diff --git "a/split_einsum_v2/Resources/TextEncoder.mlmodelc/model.mil" "b/split_einsum_v2/Resources/TextEncoder.mlmodelc/model.mil" new file mode 100644--- /dev/null +++ "b/split_einsum_v2/Resources/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"}, {"coremltools-component-torch", "2.2.0"}, {"coremltools-source-dialect", "TorchScript"}, {"coremltools-version", "7.2"}})] +{ + 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_127 = cast(dtype = cast_1_dtype_0, x = input_ids)[name = tensor("cast_127")]; + tensor inputs_embeds_cast_fp16 = gather(axis = inputs_embeds_axis_0, batch_dims = inputs_embeds_batch_dims_0, indices = cast_127, x = text_encoder_text_model_embeddings_token_embedding_weight_to_fp16)[name = tensor("inputs_embeds_cast_fp16")]; + tensor position_embeddings_to_fp16 = const()[name = tensor("position_embeddings_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(75890816)))]; + tensor input_3_cast_fp16 = add(x = inputs_embeds_cast_fp16, y = position_embeddings_to_fp16)[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(76009152)))]; + 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(76010752)))]; + tensor var_13_to_fp16 = const()[name = tensor("op_13_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_13_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 = const()[name = tensor("text_encoder_text_model_encoder_layers_0_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(76012352)))]; + 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(77192064)))]; + 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, x = hidden_states_1_cast_fp16)[name = tensor("linear_0_cast_fp16")]; + tensor var_87_to_fp16 = const()[name = tensor("op_87_to_fp16"), val = tensor(0x1p-3)]; + tensor tensor_5_cast_fp16 = mul(x = linear_0_cast_fp16, y = var_87_to_fp16)[name = tensor("tensor_5_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_0_self_attn_k_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_0_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(77193664)))]; + 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(78373376)))]; + 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, x = hidden_states_1_cast_fp16)[name = tensor("linear_1_cast_fp16")]; + tensor var_92 = const()[name = tensor("op_92"), val = tensor([1, -1, 12, 64])]; + tensor var_93_cast_fp16 = reshape(shape = var_92, x = linear_1_cast_fp16)[name = tensor("op_93_cast_fp16")]; + tensor var_94_perm_0 = const()[name = tensor("op_94_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor text_encoder_text_model_encoder_layers_0_self_attn_v_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_0_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(78374976)))]; + 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(79554688)))]; + 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, x = hidden_states_1_cast_fp16)[name = tensor("linear_2_cast_fp16")]; + tensor var_99 = const()[name = tensor("op_99"), val = tensor([1, -1, 12, 64])]; + tensor var_100_cast_fp16 = reshape(shape = var_99, x = linear_2_cast_fp16)[name = tensor("op_100_cast_fp16")]; + tensor var_101_perm_0 = const()[name = tensor("op_101_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor var_108 = const()[name = tensor("op_108"), val = tensor([1, 77, 12, 64])]; + tensor var_109_cast_fp16 = reshape(shape = var_108, x = tensor_5_cast_fp16)[name = tensor("op_109_cast_fp16")]; + tensor var_110_perm_0 = const()[name = tensor("op_110_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor var_112 = const()[name = tensor("op_112"), val = tensor([12, -1, 64])]; + tensor transpose_46 = transpose(perm = var_110_perm_0, x = var_109_cast_fp16)[name = tensor("transpose_46")]; + tensor query_states_1_cast_fp16 = reshape(shape = var_112, x = transpose_46)[name = tensor("query_states_1_cast_fp16")]; + tensor var_114 = const()[name = tensor("op_114"), val = tensor([12, -1, 64])]; + tensor transpose_48 = transpose(perm = var_94_perm_0, x = var_93_cast_fp16)[name = tensor("transpose_48")]; + tensor key_states_3_cast_fp16 = reshape(shape = var_114, x = transpose_48)[name = tensor("key_states_3_cast_fp16")]; + tensor var_116 = const()[name = tensor("op_116"), val = tensor([12, -1, 64])]; + tensor transpose_47 = transpose(perm = var_101_perm_0, x = var_100_cast_fp16)[name = tensor("transpose_47")]; + tensor value_states_3_cast_fp16 = reshape(shape = var_116, x = transpose_47)[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_121 = const()[name = tensor("op_121"), val = tensor([1, 12, 77, 77])]; + tensor var_122_cast_fp16 = reshape(shape = var_121, x = attn_weights_1_cast_fp16)[name = tensor("op_122_cast_fp16")]; + tensor causal_attention_mask_to_fp16 = const()[name = tensor("causal_attention_mask_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(79556288)))]; + tensor attn_weights_3_cast_fp16 = add(x = var_122_cast_fp16, y = causal_attention_mask_to_fp16)[name = tensor("attn_weights_3_cast_fp16")]; + tensor var_127 = const()[name = tensor("op_127"), val = tensor([12, 77, 77])]; + tensor input_5_cast_fp16 = reshape(shape = var_127, 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_132 = const()[name = tensor("op_132"), val = tensor([1, 12, 77, 64])]; + tensor attn_output_3_cast_fp16 = reshape(shape = var_132, 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_135 = const()[name = tensor("op_135"), val = tensor([1, 77, 768])]; + tensor transpose_45 = transpose(perm = attn_output_5_perm_0, x = attn_output_3_cast_fp16)[name = tensor("transpose_45")]; + tensor input_9_cast_fp16 = reshape(shape = var_135, x = transpose_45)[name = tensor("input_9_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_0_self_attn_out_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_0_self_attn_out_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(79568256)))]; + 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(80747968)))]; + 