diff --git "a/SplitEinsum-Resources-6bit/TextEncoder.mlmodelc/model.mil" "b/SplitEinsum-Resources-6bit/TextEncoder.mlmodelc/model.mil" new file mode 100644--- /dev/null +++ "b/SplitEinsum-Resources-6bit/TextEncoder.mlmodelc/model.mil" @@ -0,0 +1,872 @@ +program(1.0) +[buildInfo = dict, tensor>({{"coremlc-component-MIL", "3304.5.2"}, {"coremlc-version", "3304.6.2"}})] +{ + func main(tensor input_ids) { + tensor var_5 = const()[name = tensor("op_5"), val = tensor(-1)]; + tensor var_6 = const()[name = tensor("op_6"), val = tensor(false)]; + tensor cast_1_dtype_0 = const()[name = tensor("cast_1_dtype_0"), val = tensor("int32")]; + tensor inputs_embeds_axis_0 = const()[name = tensor("inputs_embeds_axis_0"), val = tensor(0)]; + tensor inputs_embeds_batch_dims_0 = const()[name = tensor("inputs_embeds_batch_dims_0"), val = tensor(0)]; + tensor text_encoder_text_model_embeddings_token_embedding_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_embeddings_token_embedding_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(64)))]; + tensor cast_1 = cast(dtype = cast_1_dtype_0, x = input_ids)[name = tensor("cast_2")]; + tensor inputs_embeds_cast_fp16 = gather(axis = inputs_embeds_axis_0, batch_dims = inputs_embeds_batch_dims_0, indices = cast_1, x = text_encoder_text_model_embeddings_token_embedding_weight_to_fp16)[name = tensor("inputs_embeds_cast_fp16")]; + tensor position_embeddings_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(75890816))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(75935232))), name = tensor("position_embeddings_to_fp16_palettized"), shape = tensor([1, 77, 768])]; + tensor input_3_cast_fp16 = add(x = inputs_embeds_cast_fp16, y = position_embeddings_to_fp16_palettized)[name = tensor("input_3_cast_fp16")]; + tensor hidden_states_1_axes_0 = const()[name = tensor("hidden_states_1_axes_0"), val = tensor([-1])]; + tensor text_encoder_text_model_encoder_layers_0_layer_norm1_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_0_layer_norm1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(75935424)))]; + tensor text_encoder_text_model_encoder_layers_0_layer_norm1_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_0_layer_norm1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(75937024)))]; + tensor var_15_to_fp16 = const()[name = tensor("op_15_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_1_cast_fp16 = layer_norm(axes = hidden_states_1_axes_0, beta = text_encoder_text_model_encoder_layers_0_layer_norm1_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_0_layer_norm1_weight_to_fp16, x = input_3_cast_fp16)[name = tensor("hidden_states_1_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_0_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(75938624))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(76381056))), name = tensor("text_encoder_text_model_encoder_layers_0_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor([768, 768])]; + tensor text_encoder_text_model_encoder_layers_0_self_attn_q_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_0_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(76381248)))]; + tensor linear_0_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_0_self_attn_q_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_0_self_attn_q_proj_weight_to_fp16_palettized, x = hidden_states_1_cast_fp16)[name = tensor("linear_0_cast_fp16")]; + tensor var_107_to_fp16 = const()[name = tensor("op_107_to_fp16"), val = tensor(0x1p-3)]; + tensor tensor_5_cast_fp16 = mul(x = linear_0_cast_fp16, y = var_107_to_fp16)[name = tensor("tensor_5_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_0_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(76382848))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(76825280))), name = tensor("text_encoder_text_model_encoder_layers_0_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor([768, 768])]; + tensor text_encoder_text_model_encoder_layers_0_self_attn_k_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_0_self_attn_k_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(76825472)))]; + tensor linear_1_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_0_self_attn_k_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_0_self_attn_k_proj_weight_to_fp16_palettized, x = hidden_states_1_cast_fp16)[name = tensor("linear_1_cast_fp16")]; + tensor var_112 = const()[name = tensor("op_112"), val = tensor([1, -1, 12, 64])]; + tensor var_113_cast_fp16 = reshape(shape = var_112, x = linear_1_cast_fp16)[name = tensor("op_113_cast_fp16")]; + tensor var_114_perm_0 = const()[name = tensor("op_114_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor text_encoder_text_model_encoder_layers_0_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(76827072))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(77269504))), name = tensor("text_encoder_text_model_encoder_layers_0_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor([768, 768])]; + tensor text_encoder_text_model_encoder_layers_0_self_attn_v_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_0_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(77269696)))]; + tensor linear_2_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_0_self_attn_v_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_0_self_attn_v_proj_weight_to_fp16_palettized, x = hidden_states_1_cast_fp16)[name = tensor("linear_2_cast_fp16")]; + tensor var_119 = const()[name = tensor("op_119"), val = tensor([1, -1, 12, 64])]; + tensor var_120_cast_fp16 = reshape(shape = var_119, x = linear_2_cast_fp16)[name = tensor("op_120_cast_fp16")]; + tensor var_121_perm_0 = const()[name = tensor("op_121_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor var_128 = const()[name = tensor("op_128"), val = tensor([1, 77, 12, 64])]; + tensor var_129_cast_fp16 = reshape(shape = var_128, x = tensor_5_cast_fp16)[name = tensor("op_129_cast_fp16")]; + tensor var_130_perm_0 = const()[name = tensor("op_130_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor var_132 = const()[name = tensor("op_132"), val = tensor([12, -1, 64])]; + tensor var_130_cast_fp16 = transpose(perm = var_130_perm_0, x = var_129_cast_fp16)[name = tensor("transpose_47")]; + tensor query_states_1_cast_fp16 = reshape(shape = var_132, x = var_130_cast_fp16)[name = tensor("query_states_1_cast_fp16")]; + tensor var_134 = const()[name = tensor("op_134"), val = tensor([12, -1, 64])]; + tensor var_114_cast_fp16 = transpose(perm = var_114_perm_0, x = var_113_cast_fp16)[name = tensor("transpose_46")]; + tensor key_states_3_cast_fp16 = reshape(shape = var_134, x = var_114_cast_fp16)[name = tensor("key_states_3_cast_fp16")]; + tensor var_136 = const()[name = tensor("op_136"), val = tensor([12, -1, 64])]; + tensor var_121_cast_fp16 = transpose(perm = var_121_perm_0, x = var_120_cast_fp16)[name = tensor("transpose_45")]; + tensor value_states_3_cast_fp16 = reshape(shape = var_136, x = var_121_cast_fp16)[name = tensor("value_states_3_cast_fp16")]; + tensor attn_weights_1_transpose_x_1 = const()[name = tensor("attn_weights_1_transpose_x_1"), val = tensor(false)]; + tensor attn_weights_1_transpose_y_1 = const()[name = tensor("attn_weights_1_transpose_y_1"), val = tensor(true)]; + tensor attn_weights_1_cast_fp16 = matmul(transpose_x = attn_weights_1_transpose_x_1, transpose_y = attn_weights_1_transpose_y_1, x = query_states_1_cast_fp16, y = key_states_3_cast_fp16)[name = tensor("attn_weights_1_cast_fp16")]; + tensor var_141 = const()[name = tensor("op_141"), val = tensor([1, 12, 77, 77])]; + tensor var_142_cast_fp16 = reshape(shape = var_141, x = attn_weights_1_cast_fp16)[name = tensor("op_142_cast_fp16")]; + tensor op_56_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(77271296))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(77275840))), name = tensor("op_56_to_fp16_palettized"), shape = tensor([1, 1, 77, 77])]; + tensor attn_weights_3_cast_fp16 = add(x = var_142_cast_fp16, y = op_56_to_fp16_palettized)[name = tensor("attn_weights_3_cast_fp16")]; + tensor var_147 = const()[name = tensor("op_147"), val = tensor([12, 77, 77])]; + tensor input_5_cast_fp16 = reshape(shape = var_147, x = attn_weights_3_cast_fp16)[name = tensor("input_5_cast_fp16")]; + tensor input_7_cast_fp16 = softmax(axis = var_5, x = input_5_cast_fp16)[name = tensor("input_7_cast_fp16")]; + tensor attn_output_1_transpose_x_0 = const()[name = tensor("attn_output_1_transpose_x_0"), val = tensor(false)]; + tensor attn_output_1_transpose_y_0 = const()[name = tensor("attn_output_1_transpose_y_0"), val = tensor(false)]; + tensor attn_output_1_cast_fp16 = matmul(transpose_x = attn_output_1_transpose_x_0, transpose_y = attn_output_1_transpose_y_0, x = input_7_cast_fp16, y = value_states_3_cast_fp16)[name = tensor("attn_output_1_cast_fp16")]; + tensor var_152 = const()[name = tensor("op_152"), val = tensor([1, 12, 77, 64])]; + tensor attn_output_3_cast_fp16 = reshape(shape = var_152, x = attn_output_1_cast_fp16)[name = tensor("attn_output_3_cast_fp16")]; + tensor attn_output_5_perm_0 = const()[name = tensor("attn_output_5_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor var_155 = const()[name = tensor("op_155"), val = tensor([1, 77, 768])]; + tensor attn_output_5_cast_fp16 = transpose(perm = attn_output_5_perm_0, x = attn_output_3_cast_fp16)[name = tensor("transpose_44")]; + tensor input_9_cast_fp16 = reshape(shape = var_155, x = attn_output_5_cast_fp16)[name = tensor("input_9_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_0_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(77276032))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(77718464))), name = tensor("text_encoder_text_model_encoder_layers_0_self_attn_out_proj_weight_to_fp16_palettized"), shape = tensor([768, 768])]; + tensor text_encoder_text_model_encoder_layers_0_self_attn_out_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_0_self_attn_out_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(77718656)))]; + tensor linear_3_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_0_self_attn_out_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_0_self_attn_out_proj_weight_to_fp16_palettized, x = input_9_cast_fp16)[name = tensor("linear_3_cast_fp16")]; + tensor input_11_cast_fp16 = add(x = input_3_cast_fp16, y = linear_3_cast_fp16)[name = tensor("input_11_cast_fp16")]; + tensor input_13_axes_0 = const()[name = tensor("input_13_axes_0"), val = tensor([-1])]; + tensor text_encoder_text_model_encoder_layers_0_layer_norm2_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_0_layer_norm2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(77720256)))]; + tensor text_encoder_text_model_encoder_layers_0_layer_norm2_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_0_layer_norm2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(77721856)))]; + tensor input_13_cast_fp16 = layer_norm(axes = input_13_axes_0, beta = text_encoder_text_model_encoder_layers_0_layer_norm2_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_0_layer_norm2_weight_to_fp16, x = input_11_cast_fp16)[name = tensor("input_13_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_0_mlp_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(77723456))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(79492992))), name = tensor("text_encoder_text_model_encoder_layers_0_mlp_fc1_weight_to_fp16_palettized"), shape = tensor([3072, 768])]; + tensor text_encoder_text_model_encoder_layers_0_mlp_fc1_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_0_mlp_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(79493184)))]; + tensor linear_4_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_0_mlp_fc1_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_0_mlp_fc1_weight_to_fp16_palettized, x = input_13_cast_fp16)[name = tensor("linear_4_cast_fp16")]; + tensor var_170_to_fp16 = const()[name = tensor("op_170_to_fp16"), val = tensor(0x1.b3cp+0)]; + tensor var_171_cast_fp16 = mul(x = linear_4_cast_fp16, y = var_170_to_fp16)[name = tensor("op_171_cast_fp16")]; + tensor var_172_cast_fp16 = sigmoid(x = var_171_cast_fp16)[name = tensor("op_172_cast_fp16")]; + tensor input_17_cast_fp16 = mul(x = linear_4_cast_fp16, y = var_172_cast_fp16)[name = tensor("input_17_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_0_mlp_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(79499392))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(81268928))), name = tensor("text_encoder_text_model_encoder_layers_0_mlp_fc2_weight_to_fp16_palettized"), shape = tensor([768, 3072])]; + tensor text_encoder_text_model_encoder_layers_0_mlp_fc2_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_0_mlp_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(81269120)))]; + tensor linear_5_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_0_mlp_fc2_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_0_mlp_fc2_weight_to_fp16_palettized, x = input_17_cast_fp16)[name = tensor("linear_5_cast_fp16")]; + tensor input_19_cast_fp16 = add(x = input_11_cast_fp16, y = linear_5_cast_fp16)[name = tensor("input_19_cast_fp16")]; + tensor hidden_states_7_axes_0 = const()[name = tensor("hidden_states_7_axes_0"), val = tensor([-1])]; + tensor text_encoder_text_model_encoder_layers_1_layer_norm1_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_1_layer_norm1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(81270720)))]; + tensor text_encoder_text_model_encoder_layers_1_layer_norm1_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_1_layer_norm1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(81272320)))]; + tensor hidden_states_7_cast_fp16 = layer_norm(axes = hidden_states_7_axes_0, beta = text_encoder_text_model_encoder_layers_1_layer_norm1_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_1_layer_norm1_weight_to_fp16, x = input_19_cast_fp16)[name = tensor("hidden_states_7_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_1_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(81273920))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(81716352))), name = tensor("text_encoder_text_model_encoder_layers_1_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor([768, 768])]; + tensor text_encoder_text_model_encoder_layers_1_self_attn_q_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_1_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(81716544)))]; + tensor linear_6_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_1_self_attn_q_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_1_self_attn_q_proj_weight_to_fp16_palettized, x = hidden_states_7_cast_fp16)[name = tensor("linear_6_cast_fp16")]; + tensor var_197_to_fp16 = const()[name = tensor("op_197_to_fp16"), val = tensor(0x1p-3)]; + tensor tensor_11_cast_fp16 = mul(x = linear_6_cast_fp16, y = var_197_to_fp16)[name = tensor("tensor_11_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_1_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(81718144))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(82160576))), name = tensor("text_encoder_text_model_encoder_layers_1_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor([768, 768])]; + tensor text_encoder_text_model_encoder_layers_1_self_attn_k_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_1_self_attn_k_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(82160768)))]; + tensor linear_7_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_1_self_attn_k_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_1_self_attn_k_proj_weight_to_fp16_palettized, x = hidden_states_7_cast_fp16)[name = tensor("linear_7_cast_fp16")]; + tensor var_202 = const()[name = tensor("op_202"), val = tensor([1, -1, 12, 64])]; + tensor var_203_cast_fp16 = reshape(shape = var_202, x = linear_7_cast_fp16)[name = tensor("op_203_cast_fp16")]; + tensor var_204_perm_0 = const()[name = tensor("op_204_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor text_encoder_text_model_encoder_layers_1_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(82162368))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(82604800))), name = tensor("text_encoder_text_model_encoder_layers_1_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor([768, 768])]; + tensor text_encoder_text_model_encoder_layers_1_self_attn_v_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_1_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(82604992)))]; + tensor linear_8_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_1_self_attn_v_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_1_self_attn_v_proj_weight_to_fp16_palettized, x = hidden_states_7_cast_fp16)[name = tensor("linear_8_cast_fp16")]; + tensor var_209 = const()[name = tensor("op_209"), val = tensor([1, -1, 12, 64])]; + tensor var_210_cast_fp16 = reshape(shape = var_209, x = linear_8_cast_fp16)[name = tensor("op_210_cast_fp16")]; + tensor var_211_perm_0 = const()[name = tensor("op_211_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor var_218 = const()[name = tensor("op_218"), val = tensor([1, 77, 12, 64])]; + tensor var_219_cast_fp16 = reshape(shape = var_218, x = tensor_11_cast_fp16)[name = tensor("op_219_cast_fp16")]; + tensor var_220_perm_0 = const()[name = tensor("op_220_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor var_222 = const()[name = tensor("op_222"), val = tensor([12, -1, 64])]; + tensor var_220_cast_fp16 = transpose(perm = var_220_perm_0, x = var_219_cast_fp16)[name = tensor("transpose_43")]; + tensor query_states_3_cast_fp16 = reshape(shape = var_222, x = var_220_cast_fp16)[name = tensor("query_states_3_cast_fp16")]; + tensor var_224 = const()[name = tensor("op_224"), val = tensor([12, -1, 64])]; + tensor var_204_cast_fp16 = transpose(perm = var_204_perm_0, x = var_203_cast_fp16)[name = tensor("transpose_42")]; + tensor key_states_7_cast_fp16 = reshape(shape = var_224, x = var_204_cast_fp16)[name = tensor("key_states_7_cast_fp16")]; + tensor var_226 = const()[name = tensor("op_226"), val = tensor([12, -1, 64])]; + tensor var_211_cast_fp16 = transpose(perm = var_211_perm_0, x = var_210_cast_fp16)[name = tensor("transpose_41")]; + tensor value_states_7_cast_fp16 = reshape(shape = var_226, x = var_211_cast_fp16)[name = tensor("value_states_7_cast_fp16")]; + tensor attn_weights_7_transpose_x_1 = const()[name = tensor("attn_weights_7_transpose_x_1"), val = tensor(false)]; + tensor attn_weights_7_transpose_y_1 = const()[name = tensor("attn_weights_7_transpose_y_1"), val = tensor(true)]; + tensor attn_weights_7_cast_fp16 = matmul(transpose_x = attn_weights_7_transpose_x_1, transpose_y = attn_weights_7_transpose_y_1, x = query_states_3_cast_fp16, y = key_states_7_cast_fp16)[name = tensor("attn_weights_7_cast_fp16")]; + tensor var_231 = const()[name = tensor("op_231"), val = tensor([1, 12, 77, 77])]; + tensor var_232_cast_fp16 = reshape(shape = var_231, x = attn_weights_7_cast_fp16)[name = tensor("op_232_cast_fp16")]; + tensor attn_weights_9_cast_fp16 = add(x = var_232_cast_fp16, y = op_56_to_fp16_palettized)[name = tensor("attn_weights_9_cast_fp16")]; + tensor var_237 = const()[name = tensor("op_237"), val = tensor([12, 77, 77])]; + tensor input_21_cast_fp16 = reshape(shape = var_237, x = attn_weights_9_cast_fp16)[name = tensor("input_21_cast_fp16")]; + tensor input_23_cast_fp16 = softmax(axis = var_5, x = input_21_cast_fp16)[name = tensor("input_23_cast_fp16")]; + tensor attn_output_7_transpose_x_0 = const()[name = tensor("attn_output_7_transpose_x_0"), val = tensor(false)]; + tensor attn_output_7_transpose_y_0 = const()[name = tensor("attn_output_7_transpose_y_0"), val = tensor(false)]; + tensor attn_output_7_cast_fp16 = matmul(transpose_x = attn_output_7_transpose_x_0, transpose_y = attn_output_7_transpose_y_0, x = input_23_cast_fp16, y = value_states_7_cast_fp16)[name = tensor("attn_output_7_cast_fp16")]; + tensor var_242 = const()[name = tensor("op_242"), val = tensor([1, 12, 77, 64])]; + tensor attn_output_9_cast_fp16 = reshape(shape = var_242, x = attn_output_7_cast_fp16)[name = tensor("attn_output_9_cast_fp16")]; + tensor attn_output_11_perm_0 = const()[name = tensor("attn_output_11_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor var_245 = const()[name = tensor("op_245"), val = tensor([1, 77, 768])]; + tensor attn_output_11_cast_fp16 = transpose(perm = attn_output_11_perm_0, x = attn_output_9_cast_fp16)[name = tensor("transpose_40")]; + tensor input_25_cast_fp16 = reshape(shape = var_245, x = attn_output_11_cast_fp16)[name = tensor("input_25_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_1_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(82606592))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(83049024))), name = tensor("text_encoder_text_model_encoder_layers_1_self_attn_out_proj_weight_to_fp16_palettized"), shape = tensor([768, 768])]; + tensor text_encoder_text_model_encoder_layers_1_self_attn_out_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_1_self_attn_out_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(83049216)))]; + tensor linear_9_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_1_self_attn_out_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_1_self_attn_out_proj_weight_to_fp16_palettized, x = input_25_cast_fp16)[name = tensor("linear_9_cast_fp16")]; + tensor input_27_cast_fp16 = add(x = input_19_cast_fp16, y = linear_9_cast_fp16)[name = tensor("input_27_cast_fp16")]; + tensor input_29_axes_0 = const()[name = tensor("input_29_axes_0"), val = tensor([-1])]; + tensor text_encoder_text_model_encoder_layers_1_layer_norm2_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_1_layer_norm2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(83050816)))]; + tensor text_encoder_text_model_encoder_layers_1_layer_norm2_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_1_layer_norm2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(83052416)))]; + tensor input_29_cast_fp16 = layer_norm(axes = input_29_axes_0, beta = text_encoder_text_model_encoder_layers_1_layer_norm2_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_1_layer_norm2_weight_to_fp16, x = input_27_cast_fp16)[name = tensor("input_29_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_1_mlp_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(83054016))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(84823552))), name = tensor("text_encoder_text_model_encoder_layers_1_mlp_fc1_weight_to_fp16_palettized"), shape = tensor([3072, 768])]; + tensor text_encoder_text_model_encoder_layers_1_mlp_fc1_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_1_mlp_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(84823744)))]; + tensor linear_10_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_1_mlp_fc1_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_1_mlp_fc1_weight_to_fp16_palettized, x = input_29_cast_fp16)[name = tensor("linear_10_cast_fp16")]; + tensor var_260_to_fp16 = const()[name = tensor("op_260_to_fp16"), val = tensor(0x1.b3cp+0)]; + tensor var_261_cast_fp16 = mul(x = linear_10_cast_fp16, y = var_260_to_fp16)[name = tensor("op_261_cast_fp16")]; + tensor var_262_cast_fp16 = sigmoid(x = var_261_cast_fp16)[name = tensor("op_262_cast_fp16")]; + tensor input_33_cast_fp16 = mul(x = linear_10_cast_fp16, y = var_262_cast_fp16)[name = tensor("input_33_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_1_mlp_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(84829952))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(86599488))), name = tensor("text_encoder_text_model_encoder_layers_1_mlp_fc2_weight_to_fp16_palettized"), shape = tensor([768, 3072])]; + tensor text_encoder_text_model_encoder_layers_1_mlp_fc2_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_1_mlp_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(86599680)))]; + tensor linear_11_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_1_mlp_fc2_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_1_mlp_fc2_weight_to_fp16_palettized, x = input_33_cast_fp16)[name = tensor("linear_11_cast_fp16")]; + tensor input_35_cast_fp16 = add(x = input_27_cast_fp16, y = linear_11_cast_fp16)[name = tensor("input_35_cast_fp16")]; + tensor hidden_states_13_axes_0 = const()[name = tensor("hidden_states_13_axes_0"), val = tensor([-1])]; + tensor text_encoder_text_model_encoder_layers_2_layer_norm1_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_2_layer_norm1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(86601280)))]; + tensor text_encoder_text_model_encoder_layers_2_layer_norm1_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_2_layer_norm1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(86602880)))]; + tensor hidden_states_13_cast_fp16 = layer_norm(axes = hidden_states_13_axes_0, beta = text_encoder_text_model_encoder_layers_2_layer_norm1_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_2_layer_norm1_weight_to_fp16, x = input_35_cast_fp16)[name = tensor("hidden_states_13_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_2_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(86604480))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(87046912))), name = tensor("text_encoder_text_model_encoder_layers_2_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor([768, 768])]; + tensor text_encoder_text_model_encoder_layers_2_self_attn_q_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_2_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(87047104)))]; + tensor linear_12_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_2_self_attn_q_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_2_self_attn_q_proj_weight_to_fp16_palettized, x = hidden_states_13_cast_fp16)[name = tensor("linear_12_cast_fp16")]; + tensor var_287_to_fp16 = const()[name = tensor("op_287_to_fp16"), val = tensor(0x1p-3)]; + tensor tensor_17_cast_fp16 = mul(x = linear_12_cast_fp16, y = var_287_to_fp16)[name = tensor("tensor_17_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_2_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(87048704))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(87491136))), name = tensor("text_encoder_text_model_encoder_layers_2_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor([768, 768])]; + tensor text_encoder_text_model_encoder_layers_2_self_attn_k_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_2_self_attn_k_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(87491328)))]; + tensor linear_13_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_2_self_attn_k_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_2_self_attn_k_proj_weight_to_fp16_palettized, x = hidden_states_13_cast_fp16)[name = tensor("linear_13_cast_fp16")]; + tensor var_292 = const()[name = tensor("op_292"), val = tensor([1, -1, 12, 64])]; + tensor var_293_cast_fp16 = reshape(shape = var_292, x = linear_13_cast_fp16)[name = tensor("op_293_cast_fp16")]; + tensor var_294_perm_0 = const()[name = tensor("op_294_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor text_encoder_text_model_encoder_layers_2_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(87492928))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(87935360))), name = tensor("text_encoder_text_model_encoder_layers_2_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor([768, 768])]; + tensor text_encoder_text_model_encoder_layers_2_self_attn_v_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_2_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(87935552)))]; + tensor linear_14_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_2_self_attn_v_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_2_self_attn_v_proj_weight_to_fp16_palettized, x = hidden_states_13_cast_fp16)[name = tensor("linear_14_cast_fp16")]; + tensor var_299 = const()[name = tensor("op_299"), val = tensor([1, -1, 12, 64])]; + tensor var_300_cast_fp16 = reshape(shape = var_299, x = linear_14_cast_fp16)[name = tensor("op_300_cast_fp16")]; + tensor var_301_perm_0 = const()[name = tensor("op_301_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor var_308 = const()[name = tensor("op_308"), val = tensor([1, 77, 12, 64])]; + tensor var_309_cast_fp16 = reshape(shape = var_308, x = tensor_17_cast_fp16)[name = tensor("op_309_cast_fp16")]; + tensor var_310_perm_0 = const()[name = tensor("op_310_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor var_312 = const()[name = tensor("op_312"), val = tensor([12, -1, 64])]; + tensor var_310_cast_fp16 = transpose(perm = var_310_perm_0, x = var_309_cast_fp16)[name = tensor("transpose_39")]; + tensor query_states_5_cast_fp16 = reshape(shape = var_312, x = var_310_cast_fp16)[name = tensor("query_states_5_cast_fp16")]; + tensor var_314 = const()[name = tensor("op_314"), val = tensor([12, -1, 64])]; + tensor var_294_cast_fp16 = transpose(perm = var_294_perm_0, x = var_293_cast_fp16)[name = tensor("transpose_38")]; + tensor key_states_11_cast_fp16 = reshape(shape = var_314, x = var_294_cast_fp16)[name = tensor("key_states_11_cast_fp16")]; + tensor var_316 = const()[name = tensor("op_316"), val = tensor([12, -1, 64])]; + tensor var_301_cast_fp16 = transpose(perm = var_301_perm_0, x = var_300_cast_fp16)[name = tensor("transpose_37")]; + tensor value_states_11_cast_fp16 = reshape(shape = var_316, x = var_301_cast_fp16)[name = tensor("value_states_11_cast_fp16")]; + tensor attn_weights_13_transpose_x_1 = const()[name = tensor("attn_weights_13_transpose_x_1"), val = tensor(false)]; + tensor attn_weights_13_transpose_y_1 = const()[name = tensor("attn_weights_13_transpose_y_1"), val = tensor(true)]; + tensor attn_weights_13_cast_fp16 = matmul(transpose_x = attn_weights_13_transpose_x_1, transpose_y = attn_weights_13_transpose_y_1, x = query_states_5_cast_fp16, y = key_states_11_cast_fp16)[name = tensor("attn_weights_13_cast_fp16")]; + tensor var_321 = const()[name = tensor("op_321"), val = tensor([1, 12, 77, 77])]; + tensor var_322_cast_fp16 = reshape(shape = var_321, x = attn_weights_13_cast_fp16)[name = tensor("op_322_cast_fp16")]; + tensor attn_weights_15_cast_fp16 = add(x = var_322_cast_fp16, y = op_56_to_fp16_palettized)[name = tensor("attn_weights_15_cast_fp16")]; + tensor var_327 = const()[name = tensor("op_327"), val = tensor([12, 77, 77])]; + tensor input_37_cast_fp16 = reshape(shape = var_327, x = attn_weights_15_cast_fp16)[name = tensor("input_37_cast_fp16")]; + tensor input_39_cast_fp16 = softmax(axis = var_5, x = input_37_cast_fp16)[name = tensor("input_39_cast_fp16")]; + tensor attn_output_13_transpose_x_0 = const()[name = tensor("attn_output_13_transpose_x_0"), val = tensor(false)]; + tensor attn_output_13_transpose_y_0 = const()[name = tensor("attn_output_13_transpose_y_0"), val = tensor(false)]; + tensor attn_output_13_cast_fp16 = matmul(transpose_x = attn_output_13_transpose_x_0, transpose_y = attn_output_13_transpose_y_0, x = input_39_cast_fp16, y = value_states_11_cast_fp16)[name = tensor("attn_output_13_cast_fp16")]; + tensor var_332 = const()[name = tensor("op_332"), val = tensor([1, 12, 77, 64])]; + tensor attn_output_15_cast_fp16 = reshape(shape = var_332, x = attn_output_13_cast_fp16)[name = tensor("attn_output_15_cast_fp16")]; + tensor attn_output_17_perm_0 = const()[name = tensor("attn_output_17_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor var_335 = const()[name = tensor("op_335"), val = tensor([1, 77, 768])]; + tensor attn_output_17_cast_fp16 = transpose(perm = attn_output_17_perm_0, x = attn_output_15_cast_fp16)[name = tensor("transpose_36")]; + tensor input_41_cast_fp16 = reshape(shape = var_335, x = attn_output_17_cast_fp16)[name = tensor("input_41_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_2_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(87937152))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(88379584))), name = tensor("text_encoder_text_model_encoder_layers_2_self_attn_out_proj_weight_to_fp16_palettized"), shape = tensor([768, 768])]; + tensor text_encoder_text_model_encoder_layers_2_self_attn_out_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_2_self_attn_out_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(88379776)))]; + tensor linear_15_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_2_self_attn_out_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_2_self_attn_out_proj_weight_to_fp16_palettized, x = input_41_cast_fp16)[name = tensor("linear_15_cast_fp16")]; + tensor input_43_cast_fp16 = add(x = input_35_cast_fp16, y = linear_15_cast_fp16)[name = tensor("input_43_cast_fp16")]; + tensor input_45_axes_0 = const()[name = tensor("input_45_axes_0"), val = tensor([-1])]; + tensor text_encoder_text_model_encoder_layers_2_layer_norm2_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_2_layer_norm2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(88381376)))]; + tensor text_encoder_text_model_encoder_layers_2_layer_norm2_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_2_layer_norm2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(88382976)))]; + tensor input_45_cast_fp16 = layer_norm(axes = input_45_axes_0, beta = text_encoder_text_model_encoder_layers_2_layer_norm2_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_2_layer_norm2_weight_to_fp16, x = input_43_cast_fp16)[name = tensor("input_45_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_2_mlp_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(88384576))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(90154112))), name = tensor("text_encoder_text_model_encoder_layers_2_mlp_fc1_weight_to_fp16_palettized"), shape = tensor([3072, 768])]; + tensor text_encoder_text_model_encoder_layers_2_mlp_fc1_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_2_mlp_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(90154304)))]; + tensor linear_16_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_2_mlp_fc1_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_2_mlp_fc1_weight_to_fp16_palettized, x = input_45_cast_fp16)[name = tensor("linear_16_cast_fp16")]; + tensor var_350_to_fp16 = const()[name = tensor("op_350_to_fp16"), val = tensor(0x1.b3cp+0)]; + tensor var_351_cast_fp16 = mul(x = linear_16_cast_fp16, y = var_350_to_fp16)[name = tensor("op_351_cast_fp16")]; + tensor var_352_cast_fp16 = sigmoid(x = var_351_cast_fp16)[name = tensor("op_352_cast_fp16")]; + tensor input_49_cast_fp16 = mul(x = linear_16_cast_fp16, y = var_352_cast_fp16)[name = tensor("input_49_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_2_mlp_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(90160512))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(91930048))), name = tensor("text_encoder_text_model_encoder_layers_2_mlp_fc2_weight_to_fp16_palettized"), shape = tensor([768, 3072])]; + tensor text_encoder_text_model_encoder_layers_2_mlp_fc2_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_2_mlp_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(91930240)))]; + tensor linear_17_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_2_mlp_fc2_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_2_mlp_fc2_weight_to_fp16_palettized, x = input_49_cast_fp16)[name = tensor("linear_17_cast_fp16")]; + tensor input_51_cast_fp16 = add(x = input_43_cast_fp16, y = linear_17_cast_fp16)[name = tensor("input_51_cast_fp16")]; + tensor hidden_states_19_axes_0 = const()[name = tensor("hidden_states_19_axes_0"), val = tensor([-1])]; + tensor text_encoder_text_model_encoder_layers_3_layer_norm1_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_3_layer_norm1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(91931840)))]; + tensor text_encoder_text_model_encoder_layers_3_layer_norm1_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_3_layer_norm1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(91933440)))]; + tensor hidden_states_19_cast_fp16 = layer_norm(axes = hidden_states_19_axes_0, beta = text_encoder_text_model_encoder_layers_3_layer_norm1_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_3_layer_norm1_weight_to_fp16, x = input_51_cast_fp16)[name = tensor("hidden_states_19_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_3_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(91935040))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(92377472))), name = tensor("text_encoder_text_model_encoder_layers_3_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor([768, 768])]; + tensor text_encoder_text_model_encoder_layers_3_self_attn_q_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_3_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(92377664)))]; + tensor linear_18_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_3_self_attn_q_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_3_self_attn_q_proj_weight_to_fp16_palettized, x = hidden_states_19_cast_fp16)[name = tensor("linear_18_cast_fp16")]; + tensor var_377_to_fp16 = const()[name = tensor("op_377_to_fp16"), val = tensor(0x1p-3)]; + tensor tensor_23_cast_fp16 = mul(x = linear_18_cast_fp16, y = var_377_to_fp16)[name = tensor("tensor_23_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_3_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(92379264))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(92821696))), name = tensor("text_encoder_text_model_encoder_layers_3_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor([768, 768])]; + tensor text_encoder_text_model_encoder_layers_3_self_attn_k_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_3_self_attn_k_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(92821888)))]; + tensor linear_19_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_3_self_attn_k_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_3_self_attn_k_proj_weight_to_fp16_palettized, x = hidden_states_19_cast_fp16)[name = tensor("linear_19_cast_fp16")]; + tensor var_382 = const()[name = tensor("op_382"), val = tensor([1, -1, 12, 64])]; + tensor var_383_cast_fp16 = reshape(shape = var_382, x = linear_19_cast_fp16)[name = tensor("op_383_cast_fp16")]; + tensor var_384_perm_0 = const()[name = tensor("op_384_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor text_encoder_text_model_encoder_layers_3_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(92823488))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(93265920))), name = tensor("text_encoder_text_model_encoder_layers_3_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor([768, 768])]; + tensor text_encoder_text_model_encoder_layers_3_self_attn_v_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_3_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(93266112)))]; + tensor linear_20_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_3_self_attn_v_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_3_self_attn_v_proj_weight_to_fp16_palettized, x = hidden_states_19_cast_fp16)[name = tensor("linear_20_cast_fp16")]; + tensor var_389 = const()[name = tensor("op_389"), val = tensor([1, -1, 12, 64])]; + tensor var_390_cast_fp16 = reshape(shape = var_389, x = linear_20_cast_fp16)[name = tensor("op_390_cast_fp16")]; + tensor var_391_perm_0 = const()[name = tensor("op_391_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor var_398 = const()[name = tensor("op_398"), val = tensor([1, 77, 12, 64])]; + tensor var_399_cast_fp16 = reshape(shape = var_398, x = tensor_23_cast_fp16)[name = tensor("op_399_cast_fp16")]; + tensor var_400_perm_0 = const()[name = tensor("op_400_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor var_402 = const()[name = tensor("op_402"), val = tensor([12, -1, 64])]; + tensor var_400_cast_fp16 = transpose(perm = var_400_perm_0, x = var_399_cast_fp16)[name = tensor("transpose_35")]; + tensor query_states_7_cast_fp16 = reshape(shape = var_402, x = var_400_cast_fp16)[name = tensor("query_states_7_cast_fp16")]; + tensor var_404 = const()[name = tensor("op_404"), val = tensor([12, -1, 64])]; + tensor var_384_cast_fp16 = transpose(perm = var_384_perm_0, x = var_383_cast_fp16)[name = tensor("transpose_34")]; + tensor key_states_15_cast_fp16 = reshape(shape = var_404, x = var_384_cast_fp16)[name = tensor("key_states_15_cast_fp16")]; + tensor var_406 = const()[name = tensor("op_406"), val = tensor([12, -1, 64])]; + tensor var_391_cast_fp16 = transpose(perm = var_391_perm_0, x = var_390_cast_fp16)[name = tensor("transpose_33")]; + tensor value_states_15_cast_fp16 = reshape(shape = var_406, x = var_391_cast_fp16)[name = tensor("value_states_15_cast_fp16")]; + tensor attn_weights_19_transpose_x_1 = const()[name = tensor("attn_weights_19_transpose_x_1"), val = tensor(false)]; + tensor attn_weights_19_transpose_y_1 = const()[name = tensor("attn_weights_19_transpose_y_1"), val = tensor(true)]; + tensor attn_weights_19_cast_fp16 = matmul(transpose_x = attn_weights_19_transpose_x_1, transpose_y = attn_weights_19_transpose_y_1, x = query_states_7_cast_fp16, y = key_states_15_cast_fp16)[name = tensor("attn_weights_19_cast_fp16")]; + tensor var_411 = const()[name = tensor("op_411"), val = tensor([1, 12, 77, 77])]; + tensor var_412_cast_fp16 = reshape(shape = var_411, x = attn_weights_19_cast_fp16)[name = tensor("op_412_cast_fp16")]; + tensor attn_weights_21_cast_fp16 = add(x = var_412_cast_fp16, y = op_56_to_fp16_palettized)[name = tensor("attn_weights_21_cast_fp16")]; + tensor var_417 = const()[name = tensor("op_417"), val = tensor([12, 77, 77])]; + tensor input_53_cast_fp16 = reshape(shape = var_417, x = attn_weights_21_cast_fp16)[name = tensor("input_53_cast_fp16")]; + tensor input_55_cast_fp16 = softmax(axis = var_5, x = input_53_cast_fp16)[name = tensor("input_55_cast_fp16")]; + tensor attn_output_19_transpose_x_0 = const()[name = tensor("attn_output_19_transpose_x_0"), val = tensor(false)]; + tensor attn_output_19_transpose_y_0 = const()[name = tensor("attn_output_19_transpose_y_0"), val = tensor(false)]; + tensor attn_output_19_cast_fp16 = matmul(transpose_x = attn_output_19_transpose_x_0, transpose_y = attn_output_19_transpose_y_0, x = input_55_cast_fp16, y = value_states_15_cast_fp16)[name = tensor("attn_output_19_cast_fp16")]; + tensor var_422 = const()[name = tensor("op_422"), val = tensor([1, 12, 77, 64])]; + tensor attn_output_21_cast_fp16 = reshape(shape = var_422, x = attn_output_19_cast_fp16)[name = tensor("attn_output_21_cast_fp16")]; + tensor attn_output_23_perm_0 = const()[name = tensor("attn_output_23_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor var_425 = const()[name = tensor("op_425"), val = tensor([1, 77, 768])]; + tensor attn_output_23_cast_fp16 = transpose(perm = attn_output_23_perm_0, x = attn_output_21_cast_fp16)[name = tensor("transpose_32")]; + tensor input_57_cast_fp16 = reshape(shape = var_425, x = attn_output_23_cast_fp16)[name = tensor("input_57_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_3_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(93267712))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(93710144))), name = tensor("text_encoder_text_model_encoder_layers_3_self_attn_out_proj_weight_to_fp16_palettized"), shape = tensor([768, 768])]; + tensor text_encoder_text_model_encoder_layers_3_self_attn_out_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_3_self_attn_out_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(93710336)))]; + tensor linear_21_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_3_self_attn_out_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_3_self_attn_out_proj_weight_to_fp16_palettized, x = input_57_cast_fp16)[name = tensor("linear_21_cast_fp16")]; + tensor input_59_cast_fp16 = add(x = input_51_cast_fp16, y = linear_21_cast_fp16)[name = tensor("input_59_cast_fp16")]; + tensor input_61_axes_0 = const()[name = tensor("input_61_axes_0"), val = tensor([-1])]; + tensor text_encoder_text_model_encoder_layers_3_layer_norm2_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_3_layer_norm2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(93711936)))]; + tensor text_encoder_text_model_encoder_layers_3_layer_norm2_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_3_layer_norm2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(93713536)))]; + tensor input_61_cast_fp16 = layer_norm(axes = input_61_axes_0, beta = text_encoder_text_model_encoder_layers_3_layer_norm2_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_3_layer_norm2_weight_to_fp16, x = input_59_cast_fp16)[name = tensor("input_61_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_3_mlp_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(93715136))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(95484672))), name = tensor("text_encoder_text_model_encoder_layers_3_mlp_fc1_weight_to_fp16_palettized"), shape = tensor([3072, 768])]; + tensor text_encoder_text_model_encoder_layers_3_mlp_fc1_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_3_mlp_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(95484864)))]; + tensor linear_22_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_3_mlp_fc1_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_3_mlp_fc1_weight_to_fp16_palettized, x = input_61_cast_fp16)[name = tensor("linear_22_cast_fp16")]; + tensor var_440_to_fp16 = const()[name = tensor("op_440_to_fp16"), val = tensor(0x1.b3cp+0)]; + tensor var_441_cast_fp16 = mul(x = linear_22_cast_fp16, y = var_440_to_fp16)[name = tensor("op_441_cast_fp16")]; + tensor var_442_cast_fp16 = sigmoid(x = var_441_cast_fp16)[name = tensor("op_442_cast_fp16")]; + tensor input_65_cast_fp16 = mul(x = linear_22_cast_fp16, y = var_442_cast_fp16)[name = tensor("input_65_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_3_mlp_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(95491072))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(97260608))), name = tensor("text_encoder_text_model_encoder_layers_3_mlp_fc2_weight_to_fp16_palettized"), shape = tensor([768, 3072])]; + tensor text_encoder_text_model_encoder_layers_3_mlp_fc2_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_3_mlp_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(97260800)))]; + tensor linear_23_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_3_mlp_fc2_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_3_mlp_fc2_weight_to_fp16_palettized, x = input_65_cast_fp16)[name = tensor("linear_23_cast_fp16")]; + tensor input_67_cast_fp16 = add(x = input_59_cast_fp16, y = linear_23_cast_fp16)[name = tensor("input_67_cast_fp16")]; + tensor hidden_states_25_axes_0 = const()[name = tensor("hidden_states_25_axes_0"), val = tensor([-1])]; + tensor text_encoder_text_model_encoder_layers_4_layer_norm1_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_4_layer_norm1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(97262400)))]; + tensor text_encoder_text_model_encoder_layers_4_layer_norm1_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_4_layer_norm1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(97264000)))]; + tensor hidden_states_25_cast_fp16 = layer_norm(axes = hidden_states_25_axes_0, beta = text_encoder_text_model_encoder_layers_4_layer_norm1_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_4_layer_norm1_weight_to_fp16, x = input_67_cast_fp16)[name = tensor("hidden_states_25_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_4_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(97265600))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(97708032))), name = tensor("text_encoder_text_model_encoder_layers_4_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor([768, 768])]; + tensor text_encoder_text_model_encoder_layers_4_self_attn_q_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_4_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(97708224)))]; + tensor linear_24_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_4_self_attn_q_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_4_self_attn_q_proj_weight_to_fp16_palettized, x = hidden_states_25_cast_fp16)[name = tensor("linear_24_cast_fp16")]; + tensor var_467_to_fp16 = const()[name = tensor("op_467_to_fp16"), val = tensor(0x1p-3)]; + tensor tensor_29_cast_fp16 = mul(x = linear_24_cast_fp16, y = var_467_to_fp16)[name = tensor("tensor_29_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_4_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(97709824))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(98152256))), name = tensor("text_encoder_text_model_encoder_layers_4_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor([768, 768])]; + tensor text_encoder_text_model_encoder_layers_4_self_attn_k_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_4_self_attn_k_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(98152448)))]; + tensor linear_25_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_4_self_attn_k_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_4_self_attn_k_proj_weight_to_fp16_palettized, x = hidden_states_25_cast_fp16)[name = tensor("linear_25_cast_fp16")]; + tensor var_472 = const()[name = tensor("op_472"), val = tensor([1, -1, 12, 64])]; + tensor var_473_cast_fp16 = reshape(shape = var_472, x = linear_25_cast_fp16)[name = tensor("op_473_cast_fp16")]; + tensor var_474_perm_0 = const()[name = tensor("op_474_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor text_encoder_text_model_encoder_layers_4_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(98154048))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(98596480))), name = tensor("text_encoder_text_model_encoder_layers_4_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor([768, 768])]; + tensor text_encoder_text_model_encoder_layers_4_self_attn_v_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_4_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(98596672)))]; + tensor linear_26_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_4_self_attn_v_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_4_self_attn_v_proj_weight_to_fp16_palettized, x = hidden_states_25_cast_fp16)[name = tensor("linear_26_cast_fp16")]; + tensor var_479 = const()[name = tensor("op_479"), val = tensor([1, -1, 12, 64])]; + tensor var_480_cast_fp16 = reshape(shape = var_479, x = linear_26_cast_fp16)[name = tensor("op_480_cast_fp16")]; + tensor var_481_perm_0 = const()[name = tensor("op_481_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor var_488 = const()[name = tensor("op_488"), val = tensor([1, 77, 12, 64])]; + tensor var_489_cast_fp16 = reshape(shape = var_488, x = tensor_29_cast_fp16)[name = tensor("op_489_cast_fp16")]; + tensor var_490_perm_0 = const()[name = tensor("op_490_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor var_492 = const()[name = tensor("op_492"), val = tensor([12, -1, 64])]; + tensor var_490_cast_fp16 = transpose(perm = var_490_perm_0, x = var_489_cast_fp16)[name = tensor("transpose_31")]; + tensor query_states_9_cast_fp16 = reshape(shape = var_492, x = var_490_cast_fp16)[name = tensor("query_states_9_cast_fp16")]; + tensor var_494 = const()[name = tensor("op_494"), val = tensor([12, -1, 64])]; + tensor var_474_cast_fp16 = transpose(perm = var_474_perm_0, x = var_473_cast_fp16)[name = tensor("transpose_30")]; + tensor key_states_19_cast_fp16 = reshape(shape = var_494, x = var_474_cast_fp16)[name = tensor("key_states_19_cast_fp16")]; + tensor var_496 = const()[name = tensor("op_496"), val = tensor([12, -1, 64])]; + tensor var_481_cast_fp16 = transpose(perm = var_481_perm_0, x = var_480_cast_fp16)[name = tensor("transpose_29")]; + tensor value_states_19_cast_fp16 = reshape(shape = var_496, x = var_481_cast_fp16)[name = tensor("value_states_19_cast_fp16")]; + tensor attn_weights_25_transpose_x_1 = const()[name = tensor("attn_weights_25_transpose_x_1"), val = tensor(false)]; + tensor