end_mask_0, x = var_5665_cast_fp16)[name = tensor("op_5696_cast_fp16")]; tensor var_5697_begin_0 = const()[name = tensor("op_5697_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_5697_end_0 = const()[name = tensor("op_5697_end_0"), val = tensor([2, 40, 1, 1024])]; tensor var_5697_end_mask_0 = const()[name = tensor("op_5697_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5697_cast_fp16 = slice_by_index(begin = var_5697_begin_0, end = var_5697_end_0, end_mask = var_5697_end_mask_0, x = var_5665_cast_fp16)[name = tensor("op_5697_cast_fp16")]; tensor var_5698_begin_0 = const()[name = tensor("op_5698_begin_0"), val = tensor([0, 0, 0, 1024])]; tensor var_5698_end_0 = const()[name = tensor("op_5698_end_0"), val = tensor([2, 40, 1, 1536])]; tensor var_5698_end_mask_0 = const()[name = tensor("op_5698_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5698_cast_fp16 = slice_by_index(begin = var_5698_begin_0, end = var_5698_end_0, end_mask = var_5698_end_mask_0, x = var_5665_cast_fp16)[name = tensor("op_5698_cast_fp16")]; tensor var_5699_begin_0 = const()[name = tensor("op_5699_begin_0"), val = tensor([0, 0, 0, 1536])]; tensor var_5699_end_0 = const()[name = tensor("op_5699_end_0"), val = tensor([2, 40, 1, 2048])]; tensor var_5699_end_mask_0 = const()[name = tensor("op_5699_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5699_cast_fp16 = slice_by_index(begin = var_5699_begin_0, end = var_5699_end_0, end_mask = var_5699_end_mask_0, x = var_5665_cast_fp16)[name = tensor("op_5699_cast_fp16")]; tensor var_5700_begin_0 = const()[name = tensor("op_5700_begin_0"), val = tensor([0, 0, 0, 2048])]; tensor var_5700_end_0 = const()[name = tensor("op_5700_end_0"), val = tensor([2, 40, 1, 2560])]; tensor var_5700_end_mask_0 = const()[name = tensor("op_5700_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5700_cast_fp16 = slice_by_index(begin = var_5700_begin_0, end = var_5700_end_0, end_mask = var_5700_end_mask_0, x = var_5665_cast_fp16)[name = tensor("op_5700_cast_fp16")]; tensor var_5701_begin_0 = const()[name = tensor("op_5701_begin_0"), val = tensor([0, 0, 0, 2560])]; tensor var_5701_end_0 = const()[name = tensor("op_5701_end_0"), val = tensor([2, 40, 1, 3072])]; tensor var_5701_end_mask_0 = const()[name = tensor("op_5701_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5701_cast_fp16 = slice_by_index(begin = var_5701_begin_0, end = var_5701_end_0, end_mask = var_5701_end_mask_0, x = var_5665_cast_fp16)[name = tensor("op_5701_cast_fp16")]; tensor var_5702_begin_0 = const()[name = tensor("op_5702_begin_0"), val = tensor([0, 0, 0, 3072])]; tensor var_5702_end_0 = const()[name = tensor("op_5702_end_0"), val = tensor([2, 40, 1, 3584])]; tensor var_5702_end_mask_0 = const()[name = tensor("op_5702_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5702_cast_fp16 = slice_by_index(begin = var_5702_begin_0, end = var_5702_end_0, end_mask = var_5702_end_mask_0, x = var_5665_cast_fp16)[name = tensor("op_5702_cast_fp16")]; tensor var_5703_begin_0 = const()[name = tensor("op_5703_begin_0"), val = tensor([0, 0, 0, 3584])]; tensor var_5703_end_0 = const()[name = tensor("op_5703_end_0"), val = tensor([2, 40, 1, 4096])]; tensor var_5703_end_mask_0 = const()[name = tensor("op_5703_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5703_cast_fp16 = slice_by_index(begin = var_5703_begin_0, end = var_5703_end_0, end_mask = var_5703_end_mask_0, x = var_5665_cast_fp16)[name = tensor("op_5703_cast_fp16")]; tensor var_5704_begin_0 = const()[name = tensor("op_5704_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_5704_end_0 = const()[name = tensor("op_5704_end_0"), val = tensor([2, 40, 1, 512])]; tensor var_5704_end_mask_0 = const()[name = tensor("op_5704_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5704_cast_fp16 = slice_by_index(begin = var_5704_begin_0, end = var_5704_end_0, end_mask = var_5704_end_mask_0, x = var_5669_cast_fp16)[name = tensor("op_5704_cast_fp16")]; tensor var_5705_begin_0 = const()[name = tensor("op_5705_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_5705_end_0 = const()[name = tensor("op_5705_end_0"), val = tensor([2, 40, 1, 1024])]; tensor var_5705_end_mask_0 = const()[name = tensor("op_5705_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5705_cast_fp16 = slice_by_index(begin = var_5705_begin_0, end = var_5705_end_0, end_mask = var_5705_end_mask_0, x = var_5669_cast_fp16)[name = tensor("op_5705_cast_fp16")]; tensor var_5706_begin_0 = const()[name = tensor("op_5706_begin_0"), val = tensor([0, 0, 0, 1024])]; tensor var_5706_end_0 = const()[name = tensor("op_5706_end_0"), val = tensor([2, 40, 1, 1536])]; tensor var_5706_end_mask_0 = const()[name = tensor("op_5706_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5706_cast_fp16 = slice_by_index(begin = var_5706_begin_0, end = var_5706_end_0, end_mask = var_5706_end_mask_0, x = var_5669_cast_fp16)[name = tensor("op_5706_cast_fp16")]; tensor var_5707_begin_0 = const()[name = tensor("op_5707_begin_0"), val = tensor([0, 0, 0, 1536])]; tensor var_5707_end_0 = const()[name = tensor("op_5707_end_0"), val = tensor([2, 40, 1, 2048])]; tensor var_5707_end_mask_0 = const()[name = tensor("op_5707_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5707_cast_fp16 = slice_by_index(begin = var_5707_begin_0, end = var_5707_end_0, end_mask = var_5707_end_mask_0, x = var_5669_cast_fp16)[name = tensor("op_5707_cast_fp16")]; tensor var_5708_begin_0 = const()[name = tensor("op_5708_begin_0"), val = tensor([0, 0, 0, 2048])]; tensor var_5708_end_0 = const()[name = tensor("op_5708_end_0"), val = tensor([2, 40, 1, 2560])]; tensor var_5708_end_mask_0 = const()[name = tensor("op_5708_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5708_cast_fp16 = slice_by_index(begin = var_5708_begin_0, end = var_5708_end_0, end_mask = var_5708_end_mask_0, x = var_5669_cast_fp16)[name = tensor("op_5708_cast_fp16")]; tensor var_5709_begin_0 = const()[name = tensor("op_5709_begin_0"), val = tensor([0, 0, 0, 2560])]; tensor var_5709_end_0 = const()[name = tensor("op_5709_end_0"), val = tensor([2, 40, 1, 3072])]; tensor var_5709_end_mask_0 = const()[name = tensor("op_5709_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5709_cast_fp16 = slice_by_index(begin = var_5709_begin_0, end = var_5709_end_0, end_mask = var_5709_end_mask_0, x = var_5669_cast_fp16)[name = tensor("op_5709_cast_fp16")]; tensor var_5710_begin_0 = const()[name = tensor("op_5710_begin_0"), val = tensor([0, 0, 0, 3072])]; tensor var_5710_end_0 = const()[name = tensor("op_5710_end_0"), val = tensor([2, 40, 1, 3584])]; tensor var_5710_end_mask_0 = const()[name = tensor("op_5710_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5710_cast_fp16 = slice_by_index(begin = var_5710_begin_0, end = var_5710_end_0, end_mask = var_5710_end_mask_0, x = var_5669_cast_fp16)[name = tensor("op_5710_cast_fp16")]; tensor var_5711_begin_0 = const()[name = tensor("op_5711_begin_0"), val = tensor([0, 0, 0, 3584])]; tensor var_5711_end_0 = const()[name = tensor("op_5711_end_0"), val = tensor([2, 40, 1, 4096])]; tensor var_5711_end_mask_0 = const()[name = tensor("op_5711_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5711_cast_fp16 = slice_by_index(begin = var_5711_begin_0, end = var_5711_end_0, end_mask = var_5711_end_mask_0, x = var_5669_cast_fp16)[name = tensor("op_5711_cast_fp16")]; tensor var_5712_begin_0 = const()[name = tensor("op_5712_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_5712_end_0 = const()[name = tensor("op_5712_end_0"), val = tensor([2, 40, 1, 512])]; tensor var_5712_end_mask_0 = const()[name = tensor("op_5712_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5712_cast_fp16 = slice_by_index(begin = var_5712_begin_0, end = var_5712_end_0, end_mask = var_5712_end_mask_0, x = var_5673_cast_fp16)[name = tensor("op_5712_cast_fp16")]; tensor var_5713_begin_0 = const()[name = tensor("op_5713_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_5713_end_0 = const()[name = tensor("op_5713_end_0"), val = tensor([2, 40, 1, 1024])]; tensor var_5713_end_mask_0 = const()[name = tensor("op_5713_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5713_cast_fp16 = slice_by_index(begin = var_5713_begin_0, end = var_5713_end_0, end_mask = var_5713_end_mask_0, x = var_5673_cast_fp16)[name = tensor("op_5713_cast_fp16")]; tensor var_5714_begin_0 = const()[name = tensor("op_5714_begin_0"), val = tensor([0, 0, 0, 1024])]; tensor var_5714_end_0 = const()[name = tensor("op_5714_end_0"), val = tensor([2, 40, 1, 1536])]; tensor var_5714_end_mask_0 = const()[name = tensor("op_5714_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5714_cast_fp16 = slice_by_index(begin = var_5714_begin_0, end = var_5714_end_0, end_mask = var_5714_end_mask_0, x = var_5673_cast_fp16)[name = tensor("op_5714_cast_fp16")]; tensor var_5715_begin_0 = const()[name = tensor("op_5715_begin_0"), val = tensor([0, 0, 0, 1536])]; tensor var_5715_end_0 = const()[name = tensor("op_5715_end_0"), val = tensor([2, 40, 1, 2048])]; tensor var_5715_end_mask_0 = const()[name = tensor("op_5715_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5715_cast_fp16 = slice_by_index(begin = var_5715_begin_0, end = var_5715_end_0, end_mask = var_5715_end_mask_0, x = var_5673_cast_fp16)[name = tensor("op_5715_cast_fp16")]; tensor var_5716_begin_0 = const()[name = tensor("op_5716_begin_0"), val = tensor([0, 0, 0, 2048])]; tensor var_5716_end_0 = const()[name = tensor("op_5716_end_0"), val = tensor([2, 40, 1, 2560])]; tensor var_5716_end_mask_0 = const()[name = tensor("op_5716_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5716_cast_fp16 = slice_by_index(begin = var_5716_begin_0, end = var_5716_end_0, end_mask = var_5716_end_mask_0, x = var_5673_cast_fp16)[name = tensor("op_5716_cast_fp16")]; tensor var_5717_begin_0 = const()[name = tensor("op_5717_begin_0"), val = tensor([0, 0, 0, 2560])]; tensor var_5717_end_0 = const()[name = tensor("op_5717_end_0"), val = tensor([2, 40, 1, 3072])]; tensor var_5717_end_mask_0 = const()[name = tensor("op_5717_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5717_cast_fp16 = slice_by_index(begin = var_5717_begin_0, end = var_5717_end_0, end_mask = var_5717_end_mask_0, x = var_5673_cast_fp16)[name = tensor("op_5717_cast_fp16")]; tensor var_5718_begin_0 = const()[name = tensor("op_5718_begin_0"), val = tensor([0, 0, 0, 3072])]; tensor var_5718_end_0 = const()[name = tensor("op_5718_end_0"), val = tensor([2, 40, 1, 3584])]; tensor var_5718_end_mask_0 = const()[name = tensor("op_5718_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5718_cast_fp16 = slice_by_index(begin = var_5718_begin_0, end = var_5718_end_0, end_mask = var_5718_end_mask_0, x = var_5673_cast_fp16)[name = tensor("op_5718_cast_fp16")]; tensor var_5719_begin_0 = const()[name = tensor("op_5719_begin_0"), val = tensor([0, 0, 0, 3584])]; tensor var_5719_end_0 = const()[name = tensor("op_5719_end_0"), val = tensor([2, 40, 1, 4096])]; tensor var_5719_end_mask_0 = const()[name = tensor("op_5719_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5719_cast_fp16 = slice_by_index(begin = var_5719_begin_0, end = var_5719_end_0, end_mask = var_5719_end_mask_0, x = var_5673_cast_fp16)[name = tensor("op_5719_cast_fp16")]; tensor var_5720_begin_0 = const()[name = tensor("op_5720_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_5720_end_0 = const()[name = tensor("op_5720_end_0"), val = tensor([2, 40, 1, 512])]; tensor var_5720_end_mask_0 = const()[name = tensor("op_5720_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5720_cast_fp16 = slice_by_index(begin = var_5720_begin_0, end = var_5720_end_0, end_mask = var_5720_end_mask_0, x = var_5677_cast_fp16)[name = tensor("op_5720_cast_fp16")]; tensor var_5721_begin_0 = const()[name = tensor("op_5721_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_5721_end_0 = const()[name = tensor("op_5721_end_0"), val = tensor([2, 40, 1, 1024])]; tensor var_5721_end_mask_0 = const()[name = tensor("op_5721_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5721_cast_fp16 = slice_by_index(begin = var_5721_begin_0, end = var_5721_end_0, end_mask = var_5721_end_mask_0, x = var_5677_cast_fp16)[name = tensor("op_5721_cast_fp16")]; tensor var_5722_begin_0 = const()[name = tensor("op_5722_begin_0"), val = tensor([0, 0, 0, 1024])]; tensor var_5722_end_0 = const()[name = tensor("op_5722_end_0"), val = tensor([2, 40, 1, 1536])]; tensor var_5722_end_mask_0 = const()[name = tensor("op_5722_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5722_cast_fp16 = slice_by_index(begin = var_5722_begin_0, end = var_5722_end_0, end_mask = var_5722_end_mask_0, x = var_5677_cast_fp16)[name = tensor("op_5722_cast_fp16")]; tensor var_5723_begin_0 = const()[name = tensor("op_5723_begin_0"), val = tensor([0, 0, 0, 1536])]; tensor var_5723_end_0 = const()[name = tensor("op_5723_end_0"), val = tensor([2, 40, 1, 2048])]; tensor var_5723_end_mask_0 = const()[name = tensor("op_5723_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5723_cast_fp16 = slice_by_index(begin = var_5723_begin_0, end = var_5723_end_0, end_mask = var_5723_end_mask_0, x = var_5677_cast_fp16)[name = tensor("op_5723_cast_fp16")]; tensor var_5724_begin_0 = const()[name = tensor("op_5724_begin_0"), val = tensor([0, 0, 0, 2048])]; tensor var_5724_end_0 = const()[name = tensor("op_5724_end_0"), val = tensor([2, 40, 1, 2560])]; tensor var_5724_end_mask_0 = const()[name = tensor("op_5724_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5724_cast_fp16 = slice_by_index(begin = var_5724_begin_0, end = var_5724_end_0, end_mask = var_5724_end_mask_0, x = var_5677_cast_fp16)[name = tensor("op_5724_cast_fp16")]; tensor var_5725_begin_0 = const()[name = tensor("op_5725_begin_0"), val = tensor([0, 0, 0, 2560])]; tensor var_5725_end_0 = const()[name = tensor("op_5725_end_0"), val = tensor([2, 40, 1, 3072])]; tensor var_5725_end_mask_0 = const()[name = tensor("op_5725_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5725_cast_fp16 = slice_by_index(begin = var_5725_begin_0, end = var_5725_end_0, end_mask = var_5725_end_mask_0, x = var_5677_cast_fp16)[name = tensor("op_5725_cast_fp16")]; tensor var_5726_begin_0 = const()[name = tensor("op_5726_begin_0"), val = tensor([0, 0, 0, 3072])]; tensor var_5726_end_0 = const()[name = tensor("op_5726_end_0"), val = tensor([2, 40, 1, 3584])]; tensor var_5726_end_mask_0 = const()[name = tensor("op_5726_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5726_cast_fp16 = slice_by_index(begin = var_5726_begin_0, end = var_5726_end_0, end_mask = var_5726_end_mask_0, x = var_5677_cast_fp16)[name = tensor("op_5726_cast_fp16")]; tensor var_5727_begin_0 = const()[name = tensor("op_5727_begin_0"), val = tensor([0, 0, 0, 3584])]; tensor var_5727_end_0 = const()[name = tensor("op_5727_end_0"), val = tensor([2, 40, 1, 4096])]; tensor var_5727_end_mask_0 = const()[name = tensor("op_5727_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5727_cast_fp16 = slice_by_index(begin = var_5727_begin_0, end = var_5727_end_0, end_mask = var_5727_end_mask_0, x = var_5677_cast_fp16)[name = tensor("op_5727_cast_fp16")]; tensor var_5728_begin_0 = const()[name = tensor("op_5728_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_5728_end_0 = const()[name = tensor("op_5728_end_0"), val = tensor([2, 40, 1, 512])]; tensor var_5728_end_mask_0 = const()[name = tensor("op_5728_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5728_cast_fp16 = slice_by_index(begin = var_5728_begin_0, end = var_5728_end_0, end_mask = var_5728_end_mask_0, x = var_5681_cast_fp16)[name = tensor("op_5728_cast_fp16")]; tensor var_5729_begin_0 = const()[name = tensor("op_5729_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_5729_end_0 = const()[name = tensor("op_5729_end_0"), val = tensor([2, 40, 1, 1024])]; tensor var_5729_end_mask_0 = const()[name = tensor("op_5729_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5729_cast_fp16 = slice_by_index(begin = var_5729_begin_0, end = var_5729_end_0, end_mask = var_5729_end_mask_0, x = var_5681_cast_fp16)[name = tensor("op_5729_cast_fp16")]; tensor var_5730_begin_0 = const()[name = tensor("op_5730_begin_0"), val = tensor([0, 0, 0, 1024])]; tensor var_5730_end_0 = const()[name = tensor("op_5730_end_0"), val = tensor([2, 40, 1, 1536])]; tensor var_5730_end_mask_0 = const()[name = tensor("op_5730_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5730_cast_fp16 = slice_by_index(begin = var_5730_begin_0, end = var_5730_end_0, end_mask = var_5730_end_mask_0, x = var_5681_cast_fp16)[name = tensor("op_5730_cast_fp16")]; tensor var_5731_begin_0 = const()[name = tensor("op_5731_begin_0"), val = tensor([0, 0, 0, 1536])]; tensor var_5731_end_0 = const()[name = tensor("op_5731_end_0"), val = tensor([2, 40, 1, 2048])]; tensor var_5731_end_mask_0 = const()[name = tensor("op_5731_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5731_cast_fp16 = slice_by_index(begin = var_5731_begin_0, end = var_5731_end_0, end_mask = var_5731_end_mask_0, x = var_5681_cast_fp16)[name = tensor("op_5731_cast_fp16")]; tensor var_5732_begin_0 = const()[name = tensor("op_5732_begin_0"), val = tensor([0, 0, 0, 2048])]; tensor var_5732_end_0 = const()[name = tensor("op_5732_end_0"), val = tensor([2, 40, 1, 2560])]; tensor var_5732_end_mask_0 = const()[name = tensor("op_5732_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5732_cast_fp16 = slice_by_index(begin = var_5732_begin_0, end = var_5732_end_0, end_mask = var_5732_end_mask_0, x = var_5681_cast_fp16)[name = tensor("op_5732_cast_fp16")]; tensor var_5733_begin_0 = const()[name = tensor("op_5733_begin_0"), val = tensor([0, 0, 0, 2560])]; tensor var_5733_end_0 = const()[name = tensor("op_5733_end_0"), val = tensor([2, 40, 1, 3072])]; tensor var_5733_end_mask_0 = const()[name = tensor("op_5733_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5733_cast_fp16 = slice_by_index(begin = var_5733_begin_0, end = var_5733_end_0, end_mask = var_5733_end_mask_0, x = var_5681_cast_fp16)[name = tensor("op_5733_cast_fp16")]; tensor var_5734_begin_0 = const()[name = tensor("op_5734_begin_0"), val = tensor([0, 0, 0, 3072])]; tensor var_5734_end_0 = const()[name = tensor("op_5734_end_0"), val = tensor([2, 40, 1, 3584])]; tensor var_5734_end_mask_0 = const()[name = tensor("op_5734_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5734_cast_fp16 = slice_by_index(begin = var_5734_begin_0, end = var_5734_end_0, end_mask = var_5734_end_mask_0, x = var_5681_cast_fp16)[name = tensor("op_5734_cast_fp16")]; tensor var_5735_begin_0 = const()[name = tensor("op_5735_begin_0"), val = tensor([0, 0, 0, 3584])]; tensor var_5735_end_0 = const()[name = tensor("op_5735_end_0"), val = tensor([2, 40, 1, 4096])]; tensor var_5735_end_mask_0 = const()[name = tensor("op_5735_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5735_cast_fp16 = slice_by_index(begin = var_5735_begin_0, end = var_5735_end_0, end_mask = var_5735_end_mask_0, x = var_5681_cast_fp16)[name = tensor("op_5735_cast_fp16")]; tensor var_5736_begin_0 = const()[name = tensor("op_5736_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_5736_end_0 = const()[name = tensor("op_5736_end_0"), val = tensor([2, 40, 1, 512])]; tensor var_5736_end_mask_0 = const()[name = tensor("op_5736_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5736_cast_fp16 = slice_by_index(begin = var_5736_begin_0, end = var_5736_end_0, end_mask = var_5736_end_mask_0, x = var_5685_cast_fp16)[name = tensor("op_5736_cast_fp16")]; tensor var_5737_begin_0 = const()[name = tensor("op_5737_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_5737_end_0 = const()[name = tensor("op_5737_end_0"), val = tensor([2, 40, 1, 1024])]; tensor var_5737_end_mask_0 = const()[name = tensor("op_5737_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5737_cast_fp16 = slice_by_index(begin = var_5737_begin_0, end = var_5737_end_0, end_mask = var_5737_end_mask_0, x = var_5685_cast_fp16)[name = tensor("op_5737_cast_fp16")]; tensor var_5738_begin_0 = const()[name = tensor("op_5738_begin_0"), val = tensor([0, 0, 0, 1024])]; tensor var_5738_end_0 = const()[name = tensor("op_5738_end_0"), val = tensor([2, 40, 1, 1536])]; tensor var_5738_end_mask_0 = const()[name = tensor("op_5738_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5738_cast_fp16 = slice_by_index(begin = var_5738_begin_0, end = var_5738_end_0, end_mask = var_5738_end_mask_0, x = var_5685_cast_fp16)[name = tensor("op_5738_cast_fp16")]; tensor var_5739_begin_0 = const()[name = tensor("op_5739_begin_0"), val = tensor([0, 0, 0, 1536])]; tensor var_5739_end_0 = const()[name = tensor("op_5739_end_0"), val = tensor([2, 40, 1, 2048])]; tensor var_5739_end_mask_0 = const()[name = tensor("op_5739_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5739_cast_fp16 = slice_by_index(begin = var_5739_begin_0, end = var_5739_end_0, end_mask = var_5739_end_mask_0, x = var_5685_cast_fp16)[name = tensor("op_5739_cast_fp16")]; tensor var_5740_begin_0 = const()[name = tensor("op_5740_begin_0"), val = tensor([0, 0, 0, 2048])]; tensor var_5740_end_0 = const()[name = tensor("op_5740_end_0"), val = tensor([2, 40, 1, 2560])]; tensor var_5740_end_mask_0 = const()[name = tensor("op_5740_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5740_cast_fp16 = slice_by_index(begin = var_5740_begin_0, end = var_5740_end_0, end_mask = var_5740_end_mask_0, x = var_5685_cast_fp16)[name = tensor("op_5740_cast_fp16")]; tensor var_5741_begin_0 = const()[name = tensor("op_5741_begin_0"), val = tensor([0, 0, 0, 2560])]; tensor var_5741_end_0 = const()[name = tensor("op_5741_end_0"), val = tensor([2, 40, 1, 3072])]; tensor var_5741_end_mask_0 = const()[name = tensor("op_5741_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5741_cast_fp16 = slice_by_index(begin = var_5741_begin_0, end = var_5741_end_0, end_mask = var_5741_end_mask_0, x = var_5685_cast_fp16)[name = tensor("op_5741_cast_fp16")]; tensor var_5742_begin_0 = const()[name = tensor("op_5742_begin_0"), val = tensor([0, 0, 0, 3072])]; tensor var_5742_end_0 = const()[name = tensor("op_5742_end_0"), val = tensor([2, 40, 1, 3584])]; tensor var_5742_end_mask_0 = const()[name = tensor("op_5742_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5742_cast_fp16 = slice_by_index(begin = var_5742_begin_0, end = var_5742_end_0, end_mask = var_5742_end_mask_0, x = var_5685_cast_fp16)[name = tensor("op_5742_cast_fp16")]; tensor var_5743_begin_0 = const()[name = tensor("op_5743_begin_0"), val = tensor([0, 0, 0, 3584])]; tensor var_5743_end_0 = const()[name = tensor("op_5743_end_0"), val = tensor([2, 40, 1, 4096])]; tensor var_5743_end_mask_0 = const()[name = tensor("op_5743_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5743_cast_fp16 = slice_by_index(begin = var_5743_begin_0, end = var_5743_end_0, end_mask = var_5743_end_mask_0, x = var_5685_cast_fp16)[name = tensor("op_5743_cast_fp16")]; tensor var_5744_begin_0 = const()[name = tensor("op_5744_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_5744_end_0 = const()[name = tensor("op_5744_end_0"), val = tensor([2, 40, 1, 512])]; tensor var_5744_end_mask_0 = const()[name = tensor("op_5744_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5744_cast_fp16 = slice_by_index(begin = var_5744_begin_0, end = var_5744_end_0, end_mask = var_5744_end_mask_0, x = var_5689_cast_fp16)[name = tensor("op_5744_cast_fp16")]; tensor var_5745_begin_0 = const()[name = tensor("op_5745_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_5745_end_0 = const()[name = tensor("op_5745_end_0"), val = tensor([2, 40, 1, 1024])]; tensor var_5745_end_mask_0 = const()[name = tensor("op_5745_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5745_cast_fp16 = slice_by_index(begin = var_5745_begin_0, end = var_5745_end_0, end_mask = var_5745_end_mask_0, x = var_5689_cast_fp16)[name = tensor("op_5745_cast_fp16")]; tensor var_5746_begin_0 = const()[name = tensor("op_5746_begin_0"), val = tensor([0, 0, 0, 1024])]; tensor var_5746_end_0 = const()[name = tensor("op_5746_end_0"), val = tensor([2, 40, 1, 1536])]; tensor var_5746_end_mask_0 = const()[name = tensor("op_5746_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5746_cast_fp16 = slice_by_index(begin = var_5746_begin_0, end = var_5746_end_0, end_mask = var_5746_end_mask_0, x = var_5689_cast_fp16)[name = tensor("op_5746_cast_fp16")]; tensor var_5747_begin_0 = const()[name = tensor("op_5747_begin_0"), val = tensor([0, 0, 0, 1536])]; tensor var_5747_end_0 = const()[name = tensor("op_5747_end_0"), val = tensor([2, 40, 1, 2048])]; tensor var_5747_end_mask_0 = const()[name = tensor("op_5747_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5747_cast_fp16 = slice_by_index(begin = var_5747_begin_0, end = var_5747_end_0, end_mask = var_5747_end_mask_0, x = var_5689_cast_fp16)[name = tensor("op_5747_cast_fp16")]; tensor var_5748_begin_0 = const()[name = tensor("op_5748_begin_0"), val = tensor([0, 0, 0, 2048])]; tensor var_5748_end_0 = const()[name = tensor("op_5748_end_0"), val = tensor([2, 40, 1, 2560])]; tensor var_5748_end_mask_0 = const()[name = tensor("op_5748_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5748_cast_fp16 = slice_by_index(begin = var_5748_begin_0, end = var_5748_end_0, end_mask = var_5748_end_mask_0, x = var_5689_cast_fp16)[name = tensor("op_5748_cast_fp16")]; tensor var_5749_begin_0 = const()[name = tensor("op_5749_begin_0"), val = tensor([0, 0, 0, 2560])]; tensor var_5749_end_0 = const()[name = tensor("op_5749_end_0"), val = tensor([2, 40, 1, 3072])]; tensor var_5749_end_mask_0 = const()[name = tensor("op_5749_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5749_cast_fp16 = slice_by_index(begin = var_5749_begin_0, end = var_5749_end_0, end_mask = var_5749_end_mask_0, x = var_5689_cast_fp16)[name = tensor("op_5749_cast_fp16")]; tensor var_5750_begin_0 = const()[name = tensor("op_5750_begin_0"), val = tensor([0, 0, 0, 3072])]; tensor var_5750_end_0 = const()[name = tensor("op_5750_end_0"), val = tensor([2, 40, 1, 3584])]; tensor var_5750_end_mask_0 = const()[name = tensor("op_5750_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5750_cast_fp16 = slice_by_index(begin = var_5750_begin_0, end = var_5750_end_0, end_mask = var_5750_end_mask_0, x = var_5689_cast_fp16)[name = tensor("op_5750_cast_fp16")]; tensor var_5751_begin_0 = const()[name = tensor("op_5751_begin_0"), val = tensor([0, 0, 0, 3584])]; tensor var_5751_end_0 = const()[name = tensor("op_5751_end_0"), val = tensor([2, 40, 1, 4096])]; tensor var_5751_end_mask_0 = const()[name = tensor("op_5751_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5751_cast_fp16 = slice_by_index(begin = var_5751_begin_0, end = var_5751_end_0, end_mask = var_5751_end_mask_0, x = var_5689_cast_fp16)[name = tensor("op_5751_cast_fp16")]; tensor var_5752_begin_0 = const()[name = tensor("op_5752_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_5752_end_0 = const()[name = tensor("op_5752_end_0"), val = tensor([2, 40, 1, 512])]; tensor var_5752_end_mask_0 = const()[name = tensor("op_5752_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5752_cast_fp16 = slice_by_index(begin = var_5752_begin_0, end = var_5752_end_0, end_mask = var_5752_end_mask_0, x = var_5693_cast_fp16)[name = tensor("op_5752_cast_fp16")]; tensor var_5753_begin_0 = const()[name = tensor("op_5753_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_5753_end_0 = const()[name = tensor("op_5753_end_0"), val = tensor([2, 40, 1, 1024])]; tensor var_5753_end_mask_0 = const()[name = tensor("op_5753_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5753_cast_fp16 = slice_by_index(begin = var_5753_begin_0, end = var_5753_end_0, end_mask = var_5753_end_mask_0, x = var_5693_cast_fp16)[name = tensor("op_5753_cast_fp16")]; tensor var_5754_begin_0 = const()[name = tensor("op_5754_begin_0"), val = tensor([0, 0, 0, 1024])]; tensor var_5754_end_0 = const()[name = tensor("op_5754_end_0"), val = tensor([2, 40, 1, 1536])]; tensor var_5754_end_mask_0 = const()[name = tensor("op_5754_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5754_cast_fp16 = slice_by_index(begin = var_5754_begin_0, end = var_5754_end_0, end_mask = var_5754_end_mask_0, x = var_5693_cast_fp16)[name = tensor("op_5754_cast_fp16")]; tensor var_5755_begin_0 = const()[name = tensor("op_5755_begin_0"), val = tensor([0, 0, 0, 1536])]; tensor var_5755_end_0 = const()[name = tensor("op_5755_end_0"), val = tensor([2, 40, 1, 2048])]; tensor var_5755_end_mask_0 = const()[name = tensor("op_5755_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5755_cast_fp16 = slice_by_index(begin = var_5755_begin_0, end = var_5755_end_0, end_mask = var_5755_end_mask_0, x = var_5693_cast_fp16)[name = tensor("op_5755_cast_fp16")]; tensor var_5756_begin_0 = const()[name = tensor("op_5756_begin_0"), val = tensor([0, 0, 0, 2048])]; tensor var_5756_end_0 = const()[name = tensor("op_5756_end_0"), val = tensor([2, 40, 1, 2560])]; tensor var_5756_end_mask_0 = const()[name = tensor("op_5756_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5756_cast_fp16 = slice_by_index(begin = var_5756_begin_0, end = var_5756_end_0, end_mask = var_5756_end_mask_0, x = var_5693_cast_fp16)[name = tensor("op_5756_cast_fp16")]; tensor var_5757_begin_0 = const()[name = tensor("op_5757_begin_0"), val = tensor([0, 0, 0, 2560])]; tensor var_5757_end_0 = const()[name = tensor("op_5757_end_0"), val = tensor([2, 40, 1, 3072])]; tensor var_5757_end_mask_0 = const()[name = tensor("op_5757_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5757_cast_fp16 = slice_by_index(begin = var_5757_begin_0, end = var_5757_end_0, end_mask = var_5757_end_mask_0, x = var_5693_cast_fp16)[name = tensor("op_5757_cast_fp16")]; tensor var_5758_begin_0 = const()[name = tensor("op_5758_begin_0"), val = tensor([0, 0, 0, 3072])]; tensor var_5758_end_0 = const()[name = tensor("op_5758_end_0"), val = tensor([2, 40, 1, 3584])]; tensor var_5758_end_mask_0 = const()[name = tensor("op_5758_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5758_cast_fp16 = slice_by_index(begin = var_5758_begin_0, end = var_5758_end_0, end_mask = var_5758_end_mask_0, x = var_5693_cast_fp16)[name = tensor("op_5758_cast_fp16")]; tensor var_5759_begin_0 = const()[name = tensor("op_5759_begin_0"), val = tensor([0, 0, 0, 3584])]; tensor var_5759_end_0 = const()[name = tensor("op_5759_end_0"), val = tensor([2, 40, 1, 4096])]; tensor var_5759_end_mask_0 = const()[name = tensor("op_5759_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5759_cast_fp16 = slice_by_index(begin = var_5759_begin_0, end = var_5759_end_0, end_mask = var_5759_end_mask_0, x = var_5693_cast_fp16)[name = tensor("op_5759_cast_fp16")]; tensor k_59_perm_0 = const()[name = tensor("k_59_perm_0"), val = tensor([0, 3, 2, 1])]; tensor var_5764_begin_0 = const()[name = tensor("op_5764_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_5764_end_0 = const()[name = tensor("op_5764_end_0"), val = tensor([2, 4096, 1, 40])]; tensor var_5764_end_mask_0 = const()[name = tensor("op_5764_end_mask_0"), val = tensor([true, true, true, false])]; tensor transpose_3 = transpose(perm = k_59_perm_0, x = k_57_cast_fp16)[name = tensor("transpose_3")]; tensor var_5764_cast_fp16 = slice_by_index(begin = var_5764_begin_0, end = var_5764_end_0, end_mask = var_5764_end_mask_0, x = transpose_3)[name = tensor("op_5764_cast_fp16")]; tensor var_5768_begin_0 = const()[name = tensor("op_5768_begin_0"), val = tensor([0, 0, 0, 40])]; tensor var_5768_end_0 = const()[name = tensor("op_5768_end_0"), val = tensor([2, 4096, 1, 80])]; tensor var_5768_end_mask_0 = const()[name = tensor("op_5768_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5768_cast_fp16 = slice_by_index(begin = var_5768_begin_0, end = var_5768_end_0, end_mask = var_5768_end_mask_0, x = transpose_3)[name = tensor("op_5768_cast_fp16")]; tensor var_5772_begin_0 = const()[name = tensor("op_5772_begin_0"), val = tensor([0, 0, 0, 80])]; tensor var_5772_end_0 = const()[name = tensor("op_5772_end_0"), val = tensor([2, 4096, 1, 120])]; tensor var_5772_end_mask_0 = const()[name = tensor("op_5772_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5772_cast_fp16 = slice_by_index(begin = var_5772_begin_0, end = var_5772_end_0, end_mask = var_5772_end_mask_0, x = transpose_3)[name = tensor("op_5772_cast_fp16")]; tensor var_5776_begin_0 = const()[name = tensor("op_5776_begin_0"), val = tensor([0, 0, 0, 120])]; tensor var_5776_end_0 = const()[name = tensor("op_5776_end_0"), val = tensor([2, 4096, 1, 160])]; tensor var_5776_end_mask_0 = const()[name = tensor("op_5776_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5776_cast_fp16 = slice_by_index(begin = var_5776_begin_0, end = var_5776_end_0, end_mask = var_5776_end_mask_0, x = transpose_3)[name = tensor("op_5776_cast_fp16")]; tensor var_5780_begin_0 = const()[name = tensor("op_5780_begin_0"), val = tensor([0, 0, 0, 160])]; tensor var_5780_end_0 = const()[name = tensor("op_5780_end_0"), val = tensor([2, 4096, 1, 200])]; tensor var_5780_end_mask_0 = const()[name = tensor("op_5780_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5780_cast_fp16 = slice_by_index(begin = var_5780_begin_0, end = var_5780_end_0, end_mask = var_5780_end_mask_0, x = transpose_3)[name = tensor("op_5780_cast_fp16")]; tensor var_5784_begin_0 = const()[name = tensor("op_5784_begin_0"), val = tensor([0, 0, 0, 200])]; tensor var_5784_end_0 = const()[name = tensor("op_5784_end_0"), val = tensor([2, 4096, 1, 240])]; tensor var_5784_end_mask_0 = const()[name = tensor("op_5784_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5784_cast_fp16 = slice_by_index(begin = var_5784_begin_0, end = var_5784_end_0, end_mask = var_5784_end_mask_0, x = transpose_3)[name = tensor("op_5784_cast_fp16")]; tensor var_5788_begin_0 = const()[name = tensor("op_5788_begin_0"), val = tensor([0, 0, 0, 240])]; tensor var_5788_end_0 = const()[name = tensor("op_5788_end_0"), val = tensor([2, 4096, 1, 280])]; tensor var_5788_end_mask_0 = const()[name = tensor("op_5788_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5788_cast_fp16 = slice_by_index(begin = var_5788_begin_0, end = var_5788_end_0, end_mask = var_5788_end_mask_0, x = transpose_3)[name = tensor("op_5788_cast_fp16")]; tensor var_5792_begin_0 = const()[name = tensor("op_5792_begin_0"), val = tensor([0, 0, 0, 280])]; tensor var_5792_end_0 = const()[name = tensor("op_5792_end_0"), val = tensor([2, 4096, 1, 320])]; tensor var_5792_end_mask_0 = const()[name = tensor("op_5792_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5792_cast_fp16 = slice_by_index(begin = var_5792_begin_0, end = var_5792_end_0, end_mask = var_5792_end_mask_0, x = transpose_3)[name = tensor("op_5792_cast_fp16")]; tensor var_5794_begin_0 = const()[name = tensor("op_5794_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_5794_end_0 = const()[name = tensor("op_5794_end_0"), val = tensor([2, 40, 1, 4096])]; tensor var_5794_end_mask_0 = const()[name = tensor("op_5794_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_5794_cast_fp16 = slice_by_index(begin = var_5794_begin_0, end = var_5794_end_0, end_mask = var_5794_end_mask_0, x = v_29_cast_fp16)[name = tensor("op_5794_cast_fp16")]; tensor var_5798_begin_0 = const()[name = tensor("op_5798_begin_0"), val = tensor([0, 40, 0, 0])]; tensor var_5798_end_0 = const()[name = tensor("op_5798_end_0"), val = tensor([2, 80, 1, 4096])]; tensor var_5798_end_mask_0 = const()[name = tensor("op_5798_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_5798_cast_fp16 = slice_by_index(begin = var_5798_begin_0, end = var_5798_end_0, end_mask = var_5798_end_mask_0, x = v_29_cast_fp16)[name = tensor("op_5798_cast_fp16")]; tensor var_5802_begin_0 = const()[name = tensor("op_5802_begin_0"), val = tensor([0, 80, 0, 0])]; tensor var_5802_end_0 = const()[name = tensor("op_5802_end_0"), val = tensor([2, 120, 1, 4096])]; tensor var_5802_end_mask_0 = const()[name = tensor("op_5802_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_5802_cast_fp16 = slice_by_index(begin = var_5802_begin_0, end = var_5802_end_0, end_mask = var_5802_end_mask_0, x = v_29_cast_fp16)[name = tensor("op_5802_cast_fp16")]; tensor var_5806_begin_0 = const()[name = tensor("op_5806_begin_0"), val = tensor([0, 120, 0, 0])]; tensor var_5806_end_0 = const()[name = tensor("op_5806_end_0"), val = tensor([2, 160, 1, 4096])]; tensor var_5806_end_mask_0 = const()[name = tensor("op_5806_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_5806_cast_fp16 = slice_by_index(begin = var_5806_begin_0, end = var_5806_end_0, end_mask = var_5806_end_mask_0, x = v_29_cast_fp16)[name = tensor("op_5806_cast_fp16")]; tensor var_5810_begin_0 = const()[name = tensor("op_5810_begin_0"), val = tensor([0, 160, 0, 0])]; tensor var_5810_end_0 = const()[name = tensor("op_5810_end_0"), val = tensor([2, 200, 1, 4096])]; tensor var_5810_end_mask_0 = const()[name = tensor("op_5810_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_5810_cast_fp16 = slice_by_index(begin = var_5810_begin_0, end = var_5810_end_0, end_mask = var_5810_end_mask_0, x = v_29_cast_fp16)[name = tensor("op_5810_cast_fp16")]; tensor var_5814_begin_0 = const()[name = tensor("op_5814_begin_0"), val = tensor([0, 200, 0, 0])]; tensor var_5814_end_0 = const()[name = tensor("op_5814_end_0"), val = tensor([2, 240, 1, 4096])]; tensor var_5814_end_mask_0 = const()[name = tensor("op_5814_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_5814_cast_fp16 = slice_by_index(begin = var_5814_begin_0, end = var_5814_end_0, end_mask = var_5814_end_mask_0, x = v_29_cast_fp16)[name = tensor("op_5814_cast_fp16")]; tensor var_5818_begin_0 = const()[name = tensor("op_5818_begin_0"), val = tensor([0, 240, 0, 0])]; tensor var_5818_end_0 = const()[name = tensor("op_5818_end_0"), val = tensor([2, 280, 1, 4096])]; tensor var_5818_end_mask_0 = const()[name = tensor("op_5818_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_5818_cast_fp16 = slice_by_index(begin = var_5818_begin_0, end = var_5818_end_0, end_mask = var_5818_end_mask_0, x = v_29_cast_fp16)[name = tensor("op_5818_cast_fp16")]; tensor var_5822_begin_0 = const()[name = tensor("op_5822_begin_0"), val = tensor([0, 280, 0, 0])]; tensor var_5822_end_0 = const()[name = tensor("op_5822_end_0"), val = tensor([2, 320, 1, 4096])]; tensor var_5822_end_mask_0 = const()[name = tensor("op_5822_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_5822_cast_fp16 = slice_by_index(begin = var_5822_begin_0, end = var_5822_end_0, end_mask = var_5822_end_mask_0, x = v_29_cast_fp16)[name = tensor("op_5822_cast_fp16")]; tensor var_5826_equation_0 = const()[name = tensor("op_5826_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_5826_cast_fp16 = einsum(equation = var_5826_equation_0, values = (var_5764_cast_fp16, var_5696_cast_fp16))[name = tensor("op_5826_cast_fp16")]; tensor var_5827_to_fp16 = const()[name = tensor("op_5827_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_449_cast_fp16 = mul(x = var_5826_cast_fp16, y = var_5827_to_fp16)[name = tensor("aw_chunk_449_cast_fp16")]; tensor var_5830_equation_0 = const()[name = tensor("op_5830_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_5830_cast_fp16 = einsum(equation = var_5830_equation_0, values = (var_5764_cast_fp16, var_5697_cast_fp16))[name = tensor("op_5830_cast_fp16")]; tensor var_5831_to_fp16 = const()[name = tensor("op_5831_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_451_cast_fp16 = mul(x = var_5830_cast_fp16, y = var_5831_to_fp16)[name = tensor("aw_chunk_451_cast_fp16")]; tensor var_5834_equation_0 = const()[name = tensor("op_5834_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_5834_cast_fp16 = einsum(equation = var_5834_equation_0, values = (var_5764_cast_fp16, var_5698_cast_fp16))[name = tensor("op_5834_cast_fp16")]; tensor var_5835_to_fp16 = const()[name = tensor("op_5835_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_453_cast_fp16 = mul(x = var_5834_cast_fp16, y = var_5835_to_fp16)[name = tensor("aw_chunk_453_cast_fp16")]; tensor var_5838_equation_0 = const()[name = tensor("op_5838_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_5838_cast_fp16 = einsum(equation = var_5838_equation_0, values = (var_5764_cast_fp16, var_5699_cast_fp16))[name = tensor("op_5838_cast_fp16")]; tensor var_5839_to_fp16 = const()[name = tensor("op_5839_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_455_cast_fp16 = mul(x = var_5838_cast_fp16, y = var_5839_to_fp16)[name = tensor("aw_chunk_455_cast_fp16")]; tensor var_5842_equation_0 = const()[name = tensor("op_5842_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_5842_cast_fp16 = einsum(equation = var_5842_equation_0, values = (var_5764_cast_fp16, var_5700_cast_fp16))[name = tensor("op_5842_cast_fp16")]; tensor var_5843_to_fp16 = const()[name = tensor("op_5843_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_457_cast_fp16 = mul(x = var_5842_cast_fp16, y = var_5843_to_fp16)[name = tensor("aw_chunk_457_cast_fp16")]; tensor var_5846_equation_0 = const()[name = tensor("op_5846_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_5846_cast_fp16 = einsum(equation = var_5846_equation_0, values = (var_5764_cast_fp16, var_5701_cast_fp16))[name = tensor("op_5846_cast_fp16")]; tensor var_5847_to_fp16 = const()[name = tensor("op_5847_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_459_cast_fp16 = mul(x = var_5846_cast_fp16, y = var_5847_to_fp16)[name = tensor("aw_chunk_459_cast_fp16")]; tensor var_5850_equation_0 = const()[name = tensor("op_5850_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_5850_cast_fp16 = einsum(equation = var_5850_equation_0, values = (var_5764_cast_fp16, var_5702_cast_fp16))[name = tensor("op_5850_cast_fp16")]; tensor var_5851_to_fp16 = const()[name = tensor("op_5851_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_461_cast_fp16 = mul(x = var_5850_cast_fp16, y = var_5851_to_fp16)[name = tensor("aw_chunk_461_cast_fp16")]; tensor var_5854_equation_0 = const()[name = tensor("op_5854_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_5854_cast_fp16 = einsum(equation = var_5854_equation_0, values = (var_5764_cast_fp16, var_5703_cast_fp16))[name = tensor("op_5854_cast_fp16")]; tensor var_5855_to_fp16 = const()[name = tensor("op_5855_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_463_cast_fp16 = mul(x = var_5854_cast_fp16, y = var_5855_to_fp16)[name = tensor("aw_chunk_463_cast_fp16")]; tensor var_5858_equation_0 = const()[name = tensor("op_5858_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_5858_cast_fp16 = einsum(equation = var_5858_equation_0, values = (var_5768_cast_fp16, var_5704_cast_fp16))[name = tensor("op_5858_cast_fp16")]; tensor var_5859_to_fp16 = const()[name = tensor("op_5859_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_465_cast_fp16 = mul(x = var_5858_cast_fp16, y = var_5859_to_fp16)[name = tensor("aw_chunk_465_cast_fp16")]; tensor var_5862_equation_0 = const()[name = tensor("op_5862_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_5862_cast_fp16 = einsum(equation = var_5862_equation_0, values = (var_5768_cast_fp16, var_5705_cast_fp16))[name = tensor("op_5862_cast_fp16")]; tensor var_5863_to_fp16 = const()[name = tensor("op_5863_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_467_cast_fp16 = mul(x = var_5862_cast_fp16, y = var_5863_to_fp16)[name = tensor("aw_chunk_467_cast_fp16")]; tensor var_5866_equation_0 = const()[name = tensor("op_5866_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_5866_cast_fp16 = einsum(equation = var_5866_equation_0, values = (var_5768_cast_fp16, var_5706_cast_fp16))[name = tensor("op_5866_cast_fp16")]; tensor var_5867_to_fp16 = const()[name = tensor("op_5867_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_469_cast_fp16 = mul(x = var_5866_cast_fp16, y = var_5867_to_fp16)[name = tensor("aw_chunk_469_cast_fp16")]; tensor var_5870_equation_0 = const()[name = tensor("op_5870_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_5870_cast_fp16 = einsum(equation = var_5870_equation_0, values = (var_5768_cast_fp16, var_5707_cast_fp16))[name = tensor("op_5870_cast_fp16")]; tensor var_5871_to_fp16 = const()[name = tensor("op_5871_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_471_cast_fp16 = mul(x = var_5870_cast_fp16, y = var_5871_to_fp16)[name = tensor("aw_chunk_471_cast_fp16")]; tensor var_5874_equation_0 = const()[name = tensor("op_5874_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_5874_cast_fp16 = einsum(equation = var_5874_equation_0, values = (var_5768_cast_fp16, var_5708_cast_fp16))[name = tensor("op_5874_cast_fp16")]; tensor var_5875_to_fp16 = const()[name = tensor("op_5875_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_473_cast_fp16 = mul(x = var_5874_cast_fp16, y = var_5875_to_fp16)[name = tensor("aw_chunk_473_cast_fp16")]; tensor var_5878_equation_0 = const()[name = tensor("op_5878_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_5878_cast_fp16 = einsum(equation = var_5878_equation_0, values = (var_5768_cast_fp16, var_5709_cast_fp16))[name = tensor("op_5878_cast_fp16")]; tensor var_5879_to_fp16 = const()[name = tensor("op_5879_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_475_cast_fp16 = mul(x = var_5878_cast_fp16, y = var_5879_to_fp16)[name = tensor("aw_chunk_475_cast_fp16")]; tensor var_5882_equation_0 = const()[name = tensor("op_5882_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_5882_cast_fp16 = einsum(equation = var_5882_equation_0, values = (var_5768_cast_fp16, var_5710_cast_fp16))[name = tensor("op_5882_cast_fp16")]; tensor var_5883_to_fp16 = const()[name = tensor("op_5883_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_477_cast_fp16 = mul(x = var_5882_cast_fp16, y = var_5883_to_fp16)[name = tensor("aw_chunk_477_cast_fp16")]; tensor var_5886_equation_0 = const()[name = tensor("op_5886_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_5886_cast_fp16 = einsum(equation = var_5886_equation_0, values = (var_5768_cast_fp16, var_5711_cast_fp16))[name = tensor("op_5886_cast_fp16")]; tensor var_5887_to_fp16 = const()[name = tensor("op_5887_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_479_cast_fp16 = mul(x = var_5886_cast_fp16, y = var_5887_to_fp16)[name = tensor("aw_chunk_479_cast_fp16")]; tensor var_5890_equation_0 = const()[name = tensor("op_5890_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_5890_cast_fp16 = einsum(equation = var_5890_equation_0, values = (var_5772_cast_fp16, var_5712_cast_fp16))[name = tensor("op_5890_cast_fp16")]; tensor var_5891_to_fp16 = const()[name = tensor("op_5891_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_481_cast_fp16 = mul(x = var_5890_cast_fp16, y = var_5891_to_fp16)[name = tensor("aw_chunk_481_cast_fp16")]; tensor var_5894_equation_0 = const()[name = tensor("op_5894_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_5894_cast_fp16 = einsum(equation = var_5894_equation_0, values = (var_5772_cast_fp16, var_5713_cast_fp16))[name = tensor("op_5894_cast_fp16")]; tensor var_5895_to_fp16 = const()[name = tensor("op_5895_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_483_cast_fp16 = mul(x = var_5894_cast_fp16, y = var_5895_to_fp16)[name = tensor("aw_chunk_483_cast_fp16")]; tensor var_5898_equation_0 = const()[name = tensor("op_5898_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_5898_cast_fp16 = einsum(equation = var_5898_equation_0, values = (var_5772_cast_fp16, var_5714_cast_fp16))[name = tensor("op_5898_cast_fp16")]; tensor var_5899_to_fp16 = const()[name = tensor("op_5899_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_485_cast_fp16 = mul(x = var_5898_cast_fp16, y = var_5899_to_fp16)[name = tensor("aw_chunk_485_cast_fp16")]; tensor var_5902_equation_0 = const()[name = tensor("op_5902_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_5902_cast_fp16 = einsum(equation = var_5902_equation_0, values = (var_5772_cast_fp16, var_5715_cast_fp16))[name = tensor("op_5902_cast_fp16")]; tensor var_5903_to_fp16 = const()[name = tensor("op_5903_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_487_cast_fp16 = mul(x = var_5902_cast_fp16, y = var_5903_to_fp16)[name = tensor("aw_chunk_487_cast_fp16")]; tensor var_5906_equation_0 = const()[name = tensor("op_5906_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_5906_cast_fp16 = einsum(equation = var_5906_equation_0, values = (var_5772_cast_fp16, var_5716_cast_fp16))[name = tensor("op_5906_cast_fp16")]; tensor var_5907_to_fp16 = const()[name = tensor("op_5907_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_489_cast_fp16 = mul(x = var_5906_cast_fp16, y = var_5907_to_fp16)[name = tensor("aw_chunk_489_cast_fp16")]; tensor var_5910_equation_0 = const()[name = tensor("op_5910_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_5910_cast_fp16 = einsum(equation = var_5910_equation_0, values = (var_5772_cast_fp16, var_5717_cast_fp16))[name = tensor("op_5910_cast_fp16")]; tensor var_5911_to_fp16 = const()[name = tensor("op_5911_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_491_cast_fp16 = mul(x = var_5910_cast_fp16, y = var_5911_to_fp16)[name = tensor("aw_chunk_491_cast_fp16")]; tensor var_5914_equation_0 = const()[name = tensor("op_5914_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_5914_cast_fp16 = einsum(equation = var_5914_equation_0, values = (var_5772_cast_fp16, var_5718_cast_fp16))[name = tensor("op_5914_cast_fp16")]; tensor var_5915_to_fp16 = const()[name = tensor("op_5915_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_493_cast_fp16 = mul(x = var_5914_cast_fp16, y = var_5915_to_fp16)[name = tensor("aw_chunk_493_cast_fp16")]; tensor var_5918_equation_0 = const()[name = tensor("op_5918_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_5918_cast_fp16 = einsum(equation = var_5918_equation_0, values = (var_5772_cast_fp16, var_5719_cast_fp16))[name = tensor("op_5918_cast_fp16")]; tensor var_5919_to_fp16 = const()[name = tensor("op_5919_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_495_cast_fp16 = mul(x = var_5918_cast_fp16, y = var_5919_to_fp16)[name = tensor("aw_chunk_495_cast_fp16")]; tensor var_5922_equation_0 = const()[name = tensor("op_5922_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_5922_cast_fp16 = einsum(equation = var_5922_equation_0, values = (var_5776_cast_fp16, var_5720_cast_fp16))[name = tensor("op_5922_cast_fp16")]; tensor var_5923_to_fp16 = const()[name = tensor("op_5923_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_497_cast_fp16 = mul(x = var_5922_cast_fp16, y = var_5923_to_fp16)[name = tensor("aw_chunk_497_cast_fp16")]; tensor var_5926_equation_0 = const()[name = tensor("op_5926_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_5926_cast_fp16 = einsum(equation = var_5926_equation_0, values = (var_5776_cast_fp16, var_5721_cast_fp16))[name = tensor("op_5926_cast_fp16")]; tensor var_5927_to_fp16 = const()[name = tensor("op_5927_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_499_cast_fp16 = mul(x = var_5926_cast_fp16, y = var_5927_to_fp16)[name = tensor("aw_chunk_499_cast_fp16")]; tensor var_5930_equation_0 = const()[name = tensor("op_5930_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_5930_cast_fp16 = einsum(equation = var_5930_equation_0, values = (var_5776_cast_fp16, var_5722_cast_fp16))[name = tensor("op_5930_cast_fp16")]; tensor var_5931_to_fp16 = const()[name = tensor("op_5931_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_501_cast_fp16 = mul(x = var_5930_cast_fp16, y = var_5931_to_fp16)[name = tensor("aw_chunk_501_cast_fp16")]; tensor var_5934_equation_0 = const()[name = tensor("op_5934_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_5934_cast_fp16 = einsum(equation = var_5934_equation_0, values = (var_5776_cast_fp16, var_5723_cast_fp16))[name = tensor("op_5934_cast_fp16")]; tensor var_5935_to_fp16 = const()[name = tensor("op_5935_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_503_cast_fp16 = mul(x = var_5934_cast_fp16, y = var_5935_to_fp16)[name = tensor("aw_chunk_503_cast_fp16")]; tensor var_5938_equation_0 = const()[name = tensor("op_5938_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_5938_cast_fp16 = einsum(equation = var_5938_equation_0, values = (var_5776_cast_fp16, var_5724_cast_fp16))[name = tensor("op_5938_cast_fp16")]; tensor var_5939_to_fp16 = const()[name = tensor("op_5939_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_505_cast_fp16 = mul(x = var_5938_cast_fp16, y = var_5939_to_fp16)[name = tensor("aw_chunk_505_cast_fp16")]; tensor var_5942_equation_0 = const()[name = tensor("op_5942_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_5942_cast_fp16 = einsum(equation = var_5942_equation_0, values = (var_5776_cast_fp16, var_5725_cast_fp16))[name = tensor("op_5942_cast_fp16")]; tensor var_5943_to_fp16 = const()[name = tensor("op_5943_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_507_cast_fp16 = mul(x = var_5942_cast_fp16, y = var_5943_to_fp16)[name = tensor("aw_chunk_507_cast_fp16")]; tensor var_5946_equation_0 = const()[name = tensor("op_5946_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_5946_cast_fp16 = einsum(equation = var_5946_equation_0, values = (var_5776_cast_fp16, var_5726_cast_fp16))[name = tensor("op_5946_cast_fp16")]; tensor var_5947_to_fp16 = const()[name = tensor("op_5947_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_509_cast_fp16 = mul(x = var_5946_cast_fp16, y = var_5947_to_fp16)[name = tensor("aw_chunk_509_cast_fp16")]; tensor var_5950_equation_0 = const()[name = tensor("op_5950_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_5950_cast_fp16 = einsum(equation = var_5950_equation_0, values = (var_5776_cast_fp16, var_5727_cast_fp16))[name = tensor("op_5950_cast_fp16")]; tensor var_5951_to_fp16 = const()[name = tensor("op_5951_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_511_cast_fp16 = mul(x = var_5950_cast_fp16, y = var_5951_to_fp16)[name = tensor("aw_chunk_511_cast_fp16")]; tensor var_5954_equation_0 = const()[name = tensor("op_5954_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_5954_cast_fp16 = einsum(equation = var_5954_equation_0, values = (var_5780_cast_fp16, var_5728_cast_fp16))[name = tensor("op_5954_cast_fp16")]; tensor var_5955_to_fp16 = const()[name = tensor("op_5955_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_513_cast_fp16 = mul(x = var_5954_cast_fp16, y = var_5955_to_fp16)[name = tensor("aw_chunk_513_cast_fp16")]; tensor var_5958_equation_0 = const()[name = tensor("op_5958_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_5958_cast_fp16 = einsum(equation = var_5958_equation_0, values = (var_5780_cast_fp16, var_5729_cast_fp16))[name = tensor("op_5958_cast_fp16")]; tensor var_5959_to_fp16 = const()[name = tensor("op_5959_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_515_cast_fp16 = mul(x = var_5958_cast_fp16, y = var_5959_to_fp16)[name = tensor("aw_chunk_515_cast_fp16")]; tensor var_5962_equation_0 = const()[name = tensor("op_5962_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_5962_cast_fp16 = einsum(equation = var_5962_equation_0, values = (var_5780_cast_fp16, var_5730_cast_fp16))[name = tensor("op_5962_cast_fp16")]; tensor var_5963_to_fp16 = const()[name = tensor("op_5963_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_517_cast_fp16 = mul(x = var_5962_cast_fp16, y = var_5963_to_fp16)[name = tensor("aw_chunk_517_cast_fp16")]; tensor var_5966_equation_0 = const()[name = tensor("op_5966_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_5966_cast_fp16 = einsum(equation = var_5966_equation_0, values = (var_5780_cast_fp16, var_5731_cast_fp16))[name = tensor("op_5966_cast_fp16")]; tensor var_5967_to_fp16 = const()[name = tensor("op_5967_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_519_cast_fp16 = mul(x = var_5966_cast_fp16, y = var_5967_to_fp16)[name = tensor("aw_chunk_519_cast_fp16")]; tensor var_5970_equation_0 = const()[name = tensor("op_5970_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_5970_cast_fp16 = einsum(equation = var_5970_equation_0, values = (var_5780_cast_fp16, var_5732_cast_fp16))[name = tensor("op_5970_cast_fp16")]; tensor var_5971_to_fp16 = const()[name = tensor("op_5971_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_521_cast_fp16 = mul(x = var_5970_cast_fp16, y = var_5971_to_fp16)[name = tensor("aw_chunk_521_cast_fp16")]; tensor var_5974_equation_0 = const()[name = tensor("op_5974_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_5974_cast_fp16 = einsum(equation = var_5974_equation_0, values = (var_5780_cast_fp16, var_5733_cast_fp16))[name = tensor("op_5974_cast_fp16")]; tensor var_5975_to_fp16 = const()[name = tensor("op_5975_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_523_cast_fp16 = mul(x = var_5974_cast_fp16, y = var_5975_to_fp16)[name = tensor("aw_chunk_523_cast_fp16")]; tensor var_5978_equation_0 = const()[name = tensor("op_5978_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_5978_cast_fp16 = einsum(equation = var_5978_equation_0, values = (var_5780_cast_fp16, var_5734_cast_fp16))[name = tensor("op_5978_cast_fp16")]; tensor var_5979_to_fp16 = const()[name = tensor("op_5979_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_525_cast_fp16 = mul(x = var_5978_cast_fp16, y = var_5979_to_fp16)[name = tensor("aw_chunk_525_cast_fp16")]; tensor var_5982_equation_0 = const()[name = tensor("op_5982_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_5982_cast_fp16 = einsum(equation = var_5982_equation_0, values = (var_5780_cast_fp16, var_5735_cast_fp16))[name = tensor("op_5982_cast_fp16")]; tensor var_5983_to_fp16 = const()[name = tensor("op_5983_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_527_cast_fp16 = mul(x = var_5982_cast_fp16, y = var_5983_to_fp16)[name = tensor("aw_chunk_527_cast_fp16")]; tensor var_5986_equation_0 = const()[name = tensor("op_5986_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_5986_cast_fp16 = einsum(equation = var_5986_equation_0, values = (var_5784_cast_fp16, var_5736_cast_fp16))[name = tensor("op_5986_cast_fp16")]; tensor var_5987_to_fp16 = const()[name = tensor("op_5987_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_529_cast_fp16 = mul(x = var_5986_cast_fp16, y = var_5987_to_fp16)[name = tensor("aw_chunk_529_cast_fp16")]; tensor var_5990_equation_0 = const()[name = tensor("op_5990_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_5990_cast_fp16 = einsum(equation = var_5990_equation_0, values = (var_5784_cast_fp16, var_5737_cast_fp16))[name = tensor("op_5990_cast_fp16")]; tensor var_5991_to_fp16 = const()[name = tensor("op_5991_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_531_cast_fp16 = mul(x = var_5990_cast_fp16, y = var_5991_to_fp16)[name = tensor("aw_chunk_531_cast_fp16")]; tensor var_5994_equation_0 = const()[name = tensor("op_5994_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_5994_cast_fp16 = einsum(equation = var_5994_equation_0, values = (var_5784_cast_fp16, var_5738_cast_fp16))[name = tensor("op_5994_cast_fp16")]; tensor var_5995_to_fp16 = const()[name = tensor("op_5995_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_533_cast_fp16 = mul(x = var_5994_cast_fp16, y = var_5995_to_fp16)[name = tensor("aw_chunk_533_cast_fp16")]; tensor var_5998_equation_0 = const()[name = tensor("op_5998_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_5998_cast_fp16 = einsum(equation = var_5998_equation_0, values = (var_5784_cast_fp16, var_5739_cast_fp16))[name = tensor("op_5998_cast_fp16")]; tensor var_5999_to_fp16 = const()[name = tensor("op_5999_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_535_cast_fp16 = mul(x = var_5998_cast_fp16, y = var_5999_to_fp16)[name = tensor("aw_chunk_535_cast_fp16")]; tensor var_6002_equation_0 = const()[name = tensor("op_6002_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6002_cast_fp16 = einsum(equation = var_6002_equation_0, values = (var_5784_cast_fp16, var_5740_cast_fp16))[name = tensor("op_6002_cast_fp16")]; tensor var_6003_to_fp16 = const()[name = tensor("op_6003_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_537_cast_fp16 = mul(x = var_6002_cast_fp16, y = var_6003_to_fp16)[name = tensor("aw_chunk_537_cast_fp16")]; tensor var_6006_equation_0 = const()[name = tensor("op_6006_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6006_cast_fp16 = einsum(equation = var_6006_equation_0, values = (var_5784_cast_fp16, var_5741_cast_fp16))[name = tensor("op_6006_cast_fp16")]; tensor var_6007_to_fp16 = const()[name = tensor("op_6007_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_539_cast_fp16 = mul(x = var_6006_cast_fp16, y = var_6007_to_fp16)[name = tensor("aw_chunk_539_cast_fp16")]; tensor var_6010_equation_0 = const()[name = tensor("op_6010_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6010_cast_fp16 = einsum(equation = var_6010_equation_0, values = (var_5784_cast_fp16, var_5742_cast_fp16))[name = tensor("op_6010_cast_fp16")]; tensor var_6011_to_fp16 = const()[name = tensor("op_6011_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_541_cast_fp16 = mul(x = var_6010_cast_fp16, y = var_6011_to_fp16)[name = tensor("aw_chunk_541_cast_fp16")]; tensor var_6014_equation_0 = const()[name = tensor("op_6014_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6014_cast_fp16 = einsum(equation = var_6014_equation_0, values = (var_5784_cast_fp16, var_5743_cast_fp16))[name = tensor("op_6014_cast_fp16")]; tensor var_6015_to_fp16 = const()[name = tensor("op_6015_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_543_cast_fp16 = mul(x = var_6014_cast_fp16, y = var_6015_to_fp16)[name = tensor("aw_chunk_543_cast_fp16")]; tensor var_6018_equation_0 = const()[name = tensor("op_6018_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6018_cast_fp16 = einsum(equation = var_6018_equation_0, values = (var_5788_cast_fp16, var_5744_cast_fp16))[name = tensor("op_6018_cast_fp16")]; tensor var_6019_to_fp16 = const()[name = tensor("op_6019_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_545_cast_fp16 = mul(x = var_6018_cast_fp16, y = var_6019_to_fp16)[name = tensor("aw_chunk_545_cast_fp16")]; tensor var_6022_equation_0 = const()[name = tensor("op_6022_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6022_cast_fp16 = einsum(equation = var_6022_equation_0, values = (var_5788_cast_fp16, var_5745_cast_fp16))[name = tensor("op_6022_cast_fp16")]; tensor var_6023_to_fp16 = const()[name = tensor("op_6023_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_547_cast_fp16 = mul(x = var_6022_cast_fp16, y = var_6023_to_fp16)[name = tensor("aw_chunk_547_cast_fp16")]; tensor var_6026_equation_0 = const()[name = tensor("op_6026_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6026_cast_fp16 = einsum(equation = var_6026_equation_0, values = (var_5788_cast_fp16, var_5746_cast_fp16))[name = tensor("op_6026_cast_fp16")]; tensor var_6027_to_fp16 = const()[name = tensor("op_6027_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_549_cast_fp16 = mul(x = var_6026_cast_fp16, y = var_6027_to_fp16)[name = tensor("aw_chunk_549_cast_fp16")]; tensor var_6030_equation_0 = const()[name = tensor("op_6030_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6030_cast_fp16 = einsum(equation = var_6030_equation_0, values = (var_5788_cast_fp16, var_5747_cast_fp16))[name = tensor("op_6030_cast_fp16")]; tensor var_6031_to_fp16 = const()[name = tensor("op_6031_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_551_cast_fp16 = mul(x = var_6030_cast_fp16, y = var_6031_to_fp16)[name = tensor("aw_chunk_551_cast_fp16")]; tensor var_6034_equation_0 = const()[name = tensor("op_6034_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6034_cast_fp16 = einsum(equation = var_6034_equation_0, values = (var_5788_cast_fp16, var_5748_cast_fp16))[name = tensor("op_6034_cast_fp16")]; tensor var_6035_to_fp16 = const()[name = tensor("op_6035_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_553_cast_fp16 = mul(x = var_6034_cast_fp16, y = var_6035_to_fp16)[name = tensor("aw_chunk_553_cast_fp16")]; tensor var_6038_equation_0 = const()[name = tensor("op_6038_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6038_cast_fp16 = einsum(equation = var_6038_equation_0, values = (var_5788_cast_fp16, var_5749_cast_fp16))[name = tensor("op_6038_cast_fp16")]; tensor var_6039_to_fp16 = const()[name = tensor("op_6039_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_555_cast_fp16 = mul(x = var_6038_cast_fp16, y = var_6039_to_fp16)[name = tensor("aw_chunk_555_cast_fp16")]; tensor var_6042_equation_0 = const()[name = tensor("op_6042_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6042_cast_fp16 = einsum(equation = var_6042_equation_0, values = (var_5788_cast_fp16, var_5750_cast_fp16))[name = tensor("op_6042_cast_fp16")]; tensor var_6043_to_fp16 = const()[name = tensor("op_6043_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_557_cast_fp16 = mul(x = var_6042_cast_fp16, y = var_6043_to_fp16)[name = tensor("aw_chunk_557_cast_fp16")]; tensor var_6046_equation_0 = const()[name = tensor("op_6046_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6046_cast_fp16 = einsum(equation = var_6046_equation_0, values = (var_5788_cast_fp16, var_5751_cast_fp16))[name = tensor("op_6046_cast_fp16")]; tensor var_6047_to_fp16 = const()[name = tensor("op_6047_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_559_cast_fp16 = mul(x = var_6046_cast_fp16, y = var_6047_to_fp16)[name = tensor("aw_chunk_559_cast_fp16")]; tensor var_6050_equation_0 = const()[name = tensor("op_6050_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6050_cast_fp16 = einsum(equation = var_6050_equation_0, values = (var_5792_cast_fp16, var_5752_cast_fp16))[name = tensor("op_6050_cast_fp16")]; tensor var_6051_to_fp16 = const()[name = tensor("op_6051_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_561_cast_fp16 = mul(x = var_6050_cast_fp16, y = var_6051_to_fp16)[name = tensor("aw_chunk_561_cast_fp16")]; tensor var_6054_equation_0 = const()[name = tensor("op_6054_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6054_cast_fp16 = einsum(equation = var_6054_equation_0, values = (var_5792_cast_fp16, var_5753_cast_fp16))[name = tensor("op_6054_cast_fp16")]; tensor var_6055_to_fp16 = const()[name = tensor("op_6055_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_563_cast_fp16 = mul(x = var_6054_cast_fp16, y = var_6055_to_fp16)[name = tensor("aw_chunk_563_cast_fp16")]; tensor var_6058_equation_0 = const()[name = tensor("op_6058_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6058_cast_fp16 = einsum(equation = var_6058_equation_0, values = (var_5792_cast_fp16, var_5754_cast_fp16))[name = tensor("op_6058_cast_fp16")]; tensor var_6059_to_fp16 = const()[name = tensor("op_6059_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_565_cast_fp16 = mul(x = var_6058_cast_fp16, y = var_6059_to_fp16)[name = tensor("aw_chunk_565_cast_fp16")]; tensor var_6062_equation_0 = const()[name = tensor("op_6062_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6062_cast_fp16 = einsum(equation = var_6062_equation_0, values = (var_5792_cast_fp16, var_5755_cast_fp16))[name = tensor("op_6062_cast_fp16")]; tensor var_6063_to_fp16 = const()[name = tensor("op_6063_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_567_cast_fp16 = mul(x = var_6062_cast_fp16, y = var_6063_to_fp16)[name = tensor("aw_chunk_567_cast_fp16")]; tensor var_6066_equation_0 = const()[name = tensor("op_6066_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6066_cast_fp16 = einsum(equation = var_6066_equation_0, values = (var_5792_cast_fp16, var_5756_cast_fp16))[name = tensor("op_6066_cast_fp16")]; tensor var_6067_to_fp16 = const()[name = tensor("op_6067_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_569_cast_fp16 = mul(x = var_6066_cast_fp16, y = var_6067_to_fp16)[name = tensor("aw_chunk_569_cast_fp16")]; tensor var_6070_equation_0 = const()[name = tensor("op_6070_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6070_cast_fp16 = einsum(equation = var_6070_equation_0, values = (var_5792_cast_fp16, var_5757_cast_fp16))[name = tensor("op_6070_cast_fp16")]; tensor var_6071_to_fp16 = const()[name = tensor("op_6071_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_571_cast_fp16 = mul(x = var_6070_cast_fp16, y = var_6071_to_fp16)[name = tensor("aw_chunk_571_cast_fp16")]; tensor var_6074_equation_0 = const()[name = tensor("op_6074_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6074_cast_fp16 = einsum(equation = var_6074_equation_0, values = (var_5792_cast_fp16, var_5758_cast_fp16))[name = tensor("op_6074_cast_fp16")]; tensor var_6075_to_fp16 = const()[name = tensor("op_6075_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_573_cast_fp16 = mul(x = var_6074_cast_fp16, y = var_6075_to_fp16)[name = tensor("aw_chunk_573_cast_fp16")]; tensor var_6078_equation_0 = const()[name = tensor("op_6078_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6078_cast_fp16 = einsum(equation = var_6078_equation_0, values = (var_5792_cast_fp16, var_5759_cast_fp16))[name = tensor("op_6078_cast_fp16")]; tensor var_6079_to_fp16 = const()[name = tensor("op_6079_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_575_cast_fp16 = mul(x = var_6078_cast_fp16, y = var_6079_to_fp16)[name = tensor("aw_chunk_575_cast_fp16")]; tensor var_6081_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_449_cast_fp16)[name = tensor("op_6081_cast_fp16")]; tensor var_6082_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_451_cast_fp16)[name = tensor("op_6082_cast_fp16")]; tensor var_6083_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_453_cast_fp16)[name = tensor("op_6083_cast_fp16")]; tensor var_6084_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_455_cast_fp16)[name = tensor("op_6084_cast_fp16")]; tensor var_6085_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_457_cast_fp16)[name = tensor("op_6085_cast_fp16")]; tensor var_6086_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_459_cast_fp16)[name = tensor("op_6086_cast_fp16")]; tensor var_6087_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_461_cast_fp16)[name = tensor("op_6087_cast_fp16")]; tensor var_6088_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_463_cast_fp16)[name = tensor("op_6088_cast_fp16")]; tensor var_6089_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_465_cast_fp16)[name = tensor("op_6089_cast_fp16")]; tensor var_6090_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_467_cast_fp16)[name = tensor("op_6090_cast_fp16")]; tensor var_6091_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_469_cast_fp16)[name = tensor("op_6091_cast_fp16")]; tensor var_6092_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_471_cast_fp16)[name = tensor("op_6092_cast_fp16")]; tensor var_6093_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_473_cast_fp16)[name = tensor("op_6093_cast_fp16")]; tensor var_6094_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_475_cast_fp16)[name = tensor("op_6094_cast_fp16")]; tensor var_6095_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_477_cast_fp16)[name = tensor("op_6095_cast_fp16")]; tensor var_6096_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_479_cast_fp16)[name = tensor("op_6096_cast_fp16")]; tensor var_6097_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_481_cast_fp16)[name = tensor("op_6097_cast_fp16")]; tensor var_6098_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_483_cast_fp16)[name = tensor("op_6098_cast_fp16")]; tensor var_6099_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_485_cast_fp16)[name = tensor("op_6099_cast_fp16")]; tensor var_6100_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_487_cast_fp16)[name = tensor("op_6100_cast_fp16")]; tensor var_6101_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_489_cast_fp16)[name = tensor("op_6101_cast_fp16")]; tensor var_6102_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_491_cast_fp16)[name = tensor("op_6102_cast_fp16")]; tensor var_6103_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_493_cast_fp16)[name = tensor("op_6103_cast_fp16")]; tensor var_6104_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_495_cast_fp16)[name = tensor("op_6104_cast_fp16")]; tensor var_6105_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_497_cast_fp16)[name = tensor("op_6105_cast_fp16")]; tensor var_6106_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_499_cast_fp16)[name = tensor("op_6106_cast_fp16")]; tensor var_6107_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_501_cast_fp16)[name = tensor("op_6107_cast_fp16")]; tensor var_6108_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_503_cast_fp16)[name = tensor("op_6108_cast_fp16")]; tensor var_6109_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_505_cast_fp16)[name = tensor("op_6109_cast_fp16")]; tensor var_6110_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_507_cast_fp16)[name = tensor("op_6110_cast_fp16")]; tensor var_6111_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_509_cast_fp16)[name = tensor("op_6111_cast_fp16")]; tensor var_6112_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_511_cast_fp16)[name = tensor("op_6112_cast_fp16")]; tensor var_6113_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_513_cast_fp16)[name = tensor("op_6113_cast_fp16")]; tensor var_6114_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_515_cast_fp16)[name = tensor("op_6114_cast_fp16")]; tensor var_6115_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_517_cast_fp16)[name = tensor("op_6115_cast_fp16")]; tensor var_6116_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_519_cast_fp16)[name = tensor("op_6116_cast_fp16")]; tensor var_6117_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_521_cast_fp16)[name = tensor("op_6117_cast_fp16")]; tensor var_6118_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_523_cast_fp16)[name = tensor("op_6118_cast_fp16")]; tensor var_6119_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_525_cast_fp16)[name = tensor("op_6119_cast_fp16")]; tensor var_6120_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_527_cast_fp16)[name = tensor("op_6120_cast_fp16")]; tensor var_6121_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_529_cast_fp16)[name = tensor("op_6121_cast_fp16")]; tensor var_6122_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_531_cast_fp16)[name = tensor("op_6122_cast_fp16")]; tensor var_6123_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_533_cast_fp16)[name = tensor("op_6123_cast_fp16")]; tensor var_6124_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_535_cast_fp16)[name = tensor("op_6124_cast_fp16")]; tensor var_6125_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_537_cast_fp16)[name = tensor("op_6125_cast_fp16")]; tensor var_6126_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_539_cast_fp16)[name = tensor("op_6126_cast_fp16")]; tensor var_6127_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_541_cast_fp16)[name = tensor("op_6127_cast_fp16")]; tensor var_6128_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_543_cast_fp16)[name = tensor("op_6128_cast_fp16")]; tensor var_6129_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_545_cast_fp16)[name = tensor("op_6129_cast_fp16")]; tensor var_6130_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_547_cast_fp16)[name = tensor("op_6130_cast_fp16")]; tensor var_6131_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_549_cast_fp16)[name = tensor("op_6131_cast_fp16")]; tensor var_6132_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_551_cast_fp16)[name = tensor("op_6132_cast_fp16")]; tensor var_6133_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_553_cast_fp16)[name = tensor("op_6133_cast_fp16")]; tensor var_6134_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_555_cast_fp16)[name = tensor("op_6134_cast_fp16")]; tensor var_6135_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_557_cast_fp16)[name = tensor("op_6135_cast_fp16")]; tensor var_6136_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_559_cast_fp16)[name = tensor("op_6136_cast_fp16")]; tensor var_6137_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_561_cast_fp16)[name = tensor("op_6137_cast_fp16")]; tensor var_6138_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_563_cast_fp16)[name = tensor("op_6138_cast_fp16")]; tensor var_6139_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_565_cast_fp16)[name = tensor("op_6139_cast_fp16")]; tensor var_6140_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_567_cast_fp16)[name = tensor("op_6140_cast_fp16")]; tensor var_6141_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_569_cast_fp16)[name = tensor("op_6141_cast_fp16")]; tensor var_6142_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_571_cast_fp16)[name = tensor("op_6142_cast_fp16")]; tensor var_6143_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_573_cast_fp16)[name = tensor("op_6143_cast_fp16")]; tensor var_6144_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_575_cast_fp16)[name = tensor("op_6144_cast_fp16")]; tensor var_6146_equation_0 = const()[name = tensor("op_6146_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6146_cast_fp16 = einsum(equation = var_6146_equation_0, values = (var_5794_cast_fp16, var_6081_cast_fp16))[name = tensor("op_6146_cast_fp16")]; tensor var_6148_equation_0 = const()[name = tensor("op_6148_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6148_cast_fp16 = einsum(equation = var_6148_equation_0, values = (var_5794_cast_fp16, var_6082_cast_fp16))[name = tensor("op_6148_cast_fp16")]; tensor var_6150_equation_0 = const()[name = tensor("op_6150_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6150_cast_fp16 = einsum(equation = var_6150_equation_0, values = (var_5794_cast_fp16, var_6083_cast_fp16))[name = tensor("op_6150_cast_fp16")]; tensor var_6152_equation_0 = const()[name = tensor("op_6152_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6152_cast_fp16 = einsum(equation = var_6152_equation_0, values = (var_5794_cast_fp16, var_6084_cast_fp16))[name = tensor("op_6152_cast_fp16")]; tensor var_6154_equation_0 = const()[name = tensor("op_6154_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6154_cast_fp16 = einsum(equation = var_6154_equation_0, values = (var_5794_cast_fp16, var_6085_cast_fp16))[name = tensor("op_6154_cast_fp16")]; tensor var_6156_equation_0 = const()[name = tensor("op_6156_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6156_cast_fp16 = einsum(equation = var_6156_equation_0, values = (var_5794_cast_fp16, var_6086_cast_fp16))[name = tensor("op_6156_cast_fp16")]; tensor var_6158_equation_0 = const()[name = tensor("op_6158_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6158_cast_fp16 = einsum(equation = var_6158_equation_0, values = (var_5794_cast_fp16, var_6087_cast_fp16))[name = tensor("op_6158_cast_fp16")]; tensor var_6160_equation_0 = const()[name = tensor("op_6160_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6160_cast_fp16 = einsum(equation = var_6160_equation_0, values = (var_5794_cast_fp16, var_6088_cast_fp16))[name = tensor("op_6160_cast_fp16")]; tensor var_6162_equation_0 = const()[name = tensor("op_6162_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6162_cast_fp16 = einsum(equation = var_6162_equation_0, values = (var_5798_cast_fp16, var_6089_cast_fp16))[name = tensor("op_6162_cast_fp16")]; tensor var_6164_equation_0 = const()[name = tensor("op_6164_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6164_cast_fp16 = einsum(equation = var_6164_equation_0, values = (var_5798_cast_fp16, var_6090_cast_fp16))[name = tensor("op_6164_cast_fp16")]; tensor var_6166_equation_0 = const()[name = tensor("op_6166_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6166_cast_fp16 = einsum(equation = var_6166_equation_0, values = (var_5798_cast_fp16, var_6091_cast_fp16))[name = tensor("op_6166_cast_fp16")]; tensor var_6168_equation_0 = const()[name = tensor("op_6168_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6168_cast_fp16 = einsum(equation = var_6168_equation_0, values = (var_5798_cast_fp16, var_6092_cast_fp16))[name = tensor("op_6168_cast_fp16")]; tensor var_6170_equation_0 = const()[name = tensor("op_6170_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6170_cast_fp16 = einsum(equation = var_6170_equation_0, values = (var_5798_cast_fp16, var_6093_cast_fp16))[name = tensor("op_6170_cast_fp16")]; tensor var_6172_equation_0 = const()[name = tensor("op_6172_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6172_cast_fp16 = einsum(equation = var_6172_equation_0, values = (var_5798_cast_fp16, var_6094_cast_fp16))[name = tensor("op_6172_cast_fp16")]; tensor var_6174_equation_0 = const()[name = tensor("op_6174_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6174_cast_fp16 = einsum(equation = var_6174_equation_0, values = (var_5798_cast_fp16, var_6095_cast_fp16))[name = tensor("op_6174_cast_fp16")]; tensor var_6176_equation_0 = const()[name = tensor("op_6176_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6176_cast_fp16 = einsum(equation = var_6176_equation_0, values = (var_5798_cast_fp16, var_6096_cast_fp16))[name = tensor("op_6176_cast_fp16")]; tensor var_6178_equation_0 = const()[name = tensor("op_6178_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6178_cast_fp16 = einsum(equation = var_6178_equation_0, values = (var_5802_cast_fp16, var_6097_cast_fp16))[name = tensor("op_6178_cast_fp16")]; tensor var_6180_equation_0 = const()[name = tensor("op_6180_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6180_cast_fp16 = einsum(equation = var_6180_equation_0, values = (var_5802_cast_fp16, var_6098_cast_fp16))[name = tensor("op_6180_cast_fp16")]; tensor var_6182_equation_0 = const()[name = tensor("op_6182_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6182_cast_fp16 = einsum(equation = var_6182_equation_0, values = (var_5802_cast_fp16, var_6099_cast_fp16))[name = tensor("op_6182_cast_fp16")]; tensor var_6184_equation_0 = const()[name = tensor("op_6184_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6184_cast_fp16 = einsum(equation = var_6184_equation_0, values = (var_5802_cast_fp16, var_6100_cast_fp16))[name = tensor("op_6184_cast_fp16")]; tensor var_6186_equation_0 = const()[name = tensor("op_6186_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6186_cast_fp16 = einsum(equation = var_6186_equation_0, values = (var_5802_cast_fp16, var_6101_cast_fp16))[name = tensor("op_6186_cast_fp16")]; tensor var_6188_equation_0 = const()[name = tensor("op_6188_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6188_cast_fp16 = einsum(equation = var_6188_equation_0, values = (var_5802_cast_fp16, var_6102_cast_fp16))[name = tensor("op_6188_cast_fp16")]; tensor var_6190_equation_0 = const()[name = tensor("op_6190_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6190_cast_fp16 = einsum(equation = var_6190_equation_0, values = (var_5802_cast_fp16, var_6103_cast_fp16))[name = tensor("op_6190_cast_fp16")]; tensor var_6192_equation_0 = const()[name = tensor("op_6192_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6192_cast_fp16 = einsum(equation = var_6192_equation_0, values = (var_5802_cast_fp16, var_6104_cast_fp16))[name = tensor("op_6192_cast_fp16")]; tensor var_6194_equation_0 = const()[name = tensor("op_6194_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6194_cast_fp16 = einsum(equation = var_6194_equation_0, values = (var_5806_cast_fp16, var_6105_cast_fp16))[name = tensor("op_6194_cast_fp16")]; tensor var_6196_equation_0 = const()[name = tensor("op_6196_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6196_cast_fp16 = einsum(equation = var_6196_equation_0, values = (var_5806_cast_fp16, var_6106_cast_fp16))[name = tensor("op_6196_cast_fp16")]; tensor var_6198_equation_0 = const()[name = tensor("op_6198_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6198_cast_fp16 = einsum(equation = var_6198_equation_0, values = (var_5806_cast_fp16, var_6107_cast_fp16))[name = tensor("op_6198_cast_fp16")]; tensor var_6200_equation_0 = const()[name = tensor("op_6200_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6200_cast_fp16 = einsum(equation = var_6200_equation_0, values = (var_5806_cast_fp16, var_6108_cast_fp16))[name = tensor("op_6200_cast_fp16")]; tensor var_6202_equation_0 = const()[name = tensor("op_6202_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6202_cast_fp16 = einsum(equation = var_6202_equation_0, values = (var_5806_cast_fp16, var_6109_cast_fp16))[name = tensor("op_6202_cast_fp16")]; tensor var_6204_equation_0 = const()[name = tensor("op_6204_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6204_cast_fp16 = einsum(equation = var_6204_equation_0, values = (var_5806_cast_fp16, var_6110_cast_fp16))[name = tensor("op_6204_cast_fp16")]; tensor var_6206_equation_0 = const()[name = tensor("op_6206_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6206_cast_fp16 = einsum(equation = var_6206_equation_0, values = (var_5806_cast_fp16, var_6111_cast_fp16))[name = tensor("op_6206_cast_fp16")]; tensor var_6208_equation_0 = const()[name = tensor("op_6208_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6208_cast_fp16 = einsum(equation = var_6208_equation_0, values = (var_5806_cast_fp16, var_6112_cast_fp16))[name = tensor("op_6208_cast_fp16")]; tensor var_6210_equation_0 = const()[name = tensor("op_6210_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6210_cast_fp16 = einsum(equation = var_6210_equation_0, values = (var_5810_cast_fp16, var_6113_cast_fp16))[name = tensor("op_6210_cast_fp16")]; tensor var_6212_equation_0 = const()[name = tensor("op_6212_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6212_cast_fp16 = einsum(equation = var_6212_equation_0, values = (var_5810_cast_fp16, var_6114_cast_fp16))[name = tensor("op_6212_cast_fp16")]; tensor var_6214_equation_0 = const()[name = tensor("op_6214_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6214_cast_fp16 = einsum(equation = var_6214_equation_0, values = (var_5810_cast_fp16, var_6115_cast_fp16))[name = tensor("op_6214_cast_fp16")]; tensor var_6216_equation_0 = const()[name = tensor("op_6216_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6216_cast_fp16 = einsum(equation = var_6216_equation_0, values = (var_5810_cast_fp16, var_6116_cast_fp16))[name = tensor("op_6216_cast_fp16")]; tensor var_6218_equation_0 = const()[name = tensor("op_6218_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6218_cast_fp16 = einsum(equation = var_6218_equation_0, values = (var_5810_cast_fp16, var_6117_cast_fp16))[name = tensor("op_6218_cast_fp16")]; tensor var_6220_equation_0 = const()[name = tensor("op_6220_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6220_cast_fp16 = einsum(equation = var_6220_equation_0, values = (var_5810_cast_fp16, var_6118_cast_fp16))[name = tensor("op_6220_cast_fp16")]; tensor var_6222_equation_0 = const()[name = tensor("op_6222_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6222_cast_fp16 = einsum(equation = var_6222_equation_0, values = (var_5810_cast_fp16, var_6119_cast_fp16))[name = tensor("op_6222_cast_fp16")]; tensor var_6224_equation_0 = const()[name = tensor("op_6224_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6224_cast_fp16 = einsum(equation = var_6224_equation_0, values = (var_5810_cast_fp16, var_6120_cast_fp16))[name = tensor("op_6224_cast_fp16")]; tensor var_6226_equation_0 = const()[name = tensor("op_6226_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6226_cast_fp16 = einsum(equation = var_6226_equation_0, values = (var_5814_cast_fp16, var_6121_cast_fp16))[name = tensor("op_6226_cast_fp16")]; tensor var_6228_equation_0 = const()[name = tensor("op_6228_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6228_cast_fp16 = einsum(equation = var_6228_equation_0, values = (var_5814_cast_fp16, var_6122_cast_fp16))[name = tensor("op_6228_cast_fp16")]; tensor var_6230_equation_0 = const()[name = tensor("op_6230_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6230_cast_fp16 = einsum(equation = var_6230_equation_0, values = (var_5814_cast_fp16, var_6123_cast_fp16))[name = tensor("op_6230_cast_fp16")]; tensor var_6232_equation_0 = const()[name = tensor("op_6232_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6232_cast_fp16 = einsum(equation = var_6232_equation_0, values = (var_5814_cast_fp16, var_6124_cast_fp16))[name = tensor("op_6232_cast_fp16")]; tensor var_6234_equation_0 = const()[name = tensor("op_6234_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6234_cast_fp16 = einsum(equation = var_6234_equation_0, values = (var_5814_cast_fp16, var_6125_cast_fp16))[name = tensor("op_6234_cast_fp16")]; tensor var_6236_equation_0 = const()[name = tensor("op_6236_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6236_cast_fp16 = einsum(equation = var_6236_equation_0, values = (var_5814_cast_fp16, var_6126_cast_fp16))[name = tensor("op_6236_cast_fp16")]; tensor var_6238_equation_0 = const()[name = tensor("op_6238_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6238_cast_fp16 = einsum(equation = var_6238_equation_0, values = (var_5814_cast_fp16, var_6127_cast_fp16))[name = tensor("op_6238_cast_fp16")]; tensor var_6240_equation_0 = const()[name = tensor("op_6240_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6240_cast_fp16 = einsum(equation = var_6240_equation_0, values = (var_5814_cast_fp16, var_6128_cast_fp16))[name = tensor("op_6240_cast_fp16")]; tensor var_6242_equation_0 = const()[name = tensor("op_6242_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6242_cast_fp16 = einsum(equation = var_6242_equation_0, values = (var_5818_cast_fp16, var_6129_cast_fp16))[name = tensor("op_6242_cast_fp16")]; tensor var_6244_equation_0 = const()[name = tensor("op_6244_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6244_cast_fp16 = einsum(equation = var_6244_equation_0, values = (var_5818_cast_fp16, var_6130_cast_fp16))[name = tensor("op_6244_cast_fp16")]; tensor var_6246_equation_0 = const()[name = tensor("op_6246_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6246_cast_fp16 = einsum(equation = var_6246_equation_0, values = (var_5818_cast_fp16, var_6131_cast_fp16))[name = tensor("op_6246_cast_fp16")]; tensor var_6248_equation_0 = const()[name = tensor("op_6248_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6248_cast_fp16 = einsum(equation = var_6248_equation_0, values = (var_5818_cast_fp16, var_6132_cast_fp16))[name = tensor("op_6248_cast_fp16")]; tensor var_6250_equation_0 = const()[name = tensor("op_6250_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6250_cast_fp16 = einsum(equation = var_6250_equation_0, values = (var_5818_cast_fp16, var_6133_cast_fp16))[name = tensor("op_6250_cast_fp16")]; tensor var_6252_equation_0 = const()[name = tensor("op_6252_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6252_cast_fp16 = einsum(equation = var_6252_equation_0, values = (var_5818_cast_fp16, var_6134_cast_fp16))[name = tensor("op_6252_cast_fp16")]; tensor var_6254_equation_0 = const()[name = tensor("op_6254_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6254_cast_fp16 = einsum(equation = var_6254_equation_0, values = (var_5818_cast_fp16, var_6135_cast_fp16))[name = tensor("op_6254_cast_fp16")]; tensor var_6256_equation_0 = const()[name = tensor("op_6256_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6256_cast_fp16 = einsum(equation = var_6256_equation_0, values = (var_5818_cast_fp16, var_6136_cast_fp16))[name = tensor("op_6256_cast_fp16")]; tensor var_6258_equation_0 = const()[name = tensor("op_6258_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6258_cast_fp16 = einsum(equation = var_6258_equation_0, values = (var_5822_cast_fp16, var_6137_cast_fp16))[name = tensor("op_6258_cast_fp16")]; tensor var_6260_equation_0 = const()[name = tensor("op_6260_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6260_cast_fp16 = einsum(equation = var_6260_equation_0, values = (var_5822_cast_fp16, var_6138_cast_fp16))[name = tensor("op_6260_cast_fp16")]; tensor var_6262_equation_0 = const()[name = tensor("op_6262_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6262_cast_fp16 = einsum(equation = var_6262_equation_0, values = (var_5822_cast_fp16, var_6139_cast_fp16))[name = tensor("op_6262_cast_fp16")]; tensor var_6264_equation_0 = const()[name = tensor("op_6264_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6264_cast_fp16 = einsum(equation = var_6264_equation_0, values = (var_5822_cast_fp16, var_6140_cast_fp16))[name = tensor("op_6264_cast_fp16")]; tensor var_6266_equation_0 = const()[name = tensor("op_6266_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6266_cast_fp16 = einsum(equation = var_6266_equation_0, values = (var_5822_cast_fp16, var_6141_cast_fp16))[name = tensor("op_6266_cast_fp16")]; tensor var_6268_equation_0 = const()[name = tensor("op_6268_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6268_cast_fp16 = einsum(equation = var_6268_equation_0, values = (var_5822_cast_fp16, var_6142_cast_fp16))[name = tensor("op_6268_cast_fp16")]; tensor var_6270_equation_0 = const()[name = tensor("op_6270_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6270_cast_fp16 = einsum(equation = var_6270_equation_0, values = (var_5822_cast_fp16, var_6143_cast_fp16))[name = tensor("op_6270_cast_fp16")]; tensor var_6272_equation_0 = const()[name = tensor("op_6272_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6272_cast_fp16 = einsum(equation = var_6272_equation_0, values = (var_5822_cast_fp16, var_6144_cast_fp16))[name = tensor("op_6272_cast_fp16")]; tensor var_6274_interleave_0 = const()[name = tensor("op_6274_interleave_0"), val = tensor(false)]; tensor var_6274_cast_fp16 = concat(axis = var_5509, interleave = var_6274_interleave_0, values = (var_6146_cast_fp16, var_6148_cast_fp16, var_6150_cast_fp16, var_6152_cast_fp16, var_6154_cast_fp16, var_6156_cast_fp16, var_6158_cast_fp16, var_6160_cast_fp16))[name = tensor("op_6274_cast_fp16")]; tensor var_6276_interleave_0 = const()[name = tensor("op_6276_interleave_0"), val = tensor(false)]; tensor var_6276_cast_fp16 = concat(axis = var_5509, interleave = var_6276_interleave_0, values = (var_6162_cast_fp16, var_6164_cast_fp16, var_6166_cast_fp16, var_6168_cast_fp16, var_6170_cast_fp16, var_6172_cast_fp16, var_6174_cast_fp16, var_6176_cast_fp16))[name = tensor("op_6276_cast_fp16")]; tensor var_6278_interleave_0 = const()[name = tensor("op_6278_interleave_0"), val = tensor(false)]; tensor var_6278_cast_fp16 = concat(axis = var_5509, interleave = var_6278_interleave_0, values = (var_6178_cast_fp16, var_6180_cast_fp16, var_6182_cast_fp16, var_6184_cast_fp16, var_6186_cast_fp16, var_6188_cast_fp16, var_6190_cast_fp16, var_6192_cast_fp16))[name = tensor("op_6278_cast_fp16")]; tensor var_6280_interleave_0 = const()[name = tensor("op_6280_interleave_0"), val = tensor(false)]; tensor var_6280_cast_fp16 = concat(axis = var_5509, interleave = var_6280_interleave_0, values = (var_6194_cast_fp16, var_6196_cast_fp16, var_6198_cast_fp16, var_6200_cast_fp16, var_6202_cast_fp16, var_6204_cast_fp16, var_6206_cast_fp16, var_6208_cast_fp16))[name = tensor("op_6280_cast_fp16")]; tensor var_6282_interleave_0 = const()[name = tensor("op_6282_interleave_0"), val = tensor(false)]; tensor var_6282_cast_fp16 = concat(axis = var_5509, interleave = var_6282_interleave_0, values = (var_6210_cast_fp16, var_6212_cast_fp16, var_6214_cast_fp16, var_6216_cast_fp16, var_6218_cast_fp16, var_6220_cast_fp16, var_6222_cast_fp16, var_6224_cast_fp16))[name = tensor("op_6282_cast_fp16")]; tensor var_6284_interleave_0 = const()[name = tensor("op_6284_interleave_0"), val = tensor(false)]; tensor var_6284_cast_fp16 = concat(axis = var_5509, interleave = var_6284_interleave_0, values = (var_6226_cast_fp16, var_6228_cast_fp16, var_6230_cast_fp16, var_6232_cast_fp16, var_6234_cast_fp16, var_6236_cast_fp16, var_6238_cast_fp16, var_6240_cast_fp16))[name = tensor("op_6284_cast_fp16")]; tensor var_6286_interleave_0 = const()[name = tensor("op_6286_interleave_0"), val = tensor(false)]; tensor var_6286_cast_fp16 = concat(axis = var_5509, interleave = var_6286_interleave_0, values = (var_6242_cast_fp16, var_6244_cast_fp16, var_6246_cast_fp16, var_6248_cast_fp16, var_6250_cast_fp16, var_6252_cast_fp16, var_6254_cast_fp16, var_6256_cast_fp16))[name = tensor("op_6286_cast_fp16")]; tensor var_6288_interleave_0 = const()[name = tensor("op_6288_interleave_0"), val = tensor(false)]; tensor var_6288_cast_fp16 = concat(axis = var_5509, interleave = var_6288_interleave_0, values = (var_6258_cast_fp16, var_6260_cast_fp16, var_6262_cast_fp16, var_6264_cast_fp16, var_6266_cast_fp16, var_6268_cast_fp16, var_6270_cast_fp16, var_6272_cast_fp16))[name = tensor("op_6288_cast_fp16")]; tensor input_219_interleave_0 = const()[name = tensor("input_219_interleave_0"), val = tensor(false)]; tensor input_219_cast_fp16 = concat(axis = var_5537, interleave = input_219_interleave_0, values = (var_6274_cast_fp16, var_6276_cast_fp16, var_6278_cast_fp16, var_6280_cast_fp16, var_6282_cast_fp16, var_6284_cast_fp16, var_6286_cast_fp16, var_6288_cast_fp16))[name = tensor("input_219_cast_fp16")]; tensor var_6294 = const()[name = tensor("op_6294"), val = tensor([1, 1])]; tensor var_6296 = const()[name = tensor("op_6296"), val = tensor([1, 1])]; tensor var_6298_pad_type_0 = const()[name = tensor("op_6298_pad_type_0"), val = tensor("custom")]; tensor var_6298_pad_0 = const()[name = tensor("op_6298_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(236726016))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(236802880))), name = tensor("up_blocks_2_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16_palettized"), shape = tensor([320, 320, 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(236803072)))]; tensor var_6298_cast_fp16 = conv(bias = up_blocks_2_attentions_0_transformer_blocks_0_attn1_to_out_0_bias_to_fp16, dilations = var_6296, groups = var_5537, pad = var_6298_pad_0, pad_type = var_6298_pad_type_0, strides = var_6294, weight = up_blocks_2_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16_palettized, x = input_219_cast_fp16)[name = tensor("op_6298_cast_fp16")]; tensor inputs_45_cast_fp16 = add(x = var_6298_cast_fp16, y = inputs_43_cast_fp16)[name = tensor("inputs_45_cast_fp16")]; tensor var_6302 = const()[name = tensor("op_6302"), val = tensor([1])]; tensor channels_mean_45_cast_fp16 = reduce_mean(axes = var_6302, keep_dims = var_5532, x = inputs_45_cast_fp16)[name = tensor("channels_mean_45_cast_fp16")]; tensor zero_mean_45_cast_fp16 = sub(x = inputs_45_cast_fp16, y = channels_mean_45_cast_fp16)[name = tensor("zero_mean_45_cast_fp16")]; tensor zero_mean_sq_45_cast_fp16 = mul(x = zero_mean_45_cast_fp16, y = zero_mean_45_cast_fp16)[name = tensor("zero_mean_sq_45_cast_fp16")]; tensor var_6306 = const()[name = tensor("op_6306"), val = tensor([1])]; tensor var_6307_cast_fp16 = reduce_mean(axes = var_6306, keep_dims = var_5532, x = zero_mean_sq_45_cast_fp16)[name = tensor("op_6307_cast_fp16")]; tensor var_6308_to_fp16 = const()[name = tensor("op_6308_to_fp16"), val = tensor(0x1.5p-17)]; tensor var_6309_cast_fp16 = add(x = var_6307_cast_fp16, y = var_6308_to_fp16)[name = tensor("op_6309_cast_fp16")]; tensor denom_45_epsilon_0_to_fp16 = const()[name = tensor("denom_45_epsilon_0_to_fp16"), val = tensor(0x1p-24)]; tensor denom_45_cast_fp16 = rsqrt(epsilon = denom_45_epsilon_0_to_fp16, x = var_6309_cast_fp16)[name = tensor("denom_45_cast_fp16")]; tensor out_45_cast_fp16 = mul(x = zero_mean_45_cast_fp16, y = denom_45_cast_fp16)[name = tensor("out_45_cast_fp16")]; tensor var_6313_to_fp16 = const()[name = tensor("op_6313_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(236803776)))]; tensor var_6314_cast_fp16 = add(x = out_45_cast_fp16, y = var_6313_to_fp16)[name = tensor("op_6314_cast_fp16")]; tensor var_6316_to_fp16 = const()[name = tensor("op_6316_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(236804480)))]; tensor hidden_states_147_cast_fp16 = mul(x = var_6314_cast_fp16, y = var_6316_to_fp16)[name = tensor("hidden_states_147_cast_fp16")]; tensor var_6323 = const()[name = tensor("op_6323"), val = tensor([1, 1])]; tensor var_6325 = const()[name = tensor("op_6325"), 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_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(236805184))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(236882048))), name = tensor("up_blocks_2_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16_palettized"), shape = tensor([320, 320, 1, 1])]; tensor q_31_cast_fp16 = conv(dilations = var_6325, groups = var_5537, pad = q_31_pad_0, pad_type = q_31_pad_type_0, strides = var_6323, weight = up_blocks_2_attentions_0_transformer_blocks_0_attn2_to_q_weight_to_fp16_palettized, x = hidden_states_147_cast_fp16)[name = tensor("q_31_cast_fp16")]; tensor var_6329 = const()[name = tensor("op_6329"), val = tensor([1, 1])]; tensor var_6331 = const()[name = tensor("op_6331"), 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_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(236882240))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(237066624))), name = tensor("up_blocks_2_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized"), shape = tensor([320, 768, 1, 1])]; tensor k_61_cast_fp16 = conv(dilations = var_6331, groups = var_5537, pad = k_61_pad_0, pad_type = k_61_pad_type_0, strides = var_6329, weight = up_blocks_2_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor("k_61_cast_fp16")]; tensor var_6335 = const()[name = tensor("op_6335"), val = tensor([1, 1])]; tensor var_6337 = const()[name = tensor("op_6337"), 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_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(237066816))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(237251200))), name = tensor("up_blocks_2_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized"), shape = tensor([320, 768, 1, 1])]; tensor v_31_cast_fp16 = conv(dilations = var_6337, groups = var_5537, pad = v_31_pad_0, pad_type = v_31_pad_type_0, strides = var_6335, weight = up_blocks_2_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor("v_31_cast_fp16")]; tensor var_6341_begin_0 = const()[name = tensor("op_6341_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_6341_end_0 = const()[name = tensor("op_6341_end_0"), val = tensor([2, 40, 1, 4096])]; tensor var_6341_end_mask_0 = const()[name = tensor("op_6341_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6341_cast_fp16 = slice_by_index(begin = var_6341_begin_0, end = var_6341_end_0, end_mask = var_6341_end_mask_0, x = q_31_cast_fp16)[name = tensor("op_6341_cast_fp16")]; tensor var_6345_begin_0 = const()[name = tensor("op_6345_begin_0"), val = tensor([0, 40, 0, 0])]; tensor var_6345_end_0 = const()[name = tensor("op_6345_end_0"), val = tensor([2, 80, 1, 4096])]; tensor var_6345_end_mask_0 = const()[name = tensor("op_6345_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6345_cast_fp16 = slice_by_index(begin = var_6345_begin_0, end = var_6345_end_0, end_mask = var_6345_end_mask_0, x = q_31_cast_fp16)[name = tensor("op_6345_cast_fp16")]; tensor var_6349_begin_0 = const()[name = tensor("op_6349_begin_0"), val = tensor([0, 80, 0, 0])]; tensor var_6349_end_0 = const()[name = tensor("op_6349_end_0"), val = tensor([2, 120, 1, 4096])]; tensor var_6349_end_mask_0 = const()[name = tensor("op_6349_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6349_cast_fp16 = slice_by_index(begin = var_6349_begin_0, end = var_6349_end_0, end_mask = var_6349_end_mask_0, x = q_31_cast_fp16)[name = tensor("op_6349_cast_fp16")]; tensor var_6353_begin_0 = const()[name = tensor("op_6353_begin_0"), val = tensor([0, 120, 0, 0])]; tensor var_6353_end_0 = const()[name = tensor("op_6353_end_0"), val = tensor([2, 160, 1, 4096])]; tensor var_6353_end_mask_0 = const()[name = tensor("op_6353_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6353_cast_fp16 = slice_by_index(begin = var_6353_begin_0, end = var_6353_end_0, end_mask = var_6353_end_mask_0, x = q_31_cast_fp16)[name = tensor("op_6353_cast_fp16")]; tensor var_6357_begin_0 = const()[name = tensor("op_6357_begin_0"), val = tensor([0, 160, 0, 0])]; tensor var_6357_end_0 = const()[name = tensor("op_6357_end_0"), val = tensor([2, 200, 1, 4096])]; tensor var_6357_end_mask_0 = const()[name = tensor("op_6357_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6357_cast_fp16 = slice_by_index(begin = var_6357_begin_0, end = var_6357_end_0, end_mask = var_6357_end_mask_0, x = q_31_cast_fp16)[name = tensor("op_6357_cast_fp16")]; tensor var_6361_begin_0 = const()[name = tensor("op_6361_begin_0"), val = tensor([0, 200, 0, 0])]; tensor var_6361_end_0 = const()[name = tensor("op_6361_end_0"), val = tensor([2, 240, 1, 4096])]; tensor var_6361_end_mask_0 = const()[name = tensor("op_6361_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6361_cast_fp16 = slice_by_index(begin = var_6361_begin_0, end = var_6361_end_0, end_mask = var_6361_end_mask_0, x = q_31_cast_fp16)[name = tensor("op_6361_cast_fp16")]; tensor var_6365_begin_0 = const()[name = tensor("op_6365_begin_0"), val = tensor([0, 240, 0, 0])]; tensor var_6365_end_0 = const()[name = tensor("op_6365_end_0"), val = tensor([2, 280, 1, 4096])]; tensor var_6365_end_mask_0 = const()[name = tensor("op_6365_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6365_cast_fp16 = slice_by_index(begin = var_6365_begin_0, end = var_6365_end_0, end_mask = var_6365_end_mask_0, x = q_31_cast_fp16)[name = tensor("op_6365_cast_fp16")]; tensor var_6369_begin_0 = const()[name = tensor("op_6369_begin_0"), val = tensor([0, 280, 0, 0])]; tensor var_6369_end_0 = const()[name = tensor("op_6369_end_0"), val = tensor([2, 320, 1, 4096])]; tensor var_6369_end_mask_0 = const()[name = tensor("op_6369_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6369_cast_fp16 = slice_by_index(begin = var_6369_begin_0, end = var_6369_end_0, end_mask = var_6369_end_mask_0, x = q_31_cast_fp16)[name = tensor("op_6369_cast_fp16")]; tensor var_6372_begin_0 = const()[name = tensor("op_6372_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_6372_end_0 = const()[name = tensor("op_6372_end_0"), val = tensor([2, 40, 1, 512])]; tensor var_6372_end_mask_0 = const()[name = tensor("op_6372_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6372_cast_fp16 = slice_by_index(begin = var_6372_begin_0, end = var_6372_end_0, end_mask = var_6372_end_mask_0, x = var_6341_cast_fp16)[name = tensor("op_6372_cast_fp16")]; tensor var_6373_begin_0 = const()[name = tensor("op_6373_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_6373_end_0 = const()[name = tensor("op_6373_end_0"), val = tensor([2, 40, 1, 1024])]; tensor var_6373_end_mask_0 = const()[name = tensor("op_6373_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6373_cast_fp16 = slice_by_index(begin = var_6373_begin_0, end = var_6373_end_0, end_mask = var_6373_end_mask_0, x = var_6341_cast_fp16)[name = tensor("op_6373_cast_fp16")]; tensor var_6374_begin_0 = const()[name = tensor("op_6374_begin_0"), val = tensor([0, 0, 0, 1024])]; tensor var_6374_end_0 = const()[name = tensor("op_6374_end_0"), val = tensor([2, 40, 1, 1536])]; tensor var_6374_end_mask_0 = const()[name = tensor("op_6374_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6374_cast_fp16 = slice_by_index(begin = var_6374_begin_0, end = var_6374_end_0, end_mask = var_6374_end_mask_0, x = var_6341_cast_fp16)[name = tensor("op_6374_cast_fp16")]; tensor var_6375_begin_0 = const()[name = tensor("op_6375_begin_0"), val = tensor([0, 0, 0, 1536])]; tensor var_6375_end_0 = const()[name = tensor("op_6375_end_0"), val = tensor([2, 40, 1, 2048])]; tensor var_6375_end_mask_0 = const()[name = tensor("op_6375_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6375_cast_fp16 = slice_by_index(begin = var_6375_begin_0, end = var_6375_end_0, end_mask = var_6375_end_mask_0, x = var_6341_cast_fp16)[name = tensor("op_6375_cast_fp16")]; tensor var_6376_begin_0 = const()[name = tensor("op_6376_begin_0"), val = tensor([0, 0, 0, 2048])]; tensor var_6376_end_0 = const()[name = tensor("op_6376_end_0"), val = tensor([2, 40, 1, 2560])]; tensor var_6376_end_mask_0 = const()[name = tensor("op_6376_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6376_cast_fp16 = slice_by_index(begin = var_6376_begin_0, end = var_6376_end_0, end_mask = var_6376_end_mask_0, x = var_6341_cast_fp16)[name = tensor("op_6376_cast_fp16")]; tensor var_6377_begin_0 = const()[name = tensor("op_6377_begin_0"), val = tensor([0, 0, 0, 2560])]; tensor var_6377_end_0 = const()[name = tensor("op_6377_end_0"), val = tensor([2, 40, 1, 3072])]; tensor var_6377_end_mask_0 = const()[name = tensor("op_6377_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6377_cast_fp16 = slice_by_index(begin = var_6377_begin_0, end = var_6377_end_0, end_mask = var_6377_end_mask_0, x = var_6341_cast_fp16)[name = tensor("op_6377_cast_fp16")]; tensor var_6378_begin_0 = const()[name = tensor("op_6378_begin_0"), val = tensor([0, 0, 0, 3072])]; tensor var_6378_end_0 = const()[name = tensor("op_6378_end_0"), val = tensor([2, 40, 1, 3584])]; tensor var_6378_end_mask_0 = const()[name = tensor("op_6378_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6378_cast_fp16 = slice_by_index(begin = var_6378_begin_0, end = var_6378_end_0, end_mask = var_6378_end_mask_0, x = var_6341_cast_fp16)[name = tensor("op_6378_cast_fp16")]; tensor var_6379_begin_0 = const()[name = tensor("op_6379_begin_0"), val = tensor([0, 0, 0, 3584])]; tensor var_6379_end_0 = const()[name = tensor("op_6379_end_0"), val = tensor([2, 40, 1, 4096])]; tensor var_6379_end_mask_0 = const()[name = tensor("op_6379_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6379_cast_fp16 = slice_by_index(begin = var_6379_begin_0, end = var_6379_end_0, end_mask = var_6379_end_mask_0, x = var_6341_cast_fp16)[name = tensor("op_6379_cast_fp16")]; tensor var_6380_begin_0 = const()[name = tensor("op_6380_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_6380_end_0 = const()[name = tensor("op_6380_end_0"), val = tensor([2, 40, 1, 512])]; tensor var_6380_end_mask_0 = const()[name = tensor("op_6380_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6380_cast_fp16 = slice_by_index(begin = var_6380_begin_0, end = var_6380_end_0, end_mask = var_6380_end_mask_0, x = var_6345_cast_fp16)[name = tensor("op_6380_cast_fp16")]; tensor var_6381_begin_0 = const()[name = tensor("op_6381_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_6381_end_0 = const()[name = tensor("op_6381_end_0"), val = tensor([2, 40, 1, 1024])]; tensor var_6381_end_mask_0 = const()[name = tensor("op_6381_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6381_cast_fp16 = slice_by_index(begin = var_6381_begin_0, end = var_6381_end_0, end_mask = var_6381_end_mask_0, x = var_6345_cast_fp16)[name = tensor("op_6381_cast_fp16")]; tensor var_6382_begin_0 = const()[name = tensor("op_6382_begin_0"), val = tensor([0, 0, 0, 1024])]; tensor var_6382_end_0 = const()[name = tensor("op_6382_end_0"), val = tensor([2, 40, 1, 1536])]; tensor var_6382_end_mask_0 = const()[name = tensor("op_6382_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6382_cast_fp16 = slice_by_index(begin = var_6382_begin_0, end = var_6382_end_0, end_mask = var_6382_end_mask_0, x = var_6345_cast_fp16)[name = tensor("op_6382_cast_fp16")]; tensor var_6383_begin_0 = const()[name = tensor("op_6383_begin_0"), val = tensor([0, 0, 0, 1536])]; tensor var_6383_end_0 = const()[name = tensor("op_6383_end_0"), val = tensor([2, 40, 1, 2048])]; tensor var_6383_end_mask_0 = const()[name = tensor("op_6383_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6383_cast_fp16 = slice_by_index(begin = var_6383_begin_0, end = var_6383_end_0, end_mask = var_6383_end_mask_0, x = var_6345_cast_fp16)[name = tensor("op_6383_cast_fp16")]; tensor var_6384_begin_0 = const()[name = tensor("op_6384_begin_0"), val = tensor([0, 0, 0, 2048])]; tensor var_6384_end_0 = const()[name = tensor("op_6384_end_0"), val = tensor([2, 40, 1, 2560])]; tensor var_6384_end_mask_0 = const()[name = tensor("op_6384_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6384_cast_fp16 = slice_by_index(begin = var_6384_begin_0, end = var_6384_end_0, end_mask = var_6384_end_mask_0, x = var_6345_cast_fp16)[name = tensor("op_6384_cast_fp16")]; tensor var_6385_begin_0 = const()[name = tensor("op_6385_begin_0"), val = tensor([0, 0, 0, 2560])]; tensor var_6385_end_0 = const()[name = tensor("op_6385_end_0"), val = tensor([2, 40, 1, 3072])]; tensor var_6385_end_mask_0 = const()[name = tensor("op_6385_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6385_cast_fp16 = slice_by_index(begin = var_6385_begin_0, end = var_6385_end_0, end_mask = var_6385_end_mask_0, x = var_6345_cast_fp16)[name = tensor("op_6385_cast_fp16")]; tensor var_6386_begin_0 = const()[name = tensor("op_6386_begin_0"), val = tensor([0, 0, 0, 3072])]; tensor var_6386_end_0 = const()[name = tensor("op_6386_end_0"), val = tensor([2, 40, 1, 3584])]; tensor var_6386_end_mask_0 = const()[name = tensor("op_6386_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6386_cast_fp16 = slice_by_index(begin = var_6386_begin_0, end = var_6386_end_0, end_mask = var_6386_end_mask_0, x = var_6345_cast_fp16)[name = tensor("op_6386_cast_fp16")]; tensor var_6387_begin_0 = const()[name = tensor("op_6387_begin_0"), val = tensor([0, 0, 0, 3584])]; tensor var_6387_end_0 = const()[name = tensor("op_6387_end_0"), val = tensor([2, 40, 1, 4096])]; tensor var_6387_end_mask_0 = const()[name = tensor("op_6387_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6387_cast_fp16 = slice_by_index(begin = var_6387_begin_0, end = var_6387_end_0, end_mask = var_6387_end_mask_0, x = var_6345_cast_fp16)[name = tensor("op_6387_cast_fp16")]; tensor var_6388_begin_0 = const()[name = tensor("op_6388_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_6388_end_0 = const()[name = tensor("op_6388_end_0"), val = tensor([2, 40, 1, 512])]; tensor var_6388_end_mask_0 = const()[name = tensor("op_6388_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6388_cast_fp16 = slice_by_index(begin = var_6388_begin_0, end = var_6388_end_0, end_mask = var_6388_end_mask_0, x = var_6349_cast_fp16)[name = tensor("op_6388_cast_fp16")]; tensor var_6389_begin_0 = const()[name = tensor("op_6389_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_6389_end_0 = const()[name = tensor("op_6389_end_0"), val = tensor([2, 40, 1, 1024])]; tensor var_6389_end_mask_0 = const()[name = tensor("op_6389_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6389_cast_fp16 = slice_by_index(begin = var_6389_begin_0, end = var_6389_end_0, end_mask = var_6389_end_mask_0, x = var_6349_cast_fp16)[name = tensor("op_6389_cast_fp16")]; tensor var_6390_begin_0 = const()[name = tensor("op_6390_begin_0"), val = tensor([0, 0, 0, 1024])]; tensor var_6390_end_0 = const()[name = tensor("op_6390_end_0"), val = tensor([2, 40, 1, 1536])]; tensor var_6390_end_mask_0 = const()[name = tensor("op_6390_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6390_cast_fp16 = slice_by_index(begin = var_6390_begin_0, end = var_6390_end_0, end_mask = var_6390_end_mask_0, x = var_6349_cast_fp16)[name = tensor("op_6390_cast_fp16")]; tensor var_6391_begin_0 = const()[name = tensor("op_6391_begin_0"), val = tensor([0, 0, 0, 1536])]; tensor var_6391_end_0 = const()[name = tensor("op_6391_end_0"), val = tensor([2, 40, 1, 2048])]; tensor var_6391_end_mask_0 = const()[name = tensor("op_6391_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6391_cast_fp16 = slice_by_index(begin = var_6391_begin_0, end = var_6391_end_0, end_mask = var_6391_end_mask_0, x = var_6349_cast_fp16)[name = tensor("op_6391_cast_fp16")]; tensor var_6392_begin_0 = const()[name = tensor("op_6392_begin_0"), val = tensor([0, 0, 0, 2048])]; tensor var_6392_end_0 = const()[name = tensor("op_6392_end_0"), val = tensor([2, 40, 1, 2560])]; tensor var_6392_end_mask_0 = const()[name = tensor("op_6392_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6392_cast_fp16 = slice_by_index(begin = var_6392_begin_0, end = var_6392_end_0, end_mask = var_6392_end_mask_0, x = var_6349_cast_fp16)[name = tensor("op_6392_cast_fp16")]; tensor var_6393_begin_0 = const()[name = tensor("op_6393_begin_0"), val = tensor([0, 0, 0, 2560])]; tensor var_6393_end_0 = const()[name = tensor("op_6393_end_0"), val = tensor([2, 40, 1, 3072])]; tensor var_6393_end_mask_0 = const()[name = tensor("op_6393_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6393_cast_fp16 = slice_by_index(begin = var_6393_begin_0, end = var_6393_end_0, end_mask = var_6393_end_mask_0, x = var_6349_cast_fp16)[name = tensor("op_6393_cast_fp16")]; tensor var_6394_begin_0 = const()[name = tensor("op_6394_begin_0"), val = tensor([0, 0, 0, 3072])]; tensor var_6394_end_0 = const()[name = tensor("op_6394_end_0"), val = tensor([2, 40, 1, 3584])]; tensor var_6394_end_mask_0 = const()[name = tensor("op_6394_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6394_cast_fp16 = slice_by_index(begin = var_6394_begin_0, end = var_6394_end_0, end_mask = var_6394_end_mask_0, x = var_6349_cast_fp16)[name = tensor("op_6394_cast_fp16")]; tensor var_6395_begin_0 = const()[name = tensor("op_6395_begin_0"), val = tensor([0, 0, 0, 3584])]; tensor var_6395_end_0 = const()[name = tensor("op_6395_end_0"), val = tensor([2, 40, 1, 4096])]; tensor var_6395_end_mask_0 = const()[name = tensor("op_6395_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6395_cast_fp16 = slice_by_index(begin = var_6395_begin_0, end = var_6395_end_0, end_mask = var_6395_end_mask_0, x = var_6349_cast_fp16)[name = tensor("op_6395_cast_fp16")]; tensor var_6396_begin_0 = const()[name = tensor("op_6396_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_6396_end_0 = const()[name = tensor("op_6396_end_0"), val = tensor([2, 40, 1, 512])]; tensor var_6396_end_mask_0 = const()[name = tensor("op_6396_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6396_cast_fp16 = slice_by_index(begin = var_6396_begin_0, end = var_6396_end_0, end_mask = var_6396_end_mask_0, x = var_6353_cast_fp16)[name = tensor("op_6396_cast_fp16")]; tensor var_6397_begin_0 = const()[name = tensor("op_6397_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_6397_end_0 = const()[name = tensor("op_6397_end_0"), val = tensor([2, 40, 1, 1024])]; tensor var_6397_end_mask_0 = const()[name = tensor("op_6397_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6397_cast_fp16 = slice_by_index(begin = var_6397_begin_0, end = var_6397_end_0, end_mask = var_6397_end_mask_0, x = var_6353_cast_fp16)[name = tensor("op_6397_cast_fp16")]; tensor var_6398_begin_0 = const()[name = tensor("op_6398_begin_0"), val = tensor([0, 0, 0, 1024])]; tensor var_6398_end_0 = const()[name = tensor("op_6398_end_0"), val = tensor([2, 40, 1, 1536])]; tensor var_6398_end_mask_0 = const()[name = tensor("op_6398_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6398_cast_fp16 = slice_by_index(begin = var_6398_begin_0, end = var_6398_end_0, end_mask = var_6398_end_mask_0, x = var_6353_cast_fp16)[name = tensor("op_6398_cast_fp16")]; tensor var_6399_begin_0 = const()[name = tensor("op_6399_begin_0"), val = tensor([0, 0, 0, 1536])]; tensor var_6399_end_0 = const()[name = tensor("op_6399_end_0"), val = tensor([2, 40, 1, 2048])]; tensor var_6399_end_mask_0 = const()[name = tensor("op_6399_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6399_cast_fp16 = slice_by_index(begin = var_6399_begin_0, end = var_6399_end_0, end_mask = var_6399_end_mask_0, x = var_6353_cast_fp16)[name = tensor("op_6399_cast_fp16")]; tensor var_6400_begin_0 = const()[name = tensor("op_6400_begin_0"), val = tensor([0, 0, 0, 2048])]; tensor var_6400_end_0 = const()[name = tensor("op_6400_end_0"), val = tensor([2, 40, 1, 2560])]; tensor var_6400_end_mask_0 = const()[name = tensor("op_6400_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6400_cast_fp16 = slice_by_index(begin = var_6400_begin_0, end = var_6400_end_0, end_mask = var_6400_end_mask_0, x = var_6353_cast_fp16)[name = tensor("op_6400_cast_fp16")]; tensor var_6401_begin_0 = const()[name = tensor("op_6401_begin_0"), val = tensor([0, 0, 0, 2560])]; tensor var_6401_end_0 = const()[name = tensor("op_6401_end_0"), val = tensor([2, 40, 1, 3072])]; tensor var_6401_end_mask_0 = const()[name = tensor("op_6401_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6401_cast_fp16 = slice_by_index(begin = var_6401_begin_0, end = var_6401_end_0, end_mask = var_6401_end_mask_0, x = var_6353_cast_fp16)[name = tensor("op_6401_cast_fp16")]; tensor var_6402_begin_0 = const()[name = tensor("op_6402_begin_0"), val = tensor([0, 0, 0, 3072])]; tensor var_6402_end_0 = const()[name = tensor("op_6402_end_0"), val = tensor([2, 40, 1, 3584])]; tensor var_6402_end_mask_0 = const()[name = tensor("op_6402_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6402_cast_fp16 = slice_by_index(begin = var_6402_begin_0, end = var_6402_end_0, end_mask = var_6402_end_mask_0, x = var_6353_cast_fp16)[name = tensor("op_6402_cast_fp16")]; tensor var_6403_begin_0 = const()[name = tensor("op_6403_begin_0"), val = tensor([0, 0, 0, 3584])]; tensor var_6403_end_0 = const()[name = tensor("op_6403_end_0"), val = tensor([2, 40, 1, 4096])]; tensor var_6403_end_mask_0 = const()[name = tensor("op_6403_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6403_cast_fp16 = slice_by_index(begin = var_6403_begin_0, end = var_6403_end_0, end_mask = var_6403_end_mask_0, x = var_6353_cast_fp16)[name = tensor("op_6403_cast_fp16")]; tensor var_6404_begin_0 = const()[name = tensor("op_6404_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_6404_end_0 = const()[name = tensor("op_6404_end_0"), val = tensor([2, 40, 1, 512])]; tensor var_6404_end_mask_0 = const()[name = tensor("op_6404_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6404_cast_fp16 = slice_by_index(begin = var_6404_begin_0, end = var_6404_end_0, end_mask = var_6404_end_mask_0, x = var_6357_cast_fp16)[name = tensor("op_6404_cast_fp16")]; tensor var_6405_begin_0 = const()[name = tensor("op_6405_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_6405_end_0 = const()[name = tensor("op_6405_end_0"), val = tensor([2, 40, 1, 1024])]; tensor var_6405_end_mask_0 = const()[name = tensor("op_6405_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6405_cast_fp16 = slice_by_index(begin = var_6405_begin_0, end = var_6405_end_0, end_mask = var_6405_end_mask_0, x = var_6357_cast_fp16)[name = tensor("op_6405_cast_fp16")]; tensor var_6406_begin_0 = const()[name = tensor("op_6406_begin_0"), val = tensor([0, 0, 0, 1024])]; tensor var_6406_end_0 = const()[name = tensor("op_6406_end_0"), val = tensor([2, 40, 1, 1536])]; tensor var_6406_end_mask_0 = const()[name = tensor("op_6406_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6406_cast_fp16 = slice_by_index(begin = var_6406_begin_0, end = var_6406_end_0, end_mask = var_6406_end_mask_0, x = var_6357_cast_fp16)[name = tensor("op_6406_cast_fp16")]; tensor var_6407_begin_0 = const()[name = tensor("op_6407_begin_0"), val = tensor([0, 0, 0, 1536])]; tensor var_6407_end_0 = const()[name = tensor("op_6407_end_0"), val = tensor([2, 40, 1, 2048])]; tensor var_6407_end_mask_0 = const()[name = tensor("op_6407_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6407_cast_fp16 = slice_by_index(begin = var_6407_begin_0, end = var_6407_end_0, end_mask = var_6407_end_mask_0, x = var_6357_cast_fp16)[name = tensor("op_6407_cast_fp16")]; tensor var_6408_begin_0 = const()[name = tensor("op_6408_begin_0"), val = tensor([0, 0, 0, 2048])]; tensor var_6408_end_0 = const()[name = tensor("op_6408_end_0"), val = tensor([2, 40, 1, 2560])]; tensor var_6408_end_mask_0 = const()[name = tensor("op_6408_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6408_cast_fp16 = slice_by_index(begin = var_6408_begin_0, end = var_6408_end_0, end_mask = var_6408_end_mask_0, x = var_6357_cast_fp16)[name = tensor("op_6408_cast_fp16")]; tensor var_6409_begin_0 = const()[name = tensor("op_6409_begin_0"), val = tensor([0, 0, 0, 2560])]; tensor var_6409_end_0 = const()[name = tensor("op_6409_end_0"), val = tensor([2, 40, 1, 3072])]; tensor var_6409_end_mask_0 = const()[name = tensor("op_6409_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6409_cast_fp16 = slice_by_index(begin = var_6409_begin_0, end = var_6409_end_0, end_mask = var_6409_end_mask_0, x = var_6357_cast_fp16)[name = tensor("op_6409_cast_fp16")]; tensor var_6410_begin_0 = const()[name = tensor("op_6410_begin_0"), val = tensor([0, 0, 0, 3072])]; tensor var_6410_end_0 = const()[name = tensor("op_6410_end_0"), val = tensor([2, 40, 1, 3584])]; tensor var_6410_end_mask_0 = const()[name = tensor("op_6410_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6410_cast_fp16 = slice_by_index(begin = var_6410_begin_0, end = var_6410_end_0, end_mask = var_6410_end_mask_0, x = var_6357_cast_fp16)[name = tensor("op_6410_cast_fp16")]; tensor var_6411_begin_0 = const()[name = tensor("op_6411_begin_0"), val = tensor([0, 0, 0, 3584])]; tensor var_6411_end_0 = const()[name = tensor("op_6411_end_0"), val = tensor([2, 40, 1, 4096])]; tensor var_6411_end_mask_0 = const()[name = tensor("op_6411_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6411_cast_fp16 = slice_by_index(begin = var_6411_begin_0, end = var_6411_end_0, end_mask = var_6411_end_mask_0, x = var_6357_cast_fp16)[name = tensor("op_6411_cast_fp16")]; tensor var_6412_begin_0 = const()[name = tensor("op_6412_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_6412_end_0 = const()[name = tensor("op_6412_end_0"), val = tensor([2, 40, 1, 512])]; tensor var_6412_end_mask_0 = const()[name = tensor("op_6412_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6412_cast_fp16 = slice_by_index(begin = var_6412_begin_0, end = var_6412_end_0, end_mask = var_6412_end_mask_0, x = var_6361_cast_fp16)[name = tensor("op_6412_cast_fp16")]; tensor var_6413_begin_0 = const()[name = tensor("op_6413_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_6413_end_0 = const()[name = tensor("op_6413_end_0"), val = tensor([2, 40, 1, 1024])]; tensor var_6413_end_mask_0 = const()[name = tensor("op_6413_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6413_cast_fp16 = slice_by_index(begin = var_6413_begin_0, end = var_6413_end_0, end_mask = var_6413_end_mask_0, x = var_6361_cast_fp16)[name = tensor("op_6413_cast_fp16")]; tensor var_6414_begin_0 = const()[name = tensor("op_6414_begin_0"), val = tensor([0, 0, 0, 1024])]; tensor var_6414_end_0 = const()[name = tensor("op_6414_end_0"), val = tensor([2, 40, 1, 1536])]; tensor var_6414_end_mask_0 = const()[name = tensor("op_6414_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6414_cast_fp16 = slice_by_index(begin = var_6414_begin_0, end = var_6414_end_0, end_mask = var_6414_end_mask_0, x = var_6361_cast_fp16)[name = tensor("op_6414_cast_fp16")]; tensor var_6415_begin_0 = const()[name = tensor("op_6415_begin_0"), val = tensor([0, 0, 0, 1536])]; tensor var_6415_end_0 = const()[name = tensor("op_6415_end_0"), val = tensor([2, 40, 1, 2048])]; tensor var_6415_end_mask_0 = const()[name = tensor("op_6415_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6415_cast_fp16 = slice_by_index(begin = var_6415_begin_0, end = var_6415_end_0, end_mask = var_6415_end_mask_0, x = var_6361_cast_fp16)[name = tensor("op_6415_cast_fp16")]; tensor var_6416_begin_0 = const()[name = tensor("op_6416_begin_0"), val = tensor([0, 0, 0, 2048])]; tensor var_6416_end_0 = const()[name = tensor("op_6416_end_0"), val = tensor([2, 40, 1, 2560])]; tensor var_6416_end_mask_0 = const()[name = tensor("op_6416_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6416_cast_fp16 = slice_by_index(begin = var_6416_begin_0, end = var_6416_end_0, end_mask = var_6416_end_mask_0, x = var_6361_cast_fp16)[name = tensor("op_6416_cast_fp16")]; tensor var_6417_begin_0 = const()[name = tensor("op_6417_begin_0"), val = tensor([0, 0, 0, 2560])]; tensor var_6417_end_0 = const()[name = tensor("op_6417_end_0"), val = tensor([2, 40, 1, 3072])]; tensor var_6417_end_mask_0 = const()[name = tensor("op_6417_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6417_cast_fp16 = slice_by_index(begin = var_6417_begin_0, end = var_6417_end_0, end_mask = var_6417_end_mask_0, x = var_6361_cast_fp16)[name = tensor("op_6417_cast_fp16")]; tensor var_6418_begin_0 = const()[name = tensor("op_6418_begin_0"), val = tensor([0, 0, 0, 3072])]; tensor var_6418_end_0 = const()[name = tensor("op_6418_end_0"), val = tensor([2, 40, 1, 3584])]; tensor var_6418_end_mask_0 = const()[name = tensor("op_6418_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6418_cast_fp16 = slice_by_index(begin = var_6418_begin_0, end = var_6418_end_0, end_mask = var_6418_end_mask_0, x = var_6361_cast_fp16)[name = tensor("op_6418_cast_fp16")]; tensor var_6419_begin_0 = const()[name = tensor("op_6419_begin_0"), val = tensor([0, 0, 0, 3584])]; tensor var_6419_end_0 = const()[name = tensor("op_6419_end_0"), val = tensor([2, 40, 1, 4096])]; tensor var_6419_end_mask_0 = const()[name = tensor("op_6419_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6419_cast_fp16 = slice_by_index(begin = var_6419_begin_0, end = var_6419_end_0, end_mask = var_6419_end_mask_0, x = var_6361_cast_fp16)[name = tensor("op_6419_cast_fp16")]; tensor var_6420_begin_0 = const()[name = tensor("op_6420_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_6420_end_0 = const()[name = tensor("op_6420_end_0"), val = tensor([2, 40, 1, 512])]; tensor var_6420_end_mask_0 = const()[name = tensor("op_6420_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6420_cast_fp16 = slice_by_index(begin = var_6420_begin_0, end = var_6420_end_0, end_mask = var_6420_end_mask_0, x = var_6365_cast_fp16)[name = tensor("op_6420_cast_fp16")]; tensor var_6421_begin_0 = const()[name = tensor("op_6421_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_6421_end_0 = const()[name = tensor("op_6421_end_0"), val = tensor([2, 40, 1, 1024])]; tensor var_6421_end_mask_0 = const()[name = tensor("op_6421_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6421_cast_fp16 = slice_by_index(begin = var_6421_begin_0, end = var_6421_end_0, end_mask = var_6421_end_mask_0, x = var_6365_cast_fp16)[name = tensor("op_6421_cast_fp16")]; tensor var_6422_begin_0 = const()[name = tensor("op_6422_begin_0"), val = tensor([0, 0, 0, 1024])]; tensor var_6422_end_0 = const()[name = tensor("op_6422_end_0"), val = tensor([2, 40, 1, 1536])]; tensor var_6422_end_mask_0 = const()[name = tensor("op_6422_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6422_cast_fp16 = slice_by_index(begin = var_6422_begin_0, end = var_6422_end_0, end_mask = var_6422_end_mask_0, x = var_6365_cast_fp16)[name = tensor("op_6422_cast_fp16")]; tensor var_6423_begin_0 = const()[name = tensor("op_6423_begin_0"), val = tensor([0, 0, 0, 1536])]; tensor var_6423_end_0 = const()[name = tensor("op_6423_end_0"), val = tensor([2, 40, 1, 2048])]; tensor var_6423_end_mask_0 = const()[name = tensor("op_6423_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6423_cast_fp16 = slice_by_index(begin = var_6423_begin_0, end = var_6423_end_0, end_mask = var_6423_end_mask_0, x = var_6365_cast_fp16)[name = tensor("op_6423_cast_fp16")]; tensor var_6424_begin_0 = const()[name = tensor("op_6424_begin_0"), val = tensor([0, 0, 0, 2048])]; tensor var_6424_end_0 = const()[name = tensor("op_6424_end_0"), val = tensor([2, 40, 1, 2560])]; tensor var_6424_end_mask_0 = const()[name = tensor("op_6424_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6424_cast_fp16 = slice_by_index(begin = var_6424_begin_0, end = var_6424_end_0, end_mask = var_6424_end_mask_0, x = var_6365_cast_fp16)[name = tensor("op_6424_cast_fp16")]; tensor var_6425_begin_0 = const()[name = tensor("op_6425_begin_0"), val = tensor([0, 0, 0, 2560])]; tensor var_6425_end_0 = const()[name = tensor("op_6425_end_0"), val = tensor([2, 40, 1, 3072])]; tensor var_6425_end_mask_0 = const()[name = tensor("op_6425_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6425_cast_fp16 = slice_by_index(begin = var_6425_begin_0, end = var_6425_end_0, end_mask = var_6425_end_mask_0, x = var_6365_cast_fp16)[name = tensor("op_6425_cast_fp16")]; tensor var_6426_begin_0 = const()[name = tensor("op_6426_begin_0"), val = tensor([0, 0, 0, 3072])]; tensor var_6426_end_0 = const()[name = tensor("op_6426_end_0"), val = tensor([2, 40, 1, 3584])]; tensor var_6426_end_mask_0 = const()[name = tensor("op_6426_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6426_cast_fp16 = slice_by_index(begin = var_6426_begin_0, end = var_6426_end_0, end_mask = var_6426_end_mask_0, x = var_6365_cast_fp16)[name = tensor("op_6426_cast_fp16")]; tensor var_6427_begin_0 = const()[name = tensor("op_6427_begin_0"), val = tensor([0, 0, 0, 3584])]; tensor var_6427_end_0 = const()[name = tensor("op_6427_end_0"), val = tensor([2, 40, 1, 4096])]; tensor var_6427_end_mask_0 = const()[name = tensor("op_6427_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6427_cast_fp16 = slice_by_index(begin = var_6427_begin_0, end = var_6427_end_0, end_mask = var_6427_end_mask_0, x = var_6365_cast_fp16)[name = tensor("op_6427_cast_fp16")]; tensor var_6428_begin_0 = const()[name = tensor("op_6428_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_6428_end_0 = const()[name = tensor("op_6428_end_0"), val = tensor([2, 40, 1, 512])]; tensor var_6428_end_mask_0 = const()[name = tensor("op_6428_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6428_cast_fp16 = slice_by_index(begin = var_6428_begin_0, end = var_6428_end_0, end_mask = var_6428_end_mask_0, x = var_6369_cast_fp16)[name = tensor("op_6428_cast_fp16")]; tensor var_6429_begin_0 = const()[name = tensor("op_6429_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_6429_end_0 = const()[name = tensor("op_6429_end_0"), val = tensor([2, 40, 1, 1024])]; tensor var_6429_end_mask_0 = const()[name = tensor("op_6429_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6429_cast_fp16 = slice_by_index(begin = var_6429_begin_0, end = var_6429_end_0, end_mask = var_6429_end_mask_0, x = var_6369_cast_fp16)[name = tensor("op_6429_cast_fp16")]; tensor var_6430_begin_0 = const()[name = tensor("op_6430_begin_0"), val = tensor([0, 0, 0, 1024])]; tensor var_6430_end_0 = const()[name = tensor("op_6430_end_0"), val = tensor([2, 40, 1, 1536])]; tensor var_6430_end_mask_0 = const()[name = tensor("op_6430_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6430_cast_fp16 = slice_by_index(begin = var_6430_begin_0, end = var_6430_end_0, end_mask = var_6430_end_mask_0, x = var_6369_cast_fp16)[name = tensor("op_6430_cast_fp16")]; tensor var_6431_begin_0 = const()[name = tensor("op_6431_begin_0"), val = tensor([0, 0, 0, 1536])]; tensor var_6431_end_0 = const()[name = tensor("op_6431_end_0"), val = tensor([2, 40, 1, 2048])]; tensor var_6431_end_mask_0 = const()[name = tensor("op_6431_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6431_cast_fp16 = slice_by_index(begin = var_6431_begin_0, end = var_6431_end_0, end_mask = var_6431_end_mask_0, x = var_6369_cast_fp16)[name = tensor("op_6431_cast_fp16")]; tensor var_6432_begin_0 = const()[name = tensor("op_6432_begin_0"), val = tensor([0, 0, 0, 2048])]; tensor var_6432_end_0 = const()[name = tensor("op_6432_end_0"), val = tensor([2, 40, 1, 2560])]; tensor var_6432_end_mask_0 = const()[name = tensor("op_6432_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6432_cast_fp16 = slice_by_index(begin = var_6432_begin_0, end = var_6432_end_0, end_mask = var_6432_end_mask_0, x = var_6369_cast_fp16)[name = tensor("op_6432_cast_fp16")]; tensor var_6433_begin_0 = const()[name = tensor("op_6433_begin_0"), val = tensor([0, 0, 0, 2560])]; tensor var_6433_end_0 = const()[name = tensor("op_6433_end_0"), val = tensor([2, 40, 1, 3072])]; tensor var_6433_end_mask_0 = const()[name = tensor("op_6433_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6433_cast_fp16 = slice_by_index(begin = var_6433_begin_0, end = var_6433_end_0, end_mask = var_6433_end_mask_0, x = var_6369_cast_fp16)[name = tensor("op_6433_cast_fp16")]; tensor var_6434_begin_0 = const()[name = tensor("op_6434_begin_0"), val = tensor([0, 0, 0, 3072])]; tensor var_6434_end_0 = const()[name = tensor("op_6434_end_0"), val = tensor([2, 40, 1, 3584])]; tensor var_6434_end_mask_0 = const()[name = tensor("op_6434_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6434_cast_fp16 = slice_by_index(begin = var_6434_begin_0, end = var_6434_end_0, end_mask = var_6434_end_mask_0, x = var_6369_cast_fp16)[name = tensor("op_6434_cast_fp16")]; tensor var_6435_begin_0 = const()[name = tensor("op_6435_begin_0"), val = tensor([0, 0, 0, 3584])]; tensor var_6435_end_0 = const()[name = tensor("op_6435_end_0"), val = tensor([2, 40, 1, 4096])]; tensor var_6435_end_mask_0 = const()[name = tensor("op_6435_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6435_cast_fp16 = slice_by_index(begin = var_6435_begin_0, end = var_6435_end_0, end_mask = var_6435_end_mask_0, x = var_6369_cast_fp16)[name = tensor("op_6435_cast_fp16")]; tensor k_63_perm_0 = const()[name = tensor("k_63_perm_0"), val = tensor([0, 3, 2, 1])]; tensor var_6440_begin_0 = const()[name = tensor("op_6440_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_6440_end_0 = const()[name = tensor("op_6440_end_0"), val = tensor([2, 77, 1, 40])]; tensor var_6440_end_mask_0 = const()[name = tensor("op_6440_end_mask_0"), val = tensor([true, true, true, false])]; tensor transpose_2 = transpose(perm = k_63_perm_0, x = k_61_cast_fp16)[name = tensor("transpose_2")]; tensor var_6440_cast_fp16 = slice_by_index(begin = var_6440_begin_0, end = var_6440_end_0, end_mask = var_6440_end_mask_0, x = transpose_2)[name = tensor("op_6440_cast_fp16")]; tensor var_6444_begin_0 = const()[name = tensor("op_6444_begin_0"), val = tensor([0, 0, 0, 40])]; tensor var_6444_end_0 = const()[name = tensor("op_6444_end_0"), val = tensor([2, 77, 1, 80])]; tensor var_6444_end_mask_0 = const()[name = tensor("op_6444_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6444_cast_fp16 = slice_by_index(begin = var_6444_begin_0, end = var_6444_end_0, end_mask = var_6444_end_mask_0, x = transpose_2)[name = tensor("op_6444_cast_fp16")]; tensor var_6448_begin_0 = const()[name = tensor("op_6448_begin_0"), val = tensor([0, 0, 0, 80])]; tensor var_6448_end_0 = const()[name = tensor("op_6448_end_0"), val = tensor([2, 77, 1, 120])]; tensor var_6448_end_mask_0 = const()[name = tensor("op_6448_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6448_cast_fp16 = slice_by_index(begin = var_6448_begin_0, end = var_6448_end_0, end_mask = var_6448_end_mask_0, x = transpose_2)[name = tensor("op_6448_cast_fp16")]; tensor var_6452_begin_0 = const()[name = tensor("op_6452_begin_0"), val = tensor([0, 0, 0, 120])]; tensor var_6452_end_0 = const()[name = tensor("op_6452_end_0"), val = tensor([2, 77, 1, 160])]; tensor var_6452_end_mask_0 = const()[name = tensor("op_6452_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6452_cast_fp16 = slice_by_index(begin = var_6452_begin_0, end = var_6452_end_0, end_mask = var_6452_end_mask_0, x = transpose_2)[name = tensor("op_6452_cast_fp16")]; tensor var_6456_begin_0 = const()[name = tensor("op_6456_begin_0"), val = tensor([0, 0, 0, 160])]; tensor var_6456_end_0 = const()[name = tensor("op_6456_end_0"), val = tensor([2, 77, 1, 200])]; tensor var_6456_end_mask_0 = const()[name = tensor("op_6456_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6456_cast_fp16 = slice_by_index(begin = var_6456_begin_0, end = var_6456_end_0, end_mask = var_6456_end_mask_0, x = transpose_2)[name = tensor("op_6456_cast_fp16")]; tensor var_6460_begin_0 = const()[name = tensor("op_6460_begin_0"), val = tensor([0, 0, 0, 200])]; tensor var_6460_end_0 = const()[name = tensor("op_6460_end_0"), val = tensor([2, 77, 1, 240])]; tensor var_6460_end_mask_0 = const()[name = tensor("op_6460_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6460_cast_fp16 = slice_by_index(begin = var_6460_begin_0, end = var_6460_end_0, end_mask = var_6460_end_mask_0, x = transpose_2)[name = tensor("op_6460_cast_fp16")]; tensor var_6464_begin_0 = const()[name = tensor("op_6464_begin_0"), val = tensor([0, 0, 0, 240])]; tensor var_6464_end_0 = const()[name = tensor("op_6464_end_0"), val = tensor([2, 77, 1, 280])]; tensor var_6464_end_mask_0 = const()[name = tensor("op_6464_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6464_cast_fp16 = slice_by_index(begin = var_6464_begin_0, end = var_6464_end_0, end_mask = var_6464_end_mask_0, x = transpose_2)[name = tensor("op_6464_cast_fp16")]; tensor var_6468_begin_0 = const()[name = tensor("op_6468_begin_0"), val = tensor([0, 0, 0, 280])]; tensor var_6468_end_0 = const()[name = tensor("op_6468_end_0"), val = tensor([2, 77, 1, 320])]; tensor var_6468_end_mask_0 = const()[name = tensor("op_6468_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6468_cast_fp16 = slice_by_index(begin = var_6468_begin_0, end = var_6468_end_0, end_mask = var_6468_end_mask_0, x = transpose_2)[name = tensor("op_6468_cast_fp16")]; tensor var_6470_begin_0 = const()[name = tensor("op_6470_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_6470_end_0 = const()[name = tensor("op_6470_end_0"), val = tensor([2, 40, 1, 77])]; tensor var_6470_end_mask_0 = const()[name = tensor("op_6470_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6470_cast_fp16 = slice_by_index(begin = var_6470_begin_0, end = var_6470_end_0, end_mask = var_6470_end_mask_0, x = v_31_cast_fp16)[name = tensor("op_6470_cast_fp16")]; tensor var_6474_begin_0 = const()[name = tensor("op_6474_begin_0"), val = tensor([0, 40, 0, 0])]; tensor var_6474_end_0 = const()[name = tensor("op_6474_end_0"), val = tensor([2, 80, 1, 77])]; tensor var_6474_end_mask_0 = const()[name = tensor("op_6474_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6474_cast_fp16 = slice_by_index(begin = var_6474_begin_0, end = var_6474_end_0, end_mask = var_6474_end_mask_0, x = v_31_cast_fp16)[name = tensor("op_6474_cast_fp16")]; tensor var_6478_begin_0 = const()[name = tensor("op_6478_begin_0"), val = tensor([0, 80, 0, 0])]; tensor var_6478_end_0 = const()[name = tensor("op_6478_end_0"), val = tensor([2, 120, 1, 77])]; tensor var_6478_end_mask_0 = const()[name = tensor("op_6478_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6478_cast_fp16 = slice_by_index(begin = var_6478_begin_0, end = var_6478_end_0, end_mask = var_6478_end_mask_0, x = v_31_cast_fp16)[name = tensor("op_6478_cast_fp16")]; tensor var_6482_begin_0 = const()[name = tensor("op_6482_begin_0"), val = tensor([0, 120, 0, 0])]; tensor var_6482_end_0 = const()[name = tensor("op_6482_end_0"), val = tensor([2, 160, 1, 77])]; tensor var_6482_end_mask_0 = const()[name = tensor("op_6482_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6482_cast_fp16 = slice_by_index(begin = var_6482_begin_0, end = var_6482_end_0, end_mask = var_6482_end_mask_0, x = v_31_cast_fp16)[name = tensor("op_6482_cast_fp16")]; tensor var_6486_begin_0 = const()[name = tensor("op_6486_begin_0"), val = tensor([0, 160, 0, 0])]; tensor var_6486_end_0 = const()[name = tensor("op_6486_end_0"), val = tensor([2, 200, 1, 77])]; tensor var_6486_end_mask_0 = const()[name = tensor("op_6486_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6486_cast_fp16 = slice_by_index(begin = var_6486_begin_0, end = var_6486_end_0, end_mask = var_6486_end_mask_0, x = v_31_cast_fp16)[name = tensor("op_6486_cast_fp16")]; tensor var_6490_begin_0 = const()[name = tensor("op_6490_begin_0"), val = tensor([0, 200, 0, 0])]; tensor var_6490_end_0 = const()[name = tensor("op_6490_end_0"), val = tensor([2, 240, 1, 77])]; tensor var_6490_end_mask_0 = const()[name = tensor("op_6490_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6490_cast_fp16 = slice_by_index(begin = var_6490_begin_0, end = var_6490_end_0, end_mask = var_6490_end_mask_0, x = v_31_cast_fp16)[name = tensor("op_6490_cast_fp16")]; tensor var_6494_begin_0 = const()[name = tensor("op_6494_begin_0"), val = tensor([0, 240, 0, 0])]; tensor var_6494_end_0 = const()[name = tensor("op_6494_end_0"), val = tensor([2, 280, 1, 77])]; tensor var_6494_end_mask_0 = const()[name = tensor("op_6494_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6494_cast_fp16 = slice_by_index(begin = var_6494_begin_0, end = var_6494_end_0, end_mask = var_6494_end_mask_0, x = v_31_cast_fp16)[name = tensor("op_6494_cast_fp16")]; tensor var_6498_begin_0 = const()[name = tensor("op_6498_begin_0"), val = tensor([0, 280, 0, 0])]; tensor var_6498_end_0 = const()[name = tensor("op_6498_end_0"), val = tensor([2, 320, 1, 77])]; tensor var_6498_end_mask_0 = const()[name = tensor("op_6498_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6498_cast_fp16 = slice_by_index(begin = var_6498_begin_0, end = var_6498_end_0, end_mask = var_6498_end_mask_0, x = v_31_cast_fp16)[name = tensor("op_6498_cast_fp16")]; tensor var_6502_equation_0 = const()[name = tensor("op_6502_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6502_cast_fp16 = einsum(equation = var_6502_equation_0, values = (var_6440_cast_fp16, var_6372_cast_fp16))[name = tensor("op_6502_cast_fp16")]; tensor var_6503_to_fp16 = const()[name = tensor("op_6503_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_577_cast_fp16 = mul(x = var_6502_cast_fp16, y = var_6503_to_fp16)[name = tensor("aw_chunk_577_cast_fp16")]; tensor var_6506_equation_0 = const()[name = tensor("op_6506_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6506_cast_fp16 = einsum(equation = var_6506_equation_0, values = (var_6440_cast_fp16, var_6373_cast_fp16))[name = tensor("op_6506_cast_fp16")]; tensor var_6507_to_fp16 = const()[name = tensor("op_6507_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_579_cast_fp16 = mul(x = var_6506_cast_fp16, y = var_6507_to_fp16)[name = tensor("aw_chunk_579_cast_fp16")]; tensor var_6510_equation_0 = const()[name = tensor("op_6510_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6510_cast_fp16 = einsum(equation = var_6510_equation_0, values = (var_6440_cast_fp16, var_6374_cast_fp16))[name = tensor("op_6510_cast_fp16")]; tensor var_6511_to_fp16 = const()[name = tensor("op_6511_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_581_cast_fp16 = mul(x = var_6510_cast_fp16, y = var_6511_to_fp16)[name = tensor("aw_chunk_581_cast_fp16")]; tensor var_6514_equation_0 = const()[name = tensor("op_6514_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6514_cast_fp16 = einsum(equation = var_6514_equation_0, values = (var_6440_cast_fp16, var_6375_cast_fp16))[name = tensor("op_6514_cast_fp16")]; tensor var_6515_to_fp16 = const()[name = tensor("op_6515_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_583_cast_fp16 = mul(x = var_6514_cast_fp16, y = var_6515_to_fp16)[name = tensor("aw_chunk_583_cast_fp16")]; tensor var_6518_equation_0 = const()[name = tensor("op_6518_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6518_cast_fp16 = einsum(equation = var_6518_equation_0, values = (var_6440_cast_fp16, var_6376_cast_fp16))[name = tensor("op_6518_cast_fp16")]; tensor var_6519_to_fp16 = const()[name = tensor("op_6519_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_585_cast_fp16 = mul(x = var_6518_cast_fp16, y = var_6519_to_fp16)[name = tensor("aw_chunk_585_cast_fp16")]; tensor var_6522_equation_0 = const()[name = tensor("op_6522_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6522_cast_fp16 = einsum(equation = var_6522_equation_0, values = (var_6440_cast_fp16, var_6377_cast_fp16))[name = tensor("op_6522_cast_fp16")]; tensor var_6523_to_fp16 = const()[name = tensor("op_6523_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_587_cast_fp16 = mul(x = var_6522_cast_fp16, y = var_6523_to_fp16)[name = tensor("aw_chunk_587_cast_fp16")]; tensor var_6526_equation_0 = const()[name = tensor("op_6526_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6526_cast_fp16 = einsum(equation = var_6526_equation_0, values = (var_6440_cast_fp16, var_6378_cast_fp16))[name = tensor("op_6526_cast_fp16")]; tensor var_6527_to_fp16 = const()[name = tensor("op_6527_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_589_cast_fp16 = mul(x = var_6526_cast_fp16, y = var_6527_to_fp16)[name = tensor("aw_chunk_589_cast_fp16")]; tensor var_6530_equation_0 = const()[name = tensor("op_6530_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6530_cast_fp16 = einsum(equation = var_6530_equation_0, values = (var_6440_cast_fp16, var_6379_cast_fp16))[name = tensor("op_6530_cast_fp16")]; tensor var_6531_to_fp16 = const()[name = tensor("op_6531_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_591_cast_fp16 = mul(x = var_6530_cast_fp16, y = var_6531_to_fp16)[name = tensor("aw_chunk_591_cast_fp16")]; tensor var_6534_equation_0 = const()[name = tensor("op_6534_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6534_cast_fp16 = einsum(equation = var_6534_equation_0, values = (var_6444_cast_fp16, var_6380_cast_fp16))[name = tensor("op_6534_cast_fp16")]; tensor var_6535_to_fp16 = const()[name = tensor("op_6535_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_593_cast_fp16 = mul(x = var_6534_cast_fp16, y = var_6535_to_fp16)[name = tensor("aw_chunk_593_cast_fp16")]; tensor var_6538_equation_0 = const()[name = tensor("op_6538_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6538_cast_fp16 = einsum(equation = var_6538_equation_0, values = (var_6444_cast_fp16, var_6381_cast_fp16))[name = tensor("op_6538_cast_fp16")]; tensor var_6539_to_fp16 = const()[name = tensor("op_6539_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_595_cast_fp16 = mul(x = var_6538_cast_fp16, y = var_6539_to_fp16)[name = tensor("aw_chunk_595_cast_fp16")]; tensor var_6542_equation_0 = const()[name = tensor("op_6542_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6542_cast_fp16 = einsum(equation = var_6542_equation_0, values = (var_6444_cast_fp16, var_6382_cast_fp16))[name = tensor("op_6542_cast_fp16")]; tensor var_6543_to_fp16 = const()[name = tensor("op_6543_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_597_cast_fp16 = mul(x = var_6542_cast_fp16, y = var_6543_to_fp16)[name = tensor("aw_chunk_597_cast_fp16")]; tensor var_6546_equation_0 = const()[name = tensor("op_6546_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6546_cast_fp16 = einsum(equation = var_6546_equation_0, values = (var_6444_cast_fp16, var_6383_cast_fp16))[name = tensor("op_6546_cast_fp16")]; tensor var_6547_to_fp16 = const()[name = tensor("op_6547_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_599_cast_fp16 = mul(x = var_6546_cast_fp16, y = var_6547_to_fp16)[name = tensor("aw_chunk_599_cast_fp16")]; tensor var_6550_equation_0 = const()[name = tensor("op_6550_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6550_cast_fp16 = einsum(equation = var_6550_equation_0, values = (var_6444_cast_fp16, var_6384_cast_fp16))[name = tensor("op_6550_cast_fp16")]; tensor var_6551_to_fp16 = const()[name = tensor("op_6551_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_601_cast_fp16 = mul(x = var_6550_cast_fp16, y = var_6551_to_fp16)[name = tensor("aw_chunk_601_cast_fp16")]; tensor var_6554_equation_0 = const()[name = tensor("op_6554_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6554_cast_fp16 = einsum(equation = var_6554_equation_0, values = (var_6444_cast_fp16, var_6385_cast_fp16))[name = tensor("op_6554_cast_fp16")]; tensor var_6555_to_fp16 = const()[name = tensor("op_6555_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_603_cast_fp16 = mul(x = var_6554_cast_fp16, y = var_6555_to_fp16)[name = tensor("aw_chunk_603_cast_fp16")]; tensor var_6558_equation_0 = const()[name = tensor("op_6558_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6558_cast_fp16 = einsum(equation = var_6558_equation_0, values = (var_6444_cast_fp16, var_6386_cast_fp16))[name = tensor("op_6558_cast_fp16")]; tensor var_6559_to_fp16 = const()[name = tensor("op_6559_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_605_cast_fp16 = mul(x = var_6558_cast_fp16, y = var_6559_to_fp16)[name = tensor("aw_chunk_605_cast_fp16")]; tensor var_6562_equation_0 = const()[name = tensor("op_6562_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6562_cast_fp16 = einsum(equation = var_6562_equation_0, values = (var_6444_cast_fp16, var_6387_cast_fp16))[name = tensor("op_6562_cast_fp16")]; tensor var_6563_to_fp16 = const()[name = tensor("op_6563_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_607_cast_fp16 = mul(x = var_6562_cast_fp16, y = var_6563_to_fp16)[name = tensor("aw_chunk_607_cast_fp16")]; tensor var_6566_equation_0 = const()[name = tensor("op_6566_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6566_cast_fp16 = einsum(equation = var_6566_equation_0, values = (var_6448_cast_fp16, var_6388_cast_fp16))[name = tensor("op_6566_cast_fp16")]; tensor var_6567_to_fp16 = const()[name = tensor("op_6567_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_609_cast_fp16 = mul(x = var_6566_cast_fp16, y = var_6567_to_fp16)[name = tensor("aw_chunk_609_cast_fp16")]; tensor var_6570_equation_0 = const()[name = tensor("op_6570_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6570_cast_fp16 = einsum(equation = var_6570_equation_0, values = (var_6448_cast_fp16, var_6389_cast_fp16))[name = tensor("op_6570_cast_fp16")]; tensor var_6571_to_fp16 = const()[name = tensor("op_6571_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_611_cast_fp16 = mul(x = var_6570_cast_fp16, y = var_6571_to_fp16)[name = tensor("aw_chunk_611_cast_fp16")]; tensor var_6574_equation_0 = const()[name = tensor("op_6574_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6574_cast_fp16 = einsum(equation = var_6574_equation_0, values = (var_6448_cast_fp16, var_6390_cast_fp16))[name = tensor("op_6574_cast_fp16")]; tensor var_6575_to_fp16 = const()[name = tensor("op_6575_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_613_cast_fp16 = mul(x = var_6574_cast_fp16, y = var_6575_to_fp16)[name = tensor("aw_chunk_613_cast_fp16")]; tensor var_6578_equation_0 = const()[name = tensor("op_6578_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6578_cast_fp16 = einsum(equation = var_6578_equation_0, values = (var_6448_cast_fp16, var_6391_cast_fp16))[name = tensor("op_6578_cast_fp16")]; tensor var_6579_to_fp16 = const()[name = tensor("op_6579_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_615_cast_fp16 = mul(x = var_6578_cast_fp16, y = var_6579_to_fp16)[name = tensor("aw_chunk_615_cast_fp16")]; tensor var_6582_equation_0 = const()[name = tensor("op_6582_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6582_cast_fp16 = einsum(equation = var_6582_equation_0, values = (var_6448_cast_fp16, var_6392_cast_fp16))[name = tensor("op_6582_cast_fp16")]; tensor var_6583_to_fp16 = const()[name = tensor("op_6583_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_617_cast_fp16 = mul(x = var_6582_cast_fp16, y = var_6583_to_fp16)[name = tensor("aw_chunk_617_cast_fp16")]; tensor var_6586_equation_0 = const()[name = tensor("op_6586_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6586_cast_fp16 = einsum(equation = var_6586_equation_0, values = (var_6448_cast_fp16, var_6393_cast_fp16))[name = tensor("op_6586_cast_fp16")]; tensor var_6587_to_fp16 = const()[name = tensor("op_6587_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_619_cast_fp16 = mul(x = var_6586_cast_fp16, y = var_6587_to_fp16)[name = tensor("aw_chunk_619_cast_fp16")]; tensor var_6590_equation_0 = const()[name = tensor("op_6590_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6590_cast_fp16 = einsum(equation = var_6590_equation_0, values = (var_6448_cast_fp16, var_6394_cast_fp16))[name = tensor("op_6590_cast_fp16")]; tensor var_6591_to_fp16 = const()[name = tensor("op_6591_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_621_cast_fp16 = mul(x = var_6590_cast_fp16, y = var_6591_to_fp16)[name = tensor("aw_chunk_621_cast_fp16")]; tensor var_6594_equation_0 = const()[name = tensor("op_6594_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6594_cast_fp16 = einsum(equation = var_6594_equation_0, values = (var_6448_cast_fp16, var_6395_cast_fp16))[name = tensor("op_6594_cast_fp16")]; tensor var_6595_to_fp16 = const()[name = tensor("op_6595_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_623_cast_fp16 = mul(x = var_6594_cast_fp16, y = var_6595_to_fp16)[name = tensor("aw_chunk_623_cast_fp16")]; tensor var_6598_equation_0 = const()[name = tensor("op_6598_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6598_cast_fp16 = einsum(equation = var_6598_equation_0, values = (var_6452_cast_fp16, var_6396_cast_fp16))[name = tensor("op_6598_cast_fp16")]; tensor var_6599_to_fp16 = const()[name = tensor("op_6599_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_625_cast_fp16 = mul(x = var_6598_cast_fp16, y = var_6599_to_fp16)[name = tensor("aw_chunk_625_cast_fp16")]; tensor var_6602_equation_0 = const()[name = tensor("op_6602_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6602_cast_fp16 = einsum(equation = var_6602_equation_0, values = (var_6452_cast_fp16, var_6397_cast_fp16))[name = tensor("op_6602_cast_fp16")]; tensor var_6603_to_fp16 = const()[name = tensor("op_6603_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_627_cast_fp16 = mul(x = var_6602_cast_fp16, y = var_6603_to_fp16)[name = tensor("aw_chunk_627_cast_fp16")]; tensor var_6606_equation_0 = const()[name = tensor("op_6606_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6606_cast_fp16 = einsum(equation = var_6606_equation_0, values = (var_6452_cast_fp16, var_6398_cast_fp16))[name = tensor("op_6606_cast_fp16")]; tensor var_6607_to_fp16 = const()[name = tensor("op_6607_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_629_cast_fp16 = mul(x = var_6606_cast_fp16, y = var_6607_to_fp16)[name = tensor("aw_chunk_629_cast_fp16")]; tensor var_6610_equation_0 = const()[name = tensor("op_6610_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6610_cast_fp16 = einsum(equation = var_6610_equation_0, values = (var_6452_cast_fp16, var_6399_cast_fp16))[name = tensor("op_6610_cast_fp16")]; tensor var_6611_to_fp16 = const()[name = tensor("op_6611_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_631_cast_fp16 = mul(x = var_6610_cast_fp16, y = var_6611_to_fp16)[name = tensor("aw_chunk_631_cast_fp16")]; tensor var_6614_equation_0 = const()[name = tensor("op_6614_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6614_cast_fp16 = einsum(equation = var_6614_equation_0, values = (var_6452_cast_fp16, var_6400_cast_fp16))[name = tensor("op_6614_cast_fp16")]; tensor var_6615_to_fp16 = const()[name = tensor("op_6615_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_633_cast_fp16 = mul(x = var_6614_cast_fp16, y = var_6615_to_fp16)[name = tensor("aw_chunk_633_cast_fp16")]; tensor var_6618_equation_0 = const()[name = tensor("op_6618_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6618_cast_fp16 = einsum(equation = var_6618_equation_0, values = (var_6452_cast_fp16, var_6401_cast_fp16))[name = tensor("op_6618_cast_fp16")]; tensor var_6619_to_fp16 = const()[name = tensor("op_6619_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_635_cast_fp16 = mul(x = var_6618_cast_fp16, y = var_6619_to_fp16)[name = tensor("aw_chunk_635_cast_fp16")]; tensor var_6622_equation_0 = const()[name = tensor("op_6622_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6622_cast_fp16 = einsum(equation = var_6622_equation_0, values = (var_6452_cast_fp16, var_6402_cast_fp16))[name = tensor("op_6622_cast_fp16")]; tensor var_6623_to_fp16 = const()[name = tensor("op_6623_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_637_cast_fp16 = mul(x = var_6622_cast_fp16, y = var_6623_to_fp16)[name = tensor("aw_chunk_637_cast_fp16")]; tensor var_6626_equation_0 = const()[name = tensor("op_6626_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6626_cast_fp16 = einsum(equation = var_6626_equation_0, values = (var_6452_cast_fp16, var_6403_cast_fp16))[name = tensor("op_6626_cast_fp16")]; tensor var_6627_to_fp16 = const()[name = tensor("op_6627_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_639_cast_fp16 = mul(x = var_6626_cast_fp16, y = var_6627_to_fp16)[name = tensor("aw_chunk_639_cast_fp16")]; tensor var_6630_equation_0 = const()[name = tensor("op_6630_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6630_cast_fp16 = einsum(equation = var_6630_equation_0, values = (var_6456_cast_fp16, var_6404_cast_fp16))[name = tensor("op_6630_cast_fp16")]; tensor var_6631_to_fp16 = const()[name = tensor("op_6631_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_641_cast_fp16 = mul(x = var_6630_cast_fp16, y = var_6631_to_fp16)[name = tensor("aw_chunk_641_cast_fp16")]; tensor var_6634_equation_0 = const()[name = tensor("op_6634_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6634_cast_fp16 = einsum(equation = var_6634_equation_0, values = (var_6456_cast_fp16, var_6405_cast_fp16))[name = tensor("op_6634_cast_fp16")]; tensor var_6635_to_fp16 = const()[name = tensor("op_6635_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_643_cast_fp16 = mul(x = var_6634_cast_fp16, y = var_6635_to_fp16)[name = tensor("aw_chunk_643_cast_fp16")]; tensor var_6638_equation_0 = const()[name = tensor("op_6638_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6638_cast_fp16 = einsum(equation = var_6638_equation_0, values = (var_6456_cast_fp16, var_6406_cast_fp16))[name = tensor("op_6638_cast_fp16")]; tensor var_6639_to_fp16 = const()[name = tensor("op_6639_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_645_cast_fp16 = mul(x = var_6638_cast_fp16, y = var_6639_to_fp16)[name = tensor("aw_chunk_645_cast_fp16")]; tensor var_6642_equation_0 = const()[name = tensor("op_6642_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6642_cast_fp16 = einsum(equation = var_6642_equation_0, values = (var_6456_cast_fp16, var_6407_cast_fp16))[name = tensor("op_6642_cast_fp16")]; tensor var_6643_to_fp16 = const()[name = tensor("op_6643_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_647_cast_fp16 = mul(x = var_6642_cast_fp16, y = var_6643_to_fp16)[name = tensor("aw_chunk_647_cast_fp16")]; tensor var_6646_equation_0 = const()[name = tensor("op_6646_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6646_cast_fp16 = einsum(equation = var_6646_equation_0, values = (var_6456_cast_fp16, var_6408_cast_fp16))[name = tensor("op_6646_cast_fp16")]; tensor var_6647_to_fp16 = const()[name = tensor("op_6647_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_649_cast_fp16 = mul(x = var_6646_cast_fp16, y = var_6647_to_fp16)[name = tensor("aw_chunk_649_cast_fp16")]; tensor var_6650_equation_0 = const()[name = tensor("op_6650_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6650_cast_fp16 = einsum(equation = var_6650_equation_0, values = (var_6456_cast_fp16, var_6409_cast_fp16))[name = tensor("op_6650_cast_fp16")]; tensor var_6651_to_fp16 = const()[name = tensor("op_6651_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_651_cast_fp16 = mul(x = var_6650_cast_fp16, y = var_6651_to_fp16)[name = tensor("aw_chunk_651_cast_fp16")]; tensor var_6654_equation_0 = const()[name = tensor("op_6654_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6654_cast_fp16 = einsum(equation = var_6654_equation_0, values = (var_6456_cast_fp16, var_6410_cast_fp16))[name = tensor("op_6654_cast_fp16")]; tensor var_6655_to_fp16 = const()[name = tensor("op_6655_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_653_cast_fp16 = mul(x = var_6654_cast_fp16, y = var_6655_to_fp16)[name = tensor("aw_chunk_653_cast_fp16")]; tensor var_6658_equation_0 = const()[name = tensor("op_6658_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6658_cast_fp16 = einsum(equation = var_6658_equation_0, values = (var_6456_cast_fp16, var_6411_cast_fp16))[name = tensor("op_6658_cast_fp16")]; tensor var_6659_to_fp16 = const()[name = tensor("op_6659_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_655_cast_fp16 = mul(x = var_6658_cast_fp16, y = var_6659_to_fp16)[name = tensor("aw_chunk_655_cast_fp16")]; tensor var_6662_equation_0 = const()[name = tensor("op_6662_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6662_cast_fp16 = einsum(equation = var_6662_equation_0, values = (var_6460_cast_fp16, var_6412_cast_fp16))[name = tensor("op_6662_cast_fp16")]; tensor var_6663_to_fp16 = const()[name = tensor("op_6663_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_657_cast_fp16 = mul(x = var_6662_cast_fp16, y = var_6663_to_fp16)[name = tensor("aw_chunk_657_cast_fp16")]; tensor var_6666_equation_0 = const()[name = tensor("op_6666_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6666_cast_fp16 = einsum(equation = var_6666_equation_0, values = (var_6460_cast_fp16, var_6413_cast_fp16))[name = tensor("op_6666_cast_fp16")]; tensor var_6667_to_fp16 = const()[name = tensor("op_6667_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_659_cast_fp16 = mul(x = var_6666_cast_fp16, y = var_6667_to_fp16)[name = tensor("aw_chunk_659_cast_fp16")]; tensor var_6670_equation_0 = const()[name = tensor("op_6670_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6670_cast_fp16 = einsum(equation = var_6670_equation_0, values = (var_6460_cast_fp16, var_6414_cast_fp16))[name = tensor("op_6670_cast_fp16")]; tensor var_6671_to_fp16 = const()[name = tensor("op_6671_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_661_cast_fp16 = mul(x = var_6670_cast_fp16, y = var_6671_to_fp16)[name = tensor("aw_chunk_661_cast_fp16")]; tensor var_6674_equation_0 = const()[name = tensor("op_6674_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6674_cast_fp16 = einsum(equation = var_6674_equation_0, values = (var_6460_cast_fp16, var_6415_cast_fp16))[name = tensor("op_6674_cast_fp16")]; tensor var_6675_to_fp16 = const()[name = tensor("op_6675_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_663_cast_fp16 = mul(x = var_6674_cast_fp16, y = var_6675_to_fp16)[name = tensor("aw_chunk_663_cast_fp16")]; tensor var_6678_equation_0 = const()[name = tensor("op_6678_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6678_cast_fp16 = einsum(equation = var_6678_equation_0, values = (var_6460_cast_fp16, var_6416_cast_fp16))[name = tensor("op_6678_cast_fp16")]; tensor var_6679_to_fp16 = const()[name = tensor("op_6679_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_665_cast_fp16 = mul(x = var_6678_cast_fp16, y = var_6679_to_fp16)[name = tensor("aw_chunk_665_cast_fp16")]; tensor var_6682_equation_0 = const()[name = tensor("op_6682_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6682_cast_fp16 = einsum(equation = var_6682_equation_0, values = (var_6460_cast_fp16, var_6417_cast_fp16))[name = tensor("op_6682_cast_fp16")]; tensor var_6683_to_fp16 = const()[name = tensor("op_6683_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_667_cast_fp16 = mul(x = var_6682_cast_fp16, y = var_6683_to_fp16)[name = tensor("aw_chunk_667_cast_fp16")]; tensor var_6686_equation_0 = const()[name = tensor("op_6686_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6686_cast_fp16 = einsum(equation = var_6686_equation_0, values = (var_6460_cast_fp16, var_6418_cast_fp16))[name = tensor("op_6686_cast_fp16")]; tensor var_6687_to_fp16 = const()[name = tensor("op_6687_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_669_cast_fp16 = mul(x = var_6686_cast_fp16, y = var_6687_to_fp16)[name = tensor("aw_chunk_669_cast_fp16")]; tensor var_6690_equation_0 = const()[name = tensor("op_6690_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6690_cast_fp16 = einsum(equation = var_6690_equation_0, values = (var_6460_cast_fp16, var_6419_cast_fp16))[name = tensor("op_6690_cast_fp16")]; tensor var_6691_to_fp16 = const()[name = tensor("op_6691_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_671_cast_fp16 = mul(x = var_6690_cast_fp16, y = var_6691_to_fp16)[name = tensor("aw_chunk_671_cast_fp16")]; tensor var_6694_equation_0 = const()[name = tensor("op_6694_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6694_cast_fp16 = einsum(equation = var_6694_equation_0, values = (var_6464_cast_fp16, var_6420_cast_fp16))[name = tensor("op_6694_cast_fp16")]; tensor var_6695_to_fp16 = const()[name = tensor("op_6695_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_673_cast_fp16 = mul(x = var_6694_cast_fp16, y = var_6695_to_fp16)[name = tensor("aw_chunk_673_cast_fp16")]; tensor var_6698_equation_0 = const()[name = tensor("op_6698_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6698_cast_fp16 = einsum(equation = var_6698_equation_0, values = (var_6464_cast_fp16, var_6421_cast_fp16))[name = tensor("op_6698_cast_fp16")]; tensor var_6699_to_fp16 = const()[name = tensor("op_6699_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_675_cast_fp16 = mul(x = var_6698_cast_fp16, y = var_6699_to_fp16)[name = tensor("aw_chunk_675_cast_fp16")]; tensor var_6702_equation_0 = const()[name = tensor("op_6702_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6702_cast_fp16 = einsum(equation = var_6702_equation_0, values = (var_6464_cast_fp16, var_6422_cast_fp16))[name = tensor("op_6702_cast_fp16")]; tensor var_6703_to_fp16 = const()[name = tensor("op_6703_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_677_cast_fp16 = mul(x = var_6702_cast_fp16, y = var_6703_to_fp16)[name = tensor("aw_chunk_677_cast_fp16")]; tensor var_6706_equation_0 = const()[name = tensor("op_6706_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6706_cast_fp16 = einsum(equation = var_6706_equation_0, values = (var_6464_cast_fp16, var_6423_cast_fp16))[name = tensor("op_6706_cast_fp16")]; tensor var_6707_to_fp16 = const()[name = tensor("op_6707_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_679_cast_fp16 = mul(x = var_6706_cast_fp16, y = var_6707_to_fp16)[name = tensor("aw_chunk_679_cast_fp16")]; tensor var_6710_equation_0 = const()[name = tensor("op_6710_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6710_cast_fp16 = einsum(equation = var_6710_equation_0, values = (var_6464_cast_fp16, var_6424_cast_fp16))[name = tensor("op_6710_cast_fp16")]; tensor var_6711_to_fp16 = const()[name = tensor("op_6711_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_681_cast_fp16 = mul(x = var_6710_cast_fp16, y = var_6711_to_fp16)[name = tensor("aw_chunk_681_cast_fp16")]; tensor var_6714_equation_0 = const()[name = tensor("op_6714_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6714_cast_fp16 = einsum(equation = var_6714_equation_0, values = (var_6464_cast_fp16, var_6425_cast_fp16))[name = tensor("op_6714_cast_fp16")]; tensor var_6715_to_fp16 = const()[name = tensor("op_6715_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_683_cast_fp16 = mul(x = var_6714_cast_fp16, y = var_6715_to_fp16)[name = tensor("aw_chunk_683_cast_fp16")]; tensor var_6718_equation_0 = const()[name = tensor("op_6718_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6718_cast_fp16 = einsum(equation = var_6718_equation_0, values = (var_6464_cast_fp16, var_6426_cast_fp16))[name = tensor("op_6718_cast_fp16")]; tensor var_6719_to_fp16 = const()[name = tensor("op_6719_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_685_cast_fp16 = mul(x = var_6718_cast_fp16, y = var_6719_to_fp16)[name = tensor("aw_chunk_685_cast_fp16")]; tensor var_6722_equation_0 = const()[name = tensor("op_6722_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6722_cast_fp16 = einsum(equation = var_6722_equation_0, values = (var_6464_cast_fp16, var_6427_cast_fp16))[name = tensor("op_6722_cast_fp16")]; tensor var_6723_to_fp16 = const()[name = tensor("op_6723_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_687_cast_fp16 = mul(x = var_6722_cast_fp16, y = var_6723_to_fp16)[name = tensor("aw_chunk_687_cast_fp16")]; tensor var_6726_equation_0 = const()[name = tensor("op_6726_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6726_cast_fp16 = einsum(equation = var_6726_equation_0, values = (var_6468_cast_fp16, var_6428_cast_fp16))[name = tensor("op_6726_cast_fp16")]; tensor var_6727_to_fp16 = const()[name = tensor("op_6727_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_689_cast_fp16 = mul(x = var_6726_cast_fp16, y = var_6727_to_fp16)[name = tensor("aw_chunk_689_cast_fp16")]; tensor var_6730_equation_0 = const()[name = tensor("op_6730_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6730_cast_fp16 = einsum(equation = var_6730_equation_0, values = (var_6468_cast_fp16, var_6429_cast_fp16))[name = tensor("op_6730_cast_fp16")]; tensor var_6731_to_fp16 = const()[name = tensor("op_6731_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_691_cast_fp16 = mul(x = var_6730_cast_fp16, y = var_6731_to_fp16)[name = tensor("aw_chunk_691_cast_fp16")]; tensor var_6734_equation_0 = const()[name = tensor("op_6734_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6734_cast_fp16 = einsum(equation = var_6734_equation_0, values = (var_6468_cast_fp16, var_6430_cast_fp16))[name = tensor("op_6734_cast_fp16")]; tensor var_6735_to_fp16 = const()[name = tensor("op_6735_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_693_cast_fp16 = mul(x = var_6734_cast_fp16, y = var_6735_to_fp16)[name = tensor("aw_chunk_693_cast_fp16")]; tensor var_6738_equation_0 = const()[name = tensor("op_6738_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6738_cast_fp16 = einsum(equation = var_6738_equation_0, values = (var_6468_cast_fp16, var_6431_cast_fp16))[name = tensor("op_6738_cast_fp16")]; tensor var_6739_to_fp16 = const()[name = tensor("op_6739_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_695_cast_fp16 = mul(x = var_6738_cast_fp16, y = var_6739_to_fp16)[name = tensor("aw_chunk_695_cast_fp16")]; tensor var_6742_equation_0 = const()[name = tensor("op_6742_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6742_cast_fp16 = einsum(equation = var_6742_equation_0, values = (var_6468_cast_fp16, var_6432_cast_fp16))[name = tensor("op_6742_cast_fp16")]; tensor var_6743_to_fp16 = const()[name = tensor("op_6743_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_697_cast_fp16 = mul(x = var_6742_cast_fp16, y = var_6743_to_fp16)[name = tensor("aw_chunk_697_cast_fp16")]; tensor var_6746_equation_0 = const()[name = tensor("op_6746_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6746_cast_fp16 = einsum(equation = var_6746_equation_0, values = (var_6468_cast_fp16, var_6433_cast_fp16))[name = tensor("op_6746_cast_fp16")]; tensor var_6747_to_fp16 = const()[name = tensor("op_6747_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_699_cast_fp16 = mul(x = var_6746_cast_fp16, y = var_6747_to_fp16)[name = tensor("aw_chunk_699_cast_fp16")]; tensor var_6750_equation_0 = const()[name = tensor("op_6750_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6750_cast_fp16 = einsum(equation = var_6750_equation_0, values = (var_6468_cast_fp16, var_6434_cast_fp16))[name = tensor("op_6750_cast_fp16")]; tensor var_6751_to_fp16 = const()[name = tensor("op_6751_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_701_cast_fp16 = mul(x = var_6750_cast_fp16, y = var_6751_to_fp16)[name = tensor("aw_chunk_701_cast_fp16")]; tensor var_6754_equation_0 = const()[name = tensor("op_6754_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6754_cast_fp16 = einsum(equation = var_6754_equation_0, values = (var_6468_cast_fp16, var_6435_cast_fp16))[name = tensor("op_6754_cast_fp16")]; tensor var_6755_to_fp16 = const()[name = tensor("op_6755_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_703_cast_fp16 = mul(x = var_6754_cast_fp16, y = var_6755_to_fp16)[name = tensor("aw_chunk_703_cast_fp16")]; tensor var_6757_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_577_cast_fp16)[name = tensor("op_6757_cast_fp16")]; tensor var_6758_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_579_cast_fp16)[name = tensor("op_6758_cast_fp16")]; tensor var_6759_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_581_cast_fp16)[name = tensor("op_6759_cast_fp16")]; tensor var_6760_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_583_cast_fp16)[name = tensor("op_6760_cast_fp16")]; tensor var_6761_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_585_cast_fp16)[name = tensor("op_6761_cast_fp16")]; tensor var_6762_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_587_cast_fp16)[name = tensor("op_6762_cast_fp16")]; tensor var_6763_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_589_cast_fp16)[name = tensor("op_6763_cast_fp16")]; tensor var_6764_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_591_cast_fp16)[name = tensor("op_6764_cast_fp16")]; tensor var_6765_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_593_cast_fp16)[name = tensor("op_6765_cast_fp16")]; tensor var_6766_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_595_cast_fp16)[name = tensor("op_6766_cast_fp16")]; tensor var_6767_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_597_cast_fp16)[name = tensor("op_6767_cast_fp16")]; tensor var_6768_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_599_cast_fp16)[name = tensor("op_6768_cast_fp16")]; tensor var_6769_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_601_cast_fp16)[name = tensor("op_6769_cast_fp16")]; tensor var_6770_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_603_cast_fp16)[name = tensor("op_6770_cast_fp16")]; tensor var_6771_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_605_cast_fp16)[name = tensor("op_6771_cast_fp16")]; tensor var_6772_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_607_cast_fp16)[name = tensor("op_6772_cast_fp16")]; tensor var_6773_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_609_cast_fp16)[name = tensor("op_6773_cast_fp16")]; tensor var_6774_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_611_cast_fp16)[name = tensor("op_6774_cast_fp16")]; tensor var_6775_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_613_cast_fp16)[name = tensor("op_6775_cast_fp16")]; tensor var_6776_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_615_cast_fp16)[name = tensor("op_6776_cast_fp16")]; tensor var_6777_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_617_cast_fp16)[name = tensor("op_6777_cast_fp16")]; tensor var_6778_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_619_cast_fp16)[name = tensor("op_6778_cast_fp16")]; tensor var_6779_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_621_cast_fp16)[name = tensor("op_6779_cast_fp16")]; tensor var_6780_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_623_cast_fp16)[name = tensor("op_6780_cast_fp16")]; tensor var_6781_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_625_cast_fp16)[name = tensor("op_6781_cast_fp16")]; tensor var_6782_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_627_cast_fp16)[name = tensor("op_6782_cast_fp16")]; tensor var_6783_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_629_cast_fp16)[name = tensor("op_6783_cast_fp16")]; tensor var_6784_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_631_cast_fp16)[name = tensor("op_6784_cast_fp16")]; tensor var_6785_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_633_cast_fp16)[name = tensor("op_6785_cast_fp16")]; tensor var_6786_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_635_cast_fp16)[name = tensor("op_6786_cast_fp16")]; tensor var_6787_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_637_cast_fp16)[name = tensor("op_6787_cast_fp16")]; tensor var_6788_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_639_cast_fp16)[name = tensor("op_6788_cast_fp16")]; tensor var_6789_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_641_cast_fp16)[name = tensor("op_6789_cast_fp16")]; tensor var_6790_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_643_cast_fp16)[name = tensor("op_6790_cast_fp16")]; tensor var_6791_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_645_cast_fp16)[name = tensor("op_6791_cast_fp16")]; tensor var_6792_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_647_cast_fp16)[name = tensor("op_6792_cast_fp16")]; tensor var_6793_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_649_cast_fp16)[name = tensor("op_6793_cast_fp16")]; tensor var_6794_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_651_cast_fp16)[name = tensor("op_6794_cast_fp16")]; tensor var_6795_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_653_cast_fp16)[name = tensor("op_6795_cast_fp16")]; tensor var_6796_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_655_cast_fp16)[name = tensor("op_6796_cast_fp16")]; tensor var_6797_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_657_cast_fp16)[name = tensor("op_6797_cast_fp16")]; tensor var_6798_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_659_cast_fp16)[name = tensor("op_6798_cast_fp16")]; tensor var_6799_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_661_cast_fp16)[name = tensor("op_6799_cast_fp16")]; tensor var_6800_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_663_cast_fp16)[name = tensor("op_6800_cast_fp16")]; tensor var_6801_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_665_cast_fp16)[name = tensor("op_6801_cast_fp16")]; tensor var_6802_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_667_cast_fp16)[name = tensor("op_6802_cast_fp16")]; tensor var_6803_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_669_cast_fp16)[name = tensor("op_6803_cast_fp16")]; tensor var_6804_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_671_cast_fp16)[name = tensor("op_6804_cast_fp16")]; tensor var_6805_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_673_cast_fp16)[name = tensor("op_6805_cast_fp16")]; tensor var_6806_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_675_cast_fp16)[name = tensor("op_6806_cast_fp16")]; tensor var_6807_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_677_cast_fp16)[name = tensor("op_6807_cast_fp16")]; tensor var_6808_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_679_cast_fp16)[name = tensor("op_6808_cast_fp16")]; tensor var_6809_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_681_cast_fp16)[name = tensor("op_6809_cast_fp16")]; tensor var_6810_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_683_cast_fp16)[name = tensor("op_6810_cast_fp16")]; tensor var_6811_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_685_cast_fp16)[name = tensor("op_6811_cast_fp16")]; tensor var_6812_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_687_cast_fp16)[name = tensor("op_6812_cast_fp16")]; tensor var_6813_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_689_cast_fp16)[name = tensor("op_6813_cast_fp16")]; tensor var_6814_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_691_cast_fp16)[name = tensor("op_6814_cast_fp16")]; tensor var_6815_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_693_cast_fp16)[name = tensor("op_6815_cast_fp16")]; tensor var_6816_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_695_cast_fp16)[name = tensor("op_6816_cast_fp16")]; tensor var_6817_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_697_cast_fp16)[name = tensor("op_6817_cast_fp16")]; tensor var_6818_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_699_cast_fp16)[name = tensor("op_6818_cast_fp16")]; tensor var_6819_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_701_cast_fp16)[name = tensor("op_6819_cast_fp16")]; tensor var_6820_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_703_cast_fp16)[name = tensor("op_6820_cast_fp16")]; tensor var_6822_equation_0 = const()[name = tensor("op_6822_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6822_cast_fp16 = einsum(equation = var_6822_equation_0, values = (var_6470_cast_fp16, var_6757_cast_fp16))[name = tensor("op_6822_cast_fp16")]; tensor var_6824_equation_0 = const()[name = tensor("op_6824_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6824_cast_fp16 = einsum(equation = var_6824_equation_0, values = (var_6470_cast_fp16, var_6758_cast_fp16))[name = tensor("op_6824_cast_fp16")]; tensor var_6826_equation_0 = const()[name = tensor("op_6826_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6826_cast_fp16 = einsum(equation = var_6826_equation_0, values = (var_6470_cast_fp16, var_6759_cast_fp16))[name = tensor("op_6826_cast_fp16")]; tensor var_6828_equation_0 = const()[name = tensor("op_6828_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6828_cast_fp16 = einsum(equation = var_6828_equation_0, values = (var_6470_cast_fp16, var_6760_cast_fp16))[name = tensor("op_6828_cast_fp16")]; tensor var_6830_equation_0 = const()[name = tensor("op_6830_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6830_cast_fp16 = einsum(equation = var_6830_equation_0, values = (var_6470_cast_fp16, var_6761_cast_fp16))[name = tensor("op_6830_cast_fp16")]; tensor var_6832_equation_0 = const()[name = tensor("op_6832_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6832_cast_fp16 = einsum(equation = var_6832_equation_0, values = (var_6470_cast_fp16, var_6762_cast_fp16))[name = tensor("op_6832_cast_fp16")]; tensor var_6834_equation_0 = const()[name = tensor("op_6834_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6834_cast_fp16 = einsum(equation = var_6834_equation_0, values = (var_6470_cast_fp16, var_6763_cast_fp16))[name = tensor("op_6834_cast_fp16")]; tensor var_6836_equation_0 = const()[name = tensor("op_6836_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6836_cast_fp16 = einsum(equation = var_6836_equation_0, values = (var_6470_cast_fp16, var_6764_cast_fp16))[name = tensor("op_6836_cast_fp16")]; tensor var_6838_equation_0 = const()[name = tensor("op_6838_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6838_cast_fp16 = einsum(equation = var_6838_equation_0, values = (var_6474_cast_fp16, var_6765_cast_fp16))[name = tensor("op_6838_cast_fp16")]; tensor var_6840_equation_0 = const()[name = tensor("op_6840_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6840_cast_fp16 = einsum(equation = var_6840_equation_0, values = (var_6474_cast_fp16, var_6766_cast_fp16))[name = tensor("op_6840_cast_fp16")]; tensor var_6842_equation_0 = const()[name = tensor("op_6842_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6842_cast_fp16 = einsum(equation = var_6842_equation_0, values = (var_6474_cast_fp16, var_6767_cast_fp16))[name = tensor("op_6842_cast_fp16")]; tensor var_6844_equation_0 = const()[name = tensor("op_6844_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6844_cast_fp16 = einsum(equation = var_6844_equation_0, values = (var_6474_cast_fp16, var_6768_cast_fp16))[name = tensor("op_6844_cast_fp16")]; tensor var_6846_equation_0 = const()[name = tensor("op_6846_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6846_cast_fp16 = einsum(equation = var_6846_equation_0, values = (var_6474_cast_fp16, var_6769_cast_fp16))[name = tensor("op_6846_cast_fp16")]; tensor var_6848_equation_0 = const()[name = tensor("op_6848_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6848_cast_fp16 = einsum(equation = var_6848_equation_0, values = (var_6474_cast_fp16, var_6770_cast_fp16))[name = tensor("op_6848_cast_fp16")]; tensor var_6850_equation_0 = const()[name = tensor("op_6850_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6850_cast_fp16 = einsum(equation = var_6850_equation_0, values = (var_6474_cast_fp16, var_6771_cast_fp16))[name = tensor("op_6850_cast_fp16")]; tensor var_6852_equation_0 = const()[name = tensor("op_6852_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6852_cast_fp16 = einsum(equation = var_6852_equation_0, values = (var_6474_cast_fp16, var_6772_cast_fp16))[name = tensor("op_6852_cast_fp16")]; tensor var_6854_equation_0 = const()[name = tensor("op_6854_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6854_cast_fp16 = einsum(equation = var_6854_equation_0, values = (var_6478_cast_fp16, var_6773_cast_fp16))[name = tensor("op_6854_cast_fp16")]; tensor var_6856_equation_0 = const()[name = tensor("op_6856_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6856_cast_fp16 = einsum(equation = var_6856_equation_0, values = (var_6478_cast_fp16, var_6774_cast_fp16))[name = tensor("op_6856_cast_fp16")]; tensor var_6858_equation_0 = const()[name = tensor("op_6858_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6858_cast_fp16 = einsum(equation = var_6858_equation_0, values = (var_6478_cast_fp16, var_6775_cast_fp16))[name = tensor("op_6858_cast_fp16")]; tensor var_6860_equation_0 = const()[name = tensor("op_6860_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6860_cast_fp16 = einsum(equation = var_6860_equation_0, values = (var_6478_cast_fp16, var_6776_cast_fp16))[name = tensor("op_6860_cast_fp16")]; tensor var_6862_equation_0 = const()[name = tensor("op_6862_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6862_cast_fp16 = einsum(equation = var_6862_equation_0, values = (var_6478_cast_fp16, var_6777_cast_fp16))[name = tensor("op_6862_cast_fp16")]; tensor var_6864_equation_0 = const()[name = tensor("op_6864_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6864_cast_fp16 = einsum(equation = var_6864_equation_0, values = (var_6478_cast_fp16, var_6778_cast_fp16))[name = tensor("op_6864_cast_fp16")]; tensor var_6866_equation_0 = const()[name = tensor("op_6866_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6866_cast_fp16 = einsum(equation = var_6866_equation_0, values = (var_6478_cast_fp16, var_6779_cast_fp16))[name = tensor("op_6866_cast_fp16")]; tensor var_6868_equation_0 = const()[name = tensor("op_6868_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6868_cast_fp16 = einsum(equation = var_6868_equation_0, values = (var_6478_cast_fp16, var_6780_cast_fp16))[name = tensor("op_6868_cast_fp16")]; tensor var_6870_equation_0 = const()[name = tensor("op_6870_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6870_cast_fp16 = einsum(equation = var_6870_equation_0, values = (var_6482_cast_fp16, var_6781_cast_fp16))[name = tensor("op_6870_cast_fp16")]; tensor var_6872_equation_0 = const()[name = tensor("op_6872_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6872_cast_fp16 = einsum(equation = var_6872_equation_0, values = (var_6482_cast_fp16, var_6782_cast_fp16))[name = tensor("op_6872_cast_fp16")]; tensor var_6874_equation_0 = const()[name = tensor("op_6874_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6874_cast_fp16 = einsum(equation = var_6874_equation_0, values = (var_6482_cast_fp16, var_6783_cast_fp16))[name = tensor("op_6874_cast_fp16")]; tensor var_6876_equation_0 = const()[name = tensor("op_6876_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6876_cast_fp16 = einsum(equation = var_6876_equation_0, values = (var_6482_cast_fp16, var_6784_cast_fp16))[name = tensor("op_6876_cast_fp16")]; tensor var_6878_equation_0 = const()[name = tensor("op_6878_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6878_cast_fp16 = einsum(equation = var_6878_equation_0, values = (var_6482_cast_fp16, var_6785_cast_fp16))[name = tensor("op_6878_cast_fp16")]; tensor var_6880_equation_0 = const()[name = tensor("op_6880_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6880_cast_fp16 = einsum(equation = var_6880_equation_0, values = (var_6482_cast_fp16, var_6786_cast_fp16))[name = tensor("op_6880_cast_fp16")]; tensor var_6882_equation_0 = const()[name = tensor("op_6882_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6882_cast_fp16 = einsum(equation = var_6882_equation_0, values = (var_6482_cast_fp16, var_6787_cast_fp16))[name = tensor("op_6882_cast_fp16")]; tensor var_6884_equation_0 = const()[name = tensor("op_6884_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6884_cast_fp16 = einsum(equation = var_6884_equation_0, values = (var_6482_cast_fp16, var_6788_cast_fp16))[name = tensor("op_6884_cast_fp16")]; tensor var_6886_equation_0 = const()[name = tensor("op_6886_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6886_cast_fp16 = einsum(equation = var_6886_equation_0, values = (var_6486_cast_fp16, var_6789_cast_fp16))[name = tensor("op_6886_cast_fp16")]; tensor var_6888_equation_0 = const()[name = tensor("op_6888_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6888_cast_fp16 = einsum(equation = var_6888_equation_0, values = (var_6486_cast_fp16, var_6790_cast_fp16))[name = tensor("op_6888_cast_fp16")]; tensor var_6890_equation_0 = const()[name = tensor("op_6890_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6890_cast_fp16 = einsum(equation = var_6890_equation_0, values = (var_6486_cast_fp16, var_6791_cast_fp16))[name = tensor("op_6890_cast_fp16")]; tensor var_6892_equation_0 = const()[name = tensor("op_6892_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6892_cast_fp16 = einsum(equation = var_6892_equation_0, values = (var_6486_cast_fp16, var_6792_cast_fp16))[name = tensor("op_6892_cast_fp16")]; tensor var_6894_equation_0 = const()[name = tensor("op_6894_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6894_cast_fp16 = einsum(equation = var_6894_equation_0, values = (var_6486_cast_fp16, var_6793_cast_fp16))[name = tensor("op_6894_cast_fp16")]; tensor var_6896_equation_0 = const()[name = tensor("op_6896_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6896_cast_fp16 = einsum(equation = var_6896_equation_0, values = (var_6486_cast_fp16, var_6794_cast_fp16))[name = tensor("op_6896_cast_fp16")]; tensor var_6898_equation_0 = const()[name = tensor("op_6898_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6898_cast_fp16 = einsum(equation = var_6898_equation_0, values = (var_6486_cast_fp16, var_6795_cast_fp16))[name = tensor("op_6898_cast_fp16")]; tensor var_6900_equation_0 = const()[name = tensor("op_6900_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6900_cast_fp16 = einsum(equation = var_6900_equation_0, values = (var_6486_cast_fp16, var_6796_cast_fp16))[name = tensor("op_6900_cast_fp16")]; tensor var_6902_equation_0 = const()[name = tensor("op_6902_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6902_cast_fp16 = einsum(equation = var_6902_equation_0, values = (var_6490_cast_fp16, var_6797_cast_fp16))[name = tensor("op_6902_cast_fp16")]; tensor var_6904_equation_0 = const()[name = tensor("op_6904_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6904_cast_fp16 = einsum(equation = var_6904_equation_0, values = (var_6490_cast_fp16, var_6798_cast_fp16))[name = tensor("op_6904_cast_fp16")]; tensor var_6906_equation_0 = const()[name = tensor("op_6906_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6906_cast_fp16 = einsum(equation = var_6906_equation_0, values = (var_6490_cast_fp16, var_6799_cast_fp16))[name = tensor("op_6906_cast_fp16")]; tensor var_6908_equation_0 = const()[name = tensor("op_6908_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6908_cast_fp16 = einsum(equation = var_6908_equation_0, values = (var_6490_cast_fp16, var_6800_cast_fp16))[name = tensor("op_6908_cast_fp16")]; tensor var_6910_equation_0 = const()[name = tensor("op_6910_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6910_cast_fp16 = einsum(equation = var_6910_equation_0, values = (var_6490_cast_fp16, var_6801_cast_fp16))[name = tensor("op_6910_cast_fp16")]; tensor var_6912_equation_0 = const()[name = tensor("op_6912_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6912_cast_fp16 = einsum(equation = var_6912_equation_0, values = (var_6490_cast_fp16, var_6802_cast_fp16))[name = tensor("op_6912_cast_fp16")]; tensor var_6914_equation_0 = const()[name = tensor("op_6914_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6914_cast_fp16 = einsum(equation = var_6914_equation_0, values = (var_6490_cast_fp16, var_6803_cast_fp16))[name = tensor("op_6914_cast_fp16")]; tensor var_6916_equation_0 = const()[name = tensor("op_6916_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6916_cast_fp16 = einsum(equation = var_6916_equation_0, values = (var_6490_cast_fp16, var_6804_cast_fp16))[name = tensor("op_6916_cast_fp16")]; tensor var_6918_equation_0 = const()[name = tensor("op_6918_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6918_cast_fp16 = einsum(equation = var_6918_equation_0, values = (var_6494_cast_fp16, var_6805_cast_fp16))[name = tensor("op_6918_cast_fp16")]; tensor var_6920_equation_0 = const()[name = tensor("op_6920_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6920_cast_fp16 = einsum(equation = var_6920_equation_0, values = (var_6494_cast_fp16, var_6806_cast_fp16))[name = tensor("op_6920_cast_fp16")]; tensor var_6922_equation_0 = const()[name = tensor("op_6922_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6922_cast_fp16 = einsum(equation = var_6922_equation_0, values = (var_6494_cast_fp16, var_6807_cast_fp16))[name = tensor("op_6922_cast_fp16")]; tensor var_6924_equation_0 = const()[name = tensor("op_6924_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6924_cast_fp16 = einsum(equation = var_6924_equation_0, values = (var_6494_cast_fp16, var_6808_cast_fp16))[name = tensor("op_6924_cast_fp16")]; tensor var_6926_equation_0 = const()[name = tensor("op_6926_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6926_cast_fp16 = einsum(equation = var_6926_equation_0, values = (var_6494_cast_fp16, var_6809_cast_fp16))[name = tensor("op_6926_cast_fp16")]; tensor var_6928_equation_0 = const()[name = tensor("op_6928_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6928_cast_fp16 = einsum(equation = var_6928_equation_0, values = (var_6494_cast_fp16, var_6810_cast_fp16))[name = tensor("op_6928_cast_fp16")]; tensor var_6930_equation_0 = const()[name = tensor("op_6930_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6930_cast_fp16 = einsum(equation = var_6930_equation_0, values = (var_6494_cast_fp16, var_6811_cast_fp16))[name = tensor("op_6930_cast_fp16")]; tensor var_6932_equation_0 = const()[name = tensor("op_6932_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6932_cast_fp16 = einsum(equation = var_6932_equation_0, values = (var_6494_cast_fp16, var_6812_cast_fp16))[name = tensor("op_6932_cast_fp16")]; tensor var_6934_equation_0 = const()[name = tensor("op_6934_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6934_cast_fp16 = einsum(equation = var_6934_equation_0, values = (var_6498_cast_fp16, var_6813_cast_fp16))[name = tensor("op_6934_cast_fp16")]; tensor var_6936_equation_0 = const()[name = tensor("op_6936_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6936_cast_fp16 = einsum(equation = var_6936_equation_0, values = (var_6498_cast_fp16, var_6814_cast_fp16))[name = tensor("op_6936_cast_fp16")]; tensor var_6938_equation_0 = const()[name = tensor("op_6938_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6938_cast_fp16 = einsum(equation = var_6938_equation_0, values = (var_6498_cast_fp16, var_6815_cast_fp16))[name = tensor("op_6938_cast_fp16")]; tensor var_6940_equation_0 = const()[name = tensor("op_6940_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6940_cast_fp16 = einsum(equation = var_6940_equation_0, values = (var_6498_cast_fp16, var_6816_cast_fp16))[name = tensor("op_6940_cast_fp16")]; tensor var_6942_equation_0 = const()[name = tensor("op_6942_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6942_cast_fp16 = einsum(equation = var_6942_equation_0, values = (var_6498_cast_fp16, var_6817_cast_fp16))[name = tensor("op_6942_cast_fp16")]; tensor var_6944_equation_0 = const()[name = tensor("op_6944_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6944_cast_fp16 = einsum(equation = var_6944_equation_0, values = (var_6498_cast_fp16, var_6818_cast_fp16))[name = tensor("op_6944_cast_fp16")]; tensor var_6946_equation_0 = const()[name = tensor("op_6946_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6946_cast_fp16 = einsum(equation = var_6946_equation_0, values = (var_6498_cast_fp16, var_6819_cast_fp16))[name = tensor("op_6946_cast_fp16")]; tensor var_6948_equation_0 = const()[name = tensor("op_6948_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6948_cast_fp16 = einsum(equation = var_6948_equation_0, values = (var_6498_cast_fp16, var_6820_cast_fp16))[name = tensor("op_6948_cast_fp16")]; tensor var_6950_interleave_0 = const()[name = tensor("op_6950_interleave_0"), val = tensor(false)]; tensor var_6950_cast_fp16 = concat(axis = var_5509, interleave = var_6950_interleave_0, values = (var_6822_cast_fp16, var_6824_cast_fp16, var_6826_cast_fp16, var_6828_cast_fp16, var_6830_cast_fp16, var_6832_cast_fp16, var_6834_cast_fp16, var_6836_cast_fp16))[name = tensor("op_6950_cast_fp16")]; tensor var_6952_interleave_0 = const()[name = tensor("op_6952_interleave_0"), val = tensor(false)]; tensor var_6952_cast_fp16 = concat(axis = var_5509, interleave = var_6952_interleave_0, values = (var_6838_cast_fp16, var_6840_cast_fp16, var_6842_cast_fp16, var_6844_cast_fp16, var_6846_cast_fp16, var_6848_cast_fp16, var_6850_cast_fp16, var_6852_cast_fp16))[name = tensor("op_6952_cast_fp16")]; tensor var_6954_interleave_0 = const()[name = tensor("op_6954_interleave_0"), val = tensor(false)]; tensor var_6954_cast_fp16 = concat(axis = var_5509, interleave = var_6954_interleave_0, values = (var_6854_cast_fp16, var_6856_cast_fp16, var_6858_cast_fp16, var_6860_cast_fp16, var_6862_cast_fp16, var_6864_cast_fp16, var_6866_cast_fp16, var_6868_cast_fp16))[name = tensor("op_6954_cast_fp16")]; tensor var_6956_interleave_0 = const()[name = tensor("op_6956_interleave_0"), val = tensor(false)]; tensor var_6956_cast_fp16 = concat(axis = var_5509, interleave = var_6956_interleave_0, values = (var_6870_cast_fp16, var_6872_cast_fp16, var_6874_cast_fp16, var_6876_cast_fp16, var_6878_cast_fp16, var_6880_cast_fp16, var_6882_cast_fp16, var_6884_cast_fp16))[name = tensor("op_6956_cast_fp16")]; tensor var_6958_interleave_0 = const()[name = tensor("op_6958_interleave_0"), val = tensor(false)]; tensor var_6958_cast_fp16 = concat(axis = var_5509, interleave = var_6958_interleave_0, values = (var_6886_cast_fp16, var_6888_cast_fp16, var_6890_cast_fp16, var_6892_cast_fp16, var_6894_cast_fp16, var_6896_cast_fp16, var_6898_cast_fp16, var_6900_cast_fp16))[name = tensor("op_6958_cast_fp16")]; tensor var_6960_interleave_0 = const()[name = tensor("op_6960_interleave_0"), val = tensor(false)]; tensor var_6960_cast_fp16 = concat(axis = var_5509, interleave = var_6960_interleave_0, values = (var_6902_cast_fp16, var_6904_cast_fp16, var_6906_cast_fp16, var_6908_cast_fp16, var_6910_cast_fp16, var_6912_cast_fp16, var_6914_cast_fp16, var_6916_cast_fp16))[name = tensor("op_6960_cast_fp16")]; tensor var_6962_interleave_0 = const()[name = tensor("op_6962_interleave_0"), val = tensor(false)]; tensor var_6962_cast_fp16 = concat(axis = var_5509, interleave = var_6962_interleave_0, values = (var_6918_cast_fp16, var_6920_cast_fp16, var_6922_cast_fp16, var_6924_cast_fp16, var_6926_cast_fp16, var_6928_cast_fp16, var_6930_cast_fp16, var_6932_cast_fp16))[name = tensor("op_6962_cast_fp16")]; tensor var_6964_interleave_0 = const()[name = tensor("op_6964_interleave_0"), val = tensor(false)]; tensor var_6964_cast_fp16 = concat(axis = var_5509, interleave = var_6964_interleave_0, values = (var_6934_cast_fp16, var_6936_cast_fp16, var_6938_cast_fp16, var_6940_cast_fp16, var_6942_cast_fp16, var_6944_cast_fp16, var_6946_cast_fp16, var_6948_cast_fp16))[name = tensor("op_6964_cast_fp16")]; tensor input_221_interleave_0 = const()[name = tensor("input_221_interleave_0"), val = tensor(false)]; tensor input_221_cast_fp16 = concat(axis = var_5537, interleave = input_221_interleave_0, values = (var_6950_cast_fp16, var_6952_cast_fp16, var_6954_cast_fp16, var_6956_cast_fp16, var_6958_cast_fp16, var_6960_cast_fp16, var_6962_cast_fp16, var_6964_cast_fp16))[name = tensor("input_221_cast_fp16")]; tensor var_6970 = const()[name = tensor("op_6970"), val = tensor([1, 1])]; tensor var_6972 = const()[name = tensor("op_6972"), val = tensor([1, 1])]; tensor var_6974_pad_type_0 = const()[name = tensor("op_6974_pad_type_0"), val = tensor("custom")]; tensor var_6974_pad_0 = const()[name = tensor("op_6974_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(237251392))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(237328256))), name = tensor("up_blocks_2_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16_palettized"), shape = tensor([320, 320, 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(237328448)))]; tensor var_6974_cast_fp16 = conv(bias = up_blocks_2_attentions_0_transformer_blocks_0_attn2_to_out_0_bias_to_fp16, dilations = var_6972, groups = var_5537, pad = var_6974_pad_0, pad_type = var_6974_pad_type_0, strides = var_6970, weight = up_blocks_2_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16_palettized, x = input_221_cast_fp16)[name = tensor("op_6974_cast_fp16")]; tensor inputs_47_cast_fp16 = add(x = var_6974_cast_fp16, y = inputs_45_cast_fp16)[name = tensor("inputs_47_cast_fp16")]; tensor var_6978 = const()[name = tensor("op_6978"), val = tensor([1])]; tensor channels_mean_47_cast_fp16 = reduce_mean(axes = var_6978, keep_dims = var_5532, x = inputs_47_cast_fp16)[name = tensor("channels_mean_47_cast_fp16")]; tensor zero_mean_47_cast_fp16 = sub(x = inputs_47_cast_fp16, y = channels_mean_47_cast_fp16)[name = tensor("zero_mean_47_cast_fp16")]; tensor zero_mean_sq_47_cast_fp16 = mul(x = zero_mean_47_cast_fp16, y = zero_mean_47_cast_fp16)[name = tensor("zero_mean_sq_47_cast_fp16")]; tensor var_6982 = const()[name = tensor("op_6982"), val = tensor([1])]; tensor var_6983_cast_fp16 = reduce_mean(axes = var_6982, keep_dims = var_5532, x = zero_mean_sq_47_cast_fp16)[name = tensor("op_6983_cast_fp16")]; tensor var_6984_to_fp16 = const()[name = tensor("op_6984_to_fp16"), val = tensor(0x1.5p-17)]; tensor var_6985_cast_fp16 = add(x = var_6983_cast_fp16, y = var_6984_to_fp16)[name = tensor("op_6985_cast_fp16")]; tensor denom_47_epsilon_0_to_fp16 = const()[name = tensor("denom_47_epsilon_0_to_fp16"), val = tensor(0x1p-24)]; tensor denom_47_cast_fp16 = rsqrt(epsilon = denom_47_epsilon_0_to_fp16, x = var_6985_cast_fp16)[name = tensor("denom_47_cast_fp16")]; tensor out_47_cast_fp16 = mul(x = zero_mean_47_cast_fp16, y = denom_47_cast_fp16)[name = tensor("out_47_cast_fp16")]; tensor var_6989_to_fp16 = const()[name = tensor("op_6989_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(237329152)))]; tensor var_6990_cast_fp16 = add(x = out_47_cast_fp16, y = var_6989_to_fp16)[name = tensor("op_6990_cast_fp16")]; tensor var_6992_to_fp16 = const()[name = tensor("op_6992_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(237329856)))]; tensor input_223_cast_fp16 = mul(x = var_6990_cast_fp16, y = var_6992_to_fp16)[name = tensor("input_223_cast_fp16")]; tensor var_7000 = const()[name = tensor("op_7000"), val = tensor([1, 1])]; tensor var_7002 = const()[name = tensor("op_7002"), val = tensor([1, 1])]; tensor var_7004_pad_type_0 = const()[name = tensor("op_7004_pad_type_0"), val = tensor("custom")]; tensor var_7004_pad_0 = const()[name = tensor("op_7004_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(237330560))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(237945024))), name = tensor("up_blocks_2_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16_palettized"), shape = tensor([2560, 320, 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(237945216))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(237947200))), name = tensor("up_blocks_2_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16_palettized"), shape = tensor([2560])]; tensor var_7004_cast_fp16 = conv(bias = up_blocks_2_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16_palettized, dilations = var_7002, groups = var_5537, pad = var_7004_pad_0, pad_type = var_7004_pad_type_0, strides = var_7000, weight = up_blocks_2_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16_palettized, x = input_223_cast_fp16)[name = tensor("op_7004_cast_fp16")]; tensor var_7005_split_sizes_0 = const()[name = tensor("op_7005_split_sizes_0"), val = tensor([1280, 1280])]; tensor var_7005_axis_0 = const()[name = tensor("op_7005_axis_0"), val = tensor(1)]; tensor var_7005_cast_fp16_0, tensor var_7005_cast_fp16_1 = split(axis = var_7005_axis_0, split_sizes = var_7005_split_sizes_0, x = var_7004_cast_fp16)[name = tensor("op_7005_cast_fp16")]; tensor var_7007_mode_0 = const()[name = tensor("op_7007_mode_0"), val = tensor("EXACT")]; tensor var_7007_cast_fp16 = gelu(mode = var_7007_mode_0, x = var_7005_cast_fp16_1)[name = tensor("op_7007_cast_fp16")]; tensor input_225_cast_fp16 = mul(x = var_7005_cast_fp16_0, y = var_7007_cast_fp16)[name = tensor("input_225_cast_fp16")]; tensor var_7011 = const()[name = tensor("op_7011"), val = tensor([1, 1])]; tensor var_7013 = const()[name = tensor("op_7013"), val = tensor([1, 1])]; tensor var_7015_pad_type_0 = const()[name = tensor("op_7015_pad_type_0"), val = tensor("custom")]; tensor var_7015_pad_0 = const()[name = tensor("op_7015_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(237947392))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(238254656))), name = tensor("up_blocks_2_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16_palettized"), shape = tensor([320, 1280, 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(238254848)))]; tensor var_7015_cast_fp16 = conv(bias = up_blocks_2_attentions_0_transformer_blocks_0_ff_net_2_bias_to_fp16, dilations = var_7013, groups = var_5537, pad = var_7015_pad_0, pad_type = var_7015_pad_type_0, strides = var_7011, weight = up_blocks_2_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16_palettized, x = input_225_cast_fp16)[name = tensor("op_7015_cast_fp16")]; tensor hidden_states_151_cast_fp16 = add(x = var_7015_cast_fp16, y = inputs_47_cast_fp16)[name = tensor("hidden_states_151_cast_fp16")]; tensor var_7017 = const()[name = tensor("op_7017"), val = tensor([2, 320, 64, 64])]; tensor input_227_cast_fp16 = reshape(shape = var_7017, x = hidden_states_151_cast_fp16)[name = tensor("input_227_cast_fp16")]; 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_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([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(238255552))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(238332416))), name = tensor("up_blocks_2_attentions_0_proj_out_weight_to_fp16_palettized"), shape = tensor([320, 320, 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(238332608)))]; tensor hidden_states_153_cast_fp16 = conv(bias = up_blocks_2_attentions_0_proj_out_bias_to_fp16, dilations = var_7023, groups = var_5537, pad = hidden_states_153_pad_0, pad_type = hidden_states_153_pad_type_0, strides = var_7021, weight = up_blocks_2_attentions_0_proj_out_weight_to_fp16_palettized, x = input_227_cast_fp16)[name = tensor("hidden_states_153_cast_fp16")]; tensor hidden_states_155_cast_fp16 = add(x = hidden_states_153_cast_fp16, y = hidden_states_141_cast_fp16)[name = tensor("hidden_states_155_cast_fp16")]; tensor input_229_interleave_0 = const()[name = tensor("input_229_interleave_0"), val = tensor(false)]; tensor input_229_cast_fp16 = concat(axis = var_5537, interleave = input_229_interleave_0, values = (hidden_states_155_cast_fp16, input_7_cast_fp16))[name = tensor("input_229_cast_fp16")]; tensor reshape_96_shape_0 = const()[name = tensor("reshape_96_shape_0"), val = tensor([2, 32, 20, 64, 64])]; tensor reshape_96_cast_fp16 = reshape(shape = reshape_96_shape_0, x = input_229_cast_fp16)[name = tensor("reshape_96_cast_fp16")]; 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_fp16 = reduce_mean(axes = reduce_mean_72_axes_0, keep_dims = reduce_mean_72_keep_dims_0, x = reshape_96_cast_fp16)[name = tensor("reduce_mean_72_cast_fp16")]; tensor sub_48_cast_fp16 = sub(x = reshape_96_cast_fp16, y = reduce_mean_72_cast_fp16)[name = tensor("sub_48_cast_fp16")]; tensor square_24_cast_fp16 = square(x = sub_48_cast_fp16)[name = tensor("square_24_cast_fp16")]; 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_fp16 = reduce_mean(axes = reduce_mean_74_axes_0, keep_dims = reduce_mean_74_keep_dims_0, x = square_24_cast_fp16)[name = tensor("reduce_mean_74_cast_fp16")]; tensor add_48_y_0_to_fp16 = const()[name = tensor("add_48_y_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor add_48_cast_fp16 = add(x = reduce_mean_74_cast_fp16, y = add_48_y_0_to_fp16)[name = tensor("add_48_cast_fp16")]; tensor sqrt_24_cast_fp16 = sqrt(x = add_48_cast_fp16)[name = tensor("sqrt_24_cast_fp16")]; tensor real_div_24_cast_fp16 = real_div(x = sub_48_cast_fp16, y = sqrt_24_cast_fp16)[name = tensor("real_div_24_cast_fp16")]; tensor reshape_97_shape_0 = const()[name = tensor("reshape_97_shape_0"), val = tensor([2, 640, 64, 64])]; tensor reshape_97_cast_fp16 = reshape(shape = reshape_97_shape_0, x = real_div_24_cast_fp16)[name = tensor("reshape_97_cast_fp16")]; 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(238333312)))]; 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(238334656)))]; 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_fp16 = 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_9_mean_0_to_fp16, variance = add_9_variance_0_to_fp16, x = reshape_97_cast_fp16)[name = tensor("add_49_cast_fp16")]; tensor input_233_cast_fp16 = silu(x = add_49_cast_fp16)[name = tensor("input_233_cast_fp16")]; tensor var_7041 = const()[name = tensor("op_7041"), val = tensor([1, 1])]; tensor var_7043 = const()[name = tensor("op_7043"), 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_2_resnets_1_conv1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(238336000))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(239718464))), name = tensor("up_blocks_2_resnets_1_conv1_weight_to_fp16_palettized"), shape = tensor([320, 640, 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(239718656)))]; tensor hidden_states_157_cast_fp16 = conv(bias = up_blocks_2_resnets_1_conv1_bias_to_fp16, dilations = var_7043, groups = var_5537, pad = hidden_states_157_pad_0, pad_type = hidden_states_157_pad_type_0, strides = var_7041, weight = up_blocks_2_resnets_1_conv1_weight_to_fp16_palettized, x = input_233_cast_fp16)[name = tensor("hidden_states_157_cast_fp16")]; tensor var_7049 = const()[name = tensor("op_7049"), val = tensor([1, 1])]; tensor var_7051 = const()[name = tensor("op_7051"), 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_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(239719360))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(240026624))), name = tensor("up_blocks_2_resnets_1_time_emb_proj_weight_to_fp16_palettized"), shape = tensor([320, 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(240026816)))]; tensor temb_cast_fp16 = conv(bias = up_blocks_2_resnets_1_time_emb_proj_bias_to_fp16, dilations = var_7051, groups = var_5537, pad = temb_pad_0, pad_type = temb_pad_type_0, strides = var_7049, weight = up_blocks_2_resnets_1_time_emb_proj_weight_to_fp16_palettized, x = input_15_cast_fp16)[name = tensor("temb_cast_fp16")]; tensor input_237_cast_fp16 = add(x = hidden_states_157_cast_fp16, y = temb_cast_fp16)[name = tensor("input_237_cast_fp16")]; tensor reshape_100_shape_0 = const()[name = tensor("reshape_100_shape_0"), val = tensor([2, 32, 10, 64, 64])]; tensor reshape_100_cast_fp16 = reshape(shape = reshape_100_shape_0, x = input_237_cast_fp16)[name = tensor("reshape_100_cast_fp16")]; 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_fp16 = reduce_mean(axes = reduce_mean_75_axes_0, keep_dims = reduce_mean_75_keep_dims_0, x = reshape_100_cast_fp16)[name = tensor("reduce_mean_75_cast_fp16")]; tensor sub_50_cast_fp16 = sub(x = reshape_100_cast_fp16, y = reduce_mean_75_cast_fp16)[name = tensor("sub_50_cast_fp16")]; tensor square_25_cast_fp16 = square(x = sub_50_cast_fp16)[name = tensor("square_25_cast_fp16")]; 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_fp16 = reduce_mean(axes = reduce_mean_77_axes_0, keep_dims = reduce_mean_77_keep_dims_0, x = square_25_cast_fp16)[name = tensor("reduce_mean_77_cast_fp16")]; 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_fp16 = add(x = reduce_mean_77_cast_fp16, y = add_50_y_0_to_fp16)[name = tensor("add_50_cast_fp16")]; tensor sqrt_25_cast_fp16 = sqrt(x = add_50_cast_fp16)[name = tensor("sqrt_25_cast_fp16")]; tensor real_div_25_cast_fp16 = real_div(x = sub_50_cast_fp16, y = sqrt_25_cast_fp16)[name = tensor("real_div_25_cast_fp16")]; tensor reshape_101_shape_0 = const()[name = tensor("reshape_101_shape_0"), val = tensor([2, 320, 64, 64])]; tensor reshape_101_cast_fp16 = reshape(shape = reshape_101_shape_0, x = real_div_25_cast_fp16)[name = tensor("reshape_101_cast_fp16")]; 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(240027520)))]; 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(240028224)))]; 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_fp16 = 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_1_mean_0_to_fp16, variance = add_1_variance_0_to_fp16, x = reshape_101_cast_fp16)[name = tensor("add_51_cast_fp16")]; tensor input_241_cast_fp16 = silu(x = add_51_cast_fp16)[name = tensor("input_241_cast_fp16")]; tensor var_7061 = const()[name = tensor("op_7061"), val = tensor([1, 1])]; tensor var_7063 = const()[name = tensor("op_7063"), 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_2_resnets_1_conv2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(240028928))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(240720192))), name = tensor("up_blocks_2_resnets_1_conv2_weight_to_fp16_palettized"), shape = tensor([320, 320, 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(240720384)))]; tensor hidden_states_159_cast_fp16 = conv(bias = up_blocks_2_resnets_1_conv2_bias_to_fp16, dilations = var_7063, groups = var_5537, pad = hidden_states_159_pad_0, pad_type = hidden_states_159_pad_type_0, strides = var_7061, weight = up_blocks_2_resnets_1_conv2_weight_to_fp16_palettized, x = input_241_cast_fp16)[name = tensor("hidden_states_159_cast_fp16")]; tensor var_7068 = const()[name = tensor("op_7068"), val = tensor([1, 1])]; tensor var_7070 = const()[name = tensor("op_7070"), 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_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(240721088))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(240874752))), name = tensor("up_blocks_2_resnets_1_conv_shortcut_weight_to_fp16_palettized"), shape = tensor([320, 640, 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(240874944)))]; tensor x_cast_fp16 = conv(bias = up_blocks_2_resnets_1_conv_shortcut_bias_to_fp16, dilations = var_7070, groups = var_5537, pad = x_pad_0, pad_type = x_pad_type_0, strides = var_7068, weight = up_blocks_2_resnets_1_conv_shortcut_weight_to_fp16_palettized, x = input_229_cast_fp16)[name = tensor("x_cast_fp16")]; tensor hidden_states_161_cast_fp16 = add(x = x_cast_fp16, y = hidden_states_159_cast_fp16)[name = tensor("hidden_states_161_cast_fp16")]; tensor reshape_104_shape_0 = const()[name = tensor("reshape_104_shape_0"), val = tensor([2, 32, 10, 64, 64])]; tensor reshape_104_cast_fp16 = reshape(shape = reshape_104_shape_0, x = hidden_states_161_cast_fp16)[name = tensor("reshape_104_cast_fp16")]; 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_fp16 = reduce_mean(axes = reduce_mean_78_axes_0, keep_dims = reduce_mean_78_keep_dims_0, x = reshape_104_cast_fp16)[name = tensor("reduce_mean_78_cast_fp16")]; tensor sub_52_cast_fp16 = sub(x = reshape_104_cast_fp16, y = reduce_mean_78_cast_fp16)[name = tensor("sub_52_cast_fp16")]; tensor square_26_cast_fp16 = square(x = sub_52_cast_fp16)[name = tensor("square_26_cast_fp16")]; 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_fp16 = reduce_mean(axes = reduce_mean_80_axes_0, keep_dims = reduce_mean_80_keep_dims_0, x = square_26_cast_fp16)[name = tensor("reduce_mean_80_cast_fp16")]; tensor add_52_y_0_to_fp16 = const()[name = tensor("add_52_y_0_to_fp16"), val = tensor(0x1.1p-20)]; tensor add_52_cast_fp16 = add(x = reduce_mean_80_cast_fp16, y = add_52_y_0_to_fp16)[name = tensor("add_52_cast_fp16")]; tensor sqrt_26_cast_fp16 = sqrt(x = add_52_cast_fp16)[name = tensor("sqrt_26_cast_fp16")]; tensor real_div_26_cast_fp16 = real_div(x = sub_52_cast_fp16, y = sqrt_26_cast_fp16)[name = tensor("real_div_26_cast_fp16")]; tensor reshape_105_shape_0 = const()[name = tensor("reshape_105_shape_0"), val = tensor([2, 320, 64, 64])]; tensor reshape_105_cast_fp16 = reshape(shape = reshape_105_shape_0, x = real_div_26_cast_fp16)[name = tensor("reshape_105_cast_fp16")]; 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(240875648)))]; 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(240876352)))]; 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_fp16 = 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_1_mean_0_to_fp16, variance = add_1_variance_0_to_fp16, x = reshape_105_cast_fp16)[name = tensor("add_53_cast_fp16")]; tensor var_7090 = const()[name = tensor("op_7090"), val = tensor([1, 1])]; tensor var_7092 = const()[name = tensor("op_7092"), val = tensor([1, 1])]; tensor hidden_states_163_pad_type_0 = const()[name = tensor("hidden_states_163_pad_type_0"), val = tensor("custom")]; tensor hidden_states_163_pad_0 = const()[name = tensor("hidden_states_163_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(240877056))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(240953920))), name = tensor("up_blocks_2_attentions_1_proj_in_weight_to_fp16_palettized"), shape = tensor([320, 320, 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(240954112)))]; tensor hidden_states_163_cast_fp16 = conv(bias = up_blocks_2_attentions_1_proj_in_bias_to_fp16, dilations = var_7092, groups = var_5537, pad = hidden_states_163_pad_0, pad_type = hidden_states_163_pad_type_0, strides = var_7090, weight = up_blocks_2_attentions_1_proj_in_weight_to_fp16_palettized, x = add_53_cast_fp16)[name = tensor("hidden_states_163_cast_fp16")]; tensor var_7097 = const()[name = tensor("op_7097"), val = tensor([2, 320, 1, 4096])]; tensor inputs_49_cast_fp16 = reshape(shape = var_7097, x = hidden_states_163_cast_fp16)[name = tensor("inputs_49_cast_fp16")]; tensor var_7107 = const()[name = tensor("op_7107"), val = tensor([1])]; tensor channels_mean_49_cast_fp16 = reduce_mean(axes = var_7107, keep_dims = var_5532, x = inputs_49_cast_fp16)[name = tensor("channels_mean_49_cast_fp16")]; tensor zero_mean_49_cast_fp16 = sub(x = inputs_49_cast_fp16, y = channels_mean_49_cast_fp16)[name = tensor("zero_mean_49_cast_fp16")]; tensor zero_mean_sq_49_cast_fp16 = mul(x = zero_mean_49_cast_fp16, y = zero_mean_49_cast_fp16)[name = tensor("zero_mean_sq_49_cast_fp16")]; tensor var_7111 = const()[name = tensor("op_7111"), val = tensor([1])]; tensor var_7112_cast_fp16 = reduce_mean(axes = var_7111, keep_dims = var_5532, x = zero_mean_sq_49_cast_fp16)[name = tensor("op_7112_cast_fp16")]; tensor var_7113_to_fp16 = const()[name = tensor("op_7113_to_fp16"), val = tensor(0x1.5p-17)]; tensor var_7114_cast_fp16 = add(x = var_7112_cast_fp16, y = var_7113_to_fp16)[name = tensor("op_7114_cast_fp16")]; tensor denom_49_epsilon_0_to_fp16 = const()[name = tensor("denom_49_epsilon_0_to_fp16"), val = tensor(0x1p-24)]; tensor denom_49_cast_fp16 = rsqrt(epsilon = denom_49_epsilon_0_to_fp16, x = var_7114_cast_fp16)[name = tensor("denom_49_cast_fp16")]; tensor out_49_cast_fp16 = mul(x = zero_mean_49_cast_fp16, y = denom_49_cast_fp16)[name = tensor("out_49_cast_fp16")]; tensor var_7118_to_fp16 = const()[name = tensor("op_7118_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(240954816)))]; tensor var_7119_cast_fp16 = add(x = out_49_cast_fp16, y = var_7118_to_fp16)[name = tensor("op_7119_cast_fp16")]; tensor var_7121_to_fp16 = const()[name = tensor("op_7121_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(240955520)))]; tensor hidden_states_165_cast_fp16 = mul(x = var_7119_cast_fp16, y = var_7121_to_fp16)[name = tensor("hidden_states_165_cast_fp16")]; tensor var_7128 = const()[name = tensor("op_7128"), val = tensor([1, 1])]; tensor var_7130 = const()[name = tensor("op_7130"), 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_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(240956224))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(241033088))), name = tensor("up_blocks_2_attentions_1_transformer_blocks_0_attn1_to_q_weight_to_fp16_palettized"), shape = tensor([320, 320, 1, 1])]; tensor q_33_cast_fp16 = conv(dilations = var_7130, groups = var_5537, pad = q_33_pad_0, pad_type = q_33_pad_type_0, strides = var_7128, weight = up_blocks_2_attentions_1_transformer_blocks_0_attn1_to_q_weight_to_fp16_palettized, x = hidden_states_165_cast_fp16)[name = tensor("q_33_cast_fp16")]; tensor var_7134 = const()[name = tensor("op_7134"), val = tensor([1, 1])]; tensor var_7136 = const()[name = tensor("op_7136"), 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_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(241033280))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(241110144))), name = tensor("up_blocks_2_attentions_1_transformer_blocks_0_attn1_to_k_weight_to_fp16_palettized"), shape = tensor([320, 320, 1, 1])]; tensor k_65_cast_fp16 = conv(dilations = var_7136, groups = var_5537, pad = k_65_pad_0, pad_type = k_65_pad_type_0, strides = var_7134, weight = up_blocks_2_attentions_1_transformer_blocks_0_attn1_to_k_weight_to_fp16_palettized, x = hidden_states_165_cast_fp16)[name = tensor("k_65_cast_fp16")]; tensor var_7140 = const()[name = tensor("op_7140"), val = tensor([1, 1])]; tensor var_7142 = const()[name = tensor("op_7142"), 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_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(241110336))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(241187200))), name = tensor("up_blocks_2_attentions_1_transformer_blocks_0_attn1_to_v_weight_to_fp16_palettized"), shape = tensor([320, 320, 1, 1])]; tensor v_33_cast_fp16 = conv(dilations = var_7142, groups = var_5537, pad = v_33_pad_0, pad_type = v_33_pad_type_0, strides = var_7140, weight = up_blocks_2_attentions_1_transformer_blocks_0_attn1_to_v_weight_to_fp16_palettized, x = hidden_states_165_cast_fp16)[name = tensor("v_33_cast_fp16")]; tensor var_7146_begin_0 = const()[name = tensor("op_7146_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_7146_end_0 = const()[name = tensor("op_7146_end_0"), val = tensor([2, 40, 1, 4096])]; tensor var_7146_end_mask_0 = const()[name = tensor("op_7146_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7146_cast_fp16 = slice_by_index(begin = var_7146_begin_0, end = var_7146_end_0, end_mask = var_7146_end_mask_0, x = q_33_cast_fp16)[name = tensor("op_7146_cast_fp16")]; tensor var_7150_begin_0 = const()[name = tensor("op_7150_begin_0"), val = tensor([0, 40, 0, 0])]; tensor var_7150_end_0 = const()[name = tensor("op_7150_end_0"), val = tensor([2, 80, 1, 4096])]; tensor var_7150_end_mask_0 = const()[name = tensor("op_7150_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7150_cast_fp16 = slice_by_index(begin = var_7150_begin_0, end = var_7150_end_0, end_mask = var_7150_end_mask_0, x = q_33_cast_fp16)[name = tensor("op_7150_cast_fp16")]; tensor var_7154_begin_0 = const()[name = tensor("op_7154_begin_0"), val = tensor([0, 80, 0, 0])]; tensor var_7154_end_0 = const()[name = tensor("op_7154_end_0"), val = tensor([2, 120, 1, 4096])]; tensor var_7154_end_mask_0 = const()[name = tensor("op_7154_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7154_cast_fp16 = slice_by_index(begin = var_7154_begin_0, end = var_7154_end_0, end_mask = var_7154_end_mask_0, x = q_33_cast_fp16)[name = tensor("op_7154_cast_fp16")]; tensor var_7158_begin_0 = const()[name = tensor("op_7158_begin_0"), val = tensor([0, 120, 0, 0])]; tensor var_7158_end_0 = const()[name = tensor("op_7158_end_0"), val = tensor([2, 160, 1, 4096])]; tensor var_7158_end_mask_0 = const()[name = tensor("op_7158_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7158_cast_fp16 = slice_by_index(begin = var_7158_begin_0, end = var_7158_end_0, end_mask = var_7158_end_mask_0, x = q_33_cast_fp16)[name = tensor("op_7158_cast_fp16")]; tensor var_7162_begin_0 = const()[name = tensor("op_7162_begin_0"), val = tensor([0, 160, 0, 0])]; tensor var_7162_end_0 = const()[name = tensor("op_7162_end_0"), val = tensor([2, 200, 1, 4096])]; tensor var_7162_end_mask_0 = const()[name = tensor("op_7162_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7162_cast_fp16 = slice_by_index(begin = var_7162_begin_0, end = var_7162_end_0, end_mask = var_7162_end_mask_0, x = q_33_cast_fp16)[name = tensor("op_7162_cast_fp16")]; tensor var_7166_begin_0 = const()[name = tensor("op_7166_begin_0"), val = tensor([0, 200, 0, 0])]; tensor var_7166_end_0 = const()[name = tensor("op_7166_end_0"), val = tensor([2, 240, 1, 4096])]; tensor var_7166_end_mask_0 = const()[name = tensor("op_7166_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7166_cast_fp16 = slice_by_index(begin = var_7166_begin_0, end = var_7166_end_0, end_mask = var_7166_end_mask_0, x = q_33_cast_fp16)[name = tensor("op_7166_cast_fp16")]; tensor var_7170_begin_0 = const()[name = tensor("op_7170_begin_0"), val = tensor([0, 240, 0, 0])]; tensor var_7170_end_0 = const()[name = tensor("op_7170_end_0"), val = tensor([2, 280, 1, 4096])]; tensor var_7170_end_mask_0 = const()[name = tensor("op_7170_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7170_cast_fp16 = slice_by_index(begin = var_7170_begin_0, end = var_7170_end_0, end_mask = var_7170_end_mask_0, x = q_33_cast_fp16)[name = tensor("op_7170_cast_fp16")]; tensor var_7174_begin_0 = const()[name = tensor("op_7174_begin_0"), val = tensor([0, 280, 0, 0])]; tensor var_7174_end_0 = const()[name = tensor("op_7174_end_0"), val = tensor([2, 320, 1, 4096])]; tensor var_7174_end_mask_0 = const()[name = tensor("op_7174_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7174_cast_fp16 = slice_by_index(begin = var_7174_begin_0, end = var_7174_end_0, end_mask = var_7174_end_mask_0, x = q_33_cast_fp16)[name = tensor("op_7174_cast_fp16")]; tensor var_7177_begin_0 = const()[name = tensor("op_7177_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_7177_end_0 = const()[name = tensor("op_7177_end_0"), val = tensor([2, 40, 1, 512])]; tensor var_7177_end_mask_0 = const()[name = tensor("op_7177_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7177_cast_fp16 = slice_by_index(begin = var_7177_begin_0, end = var_7177_end_0, end_mask = var_7177_end_mask_0, x = var_7146_cast_fp16)[name = tensor("op_7177_cast_fp16")]; tensor var_7178_begin_0 = const()[name = tensor("op_7178_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_7178_end_0 = const()[name = tensor("op_7178_end_0"), val = tensor([2, 40, 1, 1024])]; tensor var_7178_end_mask_0 = const()[name = tensor("op_7178_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7178_cast_fp16 = slice_by_index(begin = var_7178_begin_0, end = var_7178_end_0, end_mask = var_7178_end_mask_0, x = var_7146_cast_fp16)[name = tensor("op_7178_cast_fp16")]; tensor var_7179_begin_0 = const()[name = tensor("op_7179_begin_0"), val = tensor([0, 0, 0, 1024])]; tensor var_7179_end_0 = const()[name = tensor("op_7179_end_0"), val = tensor([2, 40, 1, 1536])]; tensor var_7179_end_mask_0 = const()[name = tensor("op_7179_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7179_cast_fp16 = slice_by_index(begin = var_7179_begin_0, end = var_7179_end_0, end_mask = var_7179_end_mask_0, x = var_7146_cast_fp16)[name = tensor("op_7179_cast_fp16")]; tensor var_7180_begin_0 = const()[name = tensor("op_7180_begin_0"), val = tensor([0, 0, 0, 1536])]; tensor var_7180_end_0 = const()[name = tensor("op_7180_end_0"), val = tensor([2, 40, 1, 2048])]; tensor var_7180_end_mask_0 = const()[name = tensor("op_7180_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7180_cast_fp16 = slice_by_index(begin = var_7180_begin_0, end = var_7180_end_0, end_mask = var_7180_end_mask_0, x = var_7146_cast_fp16)[name = tensor("op_7180_cast_fp16")]; tensor var_7181_begin_0 = const()[name = tensor("op_7181_begin_0"), val = tensor([0, 0, 0, 2048])]; tensor var_7181_end_0 = const()[name = tensor("op_7181_end_0"), val = tensor([2, 40, 1, 2560])]; tensor var_7181_end_mask_0 = const()[name = tensor("op_7181_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7181_cast_fp16 = slice_by_index(begin = var_7181_begin_0, end = var_7181_end_0, end_mask = var_7181_end_mask_0, x = var_7146_cast_fp16)[name = tensor("op_7181_cast_fp16")]; tensor var_7182_begin_0 = const()[name = tensor("op_7182_begin_0"), val = tensor([0, 0, 0, 2560])]; tensor var_7182_end_0 = const()[name = tensor("op_7182_end_0"), val = tensor([2, 40, 1, 3072])]; tensor var_7182_end_mask_0 = const()[name = tensor("op_7182_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7182_cast_fp16 = slice_by_index(begin = var_7182_begin_0, end = var_7182_end_0, end_mask = var_7182_end_mask_0, x = var_7146_cast_fp16)[name = tensor("op_7182_cast_fp16")]; tensor var_7183_begin_0 = const()[name = tensor("op_7183_begin_0"), val = tensor([0, 0, 0, 3072])]; tensor var_7183_end_0 = const()[name = tensor("op_7183_end_0"), val = tensor([2, 40, 1, 3584])]; tensor var_7183_end_mask_0 = const()[name = tensor("op_7183_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7183_cast_fp16 = slice_by_index(begin = var_7183_begin_0, end = var_7183_end_0, end_mask = var_7183_end_mask_0, x = var_7146_cast_fp16)[name = tensor("op_7183_cast_fp16")]; tensor var_7184_begin_0 = const()[name = tensor("op_7184_begin_0"), val = tensor([0, 0, 0, 3584])]; tensor var_7184_end_0 = const()[name = tensor("op_7184_end_0"), val = tensor([2, 40, 1, 4096])]; tensor var_7184_end_mask_0 = const()[name = tensor("op_7184_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7184_cast_fp16 = slice_by_index(begin = var_7184_begin_0, end = var_7184_end_0, end_mask = var_7184_end_mask_0, x = var_7146_cast_fp16)[name = tensor("op_7184_cast_fp16")]; tensor var_7185_begin_0 = const()[name = tensor("op_7185_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_7185_end_0 = const()[name = tensor("op_7185_end_0"), val = tensor([2, 40, 1, 512])]; tensor var_7185_end_mask_0 = const()[name = tensor("op_7185_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7185_cast_fp16 = slice_by_index(begin = var_7185_begin_0, end = var_7185_end_0, end_mask = var_7185_end_mask_0, x = var_7150_cast_fp16)[name = tensor("op_7185_cast_fp16")]; tensor var_7186_begin_0 = const()[name = tensor("op_7186_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_7186_end_0 = const()[name = tensor("op_7186_end_0"), val = tensor([2, 40, 1, 1024])]; tensor var_7186_end_mask_0 = const()[name = tensor("op_7186_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7186_cast_fp16 = slice_by_index(begin = var_7186_begin_0, end = var_7186_end_0, end_mask = var_7186_end_mask_0, x = var_7150_cast_fp16)[name = tensor("op_7186_cast_fp16")]; tensor var_7187_begin_0 = const()[name = tensor("op_7187_begin_0"), val = tensor([0, 0, 0, 1024])]; tensor var_7187_end_0 = const()[name = tensor("op_7187_end_0"), val = tensor([2, 40, 1, 1536])]; tensor var_7187_end_mask_0 = const()[name = tensor("op_7187_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7187_cast_fp16 = slice_by_index(begin = var_7187_begin_0, end = var_7187_end_0, end_mask = var_7187_end_mask_0, x = var_7150_cast_fp16)[name = tensor("op_7187_cast_fp16")]; tensor var_7188_begin_0 = const()[name = tensor("op_7188_begin_0"), val = tensor([0, 0, 0, 1536])]; tensor var_7188_end_0 = const()[name = tensor("op_7188_end_0"), val = tensor([2, 40, 1, 2048])]; tensor var_7188_end_mask_0 = const()[name = tensor("op_7188_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7188_cast_fp16 = slice_by_index(begin = var_7188_begin_0, end = var_7188_end_0, end_mask = var_7188_end_mask_0, x = var_7150_cast_fp16)[name = tensor("op_7188_cast_fp16")]; tensor var_7189_begin_0 = const()[name = tensor("op_7189_begin_0"), val = tensor([0, 0, 0, 2048])]; tensor var_7189_end_0 = const()[name = tensor("op_7189_end_0"), val = tensor([2, 40, 1, 2560])]; tensor var_7189_end_mask_0 = const()[name = tensor("op_7189_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7189_cast_fp16 = slice_by_index(begin = var_7189_begin_0, end = var_7189_end_0, end_mask = var_7189_end_mask_0, x = var_7150_cast_fp16)[name = tensor("op_7189_cast_fp16")]; tensor var_7190_begin_0 = const()[name = tensor("op_7190_begin_0"), val = tensor([0, 0, 0, 2560])]; tensor var_7190_end_0 = const()[name = tensor("op_7190_end_0"), val = tensor([2, 40, 1, 3072])]; tensor var_7190_end_mask_0 = const()[name = tensor("op_7190_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7190_cast_fp16 = slice_by_index(begin = var_7190_begin_0, end = var_7190_end_0, end_mask = var_7190_end_mask_0, x = var_7150_cast_fp16)[name = tensor("op_7190_cast_fp16")]; tensor var_7191_begin_0 = const()[name = tensor("op_7191_begin_0"), val = tensor([0, 0, 0, 3072])]; tensor var_7191_end_0 = const()[name = tensor("op_7191_end_0"), val = tensor([2, 40, 1, 3584])]; tensor var_7191_end_mask_0 = const()[name = tensor("op_7191_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7191_cast_fp16 = slice_by_index(begin = var_7191_begin_0, end = var_7191_end_0, end_mask = var_7191_end_mask_0, x = var_7150_cast_fp16)[name = tensor("op_7191_cast_fp16")]; tensor var_7192_begin_0 = const()[name = tensor("op_7192_begin_0"), val = tensor([0, 0, 0, 3584])]; tensor var_7192_end_0 = const()[name = tensor("op_7192_end_0"), val = tensor([2, 40, 1, 4096])]; tensor var_7192_end_mask_0 = const()[name = tensor("op_7192_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7192_cast_fp16 = slice_by_index(begin = var_7192_begin_0, end = var_7192_end_0, end_mask = var_7192_end_mask_0, x = var_7150_cast_fp16)[name = tensor("op_7192_cast_fp16")]; tensor var_7193_begin_0 = const()[name = tensor("op_7193_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_7193_end_0 = const()[name = tensor("op_7193_end_0"), val = tensor([2, 40, 1, 512])]; tensor var_7193_end_mask_0 = const()[name = tensor("op_7193_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7193_cast_fp16 = slice_by_index(begin = var_7193_begin_0, end = var_7193_end_0, end_mask = var_7193_end_mask_0, x = var_7154_cast_fp16)[name = tensor("op_7193_cast_fp16")]; tensor var_7194_begin_0 = const()[name = tensor("op_7194_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_7194_end_0 = const()[name = tensor("op_7194_end_0"), val = tensor([2, 40, 1, 1024])]; tensor var_7194_end_mask_0 = const()[name = tensor("op_7194_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7194_cast_fp16 = slice_by_index(begin = var_7194_begin_0, end = var_7194_end_0, end_mask = var_7194_end_mask_0, x = var_7154_cast_fp16)[name = tensor("op_7194_cast_fp16")]; tensor var_7195_begin_0 = const()[name = tensor("op_7195_begin_0"), val = tensor([0, 0, 0, 1024])]; tensor var_7195_end_0 = const()[name = tensor("op_7195_end_0"), val = tensor([2, 40, 1, 1536])]; tensor var_7195_end_mask_0 = const()[name = tensor("op_7195_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7195_cast_fp16 = slice_by_index(begin = var_7195_begin_0, end = var_7195_end_0, end_mask = var_7195_end_mask_0, x = var_7154_cast_fp16)[name = tensor("op_7195_cast_fp16")]; tensor var_7196_begin_0 = const()[name = tensor("op_7196_begin_0"), val = tensor([0, 0, 0, 1536])]; tensor var_7196_end_0 = const()[name = tensor("op_7196_end_0"), val = tensor([2, 40, 1, 2048])]; tensor var_7196_end_mask_0 = const()[name = tensor("op_7196_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7196_cast_fp16 = slice_by_index(begin = var_7196_begin_0, end = var_7196_end_0, end_mask = var_7196_end_mask_0, x = var_7154_cast_fp16)[name = tensor("op_7196_cast_fp16")]; tensor var_7197_begin_0 = const()[name = tensor("op_7197_begin_0"), val = tensor([0, 0, 0, 2048])]; tensor var_7197_end_0 = const()[name = tensor("op_7197_end_0"), val = tensor([2, 40, 1, 2560])]; tensor var_7197_end_mask_0 = const()[name = tensor("op_7197_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7197_cast_fp16 = slice_by_index(begin = var_7197_begin_0, end = var_7197_end_0, end_mask = var_7197_end_mask_0, x = var_7154_cast_fp16)[name = tensor("op_7197_cast_fp16")]; tensor var_7198_begin_0 = const()[name = tensor("op_7198_begin_0"), val = tensor([0, 0, 0, 2560])]; tensor var_7198_end_0 = const()[name = tensor("op_7198_end_0"), val = tensor([2, 40, 1, 3072])]; tensor var_7198_end_mask_0 = const()[name = tensor("op_7198_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7198_cast_fp16 = slice_by_index(begin = var_7198_begin_0, end = var_7198_end_0, end_mask = var_7198_end_mask_0, x = var_7154_cast_fp16)[name = tensor("op_7198_cast_fp16")]; tensor var_7199_begin_0 = const()[name = tensor("op_7199_begin_0"), val = tensor([0, 0, 0, 3072])]; tensor var_7199_end_0 = const()[name = tensor("op_7199_end_0"), val = tensor([2, 40, 1, 3584])]; tensor var_7199_end_mask_0 = const()[name = tensor("op_7199_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7199_cast_fp16 = slice_by_index(begin = var_7199_begin_0, end = var_7199_end_0, end_mask = var_7199_end_mask_0, x = var_7154_cast_fp16)[name = tensor("op_7199_cast_fp16")]; tensor var_7200_begin_0 = const()[name = tensor("op_7200_begin_0"), val = tensor([0, 0, 0, 3584])]; tensor var_7200_end_0 = const()[name = tensor("op_7200_end_0"), val = tensor([2, 40, 1, 4096])]; tensor var_7200_end_mask_0 = const()[name = tensor("op_7200_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7200_cast_fp16 = slice_by_index(begin = var_7200_begin_0, end = var_7200_end_0, end_mask = var_7200_end_mask_0, x = var_7154_cast_fp16)[name = tensor("op_7200_cast_fp16")]; tensor var_7201_begin_0 = const()[name = tensor("op_7201_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_7201_end_0 = const()[name = tensor("op_7201_end_0"), val = tensor([2, 40, 1, 512])]; tensor var_7201_end_mask_0 = const()[name = tensor("op_7201_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7201_cast_fp16 = slice_by_index(begin = var_7201_begin_0, end = var_7201_end_0, end_mask = var_7201_end_mask_0, x = var_7158_cast_fp16)[name = tensor("op_7201_cast_fp16")]; tensor var_7202_begin_0 = const()[name = tensor("op_7202_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_7202_end_0 = const()[name = tensor("op_7202_end_0"), val = tensor([2, 40, 1, 1024])]; tensor var_7202_end_mask_0 = const()[name = tensor("op_7202_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7202_cast_fp16 = slice_by_index(begin = var_7202_begin_0, end = var_7202_end_0, end_mask = var_7202_end_mask_0, x = var_7158_cast_fp16)[name = tensor("op_7202_cast_fp16")]; tensor var_7203_begin_0 = const()[name = tensor("op_7203_begin_0"), val = tensor([0, 0, 0, 1024])]; tensor var_7203_end_0 = const()[name = tensor("op_7203_end_0"), val = tensor([2, 40, 1, 1536])]; tensor var_7203_end_mask_0 = const()[name = tensor("op_7203_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7203_cast_fp16 = slice_by_index(begin = var_7203_begin_0, end = var_7203_end_0, end_mask = var_7203_end_mask_0, x = var_7158_cast_fp16)[name = tensor("op_7203_cast_fp16")]; tensor var_7204_begin_0 = const()[name = tensor("op_7204_begin_0"), val = tensor([0, 0, 0, 1536])]; tensor var_7204_end_0 = const()[name = tensor("op_7204_end_0"), val = tensor([2, 40, 1, 2048])]; tensor var_7204_end_mask_0 = const()[name = tensor("op_7204_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7204_cast_fp16 = slice_by_index(begin = var_7204_begin_0, end = var_7204_end_0, end_mask = var_7204_end_mask_0, x = var_7158_cast_fp16)[name = tensor("op_7204_cast_fp16")]; tensor var_7205_begin_0 = const()[name = tensor("op_7205_begin_0"), val = tensor([0, 0, 0, 2048])]; tensor var_7205_end_0 = const()[name = tensor("op_7205_end_0"), val = tensor([2, 40, 1, 2560])]; tensor var_7205_end_mask_0 = const()[name = tensor("op_7205_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7205_cast_fp16 = slice_by_index(begin = var_7205_begin_0, end = var_7205_end_0, end_mask = var_7205_end_mask_0, x = var_7158_cast_fp16)[name = tensor("op_7205_cast_fp16")]; tensor var_7206_begin_0 = const()[name = tensor("op_7206_begin_0"), val = tensor([0, 0, 0, 2560])]; tensor var_7206_end_0 = const()[name = tensor("op_7206_end_0"), val = tensor([2, 40, 1, 3072])]; tensor var_7206_end_mask_0 = const()[name = tensor("op_7206_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7206_cast_fp16 = slice_by_index(begin = var_7206_begin_0, end = var_7206_end_0, end_mask = var_7206_end_mask_0, x = var_7158_cast_fp16)[name = tensor("op_7206_cast_fp16")]; tensor var_7207_begin_0 = const()[name = tensor("op_7207_begin_0"), val = tensor([0, 0, 0, 3072])]; tensor var_7207_end_0 = const()[name = tensor("op_7207_end_0"), val = tensor([2, 40, 1, 3584])]; tensor var_7207_end_mask_0 = const()[name = tensor("op_7207_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7207_cast_fp16 = slice_by_index(begin = var_7207_begin_0, end = var_7207_end_0, end_mask = var_7207_end_mask_0, x = var_7158_cast_fp16)[name = tensor("op_7207_cast_fp16")]; tensor var_7208_begin_0 = const()[name = tensor("op_7208_begin_0"), val = tensor([0, 0, 0, 3584])]; tensor var_7208_end_0 = const()[name = tensor("op_7208_end_0"), val = tensor([2, 40, 1, 4096])]; tensor var_7208_end_mask_0 = const()[name = tensor("op_7208_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7208_cast_fp16 = slice_by_index(begin = var_7208_begin_0, end = var_7208_end_0, end_mask = var_7208_end_mask_0, x = var_7158_cast_fp16)[name = tensor("op_7208_cast_fp16")]; tensor var_7209_begin_0 = const()[name = tensor("op_7209_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_7209_end_0 = const()[name = tensor("op_7209_end_0"), val = tensor([2, 40, 1, 512])]; tensor var_7209_end_mask_0 = const()[name = tensor("op_7209_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7209_cast_fp16 = slice_by_index(begin = var_7209_begin_0, end = var_7209_end_0, end_mask = var_7209_end_mask_0, x = var_7162_cast_fp16)[name = tensor("op_7209_cast_fp16")]; tensor var_7210_begin_0 = const()[name = tensor("op_7210_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_7210_end_0 = const()[name = tensor("op_7210_end_0"), val = tensor([2, 40, 1, 1024])]; tensor var_7210_end_mask_0 = const()[name = tensor("op_7210_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7210_cast_fp16 = slice_by_index(begin = var_7210_begin_0, end = var_7210_end_0, end_mask = var_7210_end_mask_0, x = var_7162_cast_fp16)[name = tensor("op_7210_cast_fp16")]; tensor var_7211_begin_0 = const()[name = tensor("op_7211_begin_0"), val = tensor([0, 0, 0, 1024])]; tensor var_7211_end_0 = const()[name = tensor("op_7211_end_0"), val = tensor([2, 40, 1, 1536])]; tensor var_7211_end_mask_0 = const()[name = tensor("op_7211_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7211_cast_fp16 = slice_by_index(begin = var_7211_begin_0, end = var_7211_end_0, end_mask = var_7211_end_mask_0, x = var_7162_cast_fp16)[name = tensor("op_7211_cast_fp16")]; tensor var_7212_begin_0 = const()[name = tensor("op_7212_begin_0"), val = tensor([0, 0, 0, 1536])]; tensor var_7212_end_0 = const()[name = tensor("op_7212_end_0"), val = tensor([2, 40, 1, 2048])]; tensor var_7212_end_mask_0 = const()[name = tensor("op_7212_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7212_cast_fp16 = slice_by_index(begin = var_7212_begin_0, end = var_7212_end_0, end_mask = var_7212_end_mask_0, x = var_7162_cast_fp16)[name = tensor("op_7212_cast_fp16")]; tensor var_7213_begin_0 = const()[name = tensor("op_7213_begin_0"), val = tensor([0, 0, 0, 2048])]; tensor var_7213_end_0 = const()[name = tensor("op_7213_end_0"), val = tensor([2, 40, 1, 2560])]; tensor var_7213_end_mask_0 = const()[name = tensor("op_7213_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7213_cast_fp16 = slice_by_index(begin = var_7213_begin_0, end = var_7213_end_0, end_mask = var_7213_end_mask_0, x = var_7162_cast_fp16)[name = tensor("op_7213_cast_fp16")]; tensor var_7214_begin_0 = const()[name = tensor("op_7214_begin_0"), val = tensor([0, 0, 0, 2560])]; tensor var_7214_end_0 = const()[name = tensor("op_7214_end_0"), val = tensor([2, 40, 1, 3072])]; tensor var_7214_end_mask_0 = const()[name = tensor("op_7214_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7214_cast_fp16 = slice_by_index(begin = var_7214_begin_0, end = var_7214_end_0, end_mask = var_7214_end_mask_0, x = var_7162_cast_fp16)[name = tensor("op_7214_cast_fp16")]; tensor var_7215_begin_0 = const()[name = tensor("op_7215_begin_0"), val = tensor([0, 0, 0, 3072])]; tensor var_7215_end_0 = const()[name = tensor("op_7215_end_0"), val = tensor([2, 40, 1, 3584])]; tensor var_7215_end_mask_0 = const()[name = tensor("op_7215_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7215_cast_fp16 = slice_by_index(begin = var_7215_begin_0, end = var_7215_end_0, end_mask = var_7215_end_mask_0, x = var_7162_cast_fp16)[name = tensor("op_7215_cast_fp16")]; tensor var_7216_begin_0 = const()[name = tensor("op_7216_begin_0"), val = tensor([0, 0, 0, 3584])]; tensor var_7216_end_0 = const()[name = tensor("op_7216_end_0"), val = tensor([2, 40, 1, 4096])]; tensor var_7216_end_mask_0 = const()[name = tensor("op_7216_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7216_cast_fp16 = slice_by_index(begin = var_7216_begin_0, end = var_7216_end_0, end_mask = var_7216_end_mask_0, x = var_7162_cast_fp16)[name = tensor("op_7216_cast_fp16")]; tensor var_7217_begin_0 = const()[name = tensor("op_7217_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_7217_end_0 = const()[name = tensor("op_7217_end_0"), val = tensor([2, 40, 1, 512])]; tensor var_7217_end_mask_0 = const()[name = tensor("op_7217_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7217_cast_fp16 = slice_by_index(begin = var_7217_begin_0, end = var_7217_end_0, end_mask = var_7217_end_mask_0, x = var_7166_cast_fp16)[name = tensor("op_7217_cast_fp16")]; tensor var_7218_begin_0 = const()[name = tensor("op_7218_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_7218_end_0 = const()[name = tensor("op_7218_end_0"), val = tensor([2, 40, 1, 1024])]; tensor var_7218_end_mask_0 = const()[name = tensor("op_7218_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7218_cast_fp16 = slice_by_index(begin = var_7218_begin_0, end = var_7218_end_0, end_mask = var_7218_end_mask_0, x = var_7166_cast_fp16)[name = tensor("op_7218_cast_fp16")]; tensor var_7219_begin_0 = const()[name = tensor("op_7219_begin_0"), val = tensor([0, 0, 0, 1024])]; tensor var_7219_end_0 = const()[name = tensor("op_7219_end_0"), val = tensor([2, 40, 1, 1536])]; tensor var_7219_end_mask_0 = const()[name = tensor("op_7219_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7219_cast_fp16 = slice_by_index(begin = var_7219_begin_0, end = var_7219_end_0, end_mask = var_7219_end_mask_0, x = var_7166_cast_fp16)[name = tensor("op_7219_cast_fp16")]; tensor var_7220_begin_0 = const()[name = tensor("op_7220_begin_0"), val = tensor([0, 0, 0, 1536])]; tensor var_7220_end_0 = const()[name = tensor("op_7220_end_0"), val = tensor([2, 40, 1, 2048])]; tensor var_7220_end_mask_0 = const()[name = tensor("op_7220_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7220_cast_fp16 = slice_by_index(begin = var_7220_begin_0, end = var_7220_end_0, end_mask = var_7220_end_mask_0, x = var_7166_cast_fp16)[name = tensor("op_7220_cast_fp16")]; tensor var_7221_begin_0 = const()[name = tensor("op_7221_begin_0"), val = tensor([0, 0, 0, 2048])]; tensor var_7221_end_0 = const()[name = tensor("op_7221_end_0"), val = tensor([2, 40, 1, 2560])]; tensor var_7221_end_mask_0 = const()[name = tensor("op_7221_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7221_cast_fp16 = slice_by_index(begin = var_7221_begin_0, end = var_7221_end_0, end_mask = var_7221_end_mask_0, x = var_7166_cast_fp16)[name = tensor("op_7221_cast_fp16")]; tensor var_7222_begin_0 = const()[name = tensor("op_7222_begin_0"), val = tensor([0, 0, 0, 2560])]; tensor var_7222_end_0 = const()[name = tensor("op_7222_end_0"), val = tensor([2, 40, 1, 3072])]; tensor var_7222_end_mask_0 = const()[name = tensor("op_7222_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7222_cast_fp16 = slice_by_index(begin = var_7222_begin_0, end = var_7222_end_0, end_mask = var_7222_end_mask_0, x = var_7166_cast_fp16)[name = tensor("op_7222_cast_fp16")]; tensor var_7223_begin_0 = const()[name = tensor("op_7223_begin_0"), val = tensor([0, 0, 0, 3072])]; tensor var_7223_end_0 = const()[name = tensor("op_7223_end_0"), val = tensor([2, 40, 1, 3584])]; tensor var_7223_end_mask_0 = const()[name = tensor("op_7223_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7223_cast_fp16 = slice_by_index(begin = var_7223_begin_0, end = var_7223_end_0, end_mask = var_7223_end_mask_0, x = var_7166_cast_fp16)[name = tensor("op_7223_cast_fp16")]; tensor var_7224_begin_0 = const()[name = tensor("op_7224_begin_0"), val = tensor([0, 0, 0, 3584])]; tensor var_7224_end_0 = const()[name = tensor("op_7224_end_0"), val = tensor([2, 40, 1, 4096])]; tensor var_7224_end_mask_0 = const()[name = tensor("op_7224_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7224_cast_fp16 = slice_by_index(begin = var_7224_begin_0, end = var_7224_end_0, end_mask = var_7224_end_mask_0, x = var_7166_cast_fp16)[name = tensor("op_7224_cast_fp16")]; tensor var_7225_begin_0 = const()[name = tensor("op_7225_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_7225_end_0 = const()[name = tensor("op_7225_end_0"), val = tensor([2, 40, 1, 512])]; tensor var_7225_end_mask_0 = const()[name = tensor("op_7225_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7225_cast_fp16 = slice_by_index(begin = var_7225_begin_0, end = var_7225_end_0, end_mask = var_7225_end_mask_0, x = var_7170_cast_fp16)[name = tensor("op_7225_cast_fp16")]; tensor var_7226_begin_0 = const()[name = tensor("op_7226_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_7226_end_0 = const()[name = tensor("op_7226_end_0"), val = tensor([2, 40, 1, 1024])]; tensor var_7226_end_mask_0 = const()[name = tensor("op_7226_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7226_cast_fp16 = slice_by_index(begin = var_7226_begin_0, end = var_7226_end_0, end_mask = var_7226_end_mask_0, x = var_7170_cast_fp16)[name = tensor("op_7226_cast_fp16")]; tensor var_7227_begin_0 = const()[name = tensor("op_7227_begin_0"), val = tensor([0, 0, 0, 1024])]; tensor var_7227_end_0 = const()[name = tensor("op_7227_end_0"), val = tensor([2, 40, 1, 1536])]; tensor var_7227_end_mask_0 = const()[name = tensor("op_7227_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7227_cast_fp16 = slice_by_index(begin = var_7227_begin_0, end = var_7227_end_0, end_mask = var_7227_end_mask_0, x = var_7170_cast_fp16)[name = tensor("op_7227_cast_fp16")]; tensor var_7228_begin_0 = const()[name = tensor("op_7228_begin_0"), val = tensor([0, 0, 0, 1536])]; tensor var_7228_end_0 = const()[name = tensor("op_7228_end_0"), val = tensor([2, 40, 1, 2048])]; tensor var_7228_end_mask_0 = const()[name = tensor("op_7228_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7228_cast_fp16 = slice_by_index(begin = var_7228_begin_0, end = var_7228_end_0, end_mask = var_7228_end_mask_0, x = var_7170_cast_fp16)[name = tensor("op_7228_cast_fp16")]; tensor var_7229_begin_0 = const()[name = tensor("op_7229_begin_0"), val = tensor([0, 0, 0, 2048])]; tensor var_7229_end_0 = const()[name = tensor("op_7229_end_0"), val = tensor([2, 40, 1, 2560])]; tensor var_7229_end_mask_0 = const()[name = tensor("op_7229_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7229_cast_fp16 = slice_by_index(begin = var_7229_begin_0, end = var_7229_end_0, end_mask = var_7229_end_mask_0, x = var_7170_cast_fp16)[name = tensor("op_7229_cast_fp16")]; tensor var_7230_begin_0 = const()[name = tensor("op_7230_begin_0"), val = tensor([0, 0, 0, 2560])]; tensor var_7230_end_0 = const()[name = tensor("op_7230_end_0"), val = tensor([2, 40, 1, 3072])]; tensor var_7230_end_mask_0 = const()[name = tensor("op_7230_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7230_cast_fp16 = slice_by_index(begin = var_7230_begin_0, end = var_7230_end_0, end_mask = var_7230_end_mask_0, x = var_7170_cast_fp16)[name = tensor("op_7230_cast_fp16")]; tensor var_7231_begin_0 = const()[name = tensor("op_7231_begin_0"), val = tensor([0, 0, 0, 3072])]; tensor var_7231_end_0 = const()[name = tensor("op_7231_end_0"), val = tensor([2, 40, 1, 3584])]; tensor var_7231_end_mask_0 = const()[name = tensor("op_7231_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7231_cast_fp16 = slice_by_index(begin = var_7231_begin_0, end = var_7231_end_0, end_mask = var_7231_end_mask_0, x = var_7170_cast_fp16)[name = tensor("op_7231_cast_fp16")]; tensor var_7232_begin_0 = const()[name = tensor("op_7232_begin_0"), val = tensor([0, 0, 0, 3584])]; tensor var_7232_end_0 = const()[name = tensor("op_7232_end_0"), val = tensor([2, 40, 1, 4096])]; tensor var_7232_end_mask_0 = const()[name = tensor("op_7232_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7232_cast_fp16 = slice_by_index(begin = var_7232_begin_0, end = var_7232_end_0, end_mask = var_7232_end_mask_0, x = var_7170_cast_fp16)[name = tensor("op_7232_cast_fp16")]; tensor var_7233_begin_0 = const()[name = tensor("op_7233_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_7233_end_0 = const()[name = tensor("op_7233_end_0"), val = tensor([2, 40, 1, 512])]; tensor var_7233_end_mask_0 = const()[name = tensor("op_7233_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7233_cast_fp16 = slice_by_index(begin = var_7233_begin_0, end = var_7233_end_0, end_mask = var_7233_end_mask_0, x = var_7174_cast_fp16)[name = tensor("op_7233_cast_fp16")]; tensor var_7234_begin_0 = const()[name = tensor("op_7234_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_7234_end_0 = const()[name = tensor("op_7234_end_0"), val = tensor([2, 40, 1, 1024])]; tensor var_7234_end_mask_0 = const()[name = tensor("op_7234_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7234_cast_fp16 = slice_by_index(begin = var_7234_begin_0, end = var_7234_end_0, end_mask = var_7234_end_mask_0, x = var_7174_cast_fp16)[name = tensor("op_7234_cast_fp16")]; tensor var_7235_begin_0 = const()[name = tensor("op_7235_begin_0"), val = tensor([0, 0, 0, 1024])]; tensor var_7235_end_0 = const()[name = tensor("op_7235_end_0"), val = tensor([2, 40, 1, 1536])]; tensor var_7235_end_mask_0 = const()[name = tensor("op_7235_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7235_cast_fp16 = slice_by_index(begin = var_7235_begin_0, end = var_7235_end_0, end_mask = var_7235_end_mask_0, x = var_7174_cast_fp16)[name = tensor("op_7235_cast_fp16")]; tensor var_7236_begin_0 = const()[name = tensor("op_7236_begin_0"), val = tensor([0, 0, 0, 1536])]; tensor var_7236_end_0 = const()[name = tensor("op_7236_end_0"), val = tensor([2, 40, 1, 2048])]; tensor var_7236_end_mask_0 = const()[name = tensor("op_7236_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7236_cast_fp16 = slice_by_index(begin = var_7236_begin_0, end = var_7236_end_0, end_mask = var_7236_end_mask_0, x = var_7174_cast_fp16)[name = tensor("op_7236_cast_fp16")]; tensor var_7237_begin_0 = const()[name = tensor("op_7237_begin_0"), val = tensor([0, 0, 0, 2048])]; tensor var_7237_end_0 = const()[name = tensor("op_7237_end_0"), val = tensor([2, 40, 1, 2560])]; tensor var_7237_end_mask_0 = const()[name = tensor("op_7237_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7237_cast_fp16 = slice_by_index(begin = var_7237_begin_0, end = var_7237_end_0, end_mask = var_7237_end_mask_0, x = var_7174_cast_fp16)[name = tensor("op_7237_cast_fp16")]; tensor var_7238_begin_0 = const()[name = tensor("op_7238_begin_0"), val = tensor([0, 0, 0, 2560])]; tensor var_7238_end_0 = const()[name = tensor("op_7238_end_0"), val = tensor([2, 40, 1, 3072])]; tensor var_7238_end_mask_0 = const()[name = tensor("op_7238_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7238_cast_fp16 = slice_by_index(begin = var_7238_begin_0, end = var_7238_end_0, end_mask = var_7238_end_mask_0, x = var_7174_cast_fp16)[name = tensor("op_7238_cast_fp16")]; tensor var_7239_begin_0 = const()[name = tensor("op_7239_begin_0"), val = tensor([0, 0, 0, 3072])]; tensor var_7239_end_0 = const()[name = tensor("op_7239_end_0"), val = tensor([2, 40, 1, 3584])]; tensor var_7239_end_mask_0 = const()[name = tensor("op_7239_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7239_cast_fp16 = slice_by_index(begin = var_7239_begin_0, end = var_7239_end_0, end_mask = var_7239_end_mask_0, x = var_7174_cast_fp16)[name = tensor("op_7239_cast_fp16")]; tensor var_7240_begin_0 = const()[name = tensor("op_7240_begin_0"), val = tensor([0, 0, 0, 3584])]; tensor var_7240_end_0 = const()[name = tensor("op_7240_end_0"), val = tensor([2, 40, 1, 4096])]; tensor var_7240_end_mask_0 = const()[name = tensor("op_7240_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7240_cast_fp16 = slice_by_index(begin = var_7240_begin_0, end = var_7240_end_0, end_mask = var_7240_end_mask_0, x = var_7174_cast_fp16)[name = tensor("op_7240_cast_fp16")]; tensor k_67_perm_0 = const()[name = tensor("k_67_perm_0"), val = tensor([0, 3, 2, 1])]; tensor var_7245_begin_0 = const()[name = tensor("op_7245_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_7245_end_0 = const()[name = tensor("op_7245_end_0"), val = tensor([2, 4096, 1, 40])]; tensor var_7245_end_mask_0 = const()[name = tensor("op_7245_end_mask_0"), val = tensor([true, true, true, false])]; tensor transpose_1 = transpose(perm = k_67_perm_0, x = k_65_cast_fp16)[name = tensor("transpose_1")]; tensor var_7245_cast_fp16 = slice_by_index(begin = var_7245_begin_0, end = var_7245_end_0, end_mask = var_7245_end_mask_0, x = transpose_1)[name = tensor("op_7245_cast_fp16")]; tensor var_7249_begin_0 = const()[name = tensor("op_7249_begin_0"), val = tensor([0, 0, 0, 40])]; tensor var_7249_end_0 = const()[name = tensor("op_7249_end_0"), val = tensor([2, 4096, 1, 80])]; tensor var_7249_end_mask_0 = const()[name = tensor("op_7249_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7249_cast_fp16 = slice_by_index(begin = var_7249_begin_0, end = var_7249_end_0, end_mask = var_7249_end_mask_0, x = transpose_1)[name = tensor("op_7249_cast_fp16")]; tensor var_7253_begin_0 = const()[name = tensor("op_7253_begin_0"), val = tensor([0, 0, 0, 80])]; tensor var_7253_end_0 = const()[name = tensor("op_7253_end_0"), val = tensor([2, 4096, 1, 120])]; tensor var_7253_end_mask_0 = const()[name = tensor("op_7253_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7253_cast_fp16 = slice_by_index(begin = var_7253_begin_0, end = var_7253_end_0, end_mask = var_7253_end_mask_0, x = transpose_1)[name = tensor("op_7253_cast_fp16")]; tensor var_7257_begin_0 = const()[name = tensor("op_7257_begin_0"), val = tensor([0, 0, 0, 120])]; tensor var_7257_end_0 = const()[name = tensor("op_7257_end_0"), val = tensor([2, 4096, 1, 160])]; tensor var_7257_end_mask_0 = const()[name = tensor("op_7257_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7257_cast_fp16 = slice_by_index(begin = var_7257_begin_0, end = var_7257_end_0, end_mask = var_7257_end_mask_0, x = transpose_1)[name = tensor("op_7257_cast_fp16")]; tensor var_7261_begin_0 = const()[name = tensor("op_7261_begin_0"), val = tensor([0, 0, 0, 160])]; tensor var_7261_end_0 = const()[name = tensor("op_7261_end_0"), val = tensor([2, 4096, 1, 200])]; tensor var_7261_end_mask_0 = const()[name = tensor("op_7261_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7261_cast_fp16 = slice_by_index(begin = var_7261_begin_0, end = var_7261_end_0, end_mask = var_7261_end_mask_0, x = transpose_1)[name = tensor("op_7261_cast_fp16")]; tensor var_7265_begin_0 = const()[name = tensor("op_7265_begin_0"), val = tensor([0, 0, 0, 200])]; tensor var_7265_end_0 = const()[name = tensor("op_7265_end_0"), val = tensor([2, 4096, 1, 240])]; tensor var_7265_end_mask_0 = const()[name = tensor("op_7265_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7265_cast_fp16 = slice_by_index(begin = var_7265_begin_0, end = var_7265_end_0, end_mask = var_7265_end_mask_0, x = transpose_1)[name = tensor("op_7265_cast_fp16")]; tensor var_7269_begin_0 = const()[name = tensor("op_7269_begin_0"), val = tensor([0, 0, 0, 240])]; tensor var_7269_end_0 = const()[name = tensor("op_7269_end_0"), val = tensor([2, 4096, 1, 280])]; tensor var_7269_end_mask_0 = const()[name = tensor("op_7269_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7269_cast_fp16 = slice_by_index(begin = var_7269_begin_0, end = var_7269_end_0, end_mask = var_7269_end_mask_0, x = transpose_1)[name = tensor("op_7269_cast_fp16")]; tensor var_7273_begin_0 = const()[name = tensor("op_7273_begin_0"), val = tensor([0, 0, 0, 280])]; tensor var_7273_end_0 = const()[name = tensor("op_7273_end_0"), val = tensor([2, 4096, 1, 320])]; tensor var_7273_end_mask_0 = const()[name = tensor("op_7273_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7273_cast_fp16 = slice_by_index(begin = var_7273_begin_0, end = var_7273_end_0, end_mask = var_7273_end_mask_0, x = transpose_1)[name = tensor("op_7273_cast_fp16")]; tensor var_7275_begin_0 = const()[name = tensor("op_7275_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_7275_end_0 = const()[name = tensor("op_7275_end_0"), val = tensor([2, 40, 1, 4096])]; tensor var_7275_end_mask_0 = const()[name = tensor("op_7275_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7275_cast_fp16 = slice_by_index(begin = var_7275_begin_0, end = var_7275_end_0, end_mask = var_7275_end_mask_0, x = v_33_cast_fp16)[name = tensor("op_7275_cast_fp16")]; tensor var_7279_begin_0 = const()[name = tensor("op_7279_begin_0"), val = tensor([0, 40, 0, 0])]; tensor var_7279_end_0 = const()[name = tensor("op_7279_end_0"), val = tensor([2, 80, 1, 4096])]; tensor var_7279_end_mask_0 = const()[name = tensor("op_7279_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7279_cast_fp16 = slice_by_index(begin = var_7279_begin_0, end = var_7279_end_0, end_mask = var_7279_end_mask_0, x = v_33_cast_fp16)[name = tensor("op_7279_cast_fp16")]; tensor var_7283_begin_0 = const()[name = tensor("op_7283_begin_0"), val = tensor([0, 80, 0, 0])]; tensor var_7283_end_0 = const()[name = tensor("op_7283_end_0"), val = tensor([2, 120, 1, 4096])]; tensor var_7283_end_mask_0 = const()[name = tensor("op_7283_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7283_cast_fp16 = slice_by_index(begin = var_7283_begin_0, end = var_7283_end_0, end_mask = var_7283_end_mask_0, x = v_33_cast_fp16)[name = tensor("op_7283_cast_fp16")]; tensor var_7287_begin_0 = const()[name = tensor("op_7287_begin_0"), val = tensor([0, 120, 0, 0])]; tensor var_7287_end_0 = const()[name = tensor("op_7287_end_0"), val = tensor([2, 160, 1, 4096])]; tensor var_7287_end_mask_0 = const()[name = tensor("op_7287_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7287_cast_fp16 = slice_by_index(begin = var_7287_begin_0, end = var_7287_end_0, end_mask = var_7287_end_mask_0, x = v_33_cast_fp16)[name = tensor("op_7287_cast_fp16")]; tensor var_7291_begin_0 = const()[name = tensor("op_7291_begin_0"), val = tensor([0, 160, 0, 0])]; tensor var_7291_end_0 = const()[name = tensor("op_7291_end_0"), val = tensor([2, 200, 1, 4096])]; tensor var_7291_end_mask_0 = const()[name = tensor("op_7291_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7291_cast_fp16 = slice_by_index(begin = var_7291_begin_0, end = var_7291_end_0, end_mask = var_7291_end_mask_0, x = v_33_cast_fp16)[name = tensor("op_7291_cast_fp16")]; tensor var_7295_begin_0 = const()[name = tensor("op_7295_begin_0"), val = tensor([0, 200, 0, 0])]; tensor var_7295_end_0 = const()[name = tensor("op_7295_end_0"), val = tensor([2, 240, 1, 4096])]; tensor var_7295_end_mask_0 = const()[name = tensor("op_7295_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7295_cast_fp16 = slice_by_index(begin = var_7295_begin_0, end = var_7295_end_0, end_mask = var_7295_end_mask_0, x = v_33_cast_fp16)[name = tensor("op_7295_cast_fp16")]; tensor var_7299_begin_0 = const()[name = tensor("op_7299_begin_0"), val = tensor([0, 240, 0, 0])]; tensor var_7299_end_0 = const()[name = tensor("op_7299_end_0"), val = tensor([2, 280, 1, 4096])]; tensor var_7299_end_mask_0 = const()[name = tensor("op_7299_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7299_cast_fp16 = slice_by_index(begin = var_7299_begin_0, end = var_7299_end_0, end_mask = var_7299_end_mask_0, x = v_33_cast_fp16)[name = tensor("op_7299_cast_fp16")]; tensor var_7303_begin_0 = const()[name = tensor("op_7303_begin_0"), val = tensor([0, 280, 0, 0])]; tensor var_7303_end_0 = const()[name = tensor("op_7303_end_0"), val = tensor([2, 320, 1, 4096])]; tensor var_7303_end_mask_0 = const()[name = tensor("op_7303_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7303_cast_fp16 = slice_by_index(begin = var_7303_begin_0, end = var_7303_end_0, end_mask = var_7303_end_mask_0, x = v_33_cast_fp16)[name = tensor("op_7303_cast_fp16")]; tensor var_7307_equation_0 = const()[name = tensor("op_7307_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7307_cast_fp16 = einsum(equation = var_7307_equation_0, values = (var_7245_cast_fp16, var_7177_cast_fp16))[name = tensor("op_7307_cast_fp16")]; tensor var_7308_to_fp16 = const()[name = tensor("op_7308_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_705_cast_fp16 = mul(x = var_7307_cast_fp16, y = var_7308_to_fp16)[name = tensor("aw_chunk_705_cast_fp16")]; tensor var_7311_equation_0 = const()[name = tensor("op_7311_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7311_cast_fp16 = einsum(equation = var_7311_equation_0, values = (var_7245_cast_fp16, var_7178_cast_fp16))[name = tensor("op_7311_cast_fp16")]; tensor var_7312_to_fp16 = const()[name = tensor("op_7312_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_707_cast_fp16 = mul(x = var_7311_cast_fp16, y = var_7312_to_fp16)[name = tensor("aw_chunk_707_cast_fp16")]; tensor var_7315_equation_0 = const()[name = tensor("op_7315_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7315_cast_fp16 = einsum(equation = var_7315_equation_0, values = (var_7245_cast_fp16, var_7179_cast_fp16))[name = tensor("op_7315_cast_fp16")]; tensor var_7316_to_fp16 = const()[name = tensor("op_7316_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_709_cast_fp16 = mul(x = var_7315_cast_fp16, y = var_7316_to_fp16)[name = tensor("aw_chunk_709_cast_fp16")]; tensor var_7319_equation_0 = const()[name = tensor("op_7319_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7319_cast_fp16 = einsum(equation = var_7319_equation_0, values = (var_7245_cast_fp16, var_7180_cast_fp16))[name = tensor("op_7319_cast_fp16")]; tensor var_7320_to_fp16 = const()[name = tensor("op_7320_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_711_cast_fp16 = mul(x = var_7319_cast_fp16, y = var_7320_to_fp16)[name = tensor("aw_chunk_711_cast_fp16")]; tensor var_7323_equation_0 = const()[name = tensor("op_7323_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7323_cast_fp16 = einsum(equation = var_7323_equation_0, values = (var_7245_cast_fp16, var_7181_cast_fp16))[name = tensor("op_7323_cast_fp16")]; tensor var_7324_to_fp16 = const()[name = tensor("op_7324_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_713_cast_fp16 = mul(x = var_7323_cast_fp16, y = var_7324_to_fp16)[name = tensor("aw_chunk_713_cast_fp16")]; tensor var_7327_equation_0 = const()[name = tensor("op_7327_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7327_cast_fp16 = einsum(equation = var_7327_equation_0, values = (var_7245_cast_fp16, var_7182_cast_fp16))[name = tensor("op_7327_cast_fp16")]; tensor var_7328_to_fp16 = const()[name = tensor("op_7328_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_715_cast_fp16 = mul(x = var_7327_cast_fp16, y = var_7328_to_fp16)[name = tensor("aw_chunk_715_cast_fp16")]; tensor var_7331_equation_0 = const()[name = tensor("op_7331_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7331_cast_fp16 = einsum(equation = var_7331_equation_0, values = (var_7245_cast_fp16, var_7183_cast_fp16))[name = tensor("op_7331_cast_fp16")]; tensor var_7332_to_fp16 = const()[name = tensor("op_7332_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_717_cast_fp16 = mul(x = var_7331_cast_fp16, y = var_7332_to_fp16)[name = tensor("aw_chunk_717_cast_fp16")]; tensor var_7335_equation_0 = const()[name = tensor("op_7335_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7335_cast_fp16 = einsum(equation = var_7335_equation_0, values = (var_7245_cast_fp16, var_7184_cast_fp16))[name = tensor("op_7335_cast_fp16")]; tensor var_7336_to_fp16 = const()[name = tensor("op_7336_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_719_cast_fp16 = mul(x = var_7335_cast_fp16, y = var_7336_to_fp16)[name = tensor("aw_chunk_719_cast_fp16")]; tensor var_7339_equation_0 = const()[name = tensor("op_7339_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7339_cast_fp16 = einsum(equation = var_7339_equation_0, values = (var_7249_cast_fp16, var_7185_cast_fp16))[name = tensor("op_7339_cast_fp16")]; tensor var_7340_to_fp16 = const()[name = tensor("op_7340_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_721_cast_fp16 = mul(x = var_7339_cast_fp16, y = var_7340_to_fp16)[name = tensor("aw_chunk_721_cast_fp16")]; tensor var_7343_equation_0 = const()[name = tensor("op_7343_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7343_cast_fp16 = einsum(equation = var_7343_equation_0, values = (var_7249_cast_fp16, var_7186_cast_fp16))[name = tensor("op_7343_cast_fp16")]; tensor var_7344_to_fp16 = const()[name = tensor("op_7344_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_723_cast_fp16 = mul(x = var_7343_cast_fp16, y = var_7344_to_fp16)[name = tensor("aw_chunk_723_cast_fp16")]; tensor var_7347_equation_0 = const()[name = tensor("op_7347_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7347_cast_fp16 = einsum(equation = var_7347_equation_0, values = (var_7249_cast_fp16, var_7187_cast_fp16))[name = tensor("op_7347_cast_fp16")]; tensor var_7348_to_fp16 = const()[name = tensor("op_7348_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_725_cast_fp16 = mul(x = var_7347_cast_fp16, y = var_7348_to_fp16)[name = tensor("aw_chunk_725_cast_fp16")]; tensor var_7351_equation_0 = const()[name = tensor("op_7351_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7351_cast_fp16 = einsum(equation = var_7351_equation_0, values = (var_7249_cast_fp16, var_7188_cast_fp16))[name = tensor("op_7351_cast_fp16")]; tensor var_7352_to_fp16 = const()[name = tensor("op_7352_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_727_cast_fp16 = mul(x = var_7351_cast_fp16, y = var_7352_to_fp16)[name = tensor("aw_chunk_727_cast_fp16")]; tensor var_7355_equation_0 = const()[name = tensor("op_7355_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7355_cast_fp16 = einsum(equation = var_7355_equation_0, values = (var_7249_cast_fp16, var_7189_cast_fp16))[name = tensor("op_7355_cast_fp16")]; tensor var_7356_to_fp16 = const()[name = tensor("op_7356_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_729_cast_fp16 = mul(x = var_7355_cast_fp16, y = var_7356_to_fp16)[name = tensor("aw_chunk_729_cast_fp16")]; tensor var_7359_equation_0 = const()[name = tensor("op_7359_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7359_cast_fp16 = einsum(equation = var_7359_equation_0, values = (var_7249_cast_fp16, var_7190_cast_fp16))[name = tensor("op_7359_cast_fp16")]; tensor var_7360_to_fp16 = const()[name = tensor("op_7360_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_731_cast_fp16 = mul(x = var_7359_cast_fp16, y = var_7360_to_fp16)[name = tensor("aw_chunk_731_cast_fp16")]; tensor var_7363_equation_0 = const()[name = tensor("op_7363_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7363_cast_fp16 = einsum(equation = var_7363_equation_0, values = (var_7249_cast_fp16, var_7191_cast_fp16))[name = tensor("op_7363_cast_fp16")]; tensor var_7364_to_fp16 = const()[name = tensor("op_7364_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_733_cast_fp16 = mul(x = var_7363_cast_fp16, y = var_7364_to_fp16)[name = tensor("aw_chunk_733_cast_fp16")]; tensor var_7367_equation_0 = const()[name = tensor("op_7367_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7367_cast_fp16 = einsum(equation = var_7367_equation_0, values = (var_7249_cast_fp16, var_7192_cast_fp16))[name = tensor("op_7367_cast_fp16")]; tensor var_7368_to_fp16 = const()[name = tensor("op_7368_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_735_cast_fp16 = mul(x = var_7367_cast_fp16, y = var_7368_to_fp16)[name = tensor("aw_chunk_735_cast_fp16")]; tensor var_7371_equation_0 = const()[name = tensor("op_7371_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7371_cast_fp16 = einsum(equation = var_7371_equation_0, values = (var_7253_cast_fp16, var_7193_cast_fp16))[name = tensor("op_7371_cast_fp16")]; tensor var_7372_to_fp16 = const()[name = tensor("op_7372_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_737_cast_fp16 = mul(x = var_7371_cast_fp16, y = var_7372_to_fp16)[name = tensor("aw_chunk_737_cast_fp16")]; tensor var_7375_equation_0 = const()[name = tensor("op_7375_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7375_cast_fp16 = einsum(equation = var_7375_equation_0, values = (var_7253_cast_fp16, var_7194_cast_fp16))[name = tensor("op_7375_cast_fp16")]; tensor var_7376_to_fp16 = const()[name = tensor("op_7376_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_739_cast_fp16 = mul(x = var_7375_cast_fp16, y = var_7376_to_fp16)[name = tensor("aw_chunk_739_cast_fp16")]; tensor var_7379_equation_0 = const()[name = tensor("op_7379_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7379_cast_fp16 = einsum(equation = var_7379_equation_0, values = (var_7253_cast_fp16, var_7195_cast_fp16))[name = tensor("op_7379_cast_fp16")]; tensor var_7380_to_fp16 = const()[name = tensor("op_7380_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_741_cast_fp16 = mul(x = var_7379_cast_fp16, y = var_7380_to_fp16)[name = tensor("aw_chunk_741_cast_fp16")]; tensor var_7383_equation_0 = const()[name = tensor("op_7383_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7383_cast_fp16 = einsum(equation = var_7383_equation_0, values = (var_7253_cast_fp16, var_7196_cast_fp16))[name = tensor("op_7383_cast_fp16")]; tensor var_7384_to_fp16 = const()[name = tensor("op_7384_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_743_cast_fp16 = mul(x = var_7383_cast_fp16, y = var_7384_to_fp16)[name = tensor("aw_chunk_743_cast_fp16")]; tensor var_7387_equation_0 = const()[name = tensor("op_7387_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7387_cast_fp16 = einsum(equation = var_7387_equation_0, values = (var_7253_cast_fp16, var_7197_cast_fp16))[name = tensor("op_7387_cast_fp16")]; tensor var_7388_to_fp16 = const()[name = tensor("op_7388_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_745_cast_fp16 = mul(x = var_7387_cast_fp16, y = var_7388_to_fp16)[name = tensor("aw_chunk_745_cast_fp16")]; tensor var_7391_equation_0 = const()[name = tensor("op_7391_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7391_cast_fp16 = einsum(equation = var_7391_equation_0, values = (var_7253_cast_fp16, var_7198_cast_fp16))[name = tensor("op_7391_cast_fp16")]; tensor var_7392_to_fp16 = const()[name = tensor("op_7392_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_747_cast_fp16 = mul(x = var_7391_cast_fp16, y = var_7392_to_fp16)[name = tensor("aw_chunk_747_cast_fp16")]; tensor var_7395_equation_0 = const()[name = tensor("op_7395_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7395_cast_fp16 = einsum(equation = var_7395_equation_0, values = (var_7253_cast_fp16, var_7199_cast_fp16))[name = tensor("op_7395_cast_fp16")]; tensor var_7396_to_fp16 = const()[name = tensor("op_7396_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_749_cast_fp16 = mul(x = var_7395_cast_fp16, y = var_7396_to_fp16)[name = tensor("aw_chunk_749_cast_fp16")]; tensor var_7399_equation_0 = const()[name = tensor("op_7399_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7399_cast_fp16 = einsum(equation = var_7399_equation_0, values = (var_7253_cast_fp16, var_7200_cast_fp16))[name = tensor("op_7399_cast_fp16")]; tensor var_7400_to_fp16 = const()[name = tensor("op_7400_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_751_cast_fp16 = mul(x = var_7399_cast_fp16, y = var_7400_to_fp16)[name = tensor("aw_chunk_751_cast_fp16")]; tensor var_7403_equation_0 = const()[name = tensor("op_7403_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7403_cast_fp16 = einsum(equation = var_7403_equation_0, values = (var_7257_cast_fp16, var_7201_cast_fp16))[name = tensor("op_7403_cast_fp16")]; tensor var_7404_to_fp16 = const()[name = tensor("op_7404_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_753_cast_fp16 = mul(x = var_7403_cast_fp16, y = var_7404_to_fp16)[name = tensor("aw_chunk_753_cast_fp16")]; tensor var_7407_equation_0 = const()[name = tensor("op_7407_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7407_cast_fp16 = einsum(equation = var_7407_equation_0, values = (var_7257_cast_fp16, var_7202_cast_fp16))[name = tensor("op_7407_cast_fp16")]; tensor var_7408_to_fp16 = const()[name = tensor("op_7408_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_755_cast_fp16 = mul(x = var_7407_cast_fp16, y = var_7408_to_fp16)[name = tensor("aw_chunk_755_cast_fp16")]; tensor var_7411_equation_0 = const()[name = tensor("op_7411_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7411_cast_fp16 = einsum(equation = var_7411_equation_0, values = (var_7257_cast_fp16, var_7203_cast_fp16))[name = tensor("op_7411_cast_fp16")]; tensor var_7412_to_fp16 = const()[name = tensor("op_7412_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_757_cast_fp16 = mul(x = var_7411_cast_fp16, y = var_7412_to_fp16)[name = tensor("aw_chunk_757_cast_fp16")]; tensor var_7415_equation_0 = const()[name = tensor("op_7415_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7415_cast_fp16 = einsum(equation = var_7415_equation_0, values = (var_7257_cast_fp16, var_7204_cast_fp16))[name = tensor("op_7415_cast_fp16")]; tensor var_7416_to_fp16 = const()[name = tensor("op_7416_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_759_cast_fp16 = mul(x = var_7415_cast_fp16, y = var_7416_to_fp16)[name = tensor("aw_chunk_759_cast_fp16")]; tensor var_7419_equation_0 = const()[name = tensor("op_7419_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7419_cast_fp16 = einsum(equation = var_7419_equation_0, values = (var_7257_cast_fp16, var_7205_cast_fp16))[name = tensor("op_7419_cast_fp16")]; tensor var_7420_to_fp16 = const()[name = tensor("op_7420_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_761_cast_fp16 = mul(x = var_7419_cast_fp16, y = var_7420_to_fp16)[name = tensor("aw_chunk_761_cast_fp16")]; tensor var_7423_equation_0 = const()[name = tensor("op_7423_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7423_cast_fp16 = einsum(equation = var_7423_equation_0, values = (var_7257_cast_fp16, var_7206_cast_fp16))[name = tensor("op_7423_cast_fp16")]; tensor var_7424_to_fp16 = const()[name = tensor("op_7424_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_763_cast_fp16 = mul(x = var_7423_cast_fp16, y = var_7424_to_fp16)[name = tensor("aw_chunk_763_cast_fp16")]; tensor var_7427_equation_0 = const()[name = tensor("op_7427_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7427_cast_fp16 = einsum(equation = var_7427_equation_0, values = (var_7257_cast_fp16, var_7207_cast_fp16))[name = tensor("op_7427_cast_fp16")]; tensor var_7428_to_fp16 = const()[name = tensor("op_7428_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_765_cast_fp16 = mul(x = var_7427_cast_fp16, y = var_7428_to_fp16)[name = tensor("aw_chunk_765_cast_fp16")]; tensor var_7431_equation_0 = const()[name = tensor("op_7431_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7431_cast_fp16 = einsum(equation = var_7431_equation_0, values = (var_7257_cast_fp16, var_7208_cast_fp16))[name = tensor("op_7431_cast_fp16")]; tensor var_7432_to_fp16 = const()[name = tensor("op_7432_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_767_cast_fp16 = mul(x = var_7431_cast_fp16, y = var_7432_to_fp16)[name = tensor("aw_chunk_767_cast_fp16")]; tensor var_7435_equation_0 = const()[name = tensor("op_7435_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7435_cast_fp16 = einsum(equation = var_7435_equation_0, values = (var_7261_cast_fp16, var_7209_cast_fp16))[name = tensor("op_7435_cast_fp16")]; tensor var_7436_to_fp16 = const()[name = tensor("op_7436_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_769_cast_fp16 = mul(x = var_7435_cast_fp16, y = var_7436_to_fp16)[name = tensor("aw_chunk_769_cast_fp16")]; tensor var_7439_equation_0 = const()[name = tensor("op_7439_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7439_cast_fp16 = einsum(equation = var_7439_equation_0, values = (var_7261_cast_fp16, var_7210_cast_fp16))[name = tensor("op_7439_cast_fp16")]; tensor var_7440_to_fp16 = const()[name = tensor("op_7440_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_771_cast_fp16 = mul(x = var_7439_cast_fp16, y = var_7440_to_fp16)[name = tensor("aw_chunk_771_cast_fp16")]; tensor var_7443_equation_0 = const()[name = tensor("op_7443_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7443_cast_fp16 = einsum(equation = var_7443_equation_0, values = (var_7261_cast_fp16, var_7211_cast_fp16))[name = tensor("op_7443_cast_fp16")]; tensor var_7444_to_fp16 = const()[name = tensor("op_7444_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_773_cast_fp16 = mul(x = var_7443_cast_fp16, y = var_7444_to_fp16)[name = tensor("aw_chunk_773_cast_fp16")]; tensor var_7447_equation_0 = const()[name = tensor("op_7447_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7447_cast_fp16 = einsum(equation = var_7447_equation_0, values = (var_7261_cast_fp16, var_7212_cast_fp16))[name = tensor("op_7447_cast_fp16")]; tensor var_7448_to_fp16 = const()[name = tensor("op_7448_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_775_cast_fp16 = mul(x = var_7447_cast_fp16, y = var_7448_to_fp16)[name = tensor("aw_chunk_775_cast_fp16")]; tensor var_7451_equation_0 = const()[name = tensor("op_7451_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7451_cast_fp16 = einsum(equation = var_7451_equation_0, values = (var_7261_cast_fp16, var_7213_cast_fp16))[name = tensor("op_7451_cast_fp16")]; tensor var_7452_to_fp16 = const()[name = tensor("op_7452_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_777_cast_fp16 = mul(x = var_7451_cast_fp16, y = var_7452_to_fp16)[name = tensor("aw_chunk_777_cast_fp16")]; tensor var_7455_equation_0 = const()[name = tensor("op_7455_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7455_cast_fp16 = einsum(equation = var_7455_equation_0, values = (var_7261_cast_fp16, var_7214_cast_fp16))[name = tensor("op_7455_cast_fp16")]; tensor var_7456_to_fp16 = const()[name = tensor("op_7456_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_779_cast_fp16 = mul(x = var_7455_cast_fp16, y = var_7456_to_fp16)[name = tensor("aw_chunk_779_cast_fp16")]; tensor var_7459_equation_0 = const()[name = tensor("op_7459_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7459_cast_fp16 = einsum(equation = var_7459_equation_0, values = (var_7261_cast_fp16, var_7215_cast_fp16))[name = tensor("op_7459_cast_fp16")]; tensor var_7460_to_fp16 = const()[name = tensor("op_7460_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_781_cast_fp16 = mul(x = var_7459_cast_fp16, y = var_7460_to_fp16)[name = tensor("aw_chunk_781_cast_fp16")]; tensor var_7463_equation_0 = const()[name = tensor("op_7463_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7463_cast_fp16 = einsum(equation = var_7463_equation_0, values = (var_7261_cast_fp16, var_7216_cast_fp16))[name = tensor("op_7463_cast_fp16")]; tensor var_7464_to_fp16 = const()[name = tensor("op_7464_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_783_cast_fp16 = mul(x = var_7463_cast_fp16, y = var_7464_to_fp16)[name = tensor("aw_chunk_783_cast_fp16")]; tensor var_7467_equation_0 = const()[name = tensor("op_7467_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7467_cast_fp16 = einsum(equation = var_7467_equation_0, values = (var_7265_cast_fp16, var_7217_cast_fp16))[name = tensor("op_7467_cast_fp16")]; tensor var_7468_to_fp16 = const()[name = tensor("op_7468_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_785_cast_fp16 = mul(x = var_7467_cast_fp16, y = var_7468_to_fp16)[name = tensor("aw_chunk_785_cast_fp16")]; tensor var_7471_equation_0 = const()[name = tensor("op_7471_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7471_cast_fp16 = einsum(equation = var_7471_equation_0, values = (var_7265_cast_fp16, var_7218_cast_fp16))[name = tensor("op_7471_cast_fp16")]; tensor var_7472_to_fp16 = const()[name = tensor("op_7472_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_787_cast_fp16 = mul(x = var_7471_cast_fp16, y = var_7472_to_fp16)[name = tensor("aw_chunk_787_cast_fp16")]; tensor var_7475_equation_0 = const()[name = tensor("op_7475_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7475_cast_fp16 = einsum(equation = var_7475_equation_0, values = (var_7265_cast_fp16, var_7219_cast_fp16))[name = tensor("op_7475_cast_fp16")]; tensor var_7476_to_fp16 = const()[name = tensor("op_7476_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_789_cast_fp16 = mul(x = var_7475_cast_fp16, y = var_7476_to_fp16)[name = tensor("aw_chunk_789_cast_fp16")]; tensor var_7479_equation_0 = const()[name = tensor("op_7479_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7479_cast_fp16 = einsum(equation = var_7479_equation_0, values = (var_7265_cast_fp16, var_7220_cast_fp16))[name = tensor("op_7479_cast_fp16")]; tensor var_7480_to_fp16 = const()[name = tensor("op_7480_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_791_cast_fp16 = mul(x = var_7479_cast_fp16, y = var_7480_to_fp16)[name = tensor("aw_chunk_791_cast_fp16")]; tensor var_7483_equation_0 = const()[name = tensor("op_7483_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7483_cast_fp16 = einsum(equation = var_7483_equation_0, values = (var_7265_cast_fp16, var_7221_cast_fp16))[name = tensor("op_7483_cast_fp16")]; tensor var_7484_to_fp16 = const()[name = tensor("op_7484_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_793_cast_fp16 = mul(x = var_7483_cast_fp16, y = var_7484_to_fp16)[name = tensor("aw_chunk_793_cast_fp16")]; tensor var_7487_equation_0 = const()[name = tensor("op_7487_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7487_cast_fp16 = einsum(equation = var_7487_equation_0, values = (var_7265_cast_fp16, var_7222_cast_fp16))[name = tensor("op_7487_cast_fp16")]; tensor var_7488_to_fp16 = const()[name = tensor("op_7488_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_795_cast_fp16 = mul(x = var_7487_cast_fp16, y = var_7488_to_fp16)[name = tensor("aw_chunk_795_cast_fp16")]; tensor var_7491_equation_0 = const()[name = tensor("op_7491_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7491_cast_fp16 = einsum(equation = var_7491_equation_0, values = (var_7265_cast_fp16, var_7223_cast_fp16))[name = tensor("op_7491_cast_fp16")]; tensor var_7492_to_fp16 = const()[name = tensor("op_7492_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_797_cast_fp16 = mul(x = var_7491_cast_fp16, y = var_7492_to_fp16)[name = tensor("aw_chunk_797_cast_fp16")]; tensor var_7495_equation_0 = const()[name = tensor("op_7495_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7495_cast_fp16 = einsum(equation = var_7495_equation_0, values = (var_7265_cast_fp16, var_7224_cast_fp16))[name = tensor("op_7495_cast_fp16")]; tensor var_7496_to_fp16 = const()[name = tensor("op_7496_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_799_cast_fp16 = mul(x = var_7495_cast_fp16, y = var_7496_to_fp16)[name = tensor("aw_chunk_799_cast_fp16")]; tensor var_7499_equation_0 = const()[name = tensor("op_7499_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7499_cast_fp16 = einsum(equation = var_7499_equation_0, values = (var_7269_cast_fp16, var_7225_cast_fp16))[name = tensor("op_7499_cast_fp16")]; tensor var_7500_to_fp16 = const()[name = tensor("op_7500_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_801_cast_fp16 = mul(x = var_7499_cast_fp16, y = var_7500_to_fp16)[name = tensor("aw_chunk_801_cast_fp16")]; tensor var_7503_equation_0 = const()[name = tensor("op_7503_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7503_cast_fp16 = einsum(equation = var_7503_equation_0, values = (var_7269_cast_fp16, var_7226_cast_fp16))[name = tensor("op_7503_cast_fp16")]; tensor var_7504_to_fp16 = const()[name = tensor("op_7504_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_803_cast_fp16 = mul(x = var_7503_cast_fp16, y = var_7504_to_fp16)[name = tensor("aw_chunk_803_cast_fp16")]; tensor var_7507_equation_0 = const()[name = tensor("op_7507_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7507_cast_fp16 = einsum(equation = var_7507_equation_0, values = (var_7269_cast_fp16, var_7227_cast_fp16))[name = tensor("op_7507_cast_fp16")]; tensor var_7508_to_fp16 = const()[name = tensor("op_7508_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_805_cast_fp16 = mul(x = var_7507_cast_fp16, y = var_7508_to_fp16)[name = tensor("aw_chunk_805_cast_fp16")]; tensor var_7511_equation_0 = const()[name = tensor("op_7511_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7511_cast_fp16 = einsum(equation = var_7511_equation_0, values = (var_7269_cast_fp16, var_7228_cast_fp16))[name = tensor("op_7511_cast_fp16")]; tensor var_7512_to_fp16 = const()[name = tensor("op_7512_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_807_cast_fp16 = mul(x = var_7511_cast_fp16, y = var_7512_to_fp16)[name = tensor("aw_chunk_807_cast_fp16")]; tensor var_7515_equation_0 = const()[name = tensor("op_7515_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7515_cast_fp16 = einsum(equation = var_7515_equation_0, values = (var_7269_cast_fp16, var_7229_cast_fp16))[name = tensor("op_7515_cast_fp16")]; tensor var_7516_to_fp16 = const()[name = tensor("op_7516_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_809_cast_fp16 = mul(x = var_7515_cast_fp16, y = var_7516_to_fp16)[name = tensor("aw_chunk_809_cast_fp16")]; tensor var_7519_equation_0 = const()[name = tensor("op_7519_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7519_cast_fp16 = einsum(equation = var_7519_equation_0, values = (var_7269_cast_fp16, var_7230_cast_fp16))[name = tensor("op_7519_cast_fp16")]; tensor var_7520_to_fp16 = const()[name = tensor("op_7520_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_811_cast_fp16 = mul(x = var_7519_cast_fp16, y = var_7520_to_fp16)[name = tensor("aw_chunk_811_cast_fp16")]; tensor var_7523_equation_0 = const()[name = tensor("op_7523_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7523_cast_fp16 = einsum(equation = var_7523_equation_0, values = (var_7269_cast_fp16, var_7231_cast_fp16))[name = tensor("op_7523_cast_fp16")]; tensor var_7524_to_fp16 = const()[name = tensor("op_7524_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_813_cast_fp16 = mul(x = var_7523_cast_fp16, y = var_7524_to_fp16)[name = tensor("aw_chunk_813_cast_fp16")]; tensor var_7527_equation_0 = const()[name = tensor("op_7527_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7527_cast_fp16 = einsum(equation = var_7527_equation_0, values = (var_7269_cast_fp16, var_7232_cast_fp16))[name = tensor("op_7527_cast_fp16")]; tensor var_7528_to_fp16 = const()[name = tensor("op_7528_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_815_cast_fp16 = mul(x = var_7527_cast_fp16, y = var_7528_to_fp16)[name = tensor("aw_chunk_815_cast_fp16")]; tensor var_7531_equation_0 = const()[name = tensor("op_7531_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7531_cast_fp16 = einsum(equation = var_7531_equation_0, values = (var_7273_cast_fp16, var_7233_cast_fp16))[name = tensor("op_7531_cast_fp16")]; tensor var_7532_to_fp16 = const()[name = tensor("op_7532_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_817_cast_fp16 = mul(x = var_7531_cast_fp16, y = var_7532_to_fp16)[name = tensor("aw_chunk_817_cast_fp16")]; tensor var_7535_equation_0 = const()[name = tensor("op_7535_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7535_cast_fp16 = einsum(equation = var_7535_equation_0, values = (var_7273_cast_fp16, var_7234_cast_fp16))[name = tensor("op_7535_cast_fp16")]; tensor var_7536_to_fp16 = const()[name = tensor("op_7536_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_819_cast_fp16 = mul(x = var_7535_cast_fp16, y = var_7536_to_fp16)[name = tensor("aw_chunk_819_cast_fp16")]; tensor var_7539_equation_0 = const()[name = tensor("op_7539_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7539_cast_fp16 = einsum(equation = var_7539_equation_0, values = (var_7273_cast_fp16, var_7235_cast_fp16))[name = tensor("op_7539_cast_fp16")]; tensor var_7540_to_fp16 = const()[name = tensor("op_7540_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_821_cast_fp16 = mul(x = var_7539_cast_fp16, y = var_7540_to_fp16)[name = tensor("aw_chunk_821_cast_fp16")]; tensor var_7543_equation_0 = const()[name = tensor("op_7543_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7543_cast_fp16 = einsum(equation = var_7543_equation_0, values = (var_7273_cast_fp16, var_7236_cast_fp16))[name = tensor("op_7543_cast_fp16")]; tensor var_7544_to_fp16 = const()[name = tensor("op_7544_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_823_cast_fp16 = mul(x = var_7543_cast_fp16, y = var_7544_to_fp16)[name = tensor("aw_chunk_823_cast_fp16")]; tensor var_7547_equation_0 = const()[name = tensor("op_7547_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7547_cast_fp16 = einsum(equation = var_7547_equation_0, values = (var_7273_cast_fp16, var_7237_cast_fp16))[name = tensor("op_7547_cast_fp16")]; tensor var_7548_to_fp16 = const()[name = tensor("op_7548_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_825_cast_fp16 = mul(x = var_7547_cast_fp16, y = var_7548_to_fp16)[name = tensor("aw_chunk_825_cast_fp16")]; tensor var_7551_equation_0 = const()[name = tensor("op_7551_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7551_cast_fp16 = einsum(equation = var_7551_equation_0, values = (var_7273_cast_fp16, var_7238_cast_fp16))[name = tensor("op_7551_cast_fp16")]; tensor var_7552_to_fp16 = const()[name = tensor("op_7552_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_827_cast_fp16 = mul(x = var_7551_cast_fp16, y = var_7552_to_fp16)[name = tensor("aw_chunk_827_cast_fp16")]; tensor var_7555_equation_0 = const()[name = tensor("op_7555_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7555_cast_fp16 = einsum(equation = var_7555_equation_0, values = (var_7273_cast_fp16, var_7239_cast_fp16))[name = tensor("op_7555_cast_fp16")]; tensor var_7556_to_fp16 = const()[name = tensor("op_7556_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_829_cast_fp16 = mul(x = var_7555_cast_fp16, y = var_7556_to_fp16)[name = tensor("aw_chunk_829_cast_fp16")]; tensor var_7559_equation_0 = const()[name = tensor("op_7559_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7559_cast_fp16 = einsum(equation = var_7559_equation_0, values = (var_7273_cast_fp16, var_7240_cast_fp16))[name = tensor("op_7559_cast_fp16")]; tensor var_7560_to_fp16 = const()[name = tensor("op_7560_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_831_cast_fp16 = mul(x = var_7559_cast_fp16, y = var_7560_to_fp16)[name = tensor("aw_chunk_831_cast_fp16")]; tensor var_7562_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_705_cast_fp16)[name = tensor("op_7562_cast_fp16")]; tensor var_7563_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_707_cast_fp16)[name = tensor("op_7563_cast_fp16")]; tensor var_7564_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_709_cast_fp16)[name = tensor("op_7564_cast_fp16")]; tensor var_7565_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_711_cast_fp16)[name = tensor("op_7565_cast_fp16")]; tensor var_7566_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_713_cast_fp16)[name = tensor("op_7566_cast_fp16")]; tensor var_7567_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_715_cast_fp16)[name = tensor("op_7567_cast_fp16")]; tensor var_7568_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_717_cast_fp16)[name = tensor("op_7568_cast_fp16")]; tensor var_7569_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_719_cast_fp16)[name = tensor("op_7569_cast_fp16")]; tensor var_7570_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_721_cast_fp16)[name = tensor("op_7570_cast_fp16")]; tensor var_7571_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_723_cast_fp16)[name = tensor("op_7571_cast_fp16")]; tensor var_7572_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_725_cast_fp16)[name = tensor("op_7572_cast_fp16")]; tensor var_7573_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_727_cast_fp16)[name = tensor("op_7573_cast_fp16")]; tensor var_7574_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_729_cast_fp16)[name = tensor("op_7574_cast_fp16")]; tensor var_7575_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_731_cast_fp16)[name = tensor("op_7575_cast_fp16")]; tensor var_7576_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_733_cast_fp16)[name = tensor("op_7576_cast_fp16")]; tensor var_7577_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_735_cast_fp16)[name = tensor("op_7577_cast_fp16")]; tensor var_7578_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_737_cast_fp16)[name = tensor("op_7578_cast_fp16")]; tensor var_7579_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_739_cast_fp16)[name = tensor("op_7579_cast_fp16")]; tensor var_7580_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_741_cast_fp16)[name = tensor("op_7580_cast_fp16")]; tensor var_7581_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_743_cast_fp16)[name = tensor("op_7581_cast_fp16")]; tensor var_7582_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_745_cast_fp16)[name = tensor("op_7582_cast_fp16")]; tensor var_7583_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_747_cast_fp16)[name = tensor("op_7583_cast_fp16")]; tensor var_7584_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_749_cast_fp16)[name = tensor("op_7584_cast_fp16")]; tensor var_7585_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_751_cast_fp16)[name = tensor("op_7585_cast_fp16")]; tensor var_7586_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_753_cast_fp16)[name = tensor("op_7586_cast_fp16")]; tensor var_7587_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_755_cast_fp16)[name = tensor("op_7587_cast_fp16")]; tensor var_7588_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_757_cast_fp16)[name = tensor("op_7588_cast_fp16")]; tensor var_7589_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_759_cast_fp16)[name = tensor("op_7589_cast_fp16")]; tensor var_7590_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_761_cast_fp16)[name = tensor("op_7590_cast_fp16")]; tensor var_7591_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_763_cast_fp16)[name = tensor("op_7591_cast_fp16")]; tensor var_7592_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_765_cast_fp16)[name = tensor("op_7592_cast_fp16")]; tensor var_7593_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_767_cast_fp16)[name = tensor("op_7593_cast_fp16")]; tensor var_7594_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_769_cast_fp16)[name = tensor("op_7594_cast_fp16")]; tensor var_7595_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_771_cast_fp16)[name = tensor("op_7595_cast_fp16")]; tensor var_7596_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_773_cast_fp16)[name = tensor("op_7596_cast_fp16")]; tensor var_7597_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_775_cast_fp16)[name = tensor("op_7597_cast_fp16")]; tensor var_7598_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_777_cast_fp16)[name = tensor("op_7598_cast_fp16")]; tensor var_7599_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_779_cast_fp16)[name = tensor("op_7599_cast_fp16")]; tensor var_7600_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_781_cast_fp16)[name = tensor("op_7600_cast_fp16")]; tensor var_7601_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_783_cast_fp16)[name = tensor("op_7601_cast_fp16")]; tensor var_7602_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_785_cast_fp16)[name = tensor("op_7602_cast_fp16")]; tensor var_7603_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_787_cast_fp16)[name = tensor("op_7603_cast_fp16")]; tensor var_7604_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_789_cast_fp16)[name = tensor("op_7604_cast_fp16")]; tensor var_7605_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_791_cast_fp16)[name = tensor("op_7605_cast_fp16")]; tensor var_7606_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_793_cast_fp16)[name = tensor("op_7606_cast_fp16")]; tensor var_7607_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_795_cast_fp16)[name = tensor("op_7607_cast_fp16")]; tensor var_7608_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_797_cast_fp16)[name = tensor("op_7608_cast_fp16")]; tensor var_7609_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_799_cast_fp16)[name = tensor("op_7609_cast_fp16")]; tensor var_7610_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_801_cast_fp16)[name = tensor("op_7610_cast_fp16")]; tensor var_7611_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_803_cast_fp16)[name = tensor("op_7611_cast_fp16")]; tensor var_7612_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_805_cast_fp16)[name = tensor("op_7612_cast_fp16")]; tensor var_7613_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_807_cast_fp16)[name = tensor("op_7613_cast_fp16")]; tensor var_7614_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_809_cast_fp16)[name = tensor("op_7614_cast_fp16")]; tensor var_7615_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_811_cast_fp16)[name = tensor("op_7615_cast_fp16")]; tensor var_7616_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_813_cast_fp16)[name = tensor("op_7616_cast_fp16")]; tensor var_7617_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_815_cast_fp16)[name = tensor("op_7617_cast_fp16")]; tensor var_7618_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_817_cast_fp16)[name = tensor("op_7618_cast_fp16")]; tensor var_7619_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_819_cast_fp16)[name = tensor("op_7619_cast_fp16")]; tensor var_7620_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_821_cast_fp16)[name = tensor("op_7620_cast_fp16")]; tensor var_7621_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_823_cast_fp16)[name = tensor("op_7621_cast_fp16")]; tensor var_7622_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_825_cast_fp16)[name = tensor("op_7622_cast_fp16")]; tensor var_7623_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_827_cast_fp16)[name = tensor("op_7623_cast_fp16")]; tensor var_7624_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_829_cast_fp16)[name = tensor("op_7624_cast_fp16")]; tensor var_7625_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_831_cast_fp16)[name = tensor("op_7625_cast_fp16")]; tensor var_7627_equation_0 = const()[name = tensor("op_7627_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7627_cast_fp16 = einsum(equation = var_7627_equation_0, values = (var_7275_cast_fp16, var_7562_cast_fp16))[name = tensor("op_7627_cast_fp16")]; tensor var_7629_equation_0 = const()[name = tensor("op_7629_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7629_cast_fp16 = einsum(equation = var_7629_equation_0, values = (var_7275_cast_fp16, var_7563_cast_fp16))[name = tensor("op_7629_cast_fp16")]; tensor var_7631_equation_0 = const()[name = tensor("op_7631_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7631_cast_fp16 = einsum(equation = var_7631_equation_0, values = (var_7275_cast_fp16, var_7564_cast_fp16))[name = tensor("op_7631_cast_fp16")]; tensor var_7633_equation_0 = const()[name = tensor("op_7633_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7633_cast_fp16 = einsum(equation = var_7633_equation_0, values = (var_7275_cast_fp16, var_7565_cast_fp16))[name = tensor("op_7633_cast_fp16")]; tensor var_7635_equation_0 = const()[name = tensor("op_7635_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7635_cast_fp16 = einsum(equation = var_7635_equation_0, values = (var_7275_cast_fp16, var_7566_cast_fp16))[name = tensor("op_7635_cast_fp16")]; tensor var_7637_equation_0 = const()[name = tensor("op_7637_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7637_cast_fp16 = einsum(equation = var_7637_equation_0, values = (var_7275_cast_fp16, var_7567_cast_fp16))[name = tensor("op_7637_cast_fp16")]; tensor var_7639_equation_0 = const()[name = tensor("op_7639_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7639_cast_fp16 = einsum(equation = var_7639_equation_0, values = (var_7275_cast_fp16, var_7568_cast_fp16))[name = tensor("op_7639_cast_fp16")]; tensor var_7641_equation_0 = const()[name = tensor("op_7641_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7641_cast_fp16 = einsum(equation = var_7641_equation_0, values = (var_7275_cast_fp16, var_7569_cast_fp16))[name = tensor("op_7641_cast_fp16")]; tensor var_7643_equation_0 = const()[name = tensor("op_7643_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7643_cast_fp16 = einsum(equation = var_7643_equation_0, values = (var_7279_cast_fp16, var_7570_cast_fp16))[name = tensor("op_7643_cast_fp16")]; tensor var_7645_equation_0 = const()[name = tensor("op_7645_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7645_cast_fp16 = einsum(equation = var_7645_equation_0, values = (var_7279_cast_fp16, var_7571_cast_fp16))[name = tensor("op_7645_cast_fp16")]; tensor var_7647_equation_0 = const()[name = tensor("op_7647_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7647_cast_fp16 = einsum(equation = var_7647_equation_0, values = (var_7279_cast_fp16, var_7572_cast_fp16))[name = tensor("op_7647_cast_fp16")]; tensor var_7649_equation_0 = const()[name = tensor("op_7649_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7649_cast_fp16 = einsum(equation = var_7649_equation_0, values = (var_7279_cast_fp16, var_7573_cast_fp16))[name = tensor("op_7649_cast_fp16")]; tensor var_7651_equation_0 = const()[name = tensor("op_7651_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7651_cast_fp16 = einsum(equation = var_7651_equation_0, values = (var_7279_cast_fp16, var_7574_cast_fp16))[name = tensor("op_7651_cast_fp16")]; tensor var_7653_equation_0 = const()[name = tensor("op_7653_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7653_cast_fp16 = einsum(equation = var_7653_equation_0, values = (var_7279_cast_fp16, var_7575_cast_fp16))[name = tensor("op_7653_cast_fp16")]; tensor var_7655_equation_0 = const()[name = tensor("op_7655_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7655_cast_fp16 = einsum(equation = var_7655_equation_0, values = (var_7279_cast_fp16, var_7576_cast_fp16))[name = tensor("op_7655_cast_fp16")]; tensor var_7657_equation_0 = const()[name = tensor("op_7657_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7657_cast_fp16 = einsum(equation = var_7657_equation_0, values = (var_7279_cast_fp16, var_7577_cast_fp16))[name = tensor("op_7657_cast_fp16")]; tensor var_7659_equation_0 = const()[name = tensor("op_7659_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7659_cast_fp16 = einsum(equation = var_7659_equation_0, values = (var_7283_cast_fp16, var_7578_cast_fp16))[name = tensor("op_7659_cast_fp16")]; tensor var_7661_equation_0 = const()[name = tensor("op_7661_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7661_cast_fp16 = einsum(equation = var_7661_equation_0, values = (var_7283_cast_fp16, var_7579_cast_fp16))[name = tensor("op_7661_cast_fp16")]; tensor var_7663_equation_0 = const()[name = tensor("op_7663_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7663_cast_fp16 = einsum(equation = var_7663_equation_0, values = (var_7283_cast_fp16, var_7580_cast_fp16))[name = tensor("op_7663_cast_fp16")]; tensor var_7665_equation_0 = const()[name = tensor("op_7665_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7665_cast_fp16 = einsum(equation = var_7665_equation_0, values = (var_7283_cast_fp16, var_7581_cast_fp16))[name = tensor("op_7665_cast_fp16")]; tensor var_7667_equation_0 = const()[name = tensor("op_7667_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7667_cast_fp16 = einsum(equation = var_7667_equation_0, values = (var_7283_cast_fp16, var_7582_cast_fp16))[name = tensor("op_7667_cast_fp16")]; tensor var_7669_equation_0 = const()[name = tensor("op_7669_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7669_cast_fp16 = einsum(equation = var_7669_equation_0, values = (var_7283_cast_fp16, var_7583_cast_fp16))[name = tensor("op_7669_cast_fp16")]; tensor var_7671_equation_0 = const()[name = tensor("op_7671_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7671_cast_fp16 = einsum(equation = var_7671_equation_0, values = (var_7283_cast_fp16, var_7584_cast_fp16))[name = tensor("op_7671_cast_fp16")]; tensor var_7673_equation_0 = const()[name = tensor("op_7673_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7673_cast_fp16 = einsum(equation = var_7673_equation_0, values = (var_7283_cast_fp16, var_7585_cast_fp16))[name = tensor("op_7673_cast_fp16")]; tensor var_7675_equation_0 = const()[name = tensor("op_7675_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7675_cast_fp16 = einsum(equation = var_7675_equation_0, values = (var_7287_cast_fp16, var_7586_cast_fp16))[name = tensor("op_7675_cast_fp16")]; tensor var_7677_equation_0 = const()[name = tensor("op_7677_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7677_cast_fp16 = einsum(equation = var_7677_equation_0, values = (var_7287_cast_fp16, var_7587_cast_fp16))[name = tensor("op_7677_cast_fp16")]; tensor var_7679_equation_0 = const()[name = tensor("op_7679_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7679_cast_fp16 = einsum(equation = var_7679_equation_0, values = (var_7287_cast_fp16, var_7588_cast_fp16))[name = tensor("op_7679_cast_fp16")]; tensor var_7681_equation_0 = const()[name = tensor("op_7681_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7681_cast_fp16 = einsum(equation = var_7681_equation_0, values = (var_7287_cast_fp16, var_7589_cast_fp16))[name = tensor("op_7681_cast_fp16")]; tensor var_7683_equation_0 = const()[name = tensor("op_7683_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7683_cast_fp16 = einsum(equation = var_7683_equation_0, values = (var_7287_cast_fp16, var_7590_cast_fp16))[name = tensor("op_7683_cast_fp16")]; tensor var_7685_equation_0 = const()[name = tensor("op_7685_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7685_cast_fp16 = einsum(equation = var_7685_equation_0, values = (var_7287_cast_fp16, var_7591_cast_fp16))[name = tensor("op_7685_cast_fp16")]; tensor var_7687_equation_0 = const()[name = tensor("op_7687_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7687_cast_fp16 = einsum(equation = var_7687_equation_0, values = (var_7287_cast_fp16, var_7592_cast_fp16))[name = tensor("op_7687_cast_fp16")]; tensor var_7689_equation_0 = const()[name = tensor("op_7689_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7689_cast_fp16 = einsum(equation = var_7689_equation_0, values = (var_7287_cast_fp16, var_7593_cast_fp16))[name = tensor("op_7689_cast_fp16")]; tensor var_7691_equation_0 = const()[name = tensor("op_7691_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7691_cast_fp16 = einsum(equation = var_7691_equation_0, values = (var_7291_cast_fp16, var_7594_cast_fp16))[name = tensor("op_7691_cast_fp16")]; tensor var_7693_equation_0 = const()[name = tensor("op_7693_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7693_cast_fp16 = einsum(equation = var_7693_equation_0, values = (var_7291_cast_fp16, var_7595_cast_fp16))[name = tensor("op_7693_cast_fp16")]; tensor var_7695_equation_0 = const()[name = tensor("op_7695_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7695_cast_fp16 = einsum(equation = var_7695_equation_0, values = (var_7291_cast_fp16, var_7596_cast_fp16))[name = tensor("op_7695_cast_fp16")]; tensor var_7697_equation_0 = const()[name = tensor("op_7697_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7697_cast_fp16 = einsum(equation = var_7697_equation_0, values = (var_7291_cast_fp16, var_7597_cast_fp16))[name = tensor("op_7697_cast_fp16")]; tensor var_7699_equation_0 = const()[name = tensor("op_7699_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7699_cast_fp16 = einsum(equation = var_7699_equation_0, values = (var_7291_cast_fp16, var_7598_cast_fp16))[name = tensor("op_7699_cast_fp16")]; tensor var_7701_equation_0 = const()[name = tensor("op_7701_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7701_cast_fp16 = einsum(equation = var_7701_equation_0, values = (var_7291_cast_fp16, var_7599_cast_fp16))[name = tensor("op_7701_cast_fp16")]; tensor var_7703_equation_0 = const()[name = tensor("op_7703_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7703_cast_fp16 = einsum(equation = var_7703_equation_0, values = (var_7291_cast_fp16, var_7600_cast_fp16))[name = tensor("op_7703_cast_fp16")]; tensor var_7705_equation_0 = const()[name = tensor("op_7705_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7705_cast_fp16 = einsum(equation = var_7705_equation_0, values = (var_7291_cast_fp16, var_7601_cast_fp16))[name = tensor("op_7705_cast_fp16")]; tensor var_7707_equation_0 = const()[name = tensor("op_7707_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7707_cast_fp16 = einsum(equation = var_7707_equation_0, values = (var_7295_cast_fp16, var_7602_cast_fp16))[name = tensor("op_7707_cast_fp16")]; tensor var_7709_equation_0 = const()[name = tensor("op_7709_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7709_cast_fp16 = einsum(equation = var_7709_equation_0, values = (var_7295_cast_fp16, var_7603_cast_fp16))[name = tensor("op_7709_cast_fp16")]; tensor var_7711_equation_0 = const()[name = tensor("op_7711_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7711_cast_fp16 = einsum(equation = var_7711_equation_0, values = (var_7295_cast_fp16, var_7604_cast_fp16))[name = tensor("op_7711_cast_fp16")]; tensor var_7713_equation_0 = const()[name = tensor("op_7713_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7713_cast_fp16 = einsum(equation = var_7713_equation_0, values = (var_7295_cast_fp16, var_7605_cast_fp16))[name = tensor("op_7713_cast_fp16")]; tensor var_7715_equation_0 = const()[name = tensor("op_7715_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7715_cast_fp16 = einsum(equation = var_7715_equation_0, values = (var_7295_cast_fp16, var_7606_cast_fp16))[name = tensor("op_7715_cast_fp16")]; tensor var_7717_equation_0 = const()[name = tensor("op_7717_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7717_cast_fp16 = einsum(equation = var_7717_equation_0, values = (var_7295_cast_fp16, var_7607_cast_fp16))[name = tensor("op_7717_cast_fp16")]; tensor var_7719_equation_0 = const()[name = tensor("op_7719_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7719_cast_fp16 = einsum(equation = var_7719_equation_0, values = (var_7295_cast_fp16, var_7608_cast_fp16))[name = tensor("op_7719_cast_fp16")]; tensor var_7721_equation_0 = const()[name = tensor("op_7721_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7721_cast_fp16 = einsum(equation = var_7721_equation_0, values = (var_7295_cast_fp16, var_7609_cast_fp16))[name = tensor("op_7721_cast_fp16")]; tensor var_7723_equation_0 = const()[name = tensor("op_7723_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7723_cast_fp16 = einsum(equation = var_7723_equation_0, values = (var_7299_cast_fp16, var_7610_cast_fp16))[name = tensor("op_7723_cast_fp16")]; tensor var_7725_equation_0 = const()[name = tensor("op_7725_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7725_cast_fp16 = einsum(equation = var_7725_equation_0, values = (var_7299_cast_fp16, var_7611_cast_fp16))[name = tensor("op_7725_cast_fp16")]; tensor var_7727_equation_0 = const()[name = tensor("op_7727_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7727_cast_fp16 = einsum(equation = var_7727_equation_0, values = (var_7299_cast_fp16, var_7612_cast_fp16))[name = tensor("op_7727_cast_fp16")]; tensor var_7729_equation_0 = const()[name = tensor("op_7729_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7729_cast_fp16 = einsum(equation = var_7729_equation_0, values = (var_7299_cast_fp16, var_7613_cast_fp16))[name = tensor("op_7729_cast_fp16")]; tensor var_7731_equation_0 = const()[name = tensor("op_7731_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7731_cast_fp16 = einsum(equation = var_7731_equation_0, values = (var_7299_cast_fp16, var_7614_cast_fp16))[name = tensor("op_7731_cast_fp16")]; tensor var_7733_equation_0 = const()[name = tensor("op_7733_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7733_cast_fp16 = einsum(equation = var_7733_equation_0, values = (var_7299_cast_fp16, var_7615_cast_fp16))[name = tensor("op_7733_cast_fp16")]; tensor var_7735_equation_0 = const()[name = tensor("op_7735_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7735_cast_fp16 = einsum(equation = var_7735_equation_0, values = (var_7299_cast_fp16, var_7616_cast_fp16))[name = tensor("op_7735_cast_fp16")]; tensor var_7737_equation_0 = const()[name = tensor("op_7737_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7737_cast_fp16 = einsum(equation = var_7737_equation_0, values = (var_7299_cast_fp16, var_7617_cast_fp16))[name = tensor("op_7737_cast_fp16")]; tensor var_7739_equation_0 = const()[name = tensor("op_7739_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7739_cast_fp16 = einsum(equation = var_7739_equation_0, values = (var_7303_cast_fp16, var_7618_cast_fp16))[name = tensor("op_7739_cast_fp16")]; tensor var_7741_equation_0 = const()[name = tensor("op_7741_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7741_cast_fp16 = einsum(equation = var_7741_equation_0, values = (var_7303_cast_fp16, var_7619_cast_fp16))[name = tensor("op_7741_cast_fp16")]; tensor var_7743_equation_0 = const()[name = tensor("op_7743_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7743_cast_fp16 = einsum(equation = var_7743_equation_0, values = (var_7303_cast_fp16, var_7620_cast_fp16))[name = tensor("op_7743_cast_fp16")]; tensor var_7745_equation_0 = const()[name = tensor("op_7745_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7745_cast_fp16 = einsum(equation = var_7745_equation_0, values = (var_7303_cast_fp16, var_7621_cast_fp16))[name = tensor("op_7745_cast_fp16")]; tensor var_7747_equation_0 = const()[name = tensor("op_7747_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7747_cast_fp16 = einsum(equation = var_7747_equation_0, values = (var_7303_cast_fp16, var_7622_cast_fp16))[name = tensor("op_7747_cast_fp16")]; tensor var_7749_equation_0 = const()[name = tensor("op_7749_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7749_cast_fp16 = einsum(equation = var_7749_equation_0, values = (var_7303_cast_fp16, var_7623_cast_fp16))[name = tensor("op_7749_cast_fp16")]; tensor var_7751_equation_0 = const()[name = tensor("op_7751_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7751_cast_fp16 = einsum(equation = var_7751_equation_0, values = (var_7303_cast_fp16, var_7624_cast_fp16))[name = tensor("op_7751_cast_fp16")]; tensor var_7753_equation_0 = const()[name = tensor("op_7753_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7753_cast_fp16 = einsum(equation = var_7753_equation_0, values = (var_7303_cast_fp16, var_7625_cast_fp16))[name = tensor("op_7753_cast_fp16")]; tensor var_7755_interleave_0 = const()[name = tensor("op_7755_interleave_0"), val = tensor(false)]; tensor var_7755_cast_fp16 = concat(axis = var_5509, interleave = var_7755_interleave_0, values = (var_7627_cast_fp16, var_7629_cast_fp16, var_7631_cast_fp16, var_7633_cast_fp16, var_7635_cast_fp16, var_7637_cast_fp16, var_7639_cast_fp16, var_7641_cast_fp16))[name = tensor("op_7755_cast_fp16")]; tensor var_7757_interleave_0 = const()[name = tensor("op_7757_interleave_0"), val = tensor(false)]; tensor var_7757_cast_fp16 = concat(axis = var_5509, interleave = var_7757_interleave_0, values = (var_7643_cast_fp16, var_7645_cast_fp16, var_7647_cast_fp16, var_7649_cast_fp16, var_7651_cast_fp16, var_7653_cast_fp16, var_7655_cast_fp16, var_7657_cast_fp16))[name = tensor("op_7757_cast_fp16")]; tensor var_7759_interleave_0 = const()[name = tensor("op_7759_interleave_0"), val = tensor(false)]; tensor var_7759_cast_fp16 = concat(axis = var_5509, interleave = var_7759_interleave_0, values = (var_7659_cast_fp16, var_7661_cast_fp16, var_7663_cast_fp16, var_7665_cast_fp16, var_7667_cast_fp16, var_7669_cast_fp16, var_7671_cast_fp16, var_7673_cast_fp16))[name = tensor("op_7759_cast_fp16")]; tensor var_7761_interleave_0 = const()[name = tensor("op_7761_interleave_0"), val = tensor(false)]; tensor var_7761_cast_fp16 = concat(axis = var_5509, interleave = var_7761_interleave_0, values = (var_7675_cast_fp16, var_7677_cast_fp16, var_7679_cast_fp16, var_7681_cast_fp16, var_7683_cast_fp16, var_7685_cast_fp16, var_7687_cast_fp16, var_7689_cast_fp16))[name = tensor("op_7761_cast_fp16")]; tensor var_7763_interleave_0 = const()[name = tensor("op_7763_interleave_0"), val = tensor(false)]; tensor var_7763_cast_fp16 = concat(axis = var_5509, interleave = var_7763_interleave_0, values = (var_7691_cast_fp16, var_7693_cast_fp16, var_7695_cast_fp16, var_7697_cast_fp16, var_7699_cast_fp16, var_7701_cast_fp16, var_7703_cast_fp16, var_7705_cast_fp16))[name = tensor("op_7763_cast_fp16")]; tensor var_7765_interleave_0 = const()[name = tensor("op_7765_interleave_0"), val = tensor(false)]; tensor var_7765_cast_fp16 = concat(axis = var_5509, interleave = var_7765_interleave_0, values = (var_7707_cast_fp16, var_7709_cast_fp16, var_7711_cast_fp16, var_7713_cast_fp16, var_7715_cast_fp16, var_7717_cast_fp16, var_7719_cast_fp16, var_7721_cast_fp16))[name = tensor("op_7765_cast_fp16")]; tensor var_7767_interleave_0 = const()[name = tensor("op_7767_interleave_0"), val = tensor(false)]; tensor var_7767_cast_fp16 = concat(axis = var_5509, interleave = var_7767_interleave_0, values = (var_7723_cast_fp16, var_7725_cast_fp16, var_7727_cast_fp16, var_7729_cast_fp16, var_7731_cast_fp16, var_7733_cast_fp16, var_7735_cast_fp16, var_7737_cast_fp16))[name = tensor("op_7767_cast_fp16")]; tensor var_7769_interleave_0 = const()[name = tensor("op_7769_interleave_0"), val = tensor(false)]; tensor var_7769_cast_fp16 = concat(axis = var_5509, interleave = var_7769_interleave_0, values = (var_7739_cast_fp16, var_7741_cast_fp16, var_7743_cast_fp16, var_7745_cast_fp16, var_7747_cast_fp16, var_7749_cast_fp16, var_7751_cast_fp16, var_7753_cast_fp16))[name = tensor("op_7769_cast_fp16")]; tensor input_245_interleave_0 = const()[name = tensor("input_245_interleave_0"), val = tensor(false)]; tensor input_245_cast_fp16 = concat(axis = var_5537, interleave = input_245_interleave_0, values = (var_7755_cast_fp16, var_7757_cast_fp16, var_7759_cast_fp16, var_7761_cast_fp16, var_7763_cast_fp16, var_7765_cast_fp16, var_7767_cast_fp16, var_7769_cast_fp16))[name = tensor("input_245_cast_fp16")]; tensor var_7775 = const()[name = tensor("op_7775"), val = tensor([1, 1])]; tensor var_7777 = const()[name = tensor("op_7777"), val = tensor([1, 1])]; tensor var_7779_pad_type_0 = const()[name = tensor("op_7779_pad_type_0"), val = tensor("custom")]; tensor var_7779_pad_0 = const()[name = tensor("op_7779_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(241187392))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(241264256))), name = tensor("up_blocks_2_attentions_1_transformer_blocks_0_attn1_to_out_0_weight_to_fp16_palettized"), shape = tensor([320, 320, 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(241264448)))]; tensor var_7779_cast_fp16 = conv(bias = up_blocks_2_attentions_1_transformer_blocks_0_attn1_to_out_0_bias_to_fp16, dilations = var_7777, groups = var_5537, pad = var_7779_pad_0, pad_type = var_7779_pad_type_0, strides = var_7775, weight = up_blocks_2_attentions_1_transformer_blocks_0_attn1_to_out_0_weight_to_fp16_palettized, x = input_245_cast_fp16)[name = tensor("op_7779_cast_fp16")]; tensor inputs_51_cast_fp16 = add(x = var_7779_cast_fp16, y = inputs_49_cast_fp16)[name = tensor("inputs_51_cast_fp16")]; tensor var_7783 = const()[name = tensor("op_7783"), val = tensor([1])]; tensor channels_mean_51_cast_fp16 = reduce_mean(axes = var_7783, keep_dims = var_5532, x = inputs_51_cast_fp16)[name = tensor("channels_mean_51_cast_fp16")]; tensor zero_mean_51_cast_fp16 = sub(x = inputs_51_cast_fp16, y = channels_mean_51_cast_fp16)[name = tensor("zero_mean_51_cast_fp16")]; tensor zero_mean_sq_51_cast_fp16 = mul(x = zero_mean_51_cast_fp16, y = zero_mean_51_cast_fp16)[name = tensor("zero_mean_sq_51_cast_fp16")]; tensor var_7787 = const()[name = tensor("op_7787"), val = tensor([1])]; tensor var_7788_cast_fp16 = reduce_mean(axes = var_7787, keep_dims = var_5532, x = zero_mean_sq_51_cast_fp16)[name = tensor("op_7788_cast_fp16")]; tensor var_7789_to_fp16 = const()[name = tensor("op_7789_to_fp16"), val = tensor(0x1.5p-17)]; tensor var_7790_cast_fp16 = add(x = var_7788_cast_fp16, y = var_7789_to_fp16)[name = tensor("op_7790_cast_fp16")]; tensor denom_51_epsilon_0_to_fp16 = const()[name = tensor("denom_51_epsilon_0_to_fp16"), val = tensor(0x1p-24)]; tensor denom_51_cast_fp16 = rsqrt(epsilon = denom_51_epsilon_0_to_fp16, x = var_7790_cast_fp16)[name = tensor("denom_51_cast_fp16")]; tensor out_51_cast_fp16 = mul(x = zero_mean_51_cast_fp16, y = denom_51_cast_fp16)[name = tensor("out_51_cast_fp16")]; tensor var_7794_to_fp16 = const()[name = tensor("op_7794_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(241265152)))]; tensor var_7795_cast_fp16 = add(x = out_51_cast_fp16, y = var_7794_to_fp16)[name = tensor("op_7795_cast_fp16")]; tensor var_7797_to_fp16 = const()[name = tensor("op_7797_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(241265856)))]; tensor hidden_states_167_cast_fp16 = mul(x = var_7795_cast_fp16, y = var_7797_to_fp16)[name = tensor("hidden_states_167_cast_fp16")]; tensor var_7804 = const()[name = tensor("op_7804"), val = tensor([1, 1])]; tensor var_7806 = const()[name = tensor("op_7806"), 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_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(241266560))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(241343424))), name = tensor("up_blocks_2_attentions_1_transformer_blocks_0_attn2_to_q_weight_to_fp16_palettized"), shape = tensor([320, 320, 1, 1])]; tensor q_cast_fp16 = conv(dilations = var_7806, groups = var_5537, pad = q_pad_0, pad_type = q_pad_type_0, strides = var_7804, weight = up_blocks_2_attentions_1_transformer_blocks_0_attn2_to_q_weight_to_fp16_palettized, x = hidden_states_167_cast_fp16)[name = tensor("q_cast_fp16")]; tensor var_7810 = const()[name = tensor("op_7810"), val = tensor([1, 1])]; tensor var_7812 = const()[name = tensor("op_7812"), 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_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(241343616))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(241528000))), name = tensor("up_blocks_2_attentions_1_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized"), shape = tensor([320, 768, 1, 1])]; tensor k_69_cast_fp16 = conv(dilations = var_7812, groups = var_5537, pad = k_69_pad_0, pad_type = k_69_pad_type_0, strides = var_7810, weight = up_blocks_2_attentions_1_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor("k_69_cast_fp16")]; tensor var_7816 = const()[name = tensor("op_7816"), val = tensor([1, 1])]; tensor var_7818 = const()[name = tensor("op_7818"), 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_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(241528192))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(241712576))), name = tensor("up_blocks_2_attentions_1_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized"), shape = tensor([320, 768, 1, 1])]; tensor v_cast_fp16 = conv(dilations = var_7818, groups = var_5537, pad = v_pad_0, pad_type = v_pad_type_0, strides = var_7816, weight = up_blocks_2_attentions_1_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized, x = encoder_hidden_states)[name = tensor("v_cast_fp16")]; tensor var_7822_begin_0 = const()[name = tensor("op_7822_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_7822_end_0 = const()[name = tensor("op_7822_end_0"), val = tensor([2, 40, 1, 4096])]; tensor var_7822_end_mask_0 = const()[name = tensor("op_7822_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7822_cast_fp16 = slice_by_index(begin = var_7822_begin_0, end = var_7822_end_0, end_mask = var_7822_end_mask_0, x = q_cast_fp16)[name = tensor("op_7822_cast_fp16")]; tensor var_7826_begin_0 = const()[name = tensor("op_7826_begin_0"), val = tensor([0, 40, 0, 0])]; tensor var_7826_end_0 = const()[name = tensor("op_7826_end_0"), val = tensor([2, 80, 1, 4096])]; tensor var_7826_end_mask_0 = const()[name = tensor("op_7826_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7826_cast_fp16 = slice_by_index(begin = var_7826_begin_0, end = var_7826_end_0, end_mask = var_7826_end_mask_0, x = q_cast_fp16)[name = tensor("op_7826_cast_fp16")]; tensor var_7830_begin_0 = const()[name = tensor("op_7830_begin_0"), val = tensor([0, 80, 0, 0])]; tensor var_7830_end_0 = const()[name = tensor("op_7830_end_0"), val = tensor([2, 120, 1, 4096])]; tensor var_7830_end_mask_0 = const()[name = tensor("op_7830_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7830_cast_fp16 = slice_by_index(begin = var_7830_begin_0, end = var_7830_end_0, end_mask = var_7830_end_mask_0, x = q_cast_fp16)[name = tensor("op_7830_cast_fp16")]; tensor var_7834_begin_0 = const()[name = tensor("op_7834_begin_0"), val = tensor([0, 120, 0, 0])]; tensor var_7834_end_0 = const()[name = tensor("op_7834_end_0"), val = tensor([2, 160, 1, 4096])]; tensor var_7834_end_mask_0 = const()[name = tensor("op_7834_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7834_cast_fp16 = slice_by_index(begin = var_7834_begin_0, end = var_7834_end_0, end_mask = var_7834_end_mask_0, x = q_cast_fp16)[name = tensor("op_7834_cast_fp16")]; tensor var_7838_begin_0 = const()[name = tensor("op_7838_begin_0"), val = tensor([0, 160, 0, 0])]; tensor var_7838_end_0 = const()[name = tensor("op_7838_end_0"), val = tensor([2, 200, 1, 4096])]; tensor var_7838_end_mask_0 = const()[name = tensor("op_7838_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7838_cast_fp16 = slice_by_index(begin = var_7838_begin_0, end = var_7838_end_0, end_mask = var_7838_end_mask_0, x = q_cast_fp16)[name = tensor("op_7838_cast_fp16")]; tensor var_7842_begin_0 = const()[name = tensor("op_7842_begin_0"), val = tensor([0, 200, 0, 0])]; tensor var_7842_end_0 = const()[name = tensor("op_7842_end_0"), val = tensor([2, 240, 1, 4096])]; tensor var_7842_end_mask_0 = const()[name = tensor("op_7842_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7842_cast_fp16 = slice_by_index(begin = var_7842_begin_0, end = var_7842_end_0, end_mask = var_7842_end_mask_0, x = q_cast_fp16)[name = tensor("op_7842_cast_fp16")]; tensor var_7846_begin_0 = const()[name = tensor("op_7846_begin_0"), val = tensor([0, 240, 0, 0])]; tensor var_7846_end_0 = const()[name = tensor("op_7846_end_0"), val = tensor([2, 280, 1, 4096])]; tensor var_7846_end_mask_0 = const()[name = tensor("op_7846_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7846_cast_fp16 = slice_by_index(begin = var_7846_begin_0, end = var_7846_end_0, end_mask = var_7846_end_mask_0, x = q_cast_fp16)[name = tensor("op_7846_cast_fp16")]; tensor var_7850_begin_0 = const()[name = tensor("op_7850_begin_0"), val = tensor([0, 280, 0, 0])]; tensor var_7850_end_0 = const()[name = tensor("op_7850_end_0"), val = tensor([2, 320, 1, 4096])]; tensor var_7850_end_mask_0 = const()[name = tensor("op_7850_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7850_cast_fp16 = slice_by_index(begin = var_7850_begin_0, end = var_7850_end_0, end_mask = var_7850_end_mask_0, x = q_cast_fp16)[name = tensor("op_7850_cast_fp16")]; tensor var_7853_begin_0 = const()[name = tensor("op_7853_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_7853_end_0 = const()[name = tensor("op_7853_end_0"), val = tensor([2, 40, 1, 512])]; tensor var_7853_end_mask_0 = const()[name = tensor("op_7853_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7853_cast_fp16 = slice_by_index(begin = var_7853_begin_0, end = var_7853_end_0, end_mask = var_7853_end_mask_0, x = var_7822_cast_fp16)[name = tensor("op_7853_cast_fp16")]; tensor var_7854_begin_0 = const()[name = tensor("op_7854_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_7854_end_0 = const()[name = tensor("op_7854_end_0"), val = tensor([2, 40, 1, 1024])]; tensor var_7854_end_mask_0 = const()[name = tensor("op_7854_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7854_cast_fp16 = slice_by_index(begin = var_7854_begin_0, end = var_7854_end_0, end_mask = var_7854_end_mask_0, x = var_7822_cast_fp16)[name = tensor("op_7854_cast_fp16")]; tensor var_7855_begin_0 = const()[name = tensor("op_7855_begin_0"), val = tensor([0, 0, 0, 1024])]; tensor var_7855_end_0 = const()[name = tensor("op_7855_end_0"), val = tensor([2, 40, 1, 1536])]; tensor var_7855_end_mask_0 = const()[name = tensor("op_7855_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7855_cast_fp16 = slice_by_index(begin = var_7855_begin_0, end = var_7855_end_0, end_mask = var_7855_end_mask_0, x = var_7822_cast_fp16)[name = tensor("op_7855_cast_fp16")]; tensor var_7856_begin_0 = const()[name = tensor("op_7856_begin_0"), val = tensor([0, 0, 0, 1536])]; tensor var_7856_end_0 = const()[name = tensor("op_7856_end_0"), val = tensor([2, 40, 1, 2048])]; tensor var_7856_end_mask_0 = const()[name = tensor("op_7856_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7856_cast_fp16 = slice_by_index(begin = var_7856_begin_0, end = var_7856_end_0, end_mask = var_7856_end_mask_0, x = var_7822_cast_fp16)[name = tensor("op_7856_cast_fp16")]; tensor var_7857_begin_0 = const()[name = tensor("op_7857_begin_0"), val = tensor([0, 0, 0, 2048])]; tensor var_7857_end_0 = const()[name = tensor("op_7857_end_0"), val = tensor([2, 40, 1, 2560])]; tensor var_7857_end_mask_0 = const()[name = tensor("op_7857_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7857_cast_fp16 = slice_by_index(begin = var_7857_begin_0, end = var_7857_end_0, end_mask = var_7857_end_mask_0, x = var_7822_cast_fp16)[name = tensor("op_7857_cast_fp16")]; tensor var_7858_begin_0 = const()[name = tensor("op_7858_begin_0"), val = tensor([0, 0, 0, 2560])]; tensor var_7858_end_0 = const()[name = tensor("op_7858_end_0"), val = tensor([2, 40, 1, 3072])]; tensor var_7858_end_mask_0 = const()[name = tensor("op_7858_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7858_cast_fp16 = slice_by_index(begin = var_7858_begin_0, end = var_7858_end_0, end_mask = var_7858_end_mask_0, x = var_7822_cast_fp16)[name = tensor("op_7858_cast_fp16")]; tensor var_7859_begin_0 = const()[name = tensor("op_7859_begin_0"), val = tensor([0, 0, 0, 3072])]; tensor var_7859_end_0 = const()[name = tensor("op_7859_end_0"), val = tensor([2, 40, 1, 3584])]; tensor var_7859_end_mask_0 = const()[name = tensor("op_7859_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7859_cast_fp16 = slice_by_index(begin = var_7859_begin_0, end = var_7859_end_0, end_mask = var_7859_end_mask_0, x = var_7822_cast_fp16)[name = tensor("op_7859_cast_fp16")]; tensor var_7860_begin_0 = const()[name = tensor("op_7860_begin_0"), val = tensor([0, 0, 0, 3584])]; tensor var_7860_end_0 = const()[name = tensor("op_7860_end_0"), val = tensor([2, 40, 1, 4096])]; tensor var_7860_end_mask_0 = const()[name = tensor("op_7860_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7860_cast_fp16 = slice_by_index(begin = var_7860_begin_0, end = var_7860_end_0, end_mask = var_7860_end_mask_0, x = var_7822_cast_fp16)[name = tensor("op_7860_cast_fp16")]; tensor var_7861_begin_0 = const()[name = tensor("op_7861_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_7861_end_0 = const()[name = tensor("op_7861_end_0"), val = tensor([2, 40, 1, 512])]; tensor var_7861_end_mask_0 = const()[name = tensor("op_7861_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7861_cast_fp16 = slice_by_index(begin = var_7861_begin_0, end = var_7861_end_0, end_mask = var_7861_end_mask_0, x = var_7826_cast_fp16)[name = tensor("op_7861_cast_fp16")]; tensor var_7862_begin_0 = const()[name = tensor("op_7862_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_7862_end_0 = const()[name = tensor("op_7862_end_0"), val = tensor([2, 40, 1, 1024])]; tensor var_7862_end_mask_0 = const()[name = tensor("op_7862_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7862_cast_fp16 = slice_by_index(begin = var_7862_begin_0, end = var_7862_end_0, end_mask = var_7862_end_mask_0, x = var_7826_cast_fp16)[name = tensor("op_7862_cast_fp16")]; tensor var_7863_begin_0 = const()[name = tensor("op_7863_begin_0"), val = tensor([0, 0, 0, 1024])]; tensor var_7863_end_0 = const()[name = tensor("op_7863_end_0"), val = tensor([2, 40, 1, 1536])]; tensor var_7863_end_mask_0 = const()[name = tensor("op_7863_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7863_cast_fp16 = slice_by_index(begin = var_7863_begin_0, end = var_7863_end_0, end_mask = var_7863_end_mask_0, x = var_7826_cast_fp16)[name = tensor("op_7863_cast_fp16")]; tensor var_7864_begin_0 = const()[name = tensor("op_7864_begin_0"), val = tensor([0, 0, 0, 1536])]; tensor var_7864_end_0 = const()[name = tensor("op_7864_end_0"), val = tensor([2, 40, 1, 2048])]; tensor var_7864_end_mask_0 = const()[name = tensor("op_7864_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7864_cast_fp16 = slice_by_index(begin = var_7864_begin_0, end = var_7864_end_0, end_mask = var_7864_end_mask_0, x = var_7826_cast_fp16)[name = tensor("op_7864_cast_fp16")]; tensor var_7865_begin_0 = const()[name = tensor("op_7865_begin_0"), val = tensor([0, 0, 0, 2048])]; tensor var_7865_end_0 = const()[name = tensor("op_7865_end_0"), val = tensor([2, 40, 1, 2560])]; tensor var_7865_end_mask_0 = const()[name = tensor("op_7865_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7865_cast_fp16 = slice_by_index(begin = var_7865_begin_0, end = var_7865_end_0, end_mask = var_7865_end_mask_0, x = var_7826_cast_fp16)[name = tensor("op_7865_cast_fp16")]; tensor var_7866_begin_0 = const()[name = tensor("op_7866_begin_0"), val = tensor([0, 0, 0, 2560])]; tensor var_7866_end_0 = const()[name = tensor("op_7866_end_0"), val = tensor([2, 40, 1, 3072])]; tensor var_7866_end_mask_0 = const()[name = tensor("op_7866_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7866_cast_fp16 = slice_by_index(begin = var_7866_begin_0, end = var_7866_end_0, end_mask = var_7866_end_mask_0, x = var_7826_cast_fp16)[name = tensor("op_7866_cast_fp16")]; tensor var_7867_begin_0 = const()[name = tensor("op_7867_begin_0"), val = tensor([0, 0, 0, 3072])]; tensor var_7867_end_0 = const()[name = tensor("op_7867_end_0"), val = tensor([2, 40, 1, 3584])]; tensor var_7867_end_mask_0 = const()[name = tensor("op_7867_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7867_cast_fp16 = slice_by_index(begin = var_7867_begin_0, end = var_7867_end_0, end_mask = var_7867_end_mask_0, x = var_7826_cast_fp16)[name = tensor("op_7867_cast_fp16")]; tensor var_7868_begin_0 = const()[name = tensor("op_7868_begin_0"), val = tensor([0, 0, 0, 3584])]; tensor var_7868_end_0 = const()[name = tensor("op_7868_end_0"), val = tensor([2, 40, 1, 4096])]; tensor var_7868_end_mask_0 = const()[name = tensor("op_7868_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7868_cast_fp16 = slice_by_index(begin = var_7868_begin_0, end = var_7868_end_0, end_mask = var_7868_end_mask_0, x = var_7826_cast_fp16)[name = tensor("op_7868_cast_fp16")]; tensor var_7869_begin_0 = const()[name = tensor("op_7869_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_7869_end_0 = const()[name = tensor("op_7869_end_0"), val = tensor([2, 40, 1, 512])]; tensor var_7869_end_mask_0 = const()[name = tensor("op_7869_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7869_cast_fp16 = slice_by_index(begin = var_7869_begin_0, end = var_7869_end_0, end_mask = var_7869_end_mask_0, x = var_7830_cast_fp16)[name = tensor("op_7869_cast_fp16")]; tensor var_7870_begin_0 = const()[name = tensor("op_7870_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_7870_end_0 = const()[name = tensor("op_7870_end_0"), val = tensor([2, 40, 1, 1024])]; tensor var_7870_end_mask_0 = const()[name = tensor("op_7870_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7870_cast_fp16 = slice_by_index(begin = var_7870_begin_0, end = var_7870_end_0, end_mask = var_7870_end_mask_0, x = var_7830_cast_fp16)[name = tensor("op_7870_cast_fp16")]; tensor var_7871_begin_0 = const()[name = tensor("op_7871_begin_0"), val = tensor([0, 0, 0, 1024])]; tensor var_7871_end_0 = const()[name = tensor("op_7871_end_0"), val = tensor([2, 40, 1, 1536])]; tensor var_7871_end_mask_0 = const()[name = tensor("op_7871_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7871_cast_fp16 = slice_by_index(begin = var_7871_begin_0, end = var_7871_end_0, end_mask = var_7871_end_mask_0, x = var_7830_cast_fp16)[name = tensor("op_7871_cast_fp16")]; tensor var_7872_begin_0 = const()[name = tensor("op_7872_begin_0"), val = tensor([0, 0, 0, 1536])]; tensor var_7872_end_0 = const()[name = tensor("op_7872_end_0"), val = tensor([2, 40, 1, 2048])]; tensor var_7872_end_mask_0 = const()[name = tensor("op_7872_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7872_cast_fp16 = slice_by_index(begin = var_7872_begin_0, end = var_7872_end_0, end_mask = var_7872_end_mask_0, x = var_7830_cast_fp16)[name = tensor("op_7872_cast_fp16")]; tensor var_7873_begin_0 = const()[name = tensor("op_7873_begin_0"), val = tensor([0, 0, 0, 2048])]; tensor var_7873_end_0 = const()[name = tensor("op_7873_end_0"), val = tensor([2, 40, 1, 2560])]; tensor var_7873_end_mask_0 = const()[name = tensor("op_7873_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7873_cast_fp16 = slice_by_index(begin = var_7873_begin_0, end = var_7873_end_0, end_mask = var_7873_end_mask_0, x = var_7830_cast_fp16)[name = tensor("op_7873_cast_fp16")]; tensor var_7874_begin_0 = const()[name = tensor("op_7874_begin_0"), val = tensor([0, 0, 0, 2560])]; tensor var_7874_end_0 = const()[name = tensor("op_7874_end_0"), val = tensor([2, 40, 1, 3072])]; tensor var_7874_end_mask_0 = const()[name = tensor("op_7874_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7874_cast_fp16 = slice_by_index(begin = var_7874_begin_0, end = var_7874_end_0, end_mask = var_7874_end_mask_0, x = var_7830_cast_fp16)[name = tensor("op_7874_cast_fp16")]; tensor var_7875_begin_0 = const()[name = tensor("op_7875_begin_0"), val = tensor([0, 0, 0, 3072])]; tensor var_7875_end_0 = const()[name = tensor("op_7875_end_0"), val = tensor([2, 40, 1, 3584])]; tensor var_7875_end_mask_0 = const()[name = tensor("op_7875_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7875_cast_fp16 = slice_by_index(begin = var_7875_begin_0, end = var_7875_end_0, end_mask = var_7875_end_mask_0, x = var_7830_cast_fp16)[name = tensor("op_7875_cast_fp16")]; tensor var_7876_begin_0 = const()[name = tensor("op_7876_begin_0"), val = tensor([0, 0, 0, 3584])]; tensor var_7876_end_0 = const()[name = tensor("op_7876_end_0"), val = tensor([2, 40, 1, 4096])]; tensor var_7876_end_mask_0 = const()[name = tensor("op_7876_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7876_cast_fp16 = slice_by_index(begin = var_7876_begin_0, end = var_7876_end_0, end_mask = var_7876_end_mask_0, x = var_7830_cast_fp16)[name = tensor("op_7876_cast_fp16")]; 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, 40, 1, 512])]; tensor var_7877_end_mask_0 = const()[name = tensor("op_7877_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7877_cast_fp16 = slice_by_index(begin = var_7877_begin_0, end = var_7877_end_0, end_mask = var_7877_end_mask_0, x = var_7834_cast_fp16)[name = tensor("op_7877_cast_fp16")]; tensor var_7878_begin_0 = const()[name = tensor("op_7878_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_7878_end_0 = const()[name = tensor("op_7878_end_0"), val = tensor([2, 40, 1, 1024])]; tensor var_7878_end_mask_0 = const()[name = tensor("op_7878_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7878_cast_fp16 = slice_by_index(begin = var_7878_begin_0, end = var_7878_end_0, end_mask = var_7878_end_mask_0, x = var_7834_cast_fp16)[name = tensor("op_7878_cast_fp16")]; tensor var_7879_begin_0 = const()[name = tensor("op_7879_begin_0"), val = tensor([0, 0, 0, 1024])]; tensor var_7879_end_0 = const()[name = tensor("op_7879_end_0"), val = tensor([2, 40, 1, 1536])]; tensor var_7879_end_mask_0 = const()[name = tensor("op_7879_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7879_cast_fp16 = slice_by_index(begin = var_7879_begin_0, end = var_7879_end_0, end_mask = var_7879_end_mask_0, x = var_7834_cast_fp16)[name = tensor("op_7879_cast_fp16")]; tensor var_7880_begin_0 = const()[name = tensor("op_7880_begin_0"), val = tensor([0, 0, 0, 1536])]; tensor var_7880_end_0 = const()[name = tensor("op_7880_end_0"), val = tensor([2, 40, 1, 2048])]; tensor var_7880_end_mask_0 = const()[name = tensor("op_7880_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7880_cast_fp16 = slice_by_index(begin = var_7880_begin_0, end = var_7880_end_0, end_mask = var_7880_end_mask_0, x = var_7834_cast_fp16)[name = tensor("op_7880_cast_fp16")]; tensor var_7881_begin_0 = const()[name = tensor("op_7881_begin_0"), val = tensor([0, 0, 0, 2048])]; tensor var_7881_end_0 = const()[name = tensor("op_7881_end_0"), val = tensor([2, 40, 1, 2560])]; tensor var_7881_end_mask_0 = const()[name = tensor("op_7881_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7881_cast_fp16 = slice_by_index(begin = var_7881_begin_0, end = var_7881_end_0, end_mask = var_7881_end_mask_0, x = var_7834_cast_fp16)[name = tensor("op_7881_cast_fp16")]; tensor var_7882_begin_0 = const()[name = tensor("op_7882_begin_0"), val = tensor([0, 0, 0, 2560])]; tensor var_7882_end_0 = const()[name = tensor("op_7882_end_0"), val = tensor([2, 40, 1, 3072])]; tensor var_7882_end_mask_0 = const()[name = tensor("op_7882_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7882_cast_fp16 = slice_by_index(begin = var_7882_begin_0, end = var_7882_end_0, end_mask = var_7882_end_mask_0, x = var_7834_cast_fp16)[name = tensor("op_7882_cast_fp16")]; tensor var_7883_begin_0 = const()[name = tensor("op_7883_begin_0"), val = tensor([0, 0, 0, 3072])]; tensor var_7883_end_0 = const()[name = tensor("op_7883_end_0"), val = tensor([2, 40, 1, 3584])]; tensor var_7883_end_mask_0 = const()[name = tensor("op_7883_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7883_cast_fp16 = slice_by_index(begin = var_7883_begin_0, end = var_7883_end_0, end_mask = var_7883_end_mask_0, x = var_7834_cast_fp16)[name = tensor("op_7883_cast_fp16")]; tensor var_7884_begin_0 = const()[name = tensor("op_7884_begin_0"), val = tensor([0, 0, 0, 3584])]; tensor var_7884_end_0 = const()[name = tensor("op_7884_end_0"), val = tensor([2, 40, 1, 4096])]; tensor var_7884_end_mask_0 = const()[name = tensor("op_7884_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7884_cast_fp16 = slice_by_index(begin = var_7884_begin_0, end = var_7884_end_0, end_mask = var_7884_end_mask_0, x = var_7834_cast_fp16)[name = tensor("op_7884_cast_fp16")]; tensor var_7885_begin_0 = const()[name = tensor("op_7885_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_7885_end_0 = const()[name = tensor("op_7885_end_0"), val = tensor([2, 40, 1, 512])]; tensor var_7885_end_mask_0 = const()[name = tensor("op_7885_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7885_cast_fp16 = slice_by_index(begin = var_7885_begin_0, end = var_7885_end_0, end_mask = var_7885_end_mask_0, x = var_7838_cast_fp16)[name = tensor("op_7885_cast_fp16")]; tensor var_7886_begin_0 = const()[name = tensor("op_7886_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_7886_end_0 = const()[name = tensor("op_7886_end_0"), val = tensor([2, 40, 1, 1024])]; tensor var_7886_end_mask_0 = const()[name = tensor("op_7886_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7886_cast_fp16 = slice_by_index(begin = var_7886_begin_0, end = var_7886_end_0, end_mask = var_7886_end_mask_0, x = var_7838_cast_fp16)[name = tensor("op_7886_cast_fp16")]; tensor var_7887_begin_0 = const()[name = tensor("op_7887_begin_0"), val = tensor([0, 0, 0, 1024])]; tensor var_7887_end_0 = const()[name = tensor("op_7887_end_0"), val = tensor([2, 40, 1, 1536])]; tensor var_7887_end_mask_0 = const()[name = tensor("op_7887_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7887_cast_fp16 = slice_by_index(begin = var_7887_begin_0, end = var_7887_end_0, end_mask = var_7887_end_mask_0, x = var_7838_cast_fp16)[name = tensor("op_7887_cast_fp16")]; tensor var_7888_begin_0 = const()[name = tensor("op_7888_begin_0"), val = tensor([0, 0, 0, 1536])]; tensor var_7888_end_0 = const()[name = tensor("op_7888_end_0"), val = tensor([2, 40, 1, 2048])]; tensor var_7888_end_mask_0 = const()[name = tensor("op_7888_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7888_cast_fp16 = slice_by_index(begin = var_7888_begin_0, end = var_7888_end_0, end_mask = var_7888_end_mask_0, x = var_7838_cast_fp16)[name = tensor("op_7888_cast_fp16")]; tensor var_7889_begin_0 = const()[name = tensor("op_7889_begin_0"), val = tensor([0, 0, 0, 2048])]; tensor var_7889_end_0 = const()[name = tensor("op_7889_end_0"), val = tensor([2, 40, 1, 2560])]; tensor var_7889_end_mask_0 = const()[name = tensor("op_7889_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7889_cast_fp16 = slice_by_index(begin = var_7889_begin_0, end = var_7889_end_0, end_mask = var_7889_end_mask_0, x = var_7838_cast_fp16)[name = tensor("op_7889_cast_fp16")]; tensor var_7890_begin_0 = const()[name = tensor("op_7890_begin_0"), val = tensor([0, 0, 0, 2560])]; tensor var_7890_end_0 = const()[name = tensor("op_7890_end_0"), val = tensor([2, 40, 1, 3072])]; tensor var_7890_end_mask_0 = const()[name = tensor("op_7890_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7890_cast_fp16 = slice_by_index(begin = var_7890_begin_0, end = var_7890_end_0, end_mask = var_7890_end_mask_0, x = var_7838_cast_fp16)[name = tensor("op_7890_cast_fp16")]; tensor var_7891_begin_0 = const()[name = tensor("op_7891_begin_0"), val = tensor([0, 0, 0, 3072])]; tensor var_7891_end_0 = const()[name = tensor("op_7891_end_0"), val = tensor([2, 40, 1, 3584])]; tensor var_7891_end_mask_0 = const()[name = tensor("op_7891_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7891_cast_fp16 = slice_by_index(begin = var_7891_begin_0, end = var_7891_end_0, end_mask = var_7891_end_mask_0, x = var_7838_cast_fp16)[name = tensor("op_7891_cast_fp16")]; tensor var_7892_begin_0 = const()[name = tensor("op_7892_begin_0"), val = tensor([0, 0, 0, 3584])]; tensor var_7892_end_0 = const()[name = tensor("op_7892_end_0"), val = tensor([2, 40, 1, 4096])]; tensor var_7892_end_mask_0 = const()[name = tensor("op_7892_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7892_cast_fp16 = slice_by_index(begin = var_7892_begin_0, end = var_7892_end_0, end_mask = var_7892_end_mask_0, x = var_7838_cast_fp16)[name = tensor("op_7892_cast_fp16")]; tensor var_7893_begin_0 = const()[name = tensor("op_7893_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_7893_end_0 = const()[name = tensor("op_7893_end_0"), val = tensor([2, 40, 1, 512])]; tensor var_7893_end_mask_0 = const()[name = tensor("op_7893_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7893_cast_fp16 = slice_by_index(begin = var_7893_begin_0, end = var_7893_end_0, end_mask = var_7893_end_mask_0, x = var_7842_cast_fp16)[name = tensor("op_7893_cast_fp16")]; tensor var_7894_begin_0 = const()[name = tensor("op_7894_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_7894_end_0 = const()[name = tensor("op_7894_end_0"), val = tensor([2, 40, 1, 1024])]; tensor var_7894_end_mask_0 = const()[name = tensor("op_7894_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7894_cast_fp16 = slice_by_index(begin = var_7894_begin_0, end = var_7894_end_0, end_mask = var_7894_end_mask_0, x = var_7842_cast_fp16)[name = tensor("op_7894_cast_fp16")]; tensor var_7895_begin_0 = const()[name = tensor("op_7895_begin_0"), val = tensor([0, 0, 0, 1024])]; tensor var_7895_end_0 = const()[name = tensor("op_7895_end_0"), val = tensor([2, 40, 1, 1536])]; tensor var_7895_end_mask_0 = const()[name = tensor("op_7895_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7895_cast_fp16 = slice_by_index(begin = var_7895_begin_0, end = var_7895_end_0, end_mask = var_7895_end_mask_0, x = var_7842_cast_fp16)[name = tensor("op_7895_cast_fp16")]; tensor var_7896_begin_0 = const()[name = tensor("op_7896_begin_0"), val = tensor([0, 0, 0, 1536])]; tensor var_7896_end_0 = const()[name = tensor("op_7896_end_0"), val = tensor([2, 40, 1, 2048])]; tensor var_7896_end_mask_0 = const()[name = tensor("op_7896_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7896_cast_fp16 = slice_by_index(begin = var_7896_begin_0, end = var_7896_end_0, end_mask = var_7896_end_mask_0, x = var_7842_cast_fp16)[name = tensor("op_7896_cast_fp16")]; tensor var_7897_begin_0 = const()[name = tensor("op_7897_begin_0"), val = tensor([0, 0, 0, 2048])]; tensor var_7897_end_0 = const()[name = tensor("op_7897_end_0"), val = tensor([2, 40, 1, 2560])]; tensor var_7897_end_mask_0 = const()[name = tensor("op_7897_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7897_cast_fp16 = slice_by_index(begin = var_7897_begin_0, end = var_7897_end_0, end_mask = var_7897_end_mask_0, x = var_7842_cast_fp16)[name = tensor("op_7897_cast_fp16")]; tensor var_7898_begin_0 = const()[name = tensor("op_7898_begin_0"), val = tensor([0, 0, 0, 2560])]; tensor var_7898_end_0 = const()[name = tensor("op_7898_end_0"), val = tensor([2, 40, 1, 3072])]; tensor var_7898_end_mask_0 = const()[name = tensor("op_7898_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7898_cast_fp16 = slice_by_index(begin = var_7898_begin_0, end = var_7898_end_0, end_mask = var_7898_end_mask_0, x = var_7842_cast_fp16)[name = tensor("op_7898_cast_fp16")]; tensor var_7899_begin_0 = const()[name = tensor("op_7899_begin_0"), val = tensor([0, 0, 0, 3072])]; tensor var_7899_end_0 = const()[name = tensor("op_7899_end_0"), val = tensor([2, 40, 1, 3584])]; tensor var_7899_end_mask_0 = const()[name = tensor("op_7899_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7899_cast_fp16 = slice_by_index(begin = var_7899_begin_0, end = var_7899_end_0, end_mask = var_7899_end_mask_0, x = var_7842_cast_fp16)[name = tensor("op_7899_cast_fp16")]; tensor var_7900_begin_0 = const()[name = tensor("op_7900_begin_0"), val = tensor([0, 0, 0, 3584])]; tensor var_7900_end_0 = const()[name = tensor("op_7900_end_0"), val = tensor([2, 40, 1, 4096])]; tensor var_7900_end_mask_0 = const()[name = tensor("op_7900_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7900_cast_fp16 = slice_by_index(begin = var_7900_begin_0, end = var_7900_end_0, end_mask = var_7900_end_mask_0, x = var_7842_cast_fp16)[name = tensor("op_7900_cast_fp16")]; tensor var_7901_begin_0 = const()[name = tensor("op_7901_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_7901_end_0 = const()[name = tensor("op_7901_end_0"), val = tensor([2, 40, 1, 512])]; tensor var_7901_end_mask_0 = const()[name = tensor("op_7901_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7901_cast_fp16 = slice_by_index(begin = var_7901_begin_0, end = var_7901_end_0, end_mask = var_7901_end_mask_0, x = var_7846_cast_fp16)[name = tensor("op_7901_cast_fp16")]; tensor var_7902_begin_0 = const()[name = tensor("op_7902_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_7902_end_0 = const()[name = tensor("op_7902_end_0"), val = tensor([2, 40, 1, 1024])]; tensor var_7902_end_mask_0 = const()[name = tensor("op_7902_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7902_cast_fp16 = slice_by_index(begin = var_7902_begin_0, end = var_7902_end_0, end_mask = var_7902_end_mask_0, x = var_7846_cast_fp16)[name = tensor("op_7902_cast_fp16")]; tensor var_7903_begin_0 = const()[name = tensor("op_7903_begin_0"), val = tensor([0, 0, 0, 1024])]; tensor var_7903_end_0 = const()[name = tensor("op_7903_end_0"), val = tensor([2, 40, 1, 1536])]; tensor var_7903_end_mask_0 = const()[name = tensor("op_7903_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7903_cast_fp16 = slice_by_index(begin = var_7903_begin_0, end = var_7903_end_0, end_mask = var_7903_end_mask_0, x = var_7846_cast_fp16)[name = tensor("op_7903_cast_fp16")]; tensor var_7904_begin_0 = const()[name = tensor("op_7904_begin_0"), val = tensor([0, 0, 0, 1536])]; tensor var_7904_end_0 = const()[name = tensor("op_7904_end_0"), val = tensor([2, 40, 1, 2048])]; tensor var_7904_end_mask_0 = const()[name = tensor("op_7904_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7904_cast_fp16 = slice_by_index(begin = var_7904_begin_0, end = var_7904_end_0, end_mask = var_7904_end_mask_0, x = var_7846_cast_fp16)[name = tensor("op_7904_cast_fp16")]; tensor var_7905_begin_0 = const()[name = tensor("op_7905_begin_0"), val = tensor([0, 0, 0, 2048])]; tensor var_7905_end_0 = const()[name = tensor("op_7905_end_0"), val = tensor([2, 40, 1, 2560])]; tensor var_7905_end_mask_0 = const()[name = tensor("op_7905_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7905_cast_fp16 = slice_by_index(begin = var_7905_begin_0, end = var_7905_end_0, end_mask = var_7905_end_mask_0, x = var_7846_cast_fp16)[name = tensor("op_7905_cast_fp16")]; tensor var_7906_begin_0 = const()[name = tensor("op_7906_begin_0"), val = tensor([0, 0, 0, 2560])]; tensor var_7906_end_0 = const()[name = tensor("op_7906_end_0"), val = tensor([2, 40, 1, 3072])]; tensor var_7906_end_mask_0 = const()[name = tensor("op_7906_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7906_cast_fp16 = slice_by_index(begin = var_7906_begin_0, end = var_7906_end_0, end_mask = var_7906_end_mask_0, x = var_7846_cast_fp16)[name = tensor("op_7906_cast_fp16")]; tensor var_7907_begin_0 = const()[name = tensor("op_7907_begin_0"), val = tensor([0, 0, 0, 3072])]; tensor var_7907_end_0 = const()[name = tensor("op_7907_end_0"), val = tensor([2, 40, 1, 3584])]; tensor var_7907_end_mask_0 = const()[name = tensor("op_7907_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7907_cast_fp16 = slice_by_index(begin = var_7907_begin_0, end = var_7907_end_0, end_mask = var_7907_end_mask_0, x = var_7846_cast_fp16)[name = tensor("op_7907_cast_fp16")]; tensor var_7908_begin_0 = const()[name = tensor("op_7908_begin_0"), val = tensor([0, 0, 0, 3584])]; tensor var_7908_end_0 = const()[name = tensor("op_7908_end_0"), val = tensor([2, 40, 1, 4096])]; tensor var_7908_end_mask_0 = const()[name = tensor("op_7908_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7908_cast_fp16 = slice_by_index(begin = var_7908_begin_0, end = var_7908_end_0, end_mask = var_7908_end_mask_0, x = var_7846_cast_fp16)[name = tensor("op_7908_cast_fp16")]; tensor var_7909_begin_0 = const()[name = tensor("op_7909_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_7909_end_0 = const()[name = tensor("op_7909_end_0"), val = tensor([2, 40, 1, 512])]; tensor var_7909_end_mask_0 = const()[name = tensor("op_7909_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7909_cast_fp16 = slice_by_index(begin = var_7909_begin_0, end = var_7909_end_0, end_mask = var_7909_end_mask_0, x = var_7850_cast_fp16)[name = tensor("op_7909_cast_fp16")]; tensor var_7910_begin_0 = const()[name = tensor("op_7910_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_7910_end_0 = const()[name = tensor("op_7910_end_0"), val = tensor([2, 40, 1, 1024])]; tensor var_7910_end_mask_0 = const()[name = tensor("op_7910_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7910_cast_fp16 = slice_by_index(begin = var_7910_begin_0, end = var_7910_end_0, end_mask = var_7910_end_mask_0, x = var_7850_cast_fp16)[name = tensor("op_7910_cast_fp16")]; tensor var_7911_begin_0 = const()[name = tensor("op_7911_begin_0"), val = tensor([0, 0, 0, 1024])]; tensor var_7911_end_0 = const()[name = tensor("op_7911_end_0"), val = tensor([2, 40, 1, 1536])]; tensor var_7911_end_mask_0 = const()[name = tensor("op_7911_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7911_cast_fp16 = slice_by_index(begin = var_7911_begin_0, end = var_7911_end_0, end_mask = var_7911_end_mask_0, x = var_7850_cast_fp16)[name = tensor("op_7911_cast_fp16")]; tensor var_7912_begin_0 = const()[name = tensor("op_7912_begin_0"), val = tensor([0, 0, 0, 1536])]; tensor var_7912_end_0 = const()[name = tensor("op_7912_end_0"), val = tensor([2, 40, 1, 2048])]; tensor var_7912_end_mask_0 = const()[name = tensor("op_7912_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7912_cast_fp16 = slice_by_index(begin = var_7912_begin_0, end = var_7912_end_0, end_mask = var_7912_end_mask_0, x = var_7850_cast_fp16)[name = tensor("op_7912_cast_fp16")]; tensor var_7913_begin_0 = const()[name = tensor("op_7913_begin_0"), val = tensor([0, 0, 0, 2048])]; tensor var_7913_end_0 = const()[name = tensor("op_7913_end_0"), val = tensor([2, 40, 1, 2560])]; tensor var_7913_end_mask_0 = const()[name = tensor("op_7913_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7913_cast_fp16 = slice_by_index(begin = var_7913_begin_0, end = var_7913_end_0, end_mask = var_7913_end_mask_0, x = var_7850_cast_fp16)[name = tensor("op_7913_cast_fp16")]; tensor var_7914_begin_0 = const()[name = tensor("op_7914_begin_0"), val = tensor([0, 0, 0, 2560])]; tensor var_7914_end_0 = const()[name = tensor("op_7914_end_0"), val = tensor([2, 40, 1, 3072])]; tensor var_7914_end_mask_0 = const()[name = tensor("op_7914_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7914_cast_fp16 = slice_by_index(begin = var_7914_begin_0, end = var_7914_end_0, end_mask = var_7914_end_mask_0, x = var_7850_cast_fp16)[name = tensor("op_7914_cast_fp16")]; tensor var_7915_begin_0 = const()[name = tensor("op_7915_begin_0"), val = tensor([0, 0, 0, 3072])]; tensor var_7915_end_0 = const()[name = tensor("op_7915_end_0"), val = tensor([2, 40, 1, 3584])]; tensor var_7915_end_mask_0 = const()[name = tensor("op_7915_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7915_cast_fp16 = slice_by_index(begin = var_7915_begin_0, end = var_7915_end_0, end_mask = var_7915_end_mask_0, x = var_7850_cast_fp16)[name = tensor("op_7915_cast_fp16")]; tensor var_7916_begin_0 = const()[name = tensor("op_7916_begin_0"), val = tensor([0, 0, 0, 3584])]; tensor var_7916_end_0 = const()[name = tensor("op_7916_end_0"), val = tensor([2, 40, 1, 4096])]; tensor var_7916_end_mask_0 = const()[name = tensor("op_7916_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7916_cast_fp16 = slice_by_index(begin = var_7916_begin_0, end = var_7916_end_0, end_mask = var_7916_end_mask_0, x = var_7850_cast_fp16)[name = tensor("op_7916_cast_fp16")]; tensor k_perm_0 = const()[name = tensor("k_perm_0"), val = tensor([0, 3, 2, 1])]; tensor var_7921_begin_0 = const()[name = tensor("op_7921_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_7921_end_0 = const()[name = tensor("op_7921_end_0"), val = tensor([2, 77, 1, 40])]; tensor var_7921_end_mask_0 = const()[name = tensor("op_7921_end_mask_0"), val = tensor([true, true, true, false])]; tensor transpose_0 = transpose(perm = k_perm_0, x = k_69_cast_fp16)[name = tensor("transpose_0")]; tensor var_7921_cast_fp16 = slice_by_index(begin = var_7921_begin_0, end = var_7921_end_0, end_mask = var_7921_end_mask_0, x = transpose_0)[name = tensor("op_7921_cast_fp16")]; tensor var_7925_begin_0 = const()[name = tensor("op_7925_begin_0"), val = tensor([0, 0, 0, 40])]; tensor var_7925_end_0 = const()[name = tensor("op_7925_end_0"), val = tensor([2, 77, 1, 80])]; tensor var_7925_end_mask_0 = const()[name = tensor("op_7925_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7925_cast_fp16 = slice_by_index(begin = var_7925_begin_0, end = var_7925_end_0, end_mask = var_7925_end_mask_0, x = transpose_0)[name = tensor("op_7925_cast_fp16")]; tensor var_7929_begin_0 = const()[name = tensor("op_7929_begin_0"), val = tensor([0, 0, 0, 80])]; tensor var_7929_end_0 = const()[name = tensor("op_7929_end_0"), val = tensor([2, 77, 1, 120])]; tensor var_7929_end_mask_0 = const()[name = tensor("op_7929_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7929_cast_fp16 = slice_by_index(begin = var_7929_begin_0, end = var_7929_end_0, end_mask = var_7929_end_mask_0, x = transpose_0)[name = tensor("op_7929_cast_fp16")]; tensor var_7933_begin_0 = const()[name = tensor("op_7933_begin_0"), val = tensor([0, 0, 0, 120])]; tensor var_7933_end_0 = const()[name = tensor("op_7933_end_0"), val = tensor([2, 77, 1, 160])]; tensor var_7933_end_mask_0 = const()[name = tensor("op_7933_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7933_cast_fp16 = slice_by_index(begin = var_7933_begin_0, end = var_7933_end_0, end_mask = var_7933_end_mask_0, x = transpose_0)[name = tensor("op_7933_cast_fp16")]; tensor var_7937_begin_0 = const()[name = tensor("op_7937_begin_0"), val = tensor([0, 0, 0, 160])]; tensor var_7937_end_0 = const()[name = tensor("op_7937_end_0"), val = tensor([2, 77, 1, 200])]; tensor var_7937_end_mask_0 = const()[name = tensor("op_7937_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7937_cast_fp16 = slice_by_index(begin = var_7937_begin_0, end = var_7937_end_0, end_mask = var_7937_end_mask_0, x = transpose_0)[name = tensor("op_7937_cast_fp16")]; tensor var_7941_begin_0 = const()[name = tensor("op_7941_begin_0"), val = tensor([0, 0, 0, 200])]; tensor var_7941_end_0 = const()[name = tensor("op_7941_end_0"), val = tensor([2, 77, 1, 240])]; tensor var_7941_end_mask_0 = const()[name = tensor("op_7941_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7941_cast_fp16 = slice_by_index(begin = var_7941_begin_0, end = var_7941_end_0, end_mask = var_7941_end_mask_0, x = transpose_0)[name = tensor("op_7941_cast_fp16")]; tensor var_7945_begin_0 = const()[name = tensor("op_7945_begin_0"), val = tensor([0, 0, 0, 240])]; tensor var_7945_end_0 = const()[name = tensor("op_7945_end_0"), val = tensor([2, 77, 1, 280])]; tensor var_7945_end_mask_0 = const()[name = tensor("op_7945_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7945_cast_fp16 = slice_by_index(begin = var_7945_begin_0, end = var_7945_end_0, end_mask = var_7945_end_mask_0, x = transpose_0)[name = tensor("op_7945_cast_fp16")]; tensor var_7949_begin_0 = const()[name = tensor("op_7949_begin_0"), val = tensor([0, 0, 0, 280])]; tensor var_7949_end_0 = const()[name = tensor("op_7949_end_0"), val = tensor([2, 77, 1, 320])]; tensor var_7949_end_mask_0 = const()[name = tensor("op_7949_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7949_cast_fp16 = slice_by_index(begin = var_7949_begin_0, end = var_7949_end_0, end_mask = var_7949_end_mask_0, x = transpose_0)[name = tensor("op_7949_cast_fp16")]; tensor var_7951_begin_0 = const()[name = tensor("op_7951_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_7951_end_0 = const()[name = tensor("op_7951_end_0"), val = tensor([2, 40, 1, 77])]; tensor var_7951_end_mask_0 = const()[name = tensor("op_7951_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7951_cast_fp16 = slice_by_index(begin = var_7951_begin_0, end = var_7951_end_0, end_mask = var_7951_end_mask_0, x = v_cast_fp16)[name = tensor("op_7951_cast_fp16")]; tensor var_7955_begin_0 = const()[name = tensor("op_7955_begin_0"), val = tensor([0, 40, 0, 0])]; tensor var_7955_end_0 = const()[name = tensor("op_7955_end_0"), val = tensor([2, 80, 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_fp16 = slice_by_index(begin = var_7955_begin_0, end = var_7955_end_0, end_mask = var_7955_end_mask_0, x = v_cast_fp16)[name = tensor("op_7955_cast_fp16")]; tensor var_7959_begin_0 = const()[name = tensor("op_7959_begin_0"), val = tensor([0, 80, 0, 0])]; tensor var_7959_end_0 = const()[name = tensor("op_7959_end_0"), val = tensor([2, 120, 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_fp16 = slice_by_index(begin = var_7959_begin_0, end = var_7959_end_0, end_mask = var_7959_end_mask_0, x = v_cast_fp16)[name = tensor("op_7959_cast_fp16")]; tensor var_7963_begin_0 = const()[name = tensor("op_7963_begin_0"), val = tensor([0, 120, 0, 0])]; tensor var_7963_end_0 = const()[name = tensor("op_7963_end_0"), val = tensor([2, 160, 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_fp16 = slice_by_index(begin = var_7963_begin_0, end = var_7963_end_0, end_mask = var_7963_end_mask_0, x = v_cast_fp16)[name = tensor("op_7963_cast_fp16")]; tensor var_7967_begin_0 = const()[name = tensor("op_7967_begin_0"), val = tensor([0, 160, 0, 0])]; tensor var_7967_end_0 = const()[name = tensor("op_7967_end_0"), val = tensor([2, 200, 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_fp16 = slice_by_index(begin = var_7967_begin_0, end = var_7967_end_0, end_mask = var_7967_end_mask_0, x = v_cast_fp16)[name = tensor("op_7967_cast_fp16")]; tensor var_7971_begin_0 = const()[name = tensor("op_7971_begin_0"), val = tensor([0, 200, 0, 0])]; tensor var_7971_end_0 = const()[name = tensor("op_7971_end_0"), val = tensor([2, 240, 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_fp16 = slice_by_index(begin = var_7971_begin_0, end = var_7971_end_0, end_mask = var_7971_end_mask_0, x = v_cast_fp16)[name = tensor("op_7971_cast_fp16")]; tensor var_7975_begin_0 = const()[name = tensor("op_7975_begin_0"), val = tensor([0, 240, 0, 0])]; tensor var_7975_end_0 = const()[name = tensor("op_7975_end_0"), val = tensor([2, 280, 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_fp16 = slice_by_index(begin = var_7975_begin_0, end = var_7975_end_0, end_mask = var_7975_end_mask_0, x = v_cast_fp16)[name = tensor("op_7975_cast_fp16")]; tensor var_7979_begin_0 = const()[name = tensor("op_7979_begin_0"), val = tensor([0, 280, 0, 0])]; tensor var_7979_end_0 = const()[name = tensor("op_7979_end_0"), val = tensor([2, 320, 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_fp16 = slice_by_index(begin = var_7979_begin_0, end = var_7979_end_0, end_mask = var_7979_end_mask_0, x = v_cast_fp16)[name = tensor("op_7979_cast_fp16")]; tensor var_7983_equation_0 = const()[name = tensor("op_7983_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7983_cast_fp16 = einsum(equation = var_7983_equation_0, values = (var_7921_cast_fp16, var_7853_cast_fp16))[name = tensor("op_7983_cast_fp16")]; tensor var_7984_to_fp16 = const()[name = tensor("op_7984_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_833_cast_fp16 = mul(x = var_7983_cast_fp16, y = var_7984_to_fp16)[name = tensor("aw_chunk_833_cast_fp16")]; tensor var_7987_equation_0 = const()[name = tensor("op_7987_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7987_cast_fp16 = einsum(equation = var_7987_equation_0, values = (var_7921_cast_fp16, var_7854_cast_fp16))[name = tensor("op_7987_cast_fp16")]; tensor var_7988_to_fp16 = const()[name = tensor("op_7988_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_835_cast_fp16 = mul(x = var_7987_cast_fp16, y = var_7988_to_fp16)[name = tensor("aw_chunk_835_cast_fp16")]; tensor var_7991_equation_0 = const()[name = tensor("op_7991_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7991_cast_fp16 = einsum(equation = var_7991_equation_0, values = (var_7921_cast_fp16, var_7855_cast_fp16))[name = tensor("op_7991_cast_fp16")]; tensor var_7992_to_fp16 = const()[name = tensor("op_7992_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_837_cast_fp16 = mul(x = var_7991_cast_fp16, y = var_7992_to_fp16)[name = tensor("aw_chunk_837_cast_fp16")]; tensor var_7995_equation_0 = const()[name = tensor("op_7995_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7995_cast_fp16 = einsum(equation = var_7995_equation_0, values = (var_7921_cast_fp16, var_7856_cast_fp16))[name = tensor("op_7995_cast_fp16")]; tensor var_7996_to_fp16 = const()[name = tensor("op_7996_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_839_cast_fp16 = mul(x = var_7995_cast_fp16, y = var_7996_to_fp16)[name = tensor("aw_chunk_839_cast_fp16")]; tensor var_7999_equation_0 = const()[name = tensor("op_7999_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7999_cast_fp16 = einsum(equation = var_7999_equation_0, values = (var_7921_cast_fp16, var_7857_cast_fp16))[name = tensor("op_7999_cast_fp16")]; tensor var_8000_to_fp16 = const()[name = tensor("op_8000_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_841_cast_fp16 = mul(x = var_7999_cast_fp16, y = var_8000_to_fp16)[name = tensor("aw_chunk_841_cast_fp16")]; tensor var_8003_equation_0 = const()[name = tensor("op_8003_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8003_cast_fp16 = einsum(equation = var_8003_equation_0, values = (var_7921_cast_fp16, var_7858_cast_fp16))[name = tensor("op_8003_cast_fp16")]; tensor var_8004_to_fp16 = const()[name = tensor("op_8004_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_843_cast_fp16 = mul(x = var_8003_cast_fp16, y = var_8004_to_fp16)[name = tensor("aw_chunk_843_cast_fp16")]; tensor var_8007_equation_0 = const()[name = tensor("op_8007_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8007_cast_fp16 = einsum(equation = var_8007_equation_0, values = (var_7921_cast_fp16, var_7859_cast_fp16))[name = tensor("op_8007_cast_fp16")]; tensor var_8008_to_fp16 = const()[name = tensor("op_8008_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_845_cast_fp16 = mul(x = var_8007_cast_fp16, y = var_8008_to_fp16)[name = tensor("aw_chunk_845_cast_fp16")]; tensor var_8011_equation_0 = const()[name = tensor("op_8011_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8011_cast_fp16 = einsum(equation = var_8011_equation_0, values = (var_7921_cast_fp16, var_7860_cast_fp16))[name = tensor("op_8011_cast_fp16")]; tensor var_8012_to_fp16 = const()[name = tensor("op_8012_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_847_cast_fp16 = mul(x = var_8011_cast_fp16, y = var_8012_to_fp16)[name = tensor("aw_chunk_847_cast_fp16")]; tensor var_8015_equation_0 = const()[name = tensor("op_8015_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8015_cast_fp16 = einsum(equation = var_8015_equation_0, values = (var_7925_cast_fp16, var_7861_cast_fp16))[name = tensor("op_8015_cast_fp16")]; tensor var_8016_to_fp16 = const()[name = tensor("op_8016_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_849_cast_fp16 = mul(x = var_8015_cast_fp16, y = var_8016_to_fp16)[name = tensor("aw_chunk_849_cast_fp16")]; tensor var_8019_equation_0 = const()[name = tensor("op_8019_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8019_cast_fp16 = einsum(equation = var_8019_equation_0, values = (var_7925_cast_fp16, var_7862_cast_fp16))[name = tensor("op_8019_cast_fp16")]; tensor var_8020_to_fp16 = const()[name = tensor("op_8020_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_851_cast_fp16 = mul(x = var_8019_cast_fp16, y = var_8020_to_fp16)[name = tensor("aw_chunk_851_cast_fp16")]; tensor var_8023_equation_0 = const()[name = tensor("op_8023_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8023_cast_fp16 = einsum(equation = var_8023_equation_0, values = (var_7925_cast_fp16, var_7863_cast_fp16))[name = tensor("op_8023_cast_fp16")]; tensor var_8024_to_fp16 = const()[name = tensor("op_8024_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_853_cast_fp16 = mul(x = var_8023_cast_fp16, y = var_8024_to_fp16)[name = tensor("aw_chunk_853_cast_fp16")]; tensor var_8027_equation_0 = const()[name = tensor("op_8027_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8027_cast_fp16 = einsum(equation = var_8027_equation_0, values = (var_7925_cast_fp16, var_7864_cast_fp16))[name = tensor("op_8027_cast_fp16")]; tensor var_8028_to_fp16 = const()[name = tensor("op_8028_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_855_cast_fp16 = mul(x = var_8027_cast_fp16, y = var_8028_to_fp16)[name = tensor("aw_chunk_855_cast_fp16")]; tensor var_8031_equation_0 = const()[name = tensor("op_8031_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8031_cast_fp16 = einsum(equation = var_8031_equation_0, values = (var_7925_cast_fp16, var_7865_cast_fp16))[name = tensor("op_8031_cast_fp16")]; tensor var_8032_to_fp16 = const()[name = tensor("op_8032_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_857_cast_fp16 = mul(x = var_8031_cast_fp16, y = var_8032_to_fp16)[name = tensor("aw_chunk_857_cast_fp16")]; tensor var_8035_equation_0 = const()[name = tensor("op_8035_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8035_cast_fp16 = einsum(equation = var_8035_equation_0, values = (var_7925_cast_fp16, var_7866_cast_fp16))[name = tensor("op_8035_cast_fp16")]; tensor var_8036_to_fp16 = const()[name = tensor("op_8036_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_859_cast_fp16 = mul(x = var_8035_cast_fp16, y = var_8036_to_fp16)[name = tensor("aw_chunk_859_cast_fp16")]; tensor var_8039_equation_0 = const()[name = tensor("op_8039_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8039_cast_fp16 = einsum(equation = var_8039_equation_0, values = (var_7925_cast_fp16, var_7867_cast_fp16))[name = tensor("op_8039_cast_fp16")]; tensor var_8040_to_fp16 = const()[name = tensor("op_8040_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_861_cast_fp16 = mul(x = var_8039_cast_fp16, y = var_8040_to_fp16)[name = tensor("aw_chunk_861_cast_fp16")]; tensor var_8043_equation_0 = const()[name = tensor("op_8043_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8043_cast_fp16 = einsum(equation = var_8043_equation_0, values = (var_7925_cast_fp16, var_7868_cast_fp16))[name = tensor("op_8043_cast_fp16")]; tensor var_8044_to_fp16 = const()[name = tensor("op_8044_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_863_cast_fp16 = mul(x = var_8043_cast_fp16, y = var_8044_to_fp16)[name = tensor("aw_chunk_863_cast_fp16")]; tensor var_8047_equation_0 = const()[name = tensor("op_8047_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8047_cast_fp16 = einsum(equation = var_8047_equation_0, values = (var_7929_cast_fp16, var_7869_cast_fp16))[name = tensor("op_8047_cast_fp16")]; tensor var_8048_to_fp16 = const()[name = tensor("op_8048_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_865_cast_fp16 = mul(x = var_8047_cast_fp16, y = var_8048_to_fp16)[name = tensor("aw_chunk_865_cast_fp16")]; tensor var_8051_equation_0 = const()[name = tensor("op_8051_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8051_cast_fp16 = einsum(equation = var_8051_equation_0, values = (var_7929_cast_fp16, var_7870_cast_fp16))[name = tensor("op_8051_cast_fp16")]; tensor var_8052_to_fp16 = const()[name = tensor("op_8052_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_867_cast_fp16 = mul(x = var_8051_cast_fp16, y = var_8052_to_fp16)[name = tensor("aw_chunk_867_cast_fp16")]; tensor var_8055_equation_0 = const()[name = tensor("op_8055_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8055_cast_fp16 = einsum(equation = var_8055_equation_0, values = (var_7929_cast_fp16, var_7871_cast_fp16))[name = tensor("op_8055_cast_fp16")]; tensor var_8056_to_fp16 = const()[name = tensor("op_8056_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_869_cast_fp16 = mul(x = var_8055_cast_fp16, y = var_8056_to_fp16)[name = tensor("aw_chunk_869_cast_fp16")]; tensor var_8059_equation_0 = const()[name = tensor("op_8059_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8059_cast_fp16 = einsum(equation = var_8059_equation_0, values = (var_7929_cast_fp16, var_7872_cast_fp16))[name = tensor("op_8059_cast_fp16")]; tensor var_8060_to_fp16 = const()[name = tensor("op_8060_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_871_cast_fp16 = mul(x = var_8059_cast_fp16, y = var_8060_to_fp16)[name = tensor("aw_chunk_871_cast_fp16")]; tensor var_8063_equation_0 = const()[name = tensor("op_8063_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8063_cast_fp16 = einsum(equation = var_8063_equation_0, values = (var_7929_cast_fp16, var_7873_cast_fp16))[name = tensor("op_8063_cast_fp16")]; tensor var_8064_to_fp16 = const()[name = tensor("op_8064_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_873_cast_fp16 = mul(x = var_8063_cast_fp16, y = var_8064_to_fp16)[name = tensor("aw_chunk_873_cast_fp16")]; tensor var_8067_equation_0 = const()[name = tensor("op_8067_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8067_cast_fp16 = einsum(equation = var_8067_equation_0, values = (var_7929_cast_fp16, var_7874_cast_fp16))[name = tensor("op_8067_cast_fp16")]; tensor var_8068_to_fp16 = const()[name = tensor("op_8068_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_875_cast_fp16 = mul(x = var_8067_cast_fp16, y = var_8068_to_fp16)[name = tensor("aw_chunk_875_cast_fp16")]; tensor var_8071_equation_0 = const()[name = tensor("op_8071_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8071_cast_fp16 = einsum(equation = var_8071_equation_0, values = (var_7929_cast_fp16, var_7875_cast_fp16))[name = tensor("op_8071_cast_fp16")]; tensor var_8072_to_fp16 = const()[name = tensor("op_8072_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_877_cast_fp16 = mul(x = var_8071_cast_fp16, y = var_8072_to_fp16)[name = tensor("aw_chunk_877_cast_fp16")]; tensor var_8075_equation_0 = const()[name = tensor("op_8075_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8075_cast_fp16 = einsum(equation = var_8075_equation_0, values = (var_7929_cast_fp16, var_7876_cast_fp16))[name = tensor("op_8075_cast_fp16")]; tensor var_8076_to_fp16 = const()[name = tensor("op_8076_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_879_cast_fp16 = mul(x = var_8075_cast_fp16, y = var_8076_to_fp16)[name = tensor("aw_chunk_879_cast_fp16")]; tensor var_8079_equation_0 = const()[name = tensor("op_8079_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8079_cast_fp16 = einsum(equation = var_8079_equation_0, values = (var_7933_cast_fp16, var_7877_cast_fp16))[name = tensor("op_8079_cast_fp16")]; tensor var_8080_to_fp16 = const()[name = tensor("op_8080_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_881_cast_fp16 = mul(x = var_8079_cast_fp16, y = var_8080_to_fp16)[name = tensor("aw_chunk_881_cast_fp16")]; tensor var_8083_equation_0 = const()[name = tensor("op_8083_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8083_cast_fp16 = einsum(equation = var_8083_equation_0, values = (var_7933_cast_fp16, var_7878_cast_fp16))[name = tensor("op_8083_cast_fp16")]; tensor var_8084_to_fp16 = const()[name = tensor("op_8084_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_883_cast_fp16 = mul(x = var_8083_cast_fp16, y = var_8084_to_fp16)[name = tensor("aw_chunk_883_cast_fp16")]; tensor var_8087_equation_0 = const()[name = tensor("op_8087_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8087_cast_fp16 = einsum(equation = var_8087_equation_0, values = (var_7933_cast_fp16, var_7879_cast_fp16))[name = tensor("op_8087_cast_fp16")]; tensor var_8088_to_fp16 = const()[name = tensor("op_8088_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_885_cast_fp16 = mul(x = var_8087_cast_fp16, y = var_8088_to_fp16)[name = tensor("aw_chunk_885_cast_fp16")]; tensor var_8091_equation_0 = const()[name = tensor("op_8091_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8091_cast_fp16 = einsum(equation = var_8091_equation_0, values = (var_7933_cast_fp16, var_7880_cast_fp16))[name = tensor("op_8091_cast_fp16")]; tensor var_8092_to_fp16 = const()[name = tensor("op_8092_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_887_cast_fp16 = mul(x = var_8091_cast_fp16, y = var_8092_to_fp16)[name = tensor("aw_chunk_887_cast_fp16")]; tensor var_8095_equation_0 = const()[name = tensor("op_8095_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8095_cast_fp16 = einsum(equation = var_8095_equation_0, values = (var_7933_cast_fp16, var_7881_cast_fp16))[name = tensor("op_8095_cast_fp16")]; tensor var_8096_to_fp16 = const()[name = tensor("op_8096_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_889_cast_fp16 = mul(x = var_8095_cast_fp16, y = var_8096_to_fp16)[name = tensor("aw_chunk_889_cast_fp16")]; tensor var_8099_equation_0 = const()[name = tensor("op_8099_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8099_cast_fp16 = einsum(equation = var_8099_equation_0, values = (var_7933_cast_fp16, var_7882_cast_fp16))[name = tensor("op_8099_cast_fp16")]; tensor var_8100_to_fp16 = const()[name = tensor("op_8100_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_891_cast_fp16 = mul(x = var_8099_cast_fp16, y = var_8100_to_fp16)[name = tensor("aw_chunk_891_cast_fp16")]; tensor var_8103_equation_0 = const()[name = tensor("op_8103_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8103_cast_fp16 = einsum(equation = var_8103_equation_0, values = (var_7933_cast_fp16, var_7883_cast_fp16))[name = tensor("op_8103_cast_fp16")]; tensor var_8104_to_fp16 = const()[name = tensor("op_8104_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_893_cast_fp16 = mul(x = var_8103_cast_fp16, y = var_8104_to_fp16)[name = tensor("aw_chunk_893_cast_fp16")]; tensor var_8107_equation_0 = const()[name = tensor("op_8107_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8107_cast_fp16 = einsum(equation = var_8107_equation_0, values = (var_7933_cast_fp16, var_7884_cast_fp16))[name = tensor("op_8107_cast_fp16")]; tensor var_8108_to_fp16 = const()[name = tensor("op_8108_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_895_cast_fp16 = mul(x = var_8107_cast_fp16, y = var_8108_to_fp16)[name = tensor("aw_chunk_895_cast_fp16")]; tensor var_8111_equation_0 = const()[name = tensor("op_8111_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8111_cast_fp16 = einsum(equation = var_8111_equation_0, values = (var_7937_cast_fp16, var_7885_cast_fp16))[name = tensor("op_8111_cast_fp16")]; tensor var_8112_to_fp16 = const()[name = tensor("op_8112_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_897_cast_fp16 = mul(x = var_8111_cast_fp16, y = var_8112_to_fp16)[name = tensor("aw_chunk_897_cast_fp16")]; tensor var_8115_equation_0 = const()[name = tensor("op_8115_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8115_cast_fp16 = einsum(equation = var_8115_equation_0, values = (var_7937_cast_fp16, var_7886_cast_fp16))[name = tensor("op_8115_cast_fp16")]; tensor var_8116_to_fp16 = const()[name = tensor("op_8116_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_899_cast_fp16 = mul(x = var_8115_cast_fp16, y = var_8116_to_fp16)[name = tensor("aw_chunk_899_cast_fp16")]; tensor var_8119_equation_0 = const()[name = tensor("op_8119_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8119_cast_fp16 = einsum(equation = var_8119_equation_0, values = (var_7937_cast_fp16, var_7887_cast_fp16))[name = tensor("op_8119_cast_fp16")]; tensor var_8120_to_fp16 = const()[name = tensor("op_8120_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_901_cast_fp16 = mul(x = var_8119_cast_fp16, y = var_8120_to_fp16)[name = tensor("aw_chunk_901_cast_fp16")]; tensor var_8123_equation_0 = const()[name = tensor("op_8123_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8123_cast_fp16 = einsum(equation = var_8123_equation_0, values = (var_7937_cast_fp16, var_7888_cast_fp16))[name = tensor("op_8123_cast_fp16")]; tensor var_8124_to_fp16 = const()[name = tensor("op_8124_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_903_cast_fp16 = mul(x = var_8123_cast_fp16, y = var_8124_to_fp16)[name = tensor("aw_chunk_903_cast_fp16")]; tensor var_8127_equation_0 = const()[name = tensor("op_8127_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8127_cast_fp16 = einsum(equation = var_8127_equation_0, values = (var_7937_cast_fp16, var_7889_cast_fp16))[name = tensor("op_8127_cast_fp16")]; tensor var_8128_to_fp16 = const()[name = tensor("op_8128_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_905_cast_fp16 = mul(x = var_8127_cast_fp16, y = var_8128_to_fp16)[name = tensor("aw_chunk_905_cast_fp16")]; tensor var_8131_equation_0 = const()[name = tensor("op_8131_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8131_cast_fp16 = einsum(equation = var_8131_equation_0, values = (var_7937_cast_fp16, var_7890_cast_fp16))[name = tensor("op_8131_cast_fp16")]; tensor var_8132_to_fp16 = const()[name = tensor("op_8132_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_907_cast_fp16 = mul(x = var_8131_cast_fp16, y = var_8132_to_fp16)[name = tensor("aw_chunk_907_cast_fp16")]; tensor var_8135_equation_0 = const()[name = tensor("op_8135_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8135_cast_fp16 = einsum(equation = var_8135_equation_0, values = (var_7937_cast_fp16, var_7891_cast_fp16))[name = tensor("op_8135_cast_fp16")]; tensor var_8136_to_fp16 = const()[name = tensor("op_8136_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_909_cast_fp16 = mul(x = var_8135_cast_fp16, y = var_8136_to_fp16)[name = tensor("aw_chunk_909_cast_fp16")]; tensor var_8139_equation_0 = const()[name = tensor("op_8139_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8139_cast_fp16 = einsum(equation = var_8139_equation_0, values = (var_7937_cast_fp16, var_7892_cast_fp16))[name = tensor("op_8139_cast_fp16")]; tensor var_8140_to_fp16 = const()[name = tensor("op_8140_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_911_cast_fp16 = mul(x = var_8139_cast_fp16, y = var_8140_to_fp16)[name = tensor("aw_chunk_911_cast_fp16")]; tensor var_8143_equation_0 = const()[name = tensor("op_8143_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8143_cast_fp16 = einsum(equation = var_8143_equation_0, values = (var_7941_cast_fp16, var_7893_cast_fp16))[name = tensor("op_8143_cast_fp16")]; tensor var_8144_to_fp16 = const()[name = tensor("op_8144_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_913_cast_fp16 = mul(x = var_8143_cast_fp16, y = var_8144_to_fp16)[name = tensor("aw_chunk_913_cast_fp16")]; tensor var_8147_equation_0 = const()[name = tensor("op_8147_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8147_cast_fp16 = einsum(equation = var_8147_equation_0, values = (var_7941_cast_fp16, var_7894_cast_fp16))[name = tensor("op_8147_cast_fp16")]; tensor var_8148_to_fp16 = const()[name = tensor("op_8148_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_915_cast_fp16 = mul(x = var_8147_cast_fp16, y = var_8148_to_fp16)[name = tensor("aw_chunk_915_cast_fp16")]; tensor var_8151_equation_0 = const()[name = tensor("op_8151_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8151_cast_fp16 = einsum(equation = var_8151_equation_0, values = (var_7941_cast_fp16, var_7895_cast_fp16))[name = tensor("op_8151_cast_fp16")]; tensor var_8152_to_fp16 = const()[name = tensor("op_8152_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_917_cast_fp16 = mul(x = var_8151_cast_fp16, y = var_8152_to_fp16)[name = tensor("aw_chunk_917_cast_fp16")]; tensor var_8155_equation_0 = const()[name = tensor("op_8155_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8155_cast_fp16 = einsum(equation = var_8155_equation_0, values = (var_7941_cast_fp16, var_7896_cast_fp16))[name = tensor("op_8155_cast_fp16")]; tensor var_8156_to_fp16 = const()[name = tensor("op_8156_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_919_cast_fp16 = mul(x = var_8155_cast_fp16, y = var_8156_to_fp16)[name = tensor("aw_chunk_919_cast_fp16")]; tensor var_8159_equation_0 = const()[name = tensor("op_8159_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8159_cast_fp16 = einsum(equation = var_8159_equation_0, values = (var_7941_cast_fp16, var_7897_cast_fp16))[name = tensor("op_8159_cast_fp16")]; tensor var_8160_to_fp16 = const()[name = tensor("op_8160_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_921_cast_fp16 = mul(x = var_8159_cast_fp16, y = var_8160_to_fp16)[name = tensor("aw_chunk_921_cast_fp16")]; tensor var_8163_equation_0 = const()[name = tensor("op_8163_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8163_cast_fp16 = einsum(equation = var_8163_equation_0, values = (var_7941_cast_fp16, var_7898_cast_fp16))[name = tensor("op_8163_cast_fp16")]; tensor var_8164_to_fp16 = const()[name = tensor("op_8164_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_923_cast_fp16 = mul(x = var_8163_cast_fp16, y = var_8164_to_fp16)[name = tensor("aw_chunk_923_cast_fp16")]; tensor var_8167_equation_0 = const()[name = tensor("op_8167_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8167_cast_fp16 = einsum(equation = var_8167_equation_0, values = (var_7941_cast_fp16, var_7899_cast_fp16))[name = tensor("op_8167_cast_fp16")]; tensor var_8168_to_fp16 = const()[name = tensor("op_8168_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_925_cast_fp16 = mul(x = var_8167_cast_fp16, y = var_8168_to_fp16)[name = tensor("aw_chunk_925_cast_fp16")]; tensor var_8171_equation_0 = const()[name = tensor("op_8171_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8171_cast_fp16 = einsum(equation = var_8171_equation_0, values = (var_7941_cast_fp16, var_7900_cast_fp16))[name = tensor("op_8171_cast_fp16")]; tensor var_8172_to_fp16 = const()[name = tensor("op_8172_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_927_cast_fp16 = mul(x = var_8171_cast_fp16, y = var_8172_to_fp16)[name = tensor("aw_chunk_927_cast_fp16")]; tensor var_8175_equation_0 = const()[name = tensor("op_8175_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8175_cast_fp16 = einsum(equation = var_8175_equation_0, values = (var_7945_cast_fp16, var_7901_cast_fp16))[name = tensor("op_8175_cast_fp16")]; tensor var_8176_to_fp16 = const()[name = tensor("op_8176_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_929_cast_fp16 = mul(x = var_8175_cast_fp16, y = var_8176_to_fp16)[name = tensor("aw_chunk_929_cast_fp16")]; tensor var_8179_equation_0 = const()[name = tensor("op_8179_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8179_cast_fp16 = einsum(equation = var_8179_equation_0, values = (var_7945_cast_fp16, var_7902_cast_fp16))[name = tensor("op_8179_cast_fp16")]; tensor var_8180_to_fp16 = const()[name = tensor("op_8180_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_931_cast_fp16 = mul(x = var_8179_cast_fp16, y = var_8180_to_fp16)[name = tensor("aw_chunk_931_cast_fp16")]; tensor var_8183_equation_0 = const()[name = tensor("op_8183_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8183_cast_fp16 = einsum(equation = var_8183_equation_0, values = (var_7945_cast_fp16, var_7903_cast_fp16))[name = tensor("op_8183_cast_fp16")]; tensor var_8184_to_fp16 = const()[name = tensor("op_8184_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_933_cast_fp16 = mul(x = var_8183_cast_fp16, y = var_8184_to_fp16)[name = tensor("aw_chunk_933_cast_fp16")]; tensor var_8187_equation_0 = const()[name = tensor("op_8187_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8187_cast_fp16 = einsum(equation = var_8187_equation_0, values = (var_7945_cast_fp16, var_7904_cast_fp16))[name = tensor("op_8187_cast_fp16")]; tensor var_8188_to_fp16 = const()[name = tensor("op_8188_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_935_cast_fp16 = mul(x = var_8187_cast_fp16, y = var_8188_to_fp16)[name = tensor("aw_chunk_935_cast_fp16")]; tensor var_8191_equation_0 = const()[name = tensor("op_8191_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8191_cast_fp16 = einsum(equation = var_8191_equation_0, values = (var_7945_cast_fp16, var_7905_cast_fp16))[name = tensor("op_8191_cast_fp16")]; tensor var_8192_to_fp16 = const()[name = tensor("op_8192_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_937_cast_fp16 = mul(x = var_8191_cast_fp16, y = var_8192_to_fp16)[name = tensor("aw_chunk_937_cast_fp16")]; tensor var_8195_equation_0 = const()[name = tensor("op_8195_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8195_cast_fp16 = einsum(equation = var_8195_equation_0, values = (var_7945_cast_fp16, var_7906_cast_fp16))[name = tensor("op_8195_cast_fp16")]; tensor var_8196_to_fp16 = const()[name = tensor("op_8196_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_939_cast_fp16 = mul(x = var_8195_cast_fp16, y = var_8196_to_fp16)[name = tensor("aw_chunk_939_cast_fp16")]; tensor var_8199_equation_0 = const()[name = tensor("op_8199_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8199_cast_fp16 = einsum(equation = var_8199_equation_0, values = (var_7945_cast_fp16, var_7907_cast_fp16))[name = tensor("op_8199_cast_fp16")]; tensor var_8200_to_fp16 = const()[name = tensor("op_8200_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_941_cast_fp16 = mul(x = var_8199_cast_fp16, y = var_8200_to_fp16)[name = tensor("aw_chunk_941_cast_fp16")]; tensor var_8203_equation_0 = const()[name = tensor("op_8203_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8203_cast_fp16 = einsum(equation = var_8203_equation_0, values = (var_7945_cast_fp16, var_7908_cast_fp16))[name = tensor("op_8203_cast_fp16")]; tensor var_8204_to_fp16 = const()[name = tensor("op_8204_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_943_cast_fp16 = mul(x = var_8203_cast_fp16, y = var_8204_to_fp16)[name = tensor("aw_chunk_943_cast_fp16")]; tensor var_8207_equation_0 = const()[name = tensor("op_8207_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8207_cast_fp16 = einsum(equation = var_8207_equation_0, values = (var_7949_cast_fp16, var_7909_cast_fp16))[name = tensor("op_8207_cast_fp16")]; tensor var_8208_to_fp16 = const()[name = tensor("op_8208_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_945_cast_fp16 = mul(x = var_8207_cast_fp16, y = var_8208_to_fp16)[name = tensor("aw_chunk_945_cast_fp16")]; tensor var_8211_equation_0 = const()[name = tensor("op_8211_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8211_cast_fp16 = einsum(equation = var_8211_equation_0, values = (var_7949_cast_fp16, var_7910_cast_fp16))[name = tensor("op_8211_cast_fp16")]; tensor var_8212_to_fp16 = const()[name = tensor("op_8212_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_947_cast_fp16 = mul(x = var_8211_cast_fp16, y = var_8212_to_fp16)[name = tensor("aw_chunk_947_cast_fp16")]; tensor var_8215_equation_0 = const()[name = tensor("op_8215_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8215_cast_fp16 = einsum(equation = var_8215_equation_0, values = (var_7949_cast_fp16, var_7911_cast_fp16))[name = tensor("op_8215_cast_fp16")]; tensor var_8216_to_fp16 = const()[name = tensor("op_8216_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_949_cast_fp16 = mul(x = var_8215_cast_fp16, y = var_8216_to_fp16)[name = tensor("aw_chunk_949_cast_fp16")]; tensor var_8219_equation_0 = const()[name = tensor("op_8219_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8219_cast_fp16 = einsum(equation = var_8219_equation_0, values = (var_7949_cast_fp16, var_7912_cast_fp16))[name = tensor("op_8219_cast_fp16")]; tensor var_8220_to_fp16 = const()[name = tensor("op_8220_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_951_cast_fp16 = mul(x = var_8219_cast_fp16, y = var_8220_to_fp16)[name = tensor("aw_chunk_951_cast_fp16")]; tensor var_8223_equation_0 = const()[name = tensor("op_8223_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8223_cast_fp16 = einsum(equation = var_8223_equation_0, values = (var_7949_cast_fp16, var_7913_cast_fp16))[name = tensor("op_8223_cast_fp16")]; tensor var_8224_to_fp16 = const()[name = tensor("op_8224_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_953_cast_fp16 = mul(x = var_8223_cast_fp16, y = var_8224_to_fp16)[name = tensor("aw_chunk_953_cast_fp16")]; tensor var_8227_equation_0 = const()[name = tensor("op_8227_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8227_cast_fp16 = einsum(equation = var_8227_equation_0, values = (var_7949_cast_fp16, var_7914_cast_fp16))[name = tensor("op_8227_cast_fp16")]; tensor var_8228_to_fp16 = const()[name = tensor("op_8228_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_955_cast_fp16 = mul(x = var_8227_cast_fp16, y = var_8228_to_fp16)[name = tensor("aw_chunk_955_cast_fp16")]; tensor var_8231_equation_0 = const()[name = tensor("op_8231_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8231_cast_fp16 = einsum(equation = var_8231_equation_0, values = (var_7949_cast_fp16, var_7915_cast_fp16))[name = tensor("op_8231_cast_fp16")]; tensor var_8232_to_fp16 = const()[name = tensor("op_8232_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_957_cast_fp16 = mul(x = var_8231_cast_fp16, y = var_8232_to_fp16)[name = tensor("aw_chunk_957_cast_fp16")]; tensor var_8235_equation_0 = const()[name = tensor("op_8235_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8235_cast_fp16 = einsum(equation = var_8235_equation_0, values = (var_7949_cast_fp16, var_7916_cast_fp16))[name = tensor("op_8235_cast_fp16")]; tensor var_8236_to_fp16 = const()[name = tensor("op_8236_to_fp16"), val = tensor(0x1.43cp-3)]; tensor aw_chunk_cast_fp16 = mul(x = var_8235_cast_fp16, y = var_8236_to_fp16)[name = tensor("aw_chunk_cast_fp16")]; tensor var_8238_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_833_cast_fp16)[name = tensor("op_8238_cast_fp16")]; tensor var_8239_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_835_cast_fp16)[name = tensor("op_8239_cast_fp16")]; tensor var_8240_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_837_cast_fp16)[name = tensor("op_8240_cast_fp16")]; tensor var_8241_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_839_cast_fp16)[name = tensor("op_8241_cast_fp16")]; tensor var_8242_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_841_cast_fp16)[name = tensor("op_8242_cast_fp16")]; tensor var_8243_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_843_cast_fp16)[name = tensor("op_8243_cast_fp16")]; tensor var_8244_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_845_cast_fp16)[name = tensor("op_8244_cast_fp16")]; tensor var_8245_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_847_cast_fp16)[name = tensor("op_8245_cast_fp16")]; tensor var_8246_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_849_cast_fp16)[name = tensor("op_8246_cast_fp16")]; tensor var_8247_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_851_cast_fp16)[name = tensor("op_8247_cast_fp16")]; tensor var_8248_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_853_cast_fp16)[name = tensor("op_8248_cast_fp16")]; tensor var_8249_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_855_cast_fp16)[name = tensor("op_8249_cast_fp16")]; tensor var_8250_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_857_cast_fp16)[name = tensor("op_8250_cast_fp16")]; tensor var_8251_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_859_cast_fp16)[name = tensor("op_8251_cast_fp16")]; tensor var_8252_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_861_cast_fp16)[name = tensor("op_8252_cast_fp16")]; tensor var_8253_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_863_cast_fp16)[name = tensor("op_8253_cast_fp16")]; tensor var_8254_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_865_cast_fp16)[name = tensor("op_8254_cast_fp16")]; tensor var_8255_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_867_cast_fp16)[name = tensor("op_8255_cast_fp16")]; tensor var_8256_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_869_cast_fp16)[name = tensor("op_8256_cast_fp16")]; tensor var_8257_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_871_cast_fp16)[name = tensor("op_8257_cast_fp16")]; tensor var_8258_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_873_cast_fp16)[name = tensor("op_8258_cast_fp16")]; tensor var_8259_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_875_cast_fp16)[name = tensor("op_8259_cast_fp16")]; tensor var_8260_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_877_cast_fp16)[name = tensor("op_8260_cast_fp16")]; tensor var_8261_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_879_cast_fp16)[name = tensor("op_8261_cast_fp16")]; tensor var_8262_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_881_cast_fp16)[name = tensor("op_8262_cast_fp16")]; tensor var_8263_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_883_cast_fp16)[name = tensor("op_8263_cast_fp16")]; tensor var_8264_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_885_cast_fp16)[name = tensor("op_8264_cast_fp16")]; tensor var_8265_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_887_cast_fp16)[name = tensor("op_8265_cast_fp16")]; tensor var_8266_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_889_cast_fp16)[name = tensor("op_8266_cast_fp16")]; tensor var_8267_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_891_cast_fp16)[name = tensor("op_8267_cast_fp16")]; tensor var_8268_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_893_cast_fp16)[name = tensor("op_8268_cast_fp16")]; tensor var_8269_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_895_cast_fp16)[name = tensor("op_8269_cast_fp16")]; tensor var_8270_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_897_cast_fp16)[name = tensor("op_8270_cast_fp16")]; tensor var_8271_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_899_cast_fp16)[name = tensor("op_8271_cast_fp16")]; tensor var_8272_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_901_cast_fp16)[name = tensor("op_8272_cast_fp16")]; tensor var_8273_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_903_cast_fp16)[name = tensor("op_8273_cast_fp16")]; tensor var_8274_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_905_cast_fp16)[name = tensor("op_8274_cast_fp16")]; tensor var_8275_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_907_cast_fp16)[name = tensor("op_8275_cast_fp16")]; tensor var_8276_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_909_cast_fp16)[name = tensor("op_8276_cast_fp16")]; tensor var_8277_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_911_cast_fp16)[name = tensor("op_8277_cast_fp16")]; tensor var_8278_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_913_cast_fp16)[name = tensor("op_8278_cast_fp16")]; tensor var_8279_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_915_cast_fp16)[name = tensor("op_8279_cast_fp16")]; tensor var_8280_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_917_cast_fp16)[name = tensor("op_8280_cast_fp16")]; tensor var_8281_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_919_cast_fp16)[name = tensor("op_8281_cast_fp16")]; tensor var_8282_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_921_cast_fp16)[name = tensor("op_8282_cast_fp16")]; tensor var_8283_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_923_cast_fp16)[name = tensor("op_8283_cast_fp16")]; tensor var_8284_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_925_cast_fp16)[name = tensor("op_8284_cast_fp16")]; tensor var_8285_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_927_cast_fp16)[name = tensor("op_8285_cast_fp16")]; tensor var_8286_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_929_cast_fp16)[name = tensor("op_8286_cast_fp16")]; tensor var_8287_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_931_cast_fp16)[name = tensor("op_8287_cast_fp16")]; tensor var_8288_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_933_cast_fp16)[name = tensor("op_8288_cast_fp16")]; tensor var_8289_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_935_cast_fp16)[name = tensor("op_8289_cast_fp16")]; tensor var_8290_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_937_cast_fp16)[name = tensor("op_8290_cast_fp16")]; tensor var_8291_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_939_cast_fp16)[name = tensor("op_8291_cast_fp16")]; tensor var_8292_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_941_cast_fp16)[name = tensor("op_8292_cast_fp16")]; tensor var_8293_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_943_cast_fp16)[name = tensor("op_8293_cast_fp16")]; tensor var_8294_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_945_cast_fp16)[name = tensor("op_8294_cast_fp16")]; tensor var_8295_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_947_cast_fp16)[name = tensor("op_8295_cast_fp16")]; tensor var_8296_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_949_cast_fp16)[name = tensor("op_8296_cast_fp16")]; tensor var_8297_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_951_cast_fp16)[name = tensor("op_8297_cast_fp16")]; tensor var_8298_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_953_cast_fp16)[name = tensor("op_8298_cast_fp16")]; tensor var_8299_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_955_cast_fp16)[name = tensor("op_8299_cast_fp16")]; tensor var_8300_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_957_cast_fp16)[name = tensor("op_8300_cast_fp16")]; tensor var_8301_cast_fp16 = softmax(axis = var_5537, x = aw_chunk_cast_fp16)[name = tensor("op_8301_cast_fp16")]; tensor var_8303_equation_0 = const()[name = tensor("op_8303_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8303_cast_fp16 = einsum(equation = var_8303_equation_0, values = (var_7951_cast_fp16, var_8238_cast_fp16))[name = tensor("op_8303_cast_fp16")]; tensor var_8305_equation_0 = const()[name = tensor("op_8305_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8305_cast_fp16 = einsum(equation = var_8305_equation_0, values = (var_7951_cast_fp16, var_8239_cast_fp16))[name = tensor("op_8305_cast_fp16")]; tensor var_8307_equation_0 = const()[name = tensor("op_8307_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8307_cast_fp16 = einsum(equation = var_8307_equation_0, values = (var_7951_cast_fp16, var_8240_cast_fp16))[name = tensor("op_8307_cast_fp16")]; tensor var_8309_equation_0 = const()[name = tensor("op_8309_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8309_cast_fp16 = einsum(equation = var_8309_equation_0, values = (var_7951_cast_fp16, var_8241_cast_fp16))[name = tensor("op_8309_cast_fp16")]; tensor var_8311_equation_0 = const()[name = tensor("op_8311_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8311_cast_fp16 = einsum(equation = var_8311_equation_0, values = (var_7951_cast_fp16, var_8242_cast_fp16))[name = tensor("op_8311_cast_fp16")]; tensor var_8313_equation_0 = const()[name = tensor("op_8313_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8313_cast_fp16 = einsum(equation = var_8313_equation_0, values = (var_7951_cast_fp16, var_8243_cast_fp16))[name = tensor("op_8313_cast_fp16")]; tensor var_8315_equation_0 = const()[name = tensor("op_8315_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8315_cast_fp16 = einsum(equation = var_8315_equation_0, values = (var_7951_cast_fp16, var_8244_cast_fp16))[name = tensor("op_8315_cast_fp16")]; tensor var_8317_equation_0 = const()[name = tensor("op_8317_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8317_cast_fp16 = einsum(equation = var_8317_equation_0, values = (var_7951_cast_fp16, var_8245_cast_fp16))[name = tensor("op_8317_cast_fp16")]; tensor var_8319_equation_0 = const()[name = tensor("op_8319_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8319_cast_fp16 = einsum(equation = var_8319_equation_0, values = (var_7955_cast_fp16, var_8246_cast_fp16))[name = tensor("op_8319_cast_fp16")]; tensor var_8321_equation_0 = const()[name = tensor("op_8321_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8321_cast_fp16 = einsum(equation = var_8321_equation_0, values = (var_7955_cast_fp16, var_8247_cast_fp16))[name = tensor("op_8321_cast_fp16")]; tensor var_8323_equation_0 = const()[name = tensor("op_8323_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8323_cast_fp16 = einsum(equation = var_8323_equation_0, values = (var_7955_cast_fp16, var_8248_cast_fp16))[name = tensor("op_8323_cast_fp16")]; tensor var_8325_equation_0 = const()[name = tensor("op_8325_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8325_cast_fp16 = einsum(equation = var_8325_equation_0, values = (var_7955_cast_fp16, var_8249_cast_fp16))[name = tensor("op_8325_cast_fp16")]; tensor var_8327_equation_0 = const()[name = tensor("op_8327_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8327_cast_fp16 = einsum(equation = var_8327_equation_0, values = (var_7955_cast_fp16, var_8250_cast_fp16))[name = tensor("op_8327_cast_fp16")]; tensor var_8329_equation_0 = const()[name = tensor("op_8329_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8329_cast_fp16 = einsum(equation = var_8329_equation_0, values = (var_7955_cast_fp16, var_8251_cast_fp16))[name = tensor("op_8329_cast_fp16")]; tensor var_8331_equation_0 = const()[name = tensor("op_8331_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8331_cast_fp16 = einsum(equation = var_8331_equation_0, values = (var_7955_cast_fp16, var_8252_cast_fp16))[name = tensor("op_8331_cast_fp16")]; tensor var_8333_equation_0 = const()[name = tensor("op_8333_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8333_cast_fp16 = einsum(equation = var_8333_equation_0, values = (var_7955_cast_fp16, var_8253_cast_fp16))[name = tensor("op_8333_cast_fp16")]; tensor var_8335_equation_0 = const()[name = tensor("op_8335_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8335_cast_fp16 = einsum(equation = var_8335_equation_0, values = (var_7959_cast_fp16, var_8254_cast_fp16))[name = tensor("op_8335_cast_fp16")]; tensor var_8337_equation_0 = const()[name = tensor("op_8337_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8337_cast_fp16 = einsum(equation = var_8337_equation_0, values = (var_7959_cast_fp16, var_8255_cast_fp16))[name = tensor("op_8337_cast_fp16")]; tensor var_8339_equation_0 = const()[name = tensor("op_8339_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8339_cast_fp16 = einsum(equation = var_8339_equation_0, values = (var_7959_cast_fp16, var_8256_cast_fp16))[name = tensor("op_8339_cast_fp16")]; tensor var_8341_equation_0 = const()[name = tensor("op_8341_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8341_cast_fp16 = einsum(equation = var_8341_equation_0, values = (var_7959_cast_fp16, var_8257_cast_fp16))[name = tensor("op_8341_cast_fp16")]; tensor var_8343_equation_0 = const()[name = tensor("op_8343_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8343_cast_fp16 = einsum(equation = var_8343_equation_0, values = (var_7959_cast_fp16, var_8258_cast_fp16))[name = tensor("op_8343_cast_fp16")]; tensor var_8345_equation_0 = const()[name = tensor("op_8345_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8345_cast_fp16 = einsum(equation = var_8345_equation_0, values = (var_7959_cast_fp16, var_8259_cast_fp16))[name = tensor("op_8345_cast_fp16")]; tensor var_8347_equation_0 = const()[name = tensor("op_8347_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8347_cast_fp16 = einsum(equation = var_8347_equation_0, values = (var_7959_cast_fp16, var_8260_cast_fp16))[name = tensor("op_8347_cast_fp16")]; tensor var_8349_equation_0 = const()[name = tensor("op_8349_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8349_cast_fp16 = einsum(equation = var_8349_equation_0, values = (var_7959_cast_fp16, var_8261_cast_fp16))[name = tensor("op_8349_cast_fp16")]; tensor var_8351_equation_0 = const()[name = tensor("op_8351_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8351_cast_fp16 = einsum(equation = var_8351_equation_0, values = (var_7963_cast_fp16, var_8262_cast_fp16))[name = tensor("op_8351_cast_fp16")]; tensor var_8353_equation_0 = const()[name = tensor("op_8353_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8353_cast_fp16 = einsum(equation = var_8353_equation_0, values = (var_7963_cast_fp16, var_8263_cast_fp16))[name = tensor("op_8353_cast_fp16")]; tensor var_8355_equation_0 = const()[name = tensor("op_8355_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8355_cast_fp16 = einsum(equation = var_8355_equation_0, values = (var_7963_cast_fp16, var_8264_cast_fp16))[name = tensor("op_8355_cast_fp16")]; tensor var_8357_equation_0 = const()[name = tensor("op_8357_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8357_cast_fp16 = einsum(equation = var_8357_equation_0, values = (var_7963_cast_fp16, var_8265_cast_fp16))[name = tensor("op_8357_cast_fp16")]; tensor var_8359_equation_0 = const()[name = tensor("op_8359_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8359_cast_fp16 = einsum(equation = var_8359_equation_0, values = (var_7963_cast_fp16, var_8266_cast_fp16))[name = tensor("op_8359_cast_fp16")]; tensor var_8361_equation_0 = const()[name = tensor("op_8361_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8361_cast_fp16 = einsum(equation = var_8361_equation_0, values = (var_7963_cast_fp16, var_8267_cast_fp16))[name = tensor("op_8361_cast_fp16")]; tensor var_8363_equation_0 = const()[name = tensor("op_8363_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8363_cast_fp16 = einsum(equation = var_8363_equation_0, values = (var_7963_cast_fp16, var_8268_cast_fp16))[name = tensor("op_8363_cast_fp16")]; tensor var_8365_equation_0 = const()[name = tensor("op_8365_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8365_cast_fp16 = einsum(equation = var_8365_equation_0, values = (var_7963_cast_fp16, var_8269_cast_fp16))[name = tensor("op_8365_cast_fp16")]; tensor var_8367_equation_0 = const()[name = tensor("op_8367_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8367_cast_fp16 = einsum(equation = var_8367_equation_0, values = (var_7967_cast_fp16, var_8270_cast_fp16))[name = tensor("op_8367_cast_fp16")]; tensor var_8369_equation_0 = const()[name = tensor("op_8369_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8369_cast_fp16 = einsum(equation = var_8369_equation_0, values = (var_7967_cast_fp16, var_8271_cast_fp16))[name = tensor("op_8369_cast_fp16")]; tensor var_8371_equation_0 = const()[name = tensor("op_8371_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8371_cast_fp16 = einsum(equation = var_8371_equation_0, values = (var_7967_cast_fp16, var_8272_cast_fp16))[name = tensor("op_8371_cast_fp16")]; tensor var_8373_equation_0 = const()[name = tensor("op_8373_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8373_cast_fp16 = einsum(equation = var_8373_equation_0, values = (var_7967_cast_fp16, var_8273_cast_fp16))[name = tensor("op_8373_cast_fp16")]; tensor var_8375_equation_0 = const()[name = tensor("op_8375_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8375_cast_fp16 = einsum(equation = var_8375_equation_0, values = (var_7967_cast_fp16, var_8274_cast_fp16))[name = tensor("op_8375_cast_fp16")]; tensor var_8377_equation_0 = const()[name = tensor("op_8377_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8377_cast_fp16 = einsum(equation = var_8377_equation_0, values = (var_7967_cast_fp16, var_8275_cast_fp16))[name = tensor("op_8377_cast_fp16")]; tensor var_8379_equation_0 = const()[name = tensor("op_8379_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8379_cast_fp16 = einsum(equation = var_8379_equation_0, values = (var_7967_cast_fp16, var_8276_cast_fp16))[name = tensor("op_8379_cast_fp16")]; tensor var_8381_equation_0 = const()[name = tensor("op_8381_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8381_cast_fp16 = einsum(equation = var_8381_equation_0, values = (var_7967_cast_fp16, var_8277_cast_fp16))[name = tensor("op_8381_cast_fp16")]; tensor var_8383_equation_0 = const()[name = tensor("op_8383_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8383_cast_fp16 = einsum(equation = var_8383_equation_0, values = (var_7971_cast_fp16, var_8278_cast_fp16))[name = tensor("op_8383_cast_fp16")]; tensor var_8385_equation_0 = const()[name = tensor("op_8385_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8385_cast_fp16 = einsum(equation = var_8385_equation_0, values = (var_7971_cast_fp16, var_8279_cast_fp16))[name = tensor("op_8385_cast_fp16")]; tensor var_8387_equation_0 = const()[name = tensor("op_8387_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8387_cast_fp16 = einsum(equation = var_8387_equation_0, values = (var_7971_cast_fp16, var_8280_cast_fp16))[name = tensor("op_8387_cast_fp16")]; tensor var_8389_equation_0 = const()[name = tensor("op_8389_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8389_cast_fp16 = einsum(equation = var_8389_equation_0, values = (var_7971_cast_fp16, var_8281_cast_fp16))[name = tensor("op_8389_cast_fp16")]; tensor var_8391_equation_0 = const()[name = tensor("op_8391_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8391_cast_fp16 = einsum(equation = var_8391_equation_0, values = (var_7971_cast_fp16, var_8282_cast_fp16))[name = tensor("op_8391_cast_fp16")]; tensor var_8393_equation_0 = const()[name = tensor("op_8393_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8393_cast_fp16 = einsum(equation = var_8393_equation_0, values = (var_7971_cast_fp16, var_8283_cast_fp16))[name = tensor("op_8393_cast_fp16")]; tensor var_8395_equation_0 = const()[name = tensor("op_8395_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8395_cast_fp16 = einsum(equation = var_8395_equation_0, values = (var_7971_cast_fp16, var_8284_cast_fp16))[name = tensor("op_8395_cast_fp16")]; tensor var_8397_equation_0 = const()[name = tensor("op_8397_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8397_cast_fp16 = einsum(equation = var_8397_equation_0, values = (var_7971_cast_fp16, var_8285_cast_fp16))[name = tensor("op_8397_cast_fp16")]; tensor var_8399_equation_0 = const()[name = tensor("op_8399_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8399_cast_fp16 = einsum(equation = var_8399_equation_0, values = (var_7975_cast_fp16, var_8286_cast_fp16))[name = tensor("op_8399_cast_fp16")]; tensor var_8401_equation_0 = const()[name = tensor("op_8401_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8401_cast_fp16 = einsum(equation = var_8401_equation_0, values = (var_7975_cast_fp16, var_8287_cast_fp16))[name = tensor("op_8401_cast_fp16")]; tensor var_8403_equation_0 = const()[name = tensor("op_8403_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8403_cast_fp16 = einsum(equation = var_8403_equation_0, values = (var_7975_cast_fp16, var_8288_cast_fp16))[name = tensor("op_8403_cast_fp16")]; tensor var_8405_equation_0 = const()[name = tensor("op_8405_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8405_cast_fp16 = einsum(equation = var_8405_equation_0, values = (var_7975_cast_fp16, var_8289_cast_fp16))[name = tensor("op_8405_cast_fp16")]; tensor var_8407_equation_0 = const()[name = tensor("op_8407_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8407_cast_fp16 = einsum(equation = var_8407_equation_0, values = (var_7975_cast_fp16, var_8290_cast_fp16))[name = tensor("op_8407_cast_fp16")]; tensor var_8409_equation_0 = const()[name = tensor("op_8409_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8409_cast_fp16 = einsum(equation = var_8409_equation_0, values = (var_7975_cast_fp16, var_8291_cast_fp16))[name = tensor("op_8409_cast_fp16")]; tensor var_8411_equation_0 = const()[name = tensor("op_8411_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8411_cast_fp16 = einsum(equation = var_8411_equation_0, values = (var_7975_cast_fp16, var_8292_cast_fp16))[name = tensor("op_8411_cast_fp16")]; tensor var_8413_equation_0 = const()[name = tensor("op_8413_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8413_cast_fp16 = einsum(equation = var_8413_equation_0, values = (var_7975_cast_fp16, var_8293_cast_fp16))[name = tensor("op_8413_cast_fp16")]; tensor var_8415_equation_0 = const()[name = tensor("op_8415_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8415_cast_fp16 = einsum(equation = var_8415_equation_0, values = (var_7979_cast_fp16, var_8294_cast_fp16))[name = tensor("op_8415_cast_fp16")]; tensor var_8417_equation_0 = const()[name = tensor("op_8417_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8417_cast_fp16 = einsum(equation = var_8417_equation_0, values = (var_7979_cast_fp16, var_8295_cast_fp16))[name = tensor("op_8417_cast_fp16")]; tensor var_8419_equation_0 = const()[name = tensor("op_8419_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8419_cast_fp16 = einsum(equation = var_8419_equation_0, values = (var_7979_cast_fp16, var_8296_cast_fp16))[name = tensor("op_8419_cast_fp16")]; tensor var_8421_equation_0 = const()[name = tensor("op_8421_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8421_cast_fp16 = einsum(equation = var_8421_equation_0, values = (var_7979_cast_fp16, var_8297_cast_fp16))[name = tensor("op_8421_cast_fp16")]; tensor var_8423_equation_0 = const()[name = tensor("op_8423_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8423_cast_fp16 = einsum(equation = var_8423_equation_0, values = (var_7979_cast_fp16, var_8298_cast_fp16))[name = tensor("op_8423_cast_fp16")]; tensor var_8425_equation_0 = const()[name = tensor("op_8425_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8425_cast_fp16 = einsum(equation = var_8425_equation_0, values = (var_7979_cast_fp16, var_8299_cast_fp16))[name = tensor("op_8425_cast_fp16")]; tensor var_8427_equation_0 = const()[name = tensor("op_8427_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8427_cast_fp16 = einsum(equation = var_8427_equation_0, values = (var_7979_cast_fp16, var_8300_cast_fp16))[name = tensor("op_8427_cast_fp16")]; tensor var_8429_equation_0 = const()[name = tensor("op_8429_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8429_cast_fp16 = einsum(equation = var_8429_equation_0, values = (var_7979_cast_fp16, var_8301_cast_fp16))[name = tensor("op_8429_cast_fp16")]; tensor var_8431_interleave_0 = const()[name = tensor("op_8431_interleave_0"), val = tensor(false)]; tensor var_8431_cast_fp16 = concat(axis = var_5509, interleave = var_8431_interleave_0, values = (var_8303_cast_fp16, var_8305_cast_fp16, var_8307_cast_fp16, var_8309_cast_fp16, var_8311_cast_fp16, var_8313_cast_fp16, var_8315_cast_fp16, var_8317_cast_fp16))[name = tensor("op_8431_cast_fp16")]; tensor var_8433_interleave_0 = const()[name = tensor("op_8433_interleave_0"), val = tensor(false)]; tensor var_8433_cast_fp16 = concat(axis = var_5509, interleave = var_8433_interleave_0, values = (var_8319_cast_fp16, var_8321_cast_fp16, var_8323_cast_fp16, var_8325_cast_fp16, var_8327_cast_fp16, var_8329_cast_fp16, var_8331_cast_fp16, var_8333_cast_fp16))[name = tensor("op_8433_cast_fp16")]; tensor var_8435_interleave_0 = const()[name = tensor("op_8435_interleave_0"), val = tensor(false)]; tensor var_8435_cast_fp16 = concat(axis = var_5509, interleave = var_8435_interleave_0, values = (var_8335_cast_fp16, var_8337_cast_fp16, var_8339_cast_fp16, var_8341_cast_fp16, var_8343_cast_fp16, var_8345_cast_fp16, var_8347_cast_fp16, var_8349_cast_fp16))[name = tensor("op_8435_cast_fp16")]; tensor var_8437_interleave_0 = const()[name = tensor("op_8437_interleave_0"), val = tensor(false)]; tensor var_8437_cast_fp16 = concat(axis = var_5509, interleave = var_8437_interleave_0, values = (var_8351_cast_fp16, var_8353_cast_fp16, var_8355_cast_fp16, var_8357_cast_fp16, var_8359_cast_fp16, var_8361_cast_fp16, var_8363_cast_fp16, var_8365_cast_fp16))[name = tensor("op_8437_cast_fp16")]; tensor var_8439_interleave_0 = const()[name = tensor("op_8439_interleave_0"), val = tensor(false)]; tensor var_8439_cast_fp16 = concat(axis = var_5509, interleave = var_8439_interleave_0, values = (var_8367_cast_fp16, var_8369_cast_fp16, var_8371_cast_fp16, var_8373_cast_fp16, var_8375_cast_fp16, var_8377_cast_fp16, var_8379_cast_fp16, var_8381_cast_fp16))[name = tensor("op_8439_cast_fp16")]; tensor var_8441_interleave_0 = const()[name = tensor("op_8441_interleave_0"), val = tensor(false)]; tensor var_8441_cast_fp16 = concat(axis = var_5509, interleave = var_8441_interleave_0, values = (var_8383_cast_fp16, var_8385_cast_fp16, var_8387_cast_fp16, var_8389_cast_fp16, var_8391_cast_fp16, var_8393_cast_fp16, var_8395_cast_fp16, var_8397_cast_fp16))[name = tensor("op_8441_cast_fp16")]; tensor var_8443_interleave_0 = const()[name = tensor("op_8443_interleave_0"), val = tensor(false)]; tensor var_8443_cast_fp16 = concat(axis = var_5509, interleave = var_8443_interleave_0, values = (var_8399_cast_fp16, var_8401_cast_fp16, var_8403_cast_fp16, var_8405_cast_fp16, var_8407_cast_fp16, var_8409_cast_fp16, var_8411_cast_fp16, var_8413_cast_fp16))[name = tensor("op_8443_cast_fp16")]; tensor var_8445_interleave_0 = const()[name = tensor("op_8445_interleave_0"), val = tensor(false)]; tensor var_8445_cast_fp16 = concat(axis = var_5509, interleave = var_8445_interleave_0, values = (var_8415_cast_fp16, var_8417_cast_fp16, var_8419_cast_fp16, var_8421_cast_fp16, var_8423_cast_fp16, var_8425_cast_fp16, var_8427_cast_fp16, var_8429_cast_fp16))[name = tensor("op_8445_cast_fp16")]; tensor input_247_interleave_0 = const()[name = tensor("input_247_interleave_0"), val = tensor(false)]; tensor input_247_cast_fp16 = concat(axis = var_5537, interleave = input_247_interleave_0, values = (var_8431_cast_fp16, var_8433_cast_fp16, var_8435_cast_fp16, var_8437_cast_fp16, var_8439_cast_fp16, var_8441_cast_fp16, var_8443_cast_fp16, var_8445_cast_fp16))[name = tensor("input_247_cast_fp16")]; tensor var_8451 = const()[name = tensor("op_8451"), val = tensor([1, 1])]; tensor var_8453 = const()[name = tensor("op_8453"), val = tensor([1, 1])]; tensor var_8455_pad_type_0 = const()[name = tensor("op_8455_pad_type_0"), val = tensor("custom")]; tensor var_8455_pad_0 = const()[name = tensor("op_8455_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(241712768))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(241789632))), name = tensor("up_blocks_2_attentions_1_transformer_blocks_0_attn2_to_out_0_weight_to_fp16_palettized"), shape = tensor([320, 320, 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(241789824)))]; tensor var_8455_cast_fp16 = conv(bias = up_blocks_2_attentions_1_transformer_blocks_0_attn2_to_out_0_bias_to_fp16, dilations = var_8453, groups = var_5537, pad = var_8455_pad_0, pad_type = var_8455_pad_type_0, strides = var_8451, weight = up_blocks_2_attentions_1_transformer_blocks_0_attn2_to_out_0_weight_to_fp16_palettized, x = input_247_cast_fp16)[name = tensor("op_8455_cast_fp16")]; tensor inputs_cast_fp16 = add(x = var_8455_cast_fp16, y = inputs_51_cast_fp16)[name = tensor("inputs_cast_fp16")]; tensor var_8459 = const()[name = tensor("op_8459"), val = tensor([1])]; tensor channels_mean_cast_fp16 = reduce_mean(axes = var_8459, keep_dims = var_5532, x = inputs_cast_fp16)[name = tensor("channels_mean_cast_fp16")]; tensor zero_mean_cast_fp16 = sub(x = inputs_cast_fp16, y = channels_mean_cast_fp16)[name = tensor("zero_mean_cast_fp16")]; tensor zero_mean_sq_cast_fp16 = mul(x = zero_mean_cast_fp16, y = zero_mean_cast_fp16)[name = tensor("zero_mean_sq_cast_fp16")]; tensor var_8463 = const()[name = tensor("op_8463"), val = tensor([1])]; tensor var_8464_cast_fp16 = reduce_mean(axes = var_8463, keep_dims = var_5532, x = zero_mean_sq_cast_fp16)[name = tensor("op_8464_cast_fp16")]; tensor var_8465_to_fp16 = const()[name = tensor("op_8465_to_fp16"), val = tensor(0x1.5p-17)]; tensor var_8466_cast_fp16 = add(x = var_8464_cast_fp16, y = var_8465_to_fp16)[name = tensor("op_8466_cast_fp16")]; tensor denom_epsilon_0_to_fp16 = const()[name = tensor("denom_epsilon_0_to_fp16"), val = tensor(0x1p-24)]; tensor denom_cast_fp16 = rsqrt(epsilon = denom_epsilon_0_to_fp16, x = var_8466_cast_fp16)[name = tensor("denom_cast_fp16")]; tensor out_cast_fp16 = mul(x = zero_mean_cast_fp16, y = denom_cast_fp16)[name = tensor("out_cast_fp16")]; tensor var_8470_to_fp16 = const()[name = tensor("op_8470_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(241790528)))]; tensor var_8471_cast_fp16 = add(x = out_cast_fp16, y = var_8470_to_fp16)[name = tensor("op_8471_cast_fp16")]; tensor var_8473_to_fp16 = const()[name = tensor("op_8473_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(241791232)))]; tensor input_249_cast_fp16 = mul(x = var_8471_cast_fp16, y = var_8473_to_fp16)[name = tensor("input_249_cast_fp16")]; tensor var_8481 = const()[name = tensor("op_8481"), val = tensor([1, 1])]; tensor var_8483 = const()[name = tensor("op_8483"), val = tensor([1, 1])]; tensor var_8485_pad_type_0 = const()[name = tensor("op_8485_pad_type_0"), val = tensor("custom")]; tensor var_8485_pad_0 = const()[name = tensor("op_8485_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(241791936))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(242406400))), name = tensor("up_blocks_2_attentions_1_transformer_blocks_0_ff_net_0_proj_weight_to_fp16_palettized"), shape = tensor([2560, 320, 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(242406592))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(242408576))), name = tensor("up_blocks_2_attentions_1_transformer_blocks_0_ff_net_0_proj_bias_to_fp16_palettized"), shape = tensor([2560])]; tensor var_8485_cast_fp16 = conv(bias = up_blocks_2_attentions_1_transformer_blocks_0_ff_net_0_proj_bias_to_fp16_palettized, dilations = var_8483, groups = var_5537, pad = var_8485_pad_0, pad_type = var_8485_pad_type_0, strides = var_8481, weight = up_blocks_2_attentions_1_transformer_blocks_0_ff_net_0_proj_weight_to_fp16_palettized, x = input_249_cast_fp16)[name = tensor("op_8485_cast_fp16")]; tensor var_8486_split_sizes_0 = const()[name = tensor("op_8486_split_sizes_0"), val = tensor([1280, 1280])]; tensor var_8486_axis_0 = const()[name = tensor("op_8486_axis_0"), val = tensor(1)]; tensor var_8486_cast_fp16_0, tensor var_8486_cast_fp16_1 = split(axis = var_8486_axis_0, split_sizes = var_8486_split_sizes_0, x = var_8485_cast_fp16)[name = tensor("op_8486_cast_fp16")]; tensor var_8488_mode_0 = const()[name = tensor("op_8488_mode_0"), val = tensor("EXACT")]; tensor var_8488_cast_fp16 = gelu(mode = var_8488_mode_0, x = var_8486_cast_fp16_1)[name = tensor("op_8488_cast_fp16")]; tensor input_251_cast_fp16 = mul(x = var_8486_cast_fp16_0, y = var_8488_cast_fp16)[name = tensor("input_251_cast_fp16")]; tensor var_8492 = const()[name = tensor("op_8492"), val = tensor([1, 1])]; tensor var_8494 = const()[name = tensor("op_8494"), val = tensor([1, 1])]; tensor var_8496_pad_type_0 = const()[name = tensor("op_8496_pad_type_0"), val = tensor("custom")]; tensor var_8496_pad_0 = const()[name = tensor("op_8496_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(242408768))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(242716032))), name = tensor("up_blocks_2_attentions_1_transformer_blocks_0_ff_net_2_weight_to_fp16_palettized"), shape = tensor([320, 1280, 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(242716224)))]; tensor var_8496_cast_fp16 = conv(bias = up_blocks_2_attentions_1_transformer_blocks_0_ff_net_2_bias_to_fp16, dilations = var_8494, groups = var_5537, pad = var_8496_pad_0, pad_type = var_8496_pad_type_0, strides = var_8492, weight = up_blocks_2_attentions_1_transformer_blocks_0_ff_net_2_weight_to_fp16_palettized, x = input_251_cast_fp16)[name = tensor("op_8496_cast_fp16")]; tensor hidden_states_171_cast_fp16 = add(x = var_8496_cast_fp16, y = inputs_cast_fp16)[name = tensor("hidden_states_171_cast_fp16")]; tensor var_8498 = const()[name = tensor("op_8498"), val = tensor([2, 320, 64, 64])]; tensor input_253_cast_fp16 = reshape(shape = var_8498, x = hidden_states_171_cast_fp16)[name = tensor("input_253_cast_fp16")]; tensor var_8502 = const()[name = tensor("op_8502"), val = tensor([1, 1])]; tensor var_8504 = const()[name = tensor("op_8504"), 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_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(242716928))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(242793792))), name = tensor("up_blocks_2_attentions_1_proj_out_weight_to_fp16_palettized"), shape = tensor([320, 320, 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(242793984)))]; tensor hidden_states_cast_fp16 = conv(bias = up_blocks_2_attentions_1_proj_out_bias_to_fp16, dilations = var_8504, groups = var_5537, pad = hidden_states_pad_0, pad_type = hidden_states_pad_type_0, strides = var_8502, weight = up_blocks_2_attentions_1_proj_out_weight_to_fp16_palettized, x = input_253_cast_fp16)[name = tensor("hidden_states_cast_fp16")]; tensor input_255_cast_fp16 = add(x = hidden_states_cast_fp16, y = hidden_states_161_cast_fp16)[name = tensor("input_255_cast_fp16")]; tensor reshape_108_shape_0 = const()[name = tensor("reshape_108_shape_0"), val = tensor([2, 32, 10, 64, 64])]; tensor reshape_108_cast_fp16 = reshape(shape = reshape_108_shape_0, x = input_255_cast_fp16)[name = tensor("reshape_108_cast_fp16")]; 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_fp16 = reduce_mean(axes = reduce_mean_81_axes_0, keep_dims = reduce_mean_81_keep_dims_0, x = reshape_108_cast_fp16)[name = tensor("reduce_mean_81_cast_fp16")]; tensor sub_54_cast_fp16 = sub(x = reshape_108_cast_fp16, y = reduce_mean_81_cast_fp16)[name = tensor("sub_54_cast_fp16")]; tensor square_27_cast_fp16 = square(x = sub_54_cast_fp16)[name = tensor("square_27_cast_fp16")]; 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_fp16 = reduce_mean(axes = reduce_mean_83_axes_0, keep_dims = reduce_mean_83_keep_dims_0, x = square_27_cast_fp16)[name = tensor("reduce_mean_83_cast_fp16")]; 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_fp16 = add(x = reduce_mean_83_cast_fp16, y = add_54_y_0_to_fp16)[name = tensor("add_54_cast_fp16")]; tensor sqrt_27_cast_fp16 = sqrt(x = add_54_cast_fp16)[name = tensor("sqrt_27_cast_fp16")]; tensor real_div_27_cast_fp16 = real_div(x = sub_54_cast_fp16, y = sqrt_27_cast_fp16)[name = tensor("real_div_27_cast_fp16")]; tensor reshape_109_shape_0 = const()[name = tensor("reshape_109_shape_0"), val = tensor([2, 320, 64, 64])]; tensor reshape_109_cast_fp16 = reshape(shape = reshape_109_shape_0, x = real_div_27_cast_fp16)[name = tensor("reshape_109_cast_fp16")]; tensor add_55_gamma_0_to_fp16 = const()[name = tensor("add_55_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(242794688)))]; tensor add_55_beta_0_to_fp16 = const()[name = tensor("add_55_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(242795392)))]; 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_fp16 = batch_norm(beta = add_55_beta_0_to_fp16, epsilon = add_55_epsilon_0_to_fp16, gamma = add_55_gamma_0_to_fp16, mean = add_1_mean_0_to_fp16, variance = add_1_variance_0_to_fp16, x = reshape_109_cast_fp16)[name = tensor("add_55_cast_fp16")]; tensor input_cast_fp16 = silu(x = add_55_cast_fp16)[name = tensor("input_cast_fp16")]; tensor var_8518 = const()[name = tensor("op_8518"), val = tensor(1)]; tensor var_8521 = const()[name = tensor("op_8521"), val = tensor([1, 1])]; tensor var_8523 = const()[name = tensor("op_8523"), val = tensor([1, 1])]; tensor var_8525_pad_type_0 = const()[name = tensor("op_8525_pad_type_0"), val = tensor("custom")]; tensor var_8525_pad_0 = const()[name = tensor("op_8525_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(242796096))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(242804800))), 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.ba8p-10, -0x1.df8p-12, -0x1.538p-11, -0x1.284p-9])]; tensor var_8525_cast_fp16 = conv(bias = conv_out_bias_to_fp16, dilations = var_8523, groups = var_8518, pad = var_8525_pad_0, pad_type = var_8525_pad_type_0, strides = var_8521, weight = conv_out_weight_to_fp16_palettized, x = input_cast_fp16)[name = tensor("op_8525_cast_fp16")]; tensor var_8525_cast_fp16_to_fp32_dtype_0 = const()[name = tensor("op_8525_cast_fp16_to_fp32_dtype_0"), val = tensor("fp32")]; tensor noise_pred = cast(dtype = var_8525_cast_fp16_to_fp32_dtype_0, x = var_8525_cast_fp16)[name = tensor("cast_0")]; } -> (noise_pred); }