attn_weights_25_transpose_y_1 = const()[name = tensor("attn_weights_25_transpose_y_1"), val = tensor(true)]; + tensor attn_weights_25_cast_fp16 = matmul(transpose_x = attn_weights_25_transpose_x_1, transpose_y = attn_weights_25_transpose_y_1, x = query_states_9_cast_fp16, y = key_states_19_cast_fp16)[name = tensor("attn_weights_25_cast_fp16")]; + tensor var_501 = const()[name = tensor("op_501"), val = tensor([1, 12, 77, 77])]; + tensor var_502_cast_fp16 = reshape(shape = var_501, x = attn_weights_25_cast_fp16)[name = tensor("op_502_cast_fp16")]; + tensor attn_weights_27_cast_fp16 = add(x = var_502_cast_fp16, y = op_56_to_fp16_palettized)[name = tensor("attn_weights_27_cast_fp16")]; + tensor var_507 = const()[name = tensor("op_507"), val = tensor([12, 77, 77])]; + tensor input_69_cast_fp16 = reshape(shape = var_507, x = attn_weights_27_cast_fp16)[name = tensor("input_69_cast_fp16")]; + tensor input_71_cast_fp16 = softmax(axis = var_5, x = input_69_cast_fp16)[name = tensor("input_71_cast_fp16")]; + tensor attn_output_25_transpose_x_0 = const()[name = tensor("attn_output_25_transpose_x_0"), val = tensor(false)]; + tensor attn_output_25_transpose_y_0 = const()[name = tensor("attn_output_25_transpose_y_0"), val = tensor(false)]; + tensor attn_output_25_cast_fp16 = matmul(transpose_x = attn_output_25_transpose_x_0, transpose_y = attn_output_25_transpose_y_0, x = input_71_cast_fp16, y = value_states_19_cast_fp16)[name = tensor("attn_output_25_cast_fp16")]; + tensor var_512 = const()[name = tensor("op_512"), val = tensor([1, 12, 77, 64])]; + tensor attn_output_27_cast_fp16 = reshape(shape = var_512, x = attn_output_25_cast_fp16)[name = tensor("attn_output_27_cast_fp16")]; + tensor attn_output_29_perm_0 = const()[name = tensor("attn_output_29_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor var_515 = const()[name = tensor("op_515"), val = tensor([1, 77, 768])]; + tensor attn_output_29_cast_fp16 = transpose(perm = attn_output_29_perm_0, x = attn_output_27_cast_fp16)[name = tensor("transpose_28")]; + tensor input_73_cast_fp16 = reshape(shape = var_515, x = attn_output_29_cast_fp16)[name = tensor("input_73_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_4_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(98598272))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(99040704))), name = tensor("text_encoder_text_model_encoder_layers_4_self_attn_out_proj_weight_to_fp16_palettized"), shape = tensor([768, 768])]; + tensor text_encoder_text_model_encoder_layers_4_self_attn_out_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_4_self_attn_out_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(99040896)))]; + tensor linear_27_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_4_self_attn_out_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_4_self_attn_out_proj_weight_to_fp16_palettized, x = input_73_cast_fp16)[name = tensor("linear_27_cast_fp16")]; + tensor input_75_cast_fp16 = add(x = input_67_cast_fp16, y = linear_27_cast_fp16)[name = tensor("input_75_cast_fp16")]; + tensor input_77_axes_0 = const()[name = tensor("input_77_axes_0"), val = tensor([-1])]; + tensor text_encoder_text_model_encoder_layers_4_layer_norm2_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_4_layer_norm2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(99042496)))]; + tensor text_encoder_text_model_encoder_layers_4_layer_norm2_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_4_layer_norm2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(99044096)))]; + tensor input_77_cast_fp16 = layer_norm(axes = input_77_axes_0, beta = text_encoder_text_model_encoder_layers_4_layer_norm2_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_4_layer_norm2_weight_to_fp16, x = input_75_cast_fp16)[name = tensor("input_77_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_4_mlp_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(99045696))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(100815232))), name = tensor("text_encoder_text_model_encoder_layers_4_mlp_fc1_weight_to_fp16_palettized"), shape = tensor([3072, 768])]; + tensor text_encoder_text_model_encoder_layers_4_mlp_fc1_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_4_mlp_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(100815424)))]; + tensor linear_28_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_4_mlp_fc1_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_4_mlp_fc1_weight_to_fp16_palettized, x = input_77_cast_fp16)[name = tensor("linear_28_cast_fp16")]; + tensor var_530_to_fp16 = const()[name = tensor("op_530_to_fp16"), val = tensor(0x1.b3cp+0)]; + tensor var_531_cast_fp16 = mul(x = linear_28_cast_fp16, y = var_530_to_fp16)[name = tensor("op_531_cast_fp16")]; + tensor var_532_cast_fp16 = sigmoid(x = var_531_cast_fp16)[name = tensor("op_532_cast_fp16")]; + tensor input_81_cast_fp16 = mul(x = linear_28_cast_fp16, y = var_532_cast_fp16)[name = tensor("input_81_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_4_mlp_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(100821632))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(102591168))), name = tensor("text_encoder_text_model_encoder_layers_4_mlp_fc2_weight_to_fp16_palettized"), shape = tensor([768, 3072])]; + tensor text_encoder_text_model_encoder_layers_4_mlp_fc2_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_4_mlp_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(102591360)))]; + tensor linear_29_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_4_mlp_fc2_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_4_mlp_fc2_weight_to_fp16_palettized, x = input_81_cast_fp16)[name = tensor("linear_29_cast_fp16")]; + tensor input_83_cast_fp16 = add(x = input_75_cast_fp16, y = linear_29_cast_fp16)[name = tensor("input_83_cast_fp16")]; + tensor hidden_states_31_axes_0 = const()[name = tensor("hidden_states_31_axes_0"), val = tensor([-1])]; + tensor text_encoder_text_model_encoder_layers_5_layer_norm1_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_5_layer_norm1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(102592960)))]; + tensor text_encoder_text_model_encoder_layers_5_layer_norm1_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_5_layer_norm1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(102594560)))]; + tensor hidden_states_31_cast_fp16 = layer_norm(axes = hidden_states_31_axes_0, beta = text_encoder_text_model_encoder_layers_5_layer_norm1_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_5_layer_norm1_weight_to_fp16, x = input_83_cast_fp16)[name = tensor("hidden_states_31_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_5_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(102596160))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(103038592))), name = tensor("text_encoder_text_model_encoder_layers_5_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor([768, 768])]; + tensor text_encoder_text_model_encoder_layers_5_self_attn_q_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_5_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(103038784)))]; + tensor linear_30_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_5_self_attn_q_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_5_self_attn_q_proj_weight_to_fp16_palettized, x = hidden_states_31_cast_fp16)[name = tensor("linear_30_cast_fp16")]; + tensor var_557_to_fp16 = const()[name = tensor("op_557_to_fp16"), val = tensor(0x1p-3)]; + tensor tensor_35_cast_fp16 = mul(x = linear_30_cast_fp16, y = var_557_to_fp16)[name = tensor("tensor_35_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_5_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(103040384))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(103482816))), name = tensor("text_encoder_text_model_encoder_layers_5_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor([768, 768])]; + tensor text_encoder_text_model_encoder_layers_5_self_attn_k_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_5_self_attn_k_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(103483008)))]; + tensor linear_31_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_5_self_attn_k_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_5_self_attn_k_proj_weight_to_fp16_palettized, x = hidden_states_31_cast_fp16)[name = tensor("linear_31_cast_fp16")]; + tensor var_562 = const()[name = tensor("op_562"), val = tensor([1, -1, 12, 64])]; + tensor var_563_cast_fp16 = reshape(shape = var_562, x = linear_31_cast_fp16)[name = tensor("op_563_cast_fp16")]; + tensor var_564_perm_0 = const()[name = tensor("op_564_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor text_encoder_text_model_encoder_layers_5_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(103484608))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(103927040))), name = tensor("text_encoder_text_model_encoder_layers_5_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor([768, 768])]; + tensor text_encoder_text_model_encoder_layers_5_self_attn_v_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_5_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(103927232)))]; + tensor linear_32_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_5_self_attn_v_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_5_self_attn_v_proj_weight_to_fp16_palettized, x = hidden_states_31_cast_fp16)[name = tensor("linear_32_cast_fp16")]; + tensor var_569 = const()[name = tensor("op_569"), val = tensor([1, -1, 12, 64])]; + tensor var_570_cast_fp16 = reshape(shape = var_569, x = linear_32_cast_fp16)[name = tensor("op_570_cast_fp16")]; + tensor var_571_perm_0 = const()[name = tensor("op_571_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor var_578 = const()[name = tensor("op_578"), val = tensor([1, 77, 12, 64])]; + tensor var_579_cast_fp16 = reshape(shape = var_578, x = tensor_35_cast_fp16)[name = tensor("op_579_cast_fp16")]; + tensor var_580_perm_0 = const()[name = tensor("op_580_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor var_582 = const()[name = tensor("op_582"), val = tensor([12, -1, 64])]; + tensor var_580_cast_fp16 = transpose(perm = var_580_perm_0, x = var_579_cast_fp16)[name = tensor("transpose_27")]; + tensor query_states_11_cast_fp16 = reshape(shape = var_582, x = var_580_cast_fp16)[name = tensor("query_states_11_cast_fp16")]; + tensor var_584 = const()[name = tensor("op_584"), val = tensor([12, -1, 64])]; + tensor var_564_cast_fp16 = transpose(perm = var_564_perm_0, x = var_563_cast_fp16)[name = tensor("transpose_26")]; + tensor key_states_23_cast_fp16 = reshape(shape = var_584, x = var_564_cast_fp16)[name = tensor("key_states_23_cast_fp16")]; + tensor var_586 = const()[name = tensor("op_586"), val = tensor([12, -1, 64])]; + tensor var_571_cast_fp16 = transpose(perm = var_571_perm_0, x = var_570_cast_fp16)[name = tensor("transpose_25")]; + tensor value_states_23_cast_fp16 = reshape(shape = var_586, x = var_571_cast_fp16)[name = tensor("value_states_23_cast_fp16")]; + tensor attn_weights_31_transpose_x_1 = const()[name = tensor("attn_weights_31_transpose_x_1"), val = tensor(false)]; + tensor attn_weights_31_transpose_y_1 = const()[name = tensor("attn_weights_31_transpose_y_1"), val = tensor(true)]; + tensor attn_weights_31_cast_fp16 = matmul(transpose_x = attn_weights_31_transpose_x_1, transpose_y = attn_weights_31_transpose_y_1, x = query_states_11_cast_fp16, y = key_states_23_cast_fp16)[name = tensor("attn_weights_31_cast_fp16")]; + tensor var_591 = const()[name = tensor("op_591"), val = tensor([1, 12, 77, 77])]; + tensor var_592_cast_fp16 = reshape(shape = var_591, x = attn_weights_31_cast_fp16)[name = tensor("op_592_cast_fp16")]; + tensor attn_weights_33_cast_fp16 = add(x = var_592_cast_fp16, y = op_56_to_fp16_palettized)[name = tensor("attn_weights_33_cast_fp16")]; + tensor var_597 = const()[name = tensor("op_597"), val = tensor([12, 77, 77])]; + tensor input_85_cast_fp16 = reshape(shape = var_597, x = attn_weights_33_cast_fp16)[name = tensor("input_85_cast_fp16")]; + tensor input_87_cast_fp16 = softmax(axis = var_5, x = input_85_cast_fp16)[name = tensor("input_87_cast_fp16")]; + tensor attn_output_31_transpose_x_0 = const()[name = tensor("attn_output_31_transpose_x_0"), val = tensor(false)]; + tensor attn_output_31_transpose_y_0 = const()[name = tensor("attn_output_31_transpose_y_0"), val = tensor(false)]; + tensor attn_output_31_cast_fp16 = matmul(transpose_x = attn_output_31_transpose_x_0, transpose_y = attn_output_31_transpose_y_0, x = input_87_cast_fp16, y = value_states_23_cast_fp16)[name = tensor("attn_output_31_cast_fp16")]; + tensor var_602 = const()[name = tensor("op_602"), val = tensor([1, 12, 77, 64])]; + tensor attn_output_33_cast_fp16 = reshape(shape = var_602, x = attn_output_31_cast_fp16)[name = tensor("attn_output_33_cast_fp16")]; + tensor attn_output_35_perm_0 = const()[name = tensor("attn_output_35_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor var_605 = const()[name = tensor("op_605"), val = tensor([1, 77, 768])]; + tensor attn_output_35_cast_fp16 = transpose(perm = attn_output_35_perm_0, x = attn_output_33_cast_fp16)[name = tensor("transpose_24")]; + tensor input_89_cast_fp16 = reshape(shape = var_605, x = attn_output_35_cast_fp16)[name = tensor("input_89_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_5_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(103928832))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(104371264))), name = tensor("text_encoder_text_model_encoder_layers_5_self_attn_out_proj_weight_to_fp16_palettized"), shape = tensor([768, 768])]; + tensor text_encoder_text_model_encoder_layers_5_self_attn_out_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_5_self_attn_out_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(104371456)))]; + tensor linear_33_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_5_self_attn_out_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_5_self_attn_out_proj_weight_to_fp16_palettized, x = input_89_cast_fp16)[name = tensor("linear_33_cast_fp16")]; + tensor input_91_cast_fp16 = add(x = input_83_cast_fp16, y = linear_33_cast_fp16)[name = tensor("input_91_cast_fp16")]; + tensor input_93_axes_0 = const()[name = tensor("input_93_axes_0"), val = tensor([-1])]; + tensor text_encoder_text_model_encoder_layers_5_layer_norm2_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_5_layer_norm2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(104373056)))]; + tensor text_encoder_text_model_encoder_layers_5_layer_norm2_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_5_layer_norm2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(104374656)))]; + tensor input_93_cast_fp16 = layer_norm(axes = input_93_axes_0, beta = text_encoder_text_model_encoder_layers_5_layer_norm2_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_5_layer_norm2_weight_to_fp16, x = input_91_cast_fp16)[name = tensor("input_93_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_5_mlp_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(104376256))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(106145792))), name = tensor("text_encoder_text_model_encoder_layers_5_mlp_fc1_weight_to_fp16_palettized"), shape = tensor([3072, 768])]; + tensor text_encoder_text_model_encoder_layers_5_mlp_fc1_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_5_mlp_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(106145984)))]; + tensor linear_34_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_5_mlp_fc1_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_5_mlp_fc1_weight_to_fp16_palettized, x = input_93_cast_fp16)[name = tensor("linear_34_cast_fp16")]; + tensor var_620_to_fp16 = const()[name = tensor("op_620_to_fp16"), val = tensor(0x1.b3cp+0)]; + tensor var_621_cast_fp16 = mul(x = linear_34_cast_fp16, y = var_620_to_fp16)[name = tensor("op_621_cast_fp16")]; + tensor var_622_cast_fp16 = sigmoid(x = var_621_cast_fp16)[name = tensor("op_622_cast_fp16")]; + tensor input_97_cast_fp16 = mul(x = linear_34_cast_fp16, y = var_622_cast_fp16)[name = tensor("input_97_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_5_mlp_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(106152192))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(107921728))), name = tensor("text_encoder_text_model_encoder_layers_5_mlp_fc2_weight_to_fp16_palettized"), shape = tensor([768, 3072])]; + tensor text_encoder_text_model_encoder_layers_5_mlp_fc2_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_5_mlp_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(107921920)))]; + tensor linear_35_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_5_mlp_fc2_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_5_mlp_fc2_weight_to_fp16_palettized, x = input_97_cast_fp16)[name = tensor("linear_35_cast_fp16")]; + tensor input_99_cast_fp16 = add(x = input_91_cast_fp16, y = linear_35_cast_fp16)[name = tensor("input_99_cast_fp16")]; + tensor hidden_states_37_axes_0 = const()[name = tensor("hidden_states_37_axes_0"), val = tensor([-1])]; + tensor text_encoder_text_model_encoder_layers_6_layer_norm1_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_6_layer_norm1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(107923520)))]; + tensor text_encoder_text_model_encoder_layers_6_layer_norm1_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_6_layer_norm1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(107925120)))]; + tensor hidden_states_37_cast_fp16 = layer_norm(axes = hidden_states_37_axes_0, beta = text_encoder_text_model_encoder_layers_6_layer_norm1_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_6_layer_norm1_weight_to_fp16, x = input_99_cast_fp16)[name = tensor("hidden_states_37_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_6_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(107926720))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(108369152))), name = tensor("text_encoder_text_model_encoder_layers_6_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor([768, 768])]; + tensor text_encoder_text_model_encoder_layers_6_self_attn_q_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_6_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(108369344)))]; + tensor linear_36_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_6_self_attn_q_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_6_self_attn_q_proj_weight_to_fp16_palettized, x = hidden_states_37_cast_fp16)[name = tensor("linear_36_cast_fp16")]; + tensor var_647_to_fp16 = const()[name = tensor("op_647_to_fp16"), val = tensor(0x1p-3)]; + tensor tensor_41_cast_fp16 = mul(x = linear_36_cast_fp16, y = var_647_to_fp16)[name = tensor("tensor_41_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_6_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(108370944))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(108813376))), name = tensor("text_encoder_text_model_encoder_layers_6_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor([768, 768])]; + tensor text_encoder_text_model_encoder_layers_6_self_attn_k_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_6_self_attn_k_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(108813568)))]; + tensor linear_37_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_6_self_attn_k_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_6_self_attn_k_proj_weight_to_fp16_palettized, x = hidden_states_37_cast_fp16)[name = tensor("linear_37_cast_fp16")]; + tensor var_652 = const()[name = tensor("op_652"), val = tensor([1, -1, 12, 64])]; + tensor var_653_cast_fp16 = reshape(shape = var_652, x = linear_37_cast_fp16)[name = tensor("op_653_cast_fp16")]; + tensor var_654_perm_0 = const()[name = tensor("op_654_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor text_encoder_text_model_encoder_layers_6_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(108815168))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(109257600))), name = tensor("text_encoder_text_model_encoder_layers_6_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor([768, 768])]; + tensor text_encoder_text_model_encoder_layers_6_self_attn_v_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_6_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(109257792)))]; + tensor linear_38_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_6_self_attn_v_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_6_self_attn_v_proj_weight_to_fp16_palettized, x = hidden_states_37_cast_fp16)[name = tensor("linear_38_cast_fp16")]; + tensor var_659 = const()[name = tensor("op_659"), val = tensor([1, -1, 12, 64])]; + tensor var_660_cast_fp16 = reshape(shape = var_659, x = linear_38_cast_fp16)[name = tensor("op_660_cast_fp16")]; + tensor var_661_perm_0 = const()[name = tensor("op_661_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor var_668 = const()[name = tensor("op_668"), val = tensor([1, 77, 12, 64])]; + tensor var_669_cast_fp16 = reshape(shape = var_668, x = tensor_41_cast_fp16)[name = tensor("op_669_cast_fp16")]; + tensor var_670_perm_0 = const()[name = tensor("op_670_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor var_672 = const()[name = tensor("op_672"), val = tensor([12, -1, 64])]; + tensor var_670_cast_fp16 = transpose(perm = var_670_perm_0, x = var_669_cast_fp16)[name = tensor("transpose_23")]; + tensor query_states_13_cast_fp16 = reshape(shape = var_672, x = var_670_cast_fp16)[name = tensor("query_states_13_cast_fp16")]; + tensor var_674 = const()[name = tensor("op_674"), val = tensor([12, -1, 64])]; + tensor var_654_cast_fp16 = transpose(perm = var_654_perm_0, x = var_653_cast_fp16)[name = tensor("transpose_22")]; + tensor key_states_27_cast_fp16 = reshape(shape = var_674, x = var_654_cast_fp16)[name = tensor("key_states_27_cast_fp16")]; + tensor var_676 = const()[name = tensor("op_676"), val = tensor([12, -1, 64])]; + tensor var_661_cast_fp16 = transpose(perm = var_661_perm_0, x = var_660_cast_fp16)[name = tensor("transpose_21")]; + tensor value_states_27_cast_fp16 = reshape(shape = var_676, x = var_661_cast_fp16)[name = tensor("value_states_27_cast_fp16")]; + tensor attn_weights_37_transpose_x_1 = const()[name = tensor("attn_weights_37_transpose_x_1"), val = tensor(false)]; + tensor attn_weights_37_transpose_y_1 = const()[name = tensor("attn_weights_37_transpose_y_1"), val = tensor(true)]; + tensor attn_weights_37_cast_fp16 = matmul(transpose_x = attn_weights_37_transpose_x_1, transpose_y = attn_weights_37_transpose_y_1, x = query_states_13_cast_fp16, y = key_states_27_cast_fp16)[name = tensor("attn_weights_37_cast_fp16")]; + tensor var_681 = const()[name = tensor("op_681"), val = tensor([1, 12, 77, 77])]; + tensor var_682_cast_fp16 = reshape(shape = var_681, x = attn_weights_37_cast_fp16)[name = tensor("op_682_cast_fp16")]; + tensor attn_weights_39_cast_fp16 = add(x = var_682_cast_fp16, y = op_56_to_fp16_palettized)[name = tensor("attn_weights_39_cast_fp16")]; + tensor var_687 = const()[name = tensor("op_687"), val = tensor([12, 77, 77])]; + tensor input_101_cast_fp16 = reshape(shape = var_687, x = attn_weights_39_cast_fp16)[name = tensor("input_101_cast_fp16")]; + tensor input_103_cast_fp16 = softmax(axis = var_5, x = input_101_cast_fp16)[name = tensor("input_103_cast_fp16")]; + tensor attn_output_37_transpose_x_0 = const()[name = tensor("attn_output_37_transpose_x_0"), val = tensor(false)]; + tensor attn_output_37_transpose_y_0 = const()[name = tensor("attn_output_37_transpose_y_0"), val = tensor(false)]; + tensor attn_output_37_cast_fp16 = matmul(transpose_x = attn_output_37_transpose_x_0, transpose_y = attn_output_37_transpose_y_0, x = input_103_cast_fp16, y = value_states_27_cast_fp16)[name = tensor("attn_output_37_cast_fp16")]; + tensor var_692 = const()[name = tensor("op_692"), val = tensor([1, 12, 77, 64])]; + tensor attn_output_39_cast_fp16 = reshape(shape = var_692, x = attn_output_37_cast_fp16)[name = tensor("attn_output_39_cast_fp16")]; + tensor attn_output_41_perm_0 = const()[name = tensor("attn_output_41_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor var_695 = const()[name = tensor("op_695"), val = tensor([1, 77, 768])]; + tensor attn_output_41_cast_fp16 = transpose(perm = attn_output_41_perm_0, x = attn_output_39_cast_fp16)[name = tensor("transpose_20")]; + tensor input_105_cast_fp16 = reshape(shape = var_695, x = attn_output_41_cast_fp16)[name = tensor("input_105_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_6_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(109259392))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(109701824))), name = tensor("text_encoder_text_model_encoder_layers_6_self_attn_out_proj_weight_to_fp16_palettized"), shape = tensor([768, 768])]; + tensor text_encoder_text_model_encoder_layers_6_self_attn_out_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_6_self_attn_out_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(109702016)))]; + tensor linear_39_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_6_self_attn_out_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_6_self_attn_out_proj_weight_to_fp16_palettized, x = input_105_cast_fp16)[name = tensor("linear_39_cast_fp16")]; + tensor input_107_cast_fp16 = add(x = input_99_cast_fp16, y = linear_39_cast_fp16)[name = tensor("input_107_cast_fp16")]; + tensor input_109_axes_0 = const()[name = tensor("input_109_axes_0"), val = tensor([-1])]; + tensor text_encoder_text_model_encoder_layers_6_layer_norm2_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_6_layer_norm2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(109703616)))]; + tensor text_encoder_text_model_encoder_layers_6_layer_norm2_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_6_layer_norm2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(109705216)))]; + tensor input_109_cast_fp16 = layer_norm(axes = input_109_axes_0, beta = text_encoder_text_model_encoder_layers_6_layer_norm2_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_6_layer_norm2_weight_to_fp16, x = input_107_cast_fp16)[name = tensor("input_109_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_6_mlp_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(109706816))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(111476352))), name = tensor("text_encoder_text_model_encoder_layers_6_mlp_fc1_weight_to_fp16_palettized"), shape = tensor([3072, 768])]; + tensor text_encoder_text_model_encoder_layers_6_mlp_fc1_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_6_mlp_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(111476544)))]; + tensor linear_40_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_6_mlp_fc1_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_6_mlp_fc1_weight_to_fp16_palettized, x = input_109_cast_fp16)[name = tensor("linear_40_cast_fp16")]; + tensor var_710_to_fp16 = const()[name = tensor("op_710_to_fp16"), val = tensor(0x1.b3cp+0)]; + tensor var_711_cast_fp16 = mul(x = linear_40_cast_fp16, y = var_710_to_fp16)[name = tensor("op_711_cast_fp16")]; + tensor var_712_cast_fp16 = sigmoid(x = var_711_cast_fp16)[name = tensor("op_712_cast_fp16")]; + tensor input_113_cast_fp16 = mul(x = linear_40_cast_fp16, y = var_712_cast_fp16)[name = tensor("input_113_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_6_mlp_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(111482752))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(113252288))), name = tensor("text_encoder_text_model_encoder_layers_6_mlp_fc2_weight_to_fp16_palettized"), shape = tensor([768, 3072])]; + tensor text_encoder_text_model_encoder_layers_6_mlp_fc2_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_6_mlp_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(113252480)))]; + tensor linear_41_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_6_mlp_fc2_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_6_mlp_fc2_weight_to_fp16_palettized, x = input_113_cast_fp16)[name = tensor("linear_41_cast_fp16")]; + tensor input_115_cast_fp16 = add(x = input_107_cast_fp16, y = linear_41_cast_fp16)[name = tensor("input_115_cast_fp16")]; + tensor hidden_states_43_axes_0 = const()[name = tensor("hidden_states_43_axes_0"), val = tensor([-1])]; + tensor text_encoder_text_model_encoder_layers_7_layer_norm1_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_7_layer_norm1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(113254080)))]; + tensor text_encoder_text_model_encoder_layers_7_layer_norm1_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_7_layer_norm1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(113255680)))]; + tensor hidden_states_43_cast_fp16 = layer_norm(axes = hidden_states_43_axes_0, beta = text_encoder_text_model_encoder_layers_7_layer_norm1_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_7_layer_norm1_weight_to_fp16, x = input_115_cast_fp16)[name = tensor("hidden_states_43_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_7_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(113257280))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(113699712))), name = tensor("text_encoder_text_model_encoder_layers_7_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor([768, 768])]; + tensor text_encoder_text_model_encoder_layers_7_self_attn_q_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_7_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(113699904)))]; + tensor linear_42_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_7_self_attn_q_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_7_self_attn_q_proj_weight_to_fp16_palettized, x = hidden_states_43_cast_fp16)[name = tensor("linear_42_cast_fp16")]; + tensor var_737_to_fp16 = const()[name = tensor("op_737_to_fp16"), val = tensor(0x1p-3)]; + tensor tensor_47_cast_fp16 = mul(x = linear_42_cast_fp16, y = var_737_to_fp16)[name = tensor("tensor_47_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_7_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(113701504))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(114143936))), name = tensor("text_encoder_text_model_encoder_layers_7_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor([768, 768])]; + tensor text_encoder_text_model_encoder_layers_7_self_attn_k_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_7_self_attn_k_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(114144128)))]; + tensor linear_43_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_7_self_attn_k_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_7_self_attn_k_proj_weight_to_fp16_palettized, x = hidden_states_43_cast_fp16)[name = tensor("linear_43_cast_fp16")]; + tensor var_742 = const()[name = tensor("op_742"), val = tensor([1, -1, 12, 64])]; + tensor var_743_cast_fp16 = reshape(shape = var_742, x = linear_43_cast_fp16)[name = tensor("op_743_cast_fp16")]; + tensor var_744_perm_0 = const()[name = tensor("op_744_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor text_encoder_text_model_encoder_layers_7_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(114145728))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(114588160))), name = tensor("text_encoder_text_model_encoder_layers_7_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor([768, 768])]; + tensor text_encoder_text_model_encoder_layers_7_self_attn_v_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_7_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(114588352)))]; + tensor linear_44_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_7_self_attn_v_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_7_self_attn_v_proj_weight_to_fp16_palettized, x = hidden_states_43_cast_fp16)[name = tensor("linear_44_cast_fp16")]; + tensor var_749 = const()[name = tensor("op_749"), val = tensor([1, -1, 12, 64])]; + tensor var_750_cast_fp16 = reshape(shape = var_749, x = linear_44_cast_fp16)[name = tensor("op_750_cast_fp16")]; + tensor var_751_perm_0 = const()[name = tensor("op_751_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor var_758 = const()[name = tensor("op_758"), val = tensor([1, 77, 12, 64])]; + tensor var_759_cast_fp16 = reshape(shape = var_758, x = tensor_47_cast_fp16)[name = tensor("op_759_cast_fp16")]; + tensor var_760_perm_0 = const()[name = tensor("op_760_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor var_762 = const()[name = tensor("op_762"), val = tensor([12, -1, 64])]; + tensor var_760_cast_fp16 = transpose(perm = var_760_perm_0, x = var_759_cast_fp16)[name = tensor("transpose_19")]; + tensor query_states_15_cast_fp16 = reshape(shape = var_762, x = var_760_cast_fp16)[name = tensor("query_states_15_cast_fp16")]; + tensor var_764 = const()[name = tensor("op_764"), val = tensor([12, -1, 64])]; + tensor var_744_cast_fp16 = transpose(perm = var_744_perm_0, x = var_743_cast_fp16)[name = tensor("transpose_18")]; + tensor key_states_31_cast_fp16 = reshape(shape = var_764, x = var_744_cast_fp16)[name = tensor("key_states_31_cast_fp16")]; + tensor var_766 = const()[name = tensor("op_766"), val = tensor([12, -1, 64])]; + tensor var_751_cast_fp16 = transpose(perm = var_751_perm_0, x = var_750_cast_fp16)[name = tensor("transpose_17")]; + tensor value_states_31_cast_fp16 = reshape(shape = var_766, x = var_751_cast_fp16)[name = tensor("value_states_31_cast_fp16")]; + tensor attn_weights_43_transpose_x_1 = const()[name = tensor("attn_weights_43_transpose_x_1"), val = tensor(false)]; + tensor attn_weights_43_transpose_y_1 = const()[name = tensor("attn_weights_43_transpose_y_1"), val = tensor(true)]; + tensor attn_weights_43_cast_fp16 = matmul(transpose_x = attn_weights_43_transpose_x_1, transpose_y = attn_weights_43_transpose_y_1, x = query_states_15_cast_fp16, y = key_states_31_cast_fp16)[name = tensor("attn_weights_43_cast_fp16")]; + tensor var_771 = const()[name = tensor("op_771"), val = tensor([1, 12, 77, 77])]; + tensor var_772_cast_fp16 = reshape(shape = var_771, x = attn_weights_43_cast_fp16)[name = tensor("op_772_cast_fp16")]; + tensor attn_weights_45_cast_fp16 = add(x = var_772_cast_fp16, y = op_56_to_fp16_palettized)[name = tensor("attn_weights_45_cast_fp16")]; + tensor var_777 = const()[name = tensor("op_777"), val = tensor([12, 77, 77])]; + tensor input_117_cast_fp16 = reshape(shape = var_777, x = attn_weights_45_cast_fp16)[name = tensor("input_117_cast_fp16")]; + tensor input_119_cast_fp16 = softmax(axis = var_5, x = input_117_cast_fp16)[name = tensor("input_119_cast_fp16")]; + tensor attn_output_43_transpose_x_0 = const()[name = tensor("attn_output_43_transpose_x_0"), val = tensor(false)]; + tensor attn_output_43_transpose_y_0 = const()[name = tensor("attn_output_43_transpose_y_0"), val = tensor(false)]; + tensor attn_output_43_cast_fp16 = matmul(transpose_x = attn_output_43_transpose_x_0, transpose_y = attn_output_43_transpose_y_0, x = input_119_cast_fp16, y = value_states_31_cast_fp16)[name = tensor("attn_output_43_cast_fp16")]; + tensor var_782 = const()[name = tensor("op_782"), val = tensor([1, 12, 77, 64])]; + tensor attn_output_45_cast_fp16 = reshape(shape = var_782, x = attn_output_43_cast_fp16)[name = tensor("attn_output_45_cast_fp16")]; + tensor attn_output_47_perm_0 = const()[name = tensor("attn_output_47_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor var_785 = const()[name = tensor("op_785"), val = tensor([1, 77, 768])]; + tensor attn_output_47_cast_fp16 = transpose(perm = attn_output_47_perm_0, x = attn_output_45_cast_fp16)[name = tensor("transpose_16")]; + tensor input_121_cast_fp16 = reshape(shape = var_785, x = attn_output_47_cast_fp16)[name = tensor("input_121_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_7_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(114589952))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(115032384))), name = tensor("text_encoder_text_model_encoder_layers_7_self_attn_out_proj_weight_to_fp16_palettized"), shape = tensor([768, 768])]; + tensor text_encoder_text_model_encoder_layers_7_self_attn_out_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_7_self_attn_out_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(115032576)))]; + tensor linear_45_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_7_self_attn_out_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_7_self_attn_out_proj_weight_to_fp16_palettized, x = input_121_cast_fp16)[name = tensor("linear_45_cast_fp16")]; + tensor input_123_cast_fp16 = add(x = input_115_cast_fp16, y = linear_45_cast_fp16)[name = tensor("input_123_cast_fp16")]; + tensor input_125_axes_0 = const()[name = tensor("input_125_axes_0"), val = tensor([-1])]; + tensor text_encoder_text_model_encoder_layers_7_layer_norm2_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_7_layer_norm2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(115034176)))]; + tensor text_encoder_text_model_encoder_layers_7_layer_norm2_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_7_layer_norm2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(115035776)))]; + tensor input_125_cast_fp16 = layer_norm(axes = input_125_axes_0, beta = text_encoder_text_model_encoder_layers_7_layer_norm2_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_7_layer_norm2_weight_to_fp16, x = input_123_cast_fp16)[name = tensor("input_125_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_7_mlp_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(115037376))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(116806912))), name = tensor("text_encoder_text_model_encoder_layers_7_mlp_fc1_weight_to_fp16_palettized"), shape = tensor([3072, 768])]; + tensor text_encoder_text_model_encoder_layers_7_mlp_fc1_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_7_mlp_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(116807104)))]; + tensor linear_46_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_7_mlp_fc1_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_7_mlp_fc1_weight_to_fp16_palettized, x = input_125_cast_fp16)[name = tensor("linear_46_cast_fp16")]; + tensor var_800_to_fp16 = const()[name = tensor("op_800_to_fp16"), val = tensor(0x1.b3cp+0)]; + tensor var_801_cast_fp16 = mul(x = linear_46_cast_fp16, y = var_800_to_fp16)[name = tensor("op_801_cast_fp16")]; + tensor var_802_cast_fp16 = sigmoid(x = var_801_cast_fp16)[name = tensor("op_802_cast_fp16")]; + tensor input_129_cast_fp16 = mul(x = linear_46_cast_fp16, y = var_802_cast_fp16)[name = tensor("input_129_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_7_mlp_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(116813312))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(118582848))), name = tensor("text_encoder_text_model_encoder_layers_7_mlp_fc2_weight_to_fp16_palettized"), shape = tensor([768, 3072])]; + tensor text_encoder_text_model_encoder_layers_7_mlp_fc2_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_7_mlp_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(118583040)))]; + tensor linear_47_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_7_mlp_fc2_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_7_mlp_fc2_weight_to_fp16_palettized, x = input_129_cast_fp16)[name = tensor("linear_47_cast_fp16")]; + tensor input_131_cast_fp16 = add(x = input_123_cast_fp16, y = linear_47_cast_fp16)[name = tensor("input_131_cast_fp16")]; + tensor hidden_states_49_axes_0 = const()[name = tensor("hidden_states_49_axes_0"), val = tensor([-1])]; + tensor text_encoder_text_model_encoder_layers_8_layer_norm1_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_8_layer_norm1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(118584640)))]; + tensor text_encoder_text_model_encoder_layers_8_layer_norm1_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_8_layer_norm1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(118586240)))]; + tensor hidden_states_49_cast_fp16 = layer_norm(axes = hidden_states_49_axes_0, beta = text_encoder_text_model_encoder_layers_8_layer_norm1_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_8_layer_norm1_weight_to_fp16, x = input_131_cast_fp16)[name = tensor("hidden_states_49_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_8_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(118587840))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(119030272))), name = tensor("text_encoder_text_model_encoder_layers_8_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor([768, 768])]; + tensor text_encoder_text_model_encoder_layers_8_self_attn_q_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_8_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(119030464)))]; + tensor linear_48_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_8_self_attn_q_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_8_self_attn_q_proj_weight_to_fp16_palettized, x = hidden_states_49_cast_fp16)[name = tensor("linear_48_cast_fp16")]; + tensor var_827_to_fp16 = const()[name = tensor("op_827_to_fp16"), val = tensor(0x1p-3)]; + tensor tensor_53_cast_fp16 = mul(x = linear_48_cast_fp16, y = var_827_to_fp16)[name = tensor("tensor_53_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_8_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(119032064))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(119474496))), name = tensor("text_encoder_text_model_encoder_layers_8_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor([768, 768])]; + tensor text_encoder_text_model_encoder_layers_8_self_attn_k_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_8_self_attn_k_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(119474688)))]; + tensor linear_49_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_8_self_attn_k_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_8_self_attn_k_proj_weight_to_fp16_palettized, x = hidden_states_49_cast_fp16)[name = tensor("linear_49_cast_fp16")]; + tensor var_832 = const()[name = tensor("op_832"), val = tensor([1, -1, 12, 64])]; + tensor var_833_cast_fp16 = reshape(shape = var_832, x = linear_49_cast_fp16)[name = tensor("op_833_cast_fp16")]; + tensor var_834_perm_0 = const()[name = tensor("op_834_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor text_encoder_text_model_encoder_layers_8_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(119476288))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(119918720))), name = tensor("text_encoder_text_model_encoder_layers_8_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor([768, 768])]; + tensor text_encoder_text_model_encoder_layers_8_self_attn_v_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_8_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(119918912)))]; + tensor linear_50_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_8_self_attn_v_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_8_self_attn_v_proj_weight_to_fp16_palettized, x = hidden_states_49_cast_fp16)[name = tensor("linear_50_cast_fp16")]; + tensor var_839 = const()[name = tensor("op_839"), val = tensor([1, -1, 12, 64])]; + tensor var_840_cast_fp16 = reshape(shape = var_839, x = linear_50_cast_fp16)[name = tensor("op_840_cast_fp16")]; + tensor var_841_perm_0 = const()[name = tensor("op_841_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor var_848 = const()[name = tensor("op_848"), val = tensor([1, 77, 12, 64])]; + tensor var_849_cast_fp16 = reshape(shape = var_848, x = tensor_53_cast_fp16)[name = tensor("op_849_cast_fp16")]; + tensor var_850_perm_0 = const()[name = tensor("op_850_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor var_852 = const()[name = tensor("op_852"), val = tensor([12, -1, 64])]; + tensor var_850_cast_fp16 = transpose(perm = var_850_perm_0, x = var_849_cast_fp16)[name = tensor("transpose_15")]; + tensor query_states_17_cast_fp16 = reshape(shape = var_852, x = var_850_cast_fp16)[name = tensor("query_states_17_cast_fp16")]; + tensor var_854 = const()[name = tensor("op_854"), val = tensor([12, -1, 64])]; + tensor var_834_cast_fp16 = transpose(perm = var_834_perm_0, x = var_833_cast_fp16)[name = tensor("transpose_14")]; + tensor key_states_35_cast_fp16 = reshape(shape = var_854, x = var_834_cast_fp16)[name = tensor("key_states_35_cast_fp16")]; + tensor var_856 = const()[name = tensor("op_856"), val = tensor([12, -1, 64])]; + tensor var_841_cast_fp16 = transpose(perm = var_841_perm_0, x = var_840_cast_fp16)[name = tensor("transpose_13")]; + tensor value_states_35_cast_fp16 = reshape(shape = var_856, x = var_841_cast_fp16)[name = tensor("value_states_35_cast_fp16")]; + tensor attn_weights_49_transpose_x_1 = const()[name = tensor("attn_weights_49_transpose_x_1"), val = tensor(false)]; + tensor attn_weights_49_transpose_y_1 = const()[name = tensor("attn_weights_49_transpose_y_1"), val = tensor(true)]; + tensor attn_weights_49_cast_fp16 = matmul(transpose_x = attn_weights_49_transpose_x_1, transpose_y = attn_weights_49_transpose_y_1, x = query_states_17_cast_fp16, y = key_states_35_cast_fp16)[name = tensor("attn_weights_49_cast_fp16")]; + tensor var_861 = const()[name = tensor("op_861"), val = tensor([1, 12, 77, 77])]; + tensor var_862_cast_fp16 = reshape(shape = var_861, x = attn_weights_49_cast_fp16)[name = tensor("op_862_cast_fp16")]; + tensor attn_weights_51_cast_fp16 = add(x = var_862_cast_fp16, y = op_56_to_fp16_palettized)[name = tensor("attn_weights_51_cast_fp16")]; + tensor var_867 = const()[name = tensor("op_867"), val = tensor([12, 77, 77])]; + tensor input_133_cast_fp16 = reshape(shape = var_867, x = attn_weights_51_cast_fp16)[name = tensor("input_133_cast_fp16")]; + tensor input_135_cast_fp16 = softmax(axis = var_5, x = input_133_cast_fp16)[name = tensor("input_135_cast_fp16")]; + tensor attn_output_49_transpose_x_0 = const()[name = tensor("attn_output_49_transpose_x_0"), val = tensor(false)]; + tensor attn_output_49_transpose_y_0 = const()[name = tensor("attn_output_49_transpose_y_0"), val = tensor(false)]; + tensor attn_output_49_cast_fp16 = matmul(transpose_x = attn_output_49_transpose_x_0, transpose_y = attn_output_49_transpose_y_0, x = input_135_cast_fp16, y = value_states_35_cast_fp16)[name = tensor("attn_output_49_cast_fp16")]; + tensor var_872 = const()[name = tensor("op_872"), val = tensor([1, 12, 77, 64])]; + tensor attn_output_51_cast_fp16 = reshape(shape = var_872, x = attn_output_49_cast_fp16)[name = tensor("attn_output_51_cast_fp16")]; + tensor attn_output_53_perm_0 = const()[name = tensor("attn_output_53_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor var_875 = const()[name = tensor("op_875"), val = tensor([1, 77, 768])]; + tensor attn_output_53_cast_fp16 = transpose(perm = attn_output_53_perm_0, x = attn_output_51_cast_fp16)[name = tensor("transpose_12")]; + tensor input_137_cast_fp16 = reshape(shape = var_875, x = attn_output_53_cast_fp16)[name = tensor("input_137_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_8_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(119920512))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(120362944))), name = tensor("text_encoder_text_model_encoder_layers_8_self_attn_out_proj_weight_to_fp16_palettized"), shape = tensor([768, 768])]; + tensor text_encoder_text_model_encoder_layers_8_self_attn_out_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_8_self_attn_out_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(120363136)))]; + tensor linear_51_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_8_self_attn_out_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_8_self_attn_out_proj_weight_to_fp16_palettized, x = input_137_cast_fp16)[name = tensor("linear_51_cast_fp16")]; + tensor input_139_cast_fp16 = add(x = input_131_cast_fp16, y = linear_51_cast_fp16)[name = tensor("input_139_cast_fp16")]; + tensor input_141_axes_0 = const()[name = tensor("input_141_axes_0"), val = tensor([-1])]; + tensor text_encoder_text_model_encoder_layers_8_layer_norm2_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_8_layer_norm2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(120364736)))]; + tensor text_encoder_text_model_encoder_layers_8_layer_norm2_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_8_layer_norm2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(120366336)))]; + tensor input_141_cast_fp16 = layer_norm(axes = input_141_axes_0, beta = text_encoder_text_model_encoder_layers_8_layer_norm2_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_8_layer_norm2_weight_to_fp16, x = input_139_cast_fp16)[name = tensor("input_141_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_8_mlp_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(120367936))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(122137472))), name = tensor("text_encoder_text_model_encoder_layers_8_mlp_fc1_weight_to_fp16_palettized"), shape = tensor([3072, 768])]; + tensor text_encoder_text_model_encoder_layers_8_mlp_fc1_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_8_mlp_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(122137664)))]; + tensor linear_52_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_8_mlp_fc1_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_8_mlp_fc1_weight_to_fp16_palettized, x = input_141_cast_fp16)[name = tensor("linear_52_cast_fp16")]; + tensor var_890_to_fp16 = const()[name = tensor("op_890_to_fp16"), val = tensor(0x1.b3cp+0)]; + tensor var_891_cast_fp16 = mul(x = linear_52_cast_fp16, y = var_890_to_fp16)[name = tensor("op_891_cast_fp16")]; + tensor var_892_cast_fp16 = sigmoid(x = var_891_cast_fp16)[name = tensor("op_892_cast_fp16")]; + tensor input_145_cast_fp16 = mul(x = linear_52_cast_fp16, y = var_892_cast_fp16)[name = tensor("input_145_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_8_mlp_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(122143872))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(123913408))), name = tensor("text_encoder_text_model_encoder_layers_8_mlp_fc2_weight_to_fp16_palettized"), shape = tensor([768, 3072])]; + tensor text_encoder_text_model_encoder_layers_8_mlp_fc2_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_8_mlp_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(123913600)))]; + tensor linear_53_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_8_mlp_fc2_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_8_mlp_fc2_weight_to_fp16_palettized, x = input_145_cast_fp16)[name = tensor("linear_53_cast_fp16")]; + tensor input_147_cast_fp16 = add(x = input_139_cast_fp16, y = linear_53_cast_fp16)[name = tensor("input_147_cast_fp16")]; + tensor hidden_states_55_axes_0 = const()[name = tensor("hidden_states_55_axes_0"), val = tensor([-1])]; + tensor text_encoder_text_model_encoder_layers_9_layer_norm1_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_9_layer_norm1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(123915200)))]; + tensor text_encoder_text_model_encoder_layers_9_layer_norm1_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_9_layer_norm1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(123916800)))]; + tensor hidden_states_55_cast_fp16 = layer_norm(axes = hidden_states_55_axes_0, beta = text_encoder_text_model_encoder_layers_9_layer_norm1_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_9_layer_norm1_weight_to_fp16, x = input_147_cast_fp16)[name = tensor("hidden_states_55_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_9_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(123918400))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(124360832))), name = tensor("text_encoder_text_model_encoder_layers_9_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor([768, 768])]; + tensor text_encoder_text_model_encoder_layers_9_self_attn_q_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_9_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(124361024)))]; + tensor linear_54_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_9_self_attn_q_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_9_self_attn_q_proj_weight_to_fp16_palettized, x = hidden_states_55_cast_fp16)[name = tensor("linear_54_cast_fp16")]; + tensor var_917_to_fp16 = const()[name = tensor("op_917_to_fp16"), val = tensor(0x1p-3)]; + tensor tensor_59_cast_fp16 = mul(x = linear_54_cast_fp16, y = var_917_to_fp16)[name = tensor("tensor_59_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_9_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(124362624))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(124805056))), name = tensor("text_encoder_text_model_encoder_layers_9_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor([768, 768])]; + tensor text_encoder_text_model_encoder_layers_9_self_attn_k_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_9_self_attn_k_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(124805248)))]; + tensor linear_55_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_9_self_attn_k_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_9_self_attn_k_proj_weight_to_fp16_palettized, x = hidden_states_55_cast_fp16)[name = tensor("linear_55_cast_fp16")]; + tensor var_922 = const()[name = tensor("op_922"), val = tensor([1, -1, 12, 64])]; + tensor var_923_cast_fp16 = reshape(shape = var_922, x = linear_55_cast_fp16)[name = tensor("op_923_cast_fp16")]; + tensor var_924_perm_0 = const()[name = tensor("op_924_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor text_encoder_text_model_encoder_layers_9_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(124806848))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(125249280))), name = tensor("text_encoder_text_model_encoder_layers_9_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor([768, 768])]; + tensor text_encoder_text_model_encoder_layers_9_self_attn_v_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_9_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(125249472)))]; + tensor linear_56_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_9_self_attn_v_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_9_self_attn_v_proj_weight_to_fp16_palettized, x = hidden_states_55_cast_fp16)[name = tensor("linear_56_cast_fp16")]; + tensor var_929 = const()[name = tensor("op_929"), val = tensor([1, -1, 12, 64])]; + tensor var_930_cast_fp16 = reshape(shape = var_929, x = linear_56_cast_fp16)[name = tensor("op_930_cast_fp16")]; + tensor var_931_perm_0 = const()[name = tensor("op_931_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor var_938 = const()[name = tensor("op_938"), val = tensor([1, 77, 12, 64])]; + tensor var_939_cast_fp16 = reshape(shape = var_938, x = tensor_59_cast_fp16)[name = tensor("op_939_cast_fp16")]; + tensor var_940_perm_0 = const()[name = tensor("op_940_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor var_942 = const()[name = tensor("op_942"), val = tensor([12, -1, 64])]; + tensor var_940_cast_fp16 = transpose(perm = var_940_perm_0, x = var_939_cast_fp16)[name = tensor("transpose_11")]; + tensor query_states_19_cast_fp16 = reshape(shape = var_942, x = var_940_cast_fp16)[name = tensor("query_states_19_cast_fp16")]; + tensor var_944 = const()[name = tensor("op_944"), val = tensor([12, -1, 64])]; + tensor var_924_cast_fp16 = transpose(perm = var_924_perm_0, x = var_923_cast_fp16)[name = tensor("transpose_10")]; + tensor key_states_39_cast_fp16 = reshape(shape = var_944, x = var_924_cast_fp16)[name = tensor("key_states_39_cast_fp16")]; + tensor var_946 = const()[name = tensor("op_946"), val = tensor([12, -1, 64])]; + tensor var_931_cast_fp16 = transpose(perm = var_931_perm_0, x = var_930_cast_fp16)[name = tensor("transpose_9")]; + tensor value_states_39_cast_fp16 = reshape(shape = var_946, x = var_931_cast_fp16)[name = tensor("value_states_39_cast_fp16")]; + tensor attn_weights_55_transpose_x_1 = const()[name = tensor("attn_weights_55_transpose_x_1"), val = tensor(false)]; + tensor attn_weights_55_transpose_y_1 = const()[name = tensor("attn_weights_55_transpose_y_1"), val = tensor(true)]; + tensor attn_weights_55_cast_fp16 = matmul(transpose_x = attn_weights_55_transpose_x_1, transpose_y = attn_weights_55_transpose_y_1, x = query_states_19_cast_fp16, y = key_states_39_cast_fp16)[name = tensor("attn_weights_55_cast_fp16")]; + tensor var_951 = const()[name = tensor("op_951"), val = tensor([1, 12, 77, 77])]; + tensor var_952_cast_fp16 = reshape(shape = var_951, x = attn_weights_55_cast_fp16)[name = tensor("op_952_cast_fp16")]; + tensor attn_weights_57_cast_fp16 = add(x = var_952_cast_fp16, y = op_56_to_fp16_palettized)[name = tensor("attn_weights_57_cast_fp16")]; + tensor var_957 = const()[name = tensor("op_957"), val = tensor([12, 77, 77])]; + tensor input_149_cast_fp16 = reshape(shape = var_957, x = attn_weights_57_cast_fp16)[name = tensor("input_149_cast_fp16")]; + tensor input_151_cast_fp16 = softmax(axis = var_5, x = input_149_cast_fp16)[name = tensor("input_151_cast_fp16")]; + tensor attn_output_55_transpose_x_0 = const()[name = tensor("attn_output_55_transpose_x_0"), val = tensor(false)]; + tensor attn_output_55_transpose_y_0 = const()[name = tensor("attn_output_55_transpose_y_0"), val = tensor(false)]; + tensor attn_output_55_cast_fp16 = matmul(transpose_x = attn_output_55_transpose_x_0, transpose_y = attn_output_55_transpose_y_0, x = input_151_cast_fp16, y = value_states_39_cast_fp16)[name = tensor("attn_output_55_cast_fp16")]; + tensor var_962 = const()[name = tensor("op_962"), val = tensor([1, 12, 77, 64])]; + tensor attn_output_57_cast_fp16 = reshape(shape = var_962, x = attn_output_55_cast_fp16)[name = tensor("attn_output_57_cast_fp16")]; + tensor attn_output_59_perm_0 = const()[name = tensor("attn_output_59_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor var_965 = const()[name = tensor("op_965"), val = tensor([1, 77, 768])]; + tensor attn_output_59_cast_fp16 = transpose(perm = attn_output_59_perm_0, x = attn_output_57_cast_fp16)[name = tensor("transpose_8")]; + tensor input_153_cast_fp16 = reshape(shape = var_965, x = attn_output_59_cast_fp16)[name = tensor("input_153_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_9_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(125251072))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(125693504))), name = tensor("text_encoder_text_model_encoder_layers_9_self_attn_out_proj_weight_to_fp16_palettized"), shape = tensor([768, 768])]; + tensor text_encoder_text_model_encoder_layers_9_self_attn_out_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_9_self_attn_out_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(125693696)))]; + tensor linear_57_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_9_self_attn_out_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_9_self_attn_out_proj_weight_to_fp16_palettized, x = input_153_cast_fp16)[name = tensor("linear_57_cast_fp16")]; + tensor input_155_cast_fp16 = add(x = input_147_cast_fp16, y = linear_57_cast_fp16)[name = tensor("input_155_cast_fp16")]; + tensor input_157_axes_0 = const()[name = tensor("input_157_axes_0"), val = tensor([-1])]; + tensor text_encoder_text_model_encoder_layers_9_layer_norm2_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_9_layer_norm2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(125695296)))]; + tensor text_encoder_text_model_encoder_layers_9_layer_norm2_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_9_layer_norm2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(125696896)))]; + tensor input_157_cast_fp16 = layer_norm(axes = input_157_axes_0, beta = text_encoder_text_model_encoder_layers_9_layer_norm2_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_9_layer_norm2_weight_to_fp16, x = input_155_cast_fp16)[name = tensor("input_157_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_9_mlp_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(125698496))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(127468032))), name = tensor("text_encoder_text_model_encoder_layers_9_mlp_fc1_weight_to_fp16_palettized"), shape = tensor([3072, 768])]; + tensor text_encoder_text_model_encoder_layers_9_mlp_fc1_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_9_mlp_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(127468224)))]; + tensor linear_58_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_9_mlp_fc1_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_9_mlp_fc1_weight_to_fp16_palettized, x = input_157_cast_fp16)[name = tensor("linear_58_cast_fp16")]; + tensor var_980_to_fp16 = const()[name = tensor("op_980_to_fp16"), val = tensor(0x1.b3cp+0)]; + tensor var_981_cast_fp16 = mul(x = linear_58_cast_fp16, y = var_980_to_fp16)[name = tensor("op_981_cast_fp16")]; + tensor var_982_cast_fp16 = sigmoid(x = var_981_cast_fp16)[name = tensor("op_982_cast_fp16")]; + tensor input_161_cast_fp16 = mul(x = linear_58_cast_fp16, y = var_982_cast_fp16)[name = tensor("input_161_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_9_mlp_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(127474432))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(129243968))), name = tensor("text_encoder_text_model_encoder_layers_9_mlp_fc2_weight_to_fp16_palettized"), shape = tensor([768, 3072])]; + tensor text_encoder_text_model_encoder_layers_9_mlp_fc2_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_9_mlp_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(129244160)))]; + tensor linear_59_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_9_mlp_fc2_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_9_mlp_fc2_weight_to_fp16_palettized, x = input_161_cast_fp16)[name = tensor("linear_59_cast_fp16")]; + tensor input_163_cast_fp16 = add(x = input_155_cast_fp16, y = linear_59_cast_fp16)[name = tensor("input_163_cast_fp16")]; + tensor hidden_states_61_axes_0 = const()[name = tensor("hidden_states_61_axes_0"), val = tensor([-1])]; + tensor text_encoder_text_model_encoder_layers_10_layer_norm1_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_10_layer_norm1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(129245760)))]; + tensor text_encoder_text_model_encoder_layers_10_layer_norm1_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_10_layer_norm1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(129247360)))]; + tensor hidden_states_61_cast_fp16 = layer_norm(axes = hidden_states_61_axes_0, beta = text_encoder_text_model_encoder_layers_10_layer_norm1_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_10_layer_norm1_weight_to_fp16, x = input_163_cast_fp16)[name = tensor("hidden_states_61_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_10_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(129248960))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(129691392))), name = tensor("text_encoder_text_model_encoder_layers_10_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor([768, 768])]; + tensor text_encoder_text_model_encoder_layers_10_self_attn_q_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_10_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(129691584)))]; + tensor linear_60_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_10_self_attn_q_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_10_self_attn_q_proj_weight_to_fp16_palettized, x = hidden_states_61_cast_fp16)[name = tensor("linear_60_cast_fp16")]; + tensor var_1007_to_fp16 = const()[name = tensor("op_1007_to_fp16"), val = tensor(0x1p-3)]; + tensor tensor_65_cast_fp16 = mul(x = linear_60_cast_fp16, y = var_1007_to_fp16)[name = tensor("tensor_65_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_10_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(129693184))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(130135616))), name = tensor("text_encoder_text_model_encoder_layers_10_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor([768, 768])]; + tensor text_encoder_text_model_encoder_layers_10_self_attn_k_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_10_self_attn_k_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(130135808)))]; + tensor linear_61_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_10_self_attn_k_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_10_self_attn_k_proj_weight_to_fp16_palettized, x = hidden_states_61_cast_fp16)[name = tensor("linear_61_cast_fp16")]; + tensor var_1012 = const()[name = tensor("op_1012"), val = tensor([1, -1, 12, 64])]; + tensor var_1013_cast_fp16 = reshape(shape = var_1012, x = linear_61_cast_fp16)[name = tensor("op_1013_cast_fp16")]; + tensor var_1014_perm_0 = const()[name = tensor("op_1014_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor text_encoder_text_model_encoder_layers_10_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(130137408))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(130579840))), name = tensor("text_encoder_text_model_encoder_layers_10_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor([768, 768])]; + tensor text_encoder_text_model_encoder_layers_10_self_attn_v_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_10_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(130580032)))]; + tensor linear_62_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_10_self_attn_v_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_10_self_attn_v_proj_weight_to_fp16_palettized, x = hidden_states_61_cast_fp16)[name = tensor("linear_62_cast_fp16")]; + tensor var_1019 = const()[name = tensor("op_1019"), val = tensor([1, -1, 12, 64])]; + tensor var_1020_cast_fp16 = reshape(shape = var_1019, x = linear_62_cast_fp16)[name = tensor("op_1020_cast_fp16")]; + tensor var_1021_perm_0 = const()[name = tensor("op_1021_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor var_1028 = const()[name = tensor("op_1028"), val = tensor([1, 77, 12, 64])]; + tensor var_1029_cast_fp16 = reshape(shape = var_1028, x = tensor_65_cast_fp16)[name = tensor("op_1029_cast_fp16")]; + tensor var_1030_perm_0 = const()[name = tensor("op_1030_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor var_1032 = const()[name = tensor("op_1032"), val = tensor([12, -1, 64])]; + tensor var_1030_cast_fp16 = transpose(perm = var_1030_perm_0, x = var_1029_cast_fp16)[name = tensor("transpose_7")]; + tensor query_states_21_cast_fp16 = reshape(shape = var_1032, x = var_1030_cast_fp16)[name = tensor("query_states_21_cast_fp16")]; + tensor var_1034 = const()[name = tensor("op_1034"), val = tensor([12, -1, 64])]; + tensor var_1014_cast_fp16 = transpose(perm = var_1014_perm_0, x = var_1013_cast_fp16)[name = tensor("transpose_6")]; + tensor key_states_43_cast_fp16 = reshape(shape = var_1034, x = var_1014_cast_fp16)[name = tensor("key_states_43_cast_fp16")]; + tensor var_1036 = const()[name = tensor("op_1036"), val = tensor([12, -1, 64])]; + tensor var_1021_cast_fp16 = transpose(perm = var_1021_perm_0, x = var_1020_cast_fp16)[name = tensor("transpose_5")]; + tensor value_states_43_cast_fp16 = reshape(shape = var_1036, x = var_1021_cast_fp16)[name = tensor("value_states_43_cast_fp16")]; + tensor attn_weights_61_transpose_x_1 = const()[name = tensor("attn_weights_61_transpose_x_1"), val = tensor(false)]; + tensor attn_weights_61_transpose_y_1 = const()[name = tensor("attn_weights_61_transpose_y_1"), val = tensor(true)]; + tensor attn_weights_61_cast_fp16 = matmul(transpose_x = attn_weights_61_transpose_x_1, transpose_y = attn_weights_61_transpose_y_1, x = query_states_21_cast_fp16, y = key_states_43_cast_fp16)[name = tensor("attn_weights_61_cast_fp16")]; + tensor var_1041 = const()[name = tensor("op_1041"), val = tensor([1, 12, 77, 77])]; + tensor var_1042_cast_fp16 = reshape(shape = var_1041, x = attn_weights_61_cast_fp16)[name = tensor("op_1042_cast_fp16")]; + tensor attn_weights_63_cast_fp16 = add(x = var_1042_cast_fp16, y = op_56_to_fp16_palettized)[name = tensor("attn_weights_63_cast_fp16")]; + tensor var_1047 = const()[name = tensor("op_1047"), val = tensor([12, 77, 77])]; + tensor input_165_cast_fp16 = reshape(shape = var_1047, x = attn_weights_63_cast_fp16)[name = tensor("input_165_cast_fp16")]; + tensor input_167_cast_fp16 = softmax(axis = var_5, x = input_165_cast_fp16)[name = tensor("input_167_cast_fp16")]; + tensor attn_output_61_transpose_x_0 = const()[name = tensor("attn_output_61_transpose_x_0"), val = tensor(false)]; + tensor attn_output_61_transpose_y_0 = const()[name = tensor("attn_output_61_transpose_y_0"), val = tensor(false)]; + tensor attn_output_61_cast_fp16 = matmul(transpose_x = attn_output_61_transpose_x_0, transpose_y = attn_output_61_transpose_y_0, x = input_167_cast_fp16, y = value_states_43_cast_fp16)[name = tensor("attn_output_61_cast_fp16")]; + tensor var_1052 = const()[name = tensor("op_1052"), val = tensor([1, 12, 77, 64])]; + tensor attn_output_63_cast_fp16 = reshape(shape = var_1052, x = attn_output_61_cast_fp16)[name = tensor("attn_output_63_cast_fp16")]; + tensor attn_output_65_perm_0 = const()[name = tensor("attn_output_65_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor var_1055 = const()[name = tensor("op_1055"), val = tensor([1, 77, 768])]; + tensor attn_output_65_cast_fp16 = transpose(perm = attn_output_65_perm_0, x = attn_output_63_cast_fp16)[name = tensor("transpose_4")]; + tensor input_169_cast_fp16 = reshape(shape = var_1055, x = attn_output_65_cast_fp16)[name = tensor("input_169_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_10_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(130581632))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(131024064))), name = tensor("text_encoder_text_model_encoder_layers_10_self_attn_out_proj_weight_to_fp16_palettized"), shape = tensor([768, 768])]; + tensor text_encoder_text_model_encoder_layers_10_self_attn_out_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_10_self_attn_out_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(131024256)))]; + tensor linear_63_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_10_self_attn_out_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_10_self_attn_out_proj_weight_to_fp16_palettized, x = input_169_cast_fp16)[name = tensor("linear_63_cast_fp16")]; + tensor input_171_cast_fp16 = add(x = input_163_cast_fp16, y = linear_63_cast_fp16)[name = tensor("input_171_cast_fp16")]; + tensor input_173_axes_0 = const()[name = tensor("input_173_axes_0"), val = tensor([-1])]; + tensor text_encoder_text_model_encoder_layers_10_layer_norm2_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_10_layer_norm2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(131025856)))]; + tensor text_encoder_text_model_encoder_layers_10_layer_norm2_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_10_layer_norm2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(131027456)))]; + tensor input_173_cast_fp16 = layer_norm(axes = input_173_axes_0, beta = text_encoder_text_model_encoder_layers_10_layer_norm2_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_10_layer_norm2_weight_to_fp16, x = input_171_cast_fp16)[name = tensor("input_173_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_10_mlp_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(131029056))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(132798592))), name = tensor("text_encoder_text_model_encoder_layers_10_mlp_fc1_weight_to_fp16_palettized"), shape = tensor([3072, 768])]; + tensor text_encoder_text_model_encoder_layers_10_mlp_fc1_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_10_mlp_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(132798784)))]; + tensor linear_64_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_10_mlp_fc1_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_10_mlp_fc1_weight_to_fp16_palettized, x = input_173_cast_fp16)[name = tensor("linear_64_cast_fp16")]; + tensor var_1070_to_fp16 = const()[name = tensor("op_1070_to_fp16"), val = tensor(0x1.b3cp+0)]; + tensor var_1071_cast_fp16 = mul(x = linear_64_cast_fp16, y = var_1070_to_fp16)[name = tensor("op_1071_cast_fp16")]; + tensor var_1072_cast_fp16 = sigmoid(x = var_1071_cast_fp16)[name = tensor("op_1072_cast_fp16")]; + tensor input_177_cast_fp16 = mul(x = linear_64_cast_fp16, y = var_1072_cast_fp16)[name = tensor("input_177_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_10_mlp_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(132804992))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(134574528))), name = tensor("text_encoder_text_model_encoder_layers_10_mlp_fc2_weight_to_fp16_palettized"), shape = tensor([768, 3072])]; + tensor text_encoder_text_model_encoder_layers_10_mlp_fc2_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_10_mlp_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(134574720)))]; + tensor linear_65_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_10_mlp_fc2_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_10_mlp_fc2_weight_to_fp16_palettized, x = input_177_cast_fp16)[name = tensor("linear_65_cast_fp16")]; + tensor input_179_cast_fp16 = add(x = input_171_cast_fp16, y = linear_65_cast_fp16)[name = tensor("input_179_cast_fp16")]; + tensor input_179_cast_fp16_to_fp32_dtype_0 = const()[name = tensor("input_179_cast_fp16_to_fp32_dtype_0"), val = tensor("fp32")]; + tensor hidden_states_67_axes_0 = const()[name = tensor("hidden_states_67_axes_0"), val = tensor([-1])]; + tensor text_encoder_text_model_encoder_layers_11_layer_norm1_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_11_layer_norm1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(134576320)))]; + tensor text_encoder_text_model_encoder_layers_11_layer_norm1_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_11_layer_norm1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(134577920)))]; + tensor hidden_states_67_cast_fp16 = layer_norm(axes = hidden_states_67_axes_0, beta = text_encoder_text_model_encoder_layers_11_layer_norm1_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_11_layer_norm1_weight_to_fp16, x = input_179_cast_fp16)[name = tensor("hidden_states_67_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_11_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(134579520))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(135021952))), name = tensor("text_encoder_text_model_encoder_layers_11_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor([768, 768])]; + tensor text_encoder_text_model_encoder_layers_11_self_attn_q_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_11_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(135022144)))]; + tensor linear_66_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_11_self_attn_q_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_11_self_attn_q_proj_weight_to_fp16_palettized, x = hidden_states_67_cast_fp16)[name = tensor("linear_66_cast_fp16")]; + tensor var_1097_to_fp16 = const()[name = tensor("op_1097_to_fp16"), val = tensor(0x1p-3)]; + tensor tensor_cast_fp16 = mul(x = linear_66_cast_fp16, y = var_1097_to_fp16)[name = tensor("tensor_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_11_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(135023744))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(135466176))), name = tensor("text_encoder_text_model_encoder_layers_11_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor([768, 768])]; + tensor text_encoder_text_model_encoder_layers_11_self_attn_k_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_11_self_attn_k_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(135466368)))]; + tensor linear_67_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_11_self_attn_k_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_11_self_attn_k_proj_weight_to_fp16_palettized, x = hidden_states_67_cast_fp16)[name = tensor("linear_67_cast_fp16")]; + tensor var_1102 = const()[name = tensor("op_1102"), val = tensor([1, -1, 12, 64])]; + tensor var_1103_cast_fp16 = reshape(shape = var_1102, x = linear_67_cast_fp16)[name = tensor("op_1103_cast_fp16")]; + tensor var_1104_perm_0 = const()[name = tensor("op_1104_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor text_encoder_text_model_encoder_layers_11_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(135467968))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(135910400))), name = tensor("text_encoder_text_model_encoder_layers_11_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor([768, 768])]; + tensor text_encoder_text_model_encoder_layers_11_self_attn_v_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_11_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(135910592)))]; + tensor linear_68_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_11_self_attn_v_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_11_self_attn_v_proj_weight_to_fp16_palettized, x = hidden_states_67_cast_fp16)[name = tensor("linear_68_cast_fp16")]; + tensor var_1109 = const()[name = tensor("op_1109"), val = tensor([1, -1, 12, 64])]; + tensor var_1110_cast_fp16 = reshape(shape = var_1109, x = linear_68_cast_fp16)[name = tensor("op_1110_cast_fp16")]; + tensor var_1111_perm_0 = const()[name = tensor("op_1111_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor var_1118 = const()[name = tensor("op_1118"), val = tensor([1, 77, 12, 64])]; + tensor var_1119_cast_fp16 = reshape(shape = var_1118, x = tensor_cast_fp16)[name = tensor("op_1119_cast_fp16")]; + tensor var_1120_perm_0 = const()[name = tensor("op_1120_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor var_1122 = const()[name = tensor("op_1122"), val = tensor([12, -1, 64])]; + tensor var_1120_cast_fp16 = transpose(perm = var_1120_perm_0, x = var_1119_cast_fp16)[name = tensor("transpose_3")]; + tensor query_states_cast_fp16 = reshape(shape = var_1122, x = var_1120_cast_fp16)[name = tensor("query_states_cast_fp16")]; + tensor var_1124 = const()[name = tensor("op_1124"), val = tensor([12, -1, 64])]; + tensor var_1104_cast_fp16 = transpose(perm = var_1104_perm_0, x = var_1103_cast_fp16)[name = tensor("transpose_2")]; + tensor key_states_cast_fp16 = reshape(shape = var_1124, x = var_1104_cast_fp16)[name = tensor("key_states_cast_fp16")]; + tensor var_1126 = const()[name = tensor("op_1126"), val = tensor([12, -1, 64])]; + tensor var_1111_cast_fp16 = transpose(perm = var_1111_perm_0, x = var_1110_cast_fp16)[name = tensor("transpose_1")]; + tensor value_states_cast_fp16 = reshape(shape = var_1126, x = var_1111_cast_fp16)[name = tensor("value_states_cast_fp16")]; + tensor attn_weights_67_transpose_x_1 = const()[name = tensor("attn_weights_67_transpose_x_1"), val = tensor(false)]; + tensor attn_weights_67_transpose_y_1 = const()[name = tensor("attn_weights_67_transpose_y_1"), val = tensor(true)]; + tensor attn_weights_67_cast_fp16 = matmul(transpose_x = attn_weights_67_transpose_x_1, transpose_y = attn_weights_67_transpose_y_1, x = query_states_cast_fp16, y = key_states_cast_fp16)[name = tensor("attn_weights_67_cast_fp16")]; + tensor var_1131 = const()[name = tensor("op_1131"), val = tensor([1, 12, 77, 77])]; + tensor var_1132_cast_fp16 = reshape(shape = var_1131, x = attn_weights_67_cast_fp16)[name = tensor("op_1132_cast_fp16")]; + tensor attn_weights_69_cast_fp16 = add(x = var_1132_cast_fp16, y = op_56_to_fp16_palettized)[name = tensor("attn_weights_69_cast_fp16")]; + tensor var_1137 = const()[name = tensor("op_1137"), val = tensor([12, 77, 77])]; + tensor input_181_cast_fp16 = reshape(shape = var_1137, x = attn_weights_69_cast_fp16)[name = tensor("input_181_cast_fp16")]; + tensor input_183_cast_fp16 = softmax(axis = var_5, x = input_181_cast_fp16)[name = tensor("input_183_cast_fp16")]; + tensor attn_output_67_transpose_x_0 = const()[name = tensor("attn_output_67_transpose_x_0"), val = tensor(false)]; + tensor attn_output_67_transpose_y_0 = const()[name = tensor("attn_output_67_transpose_y_0"), val = tensor(false)]; + tensor attn_output_67_cast_fp16 = matmul(transpose_x = attn_output_67_transpose_x_0, transpose_y = attn_output_67_transpose_y_0, x = input_183_cast_fp16, y = value_states_cast_fp16)[name = tensor("attn_output_67_cast_fp16")]; + tensor var_1142 = const()[name = tensor("op_1142"), val = tensor([1, 12, 77, 64])]; + tensor attn_output_69_cast_fp16 = reshape(shape = var_1142, x = attn_output_67_cast_fp16)[name = tensor("attn_output_69_cast_fp16")]; + tensor attn_output_perm_0 = const()[name = tensor("attn_output_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor var_1145 = const()[name = tensor("op_1145"), val = tensor([1, 77, 768])]; + tensor attn_output_cast_fp16 = transpose(perm = attn_output_perm_0, x = attn_output_69_cast_fp16)[name = tensor("transpose_0")]; + tensor input_185_cast_fp16 = reshape(shape = var_1145, x = attn_output_cast_fp16)[name = tensor("input_185_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_11_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(135912192))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(136354624))), name = tensor("text_encoder_text_model_encoder_layers_11_self_attn_out_proj_weight_to_fp16_palettized"), shape = tensor([768, 768])]; + tensor text_encoder_text_model_encoder_layers_11_self_attn_out_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_11_self_attn_out_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(136354816)))]; + tensor linear_69_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_11_self_attn_out_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_11_self_attn_out_proj_weight_to_fp16_palettized, x = input_185_cast_fp16)[name = tensor("linear_69_cast_fp16")]; + tensor input_187_cast_fp16 = add(x = input_179_cast_fp16, y = linear_69_cast_fp16)[name = tensor("input_187_cast_fp16")]; + tensor input_189_axes_0 = const()[name = tensor("input_189_axes_0"), val = tensor([-1])]; + tensor text_encoder_text_model_encoder_layers_11_layer_norm2_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_11_layer_norm2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(136356416)))]; + tensor text_encoder_text_model_encoder_layers_11_layer_norm2_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_11_layer_norm2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(136358016)))]; + tensor input_189_cast_fp16 = layer_norm(axes = input_189_axes_0, beta = text_encoder_text_model_encoder_layers_11_layer_norm2_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_11_layer_norm2_weight_to_fp16, x = input_187_cast_fp16)[name = tensor("input_189_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_11_mlp_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(136359616))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(138129152))), name = tensor("text_encoder_text_model_encoder_layers_11_mlp_fc1_weight_to_fp16_palettized"), shape = tensor([3072, 768])]; + tensor text_encoder_text_model_encoder_layers_11_mlp_fc1_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_11_mlp_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(138129344)))]; + tensor linear_70_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_11_mlp_fc1_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_11_mlp_fc1_weight_to_fp16_palettized, x = input_189_cast_fp16)[name = tensor("linear_70_cast_fp16")]; + tensor var_1160_to_fp16 = const()[name = tensor("op_1160_to_fp16"), val = tensor(0x1.b3cp+0)]; + tensor var_1161_cast_fp16 = mul(x = linear_70_cast_fp16, y = var_1160_to_fp16)[name = tensor("op_1161_cast_fp16")]; + tensor var_1162_cast_fp16 = sigmoid(x = var_1161_cast_fp16)[name = tensor("op_1162_cast_fp16")]; + tensor input_193_cast_fp16 = mul(x = linear_70_cast_fp16, y = var_1162_cast_fp16)[name = tensor("input_193_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_11_mlp_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(138135552))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(139905088))), name = tensor("text_encoder_text_model_encoder_layers_11_mlp_fc2_weight_to_fp16_palettized"), shape = tensor([768, 3072])]; + tensor text_encoder_text_model_encoder_layers_11_mlp_fc2_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_11_mlp_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(139905280)))]; + tensor linear_71_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_11_mlp_fc2_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_11_mlp_fc2_weight_to_fp16_palettized, x = input_193_cast_fp16)[name = tensor("linear_71_cast_fp16")]; + tensor input_cast_fp16 = add(x = input_187_cast_fp16, y = linear_71_cast_fp16)[name = tensor("input_cast_fp16")]; + tensor last_hidden_state_axes_0 = const()[name = tensor("last_hidden_state_axes_0"), val = tensor([-1])]; + tensor text_encoder_text_model_final_layer_norm_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_final_layer_norm_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(139906880)))]; + tensor text_encoder_text_model_final_layer_norm_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_final_layer_norm_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(139908480)))]; + tensor last_hidden_state_cast_fp16 = layer_norm(axes = last_hidden_state_axes_0, beta = text_encoder_text_model_final_layer_norm_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_final_layer_norm_weight_to_fp16, x = input_cast_fp16)[name = tensor("last_hidden_state_cast_fp16")]; + tensor var_1176 = const()[name = tensor("op_1176"), val = tensor([0])]; + tensor var_1178 = reduce_argmax(axis = var_5, keep_dims = var_6, x = cast_1)[name = tensor("op_1178")]; + tensor stack_0_axis_0 = const()[name = tensor("stack_0_axis_0"), val = tensor(1)]; + tensor stack_0 = stack(axis = stack_0_axis_0, values = (var_1176, var_1178))[name = tensor("stack_0")]; + tensor var_1180_transpose_batch_dims_0 = const()[name = tensor("op_1180_transpose_batch_dims_0"), val = tensor(0)]; + tensor var_1180_transpose_cast_fp16 = gather_nd(batch_dims = var_1180_transpose_batch_dims_0, indices = stack_0, x = last_hidden_state_cast_fp16)[name = tensor("op_1180_transpose_cast_fp16")]; + tensor var_1180_cast_fp16_to_fp32_dtype_0 = const()[name = tensor("op_1180_cast_fp16_to_fp32_dtype_0"), val = tensor("fp32")]; + tensor hidden_embeds = cast(dtype = input_179_cast_fp16_to_fp32_dtype_0, x = input_179_cast_fp16)[name = tensor("cast_0")]; + tensor pooled_outputs = cast(dtype = var_1180_cast_fp16_to_fp32_dtype_0, x = var_1180_transpose_cast_fp16)[name = tensor("cast_1")]; + } -> (hidden_embeds, pooled_outputs); +} \ No newline at end of file