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, 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(80749568)))]; + 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(80751168)))]; + 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_13_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 = const()[name = tensor("text_encoder_text_model_encoder_layers_0_mlp_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(80752768)))]; + 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(85471424)))]; + 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, x = input_13_cast_fp16)[name = tensor("linear_4_cast_fp16")]; + tensor var_150_to_fp16 = const()[name = tensor("op_150_to_fp16"), val = tensor(0x1.b3cp+0)]; + tensor var_151_cast_fp16 = mul(x = linear_4_cast_fp16, y = var_150_to_fp16)[name = tensor("op_151_cast_fp16")]; + tensor var_152_cast_fp16 = sigmoid(x = var_151_cast_fp16)[name = tensor("op_152_cast_fp16")]; + tensor input_17_cast_fp16 = mul(x = linear_4_cast_fp16, y = var_152_cast_fp16)[name = tensor("input_17_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_0_mlp_fc2_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_0_mlp_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(85477632)))]; + 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(90196288)))]; + 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, 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(90197888)))]; + 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(90199488)))]; + 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_13_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 = const()[name = tensor("text_encoder_text_model_encoder_layers_1_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(90201088)))]; + 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(91380800)))]; + 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, x = hidden_states_7_cast_fp16)[name = tensor("linear_6_cast_fp16")]; + tensor var_177_to_fp16 = const()[name = tensor("op_177_to_fp16"), val = tensor(0x1p-3)]; + tensor tensor_11_cast_fp16 = mul(x = linear_6_cast_fp16, y = var_177_to_fp16)[name = tensor("tensor_11_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_1_self_attn_k_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_1_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(91382400)))]; + 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(92562112)))]; + 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, x = hidden_states_7_cast_fp16)[name = tensor("linear_7_cast_fp16")]; + tensor var_182 = const()[name = tensor("op_182"), val = tensor([1, -1, 12, 64])]; + tensor var_183_cast_fp16 = reshape(shape = var_182, x = linear_7_cast_fp16)[name = tensor("op_183_cast_fp16")]; + tensor var_184_perm_0 = const()[name = tensor("op_184_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor text_encoder_text_model_encoder_layers_1_self_attn_v_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_1_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(92563712)))]; + 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(93743424)))]; + 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, x = hidden_states_7_cast_fp16)[name = tensor("linear_8_cast_fp16")]; + tensor var_189 = const()[name = tensor("op_189"), val = tensor([1, -1, 12, 64])]; + tensor var_190_cast_fp16 = reshape(shape = var_189, x = linear_8_cast_fp16)[name = tensor("op_190_cast_fp16")]; + tensor var_191_perm_0 = const()[name = tensor("op_191_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor var_198 = const()[name = tensor("op_198"), val = tensor([1, 77, 12, 64])]; + tensor var_199_cast_fp16 = reshape(shape = var_198, x = tensor_11_cast_fp16)[name = tensor("op_199_cast_fp16")]; + tensor var_200_perm_0 = const()[name = tensor("op_200_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor var_202 = const()[name = tensor("op_202"), val = tensor([12, -1, 64])]; + tensor transpose_42 = transpose(perm = var_200_perm_0, x = var_199_cast_fp16)[name = tensor("transpose_42")]; + tensor query_states_3_cast_fp16 = reshape(shape = var_202, x = transpose_42)[name = tensor("query_states_3_cast_fp16")]; + tensor var_204 = const()[name = tensor("op_204"), val = tensor([12, -1, 64])]; + tensor transpose_44 = transpose(perm = var_184_perm_0, x = var_183_cast_fp16)[name = tensor("transpose_44")]; + tensor key_states_7_cast_fp16 = reshape(shape = var_204, x = transpose_44)[name = tensor("key_states_7_cast_fp16")]; + tensor var_206 = const()[name = tensor("op_206"), val = tensor([12, -1, 64])]; + tensor transpose_43 = transpose(perm = var_191_perm_0, x = var_190_cast_fp16)[name = tensor("transpose_43")]; + tensor value_states_7_cast_fp16 = reshape(shape = var_206, x = transpose_43)[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_211 = const()[name = tensor("op_211"), val = tensor([1, 12, 77, 77])]; + tensor var_212_cast_fp16 = reshape(shape = var_211, x = attn_weights_7_cast_fp16)[name = tensor("op_212_cast_fp16")]; + tensor attn_weights_9_cast_fp16 = add(x = var_212_cast_fp16, y = causal_attention_mask_to_fp16)[name = tensor("attn_weights_9_cast_fp16")]; + tensor var_217 = const()[name = tensor("op_217"), val = tensor([12, 77, 77])]; + tensor input_21_cast_fp16 = reshape(shape = var_217, 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_222 = const()[name = tensor("op_222"), val = tensor([1, 12, 77, 64])]; + tensor attn_output_9_cast_fp16 = reshape(shape = var_222, 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_225 = const()[name = tensor("op_225"), val = tensor([1, 77, 768])]; + tensor transpose_41 = transpose(perm = attn_output_11_perm_0, x = attn_output_9_cast_fp16)[name = tensor("transpose_41")]; + tensor input_25_cast_fp16 = reshape(shape = var_225, x = transpose_41)[name = tensor("input_25_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_1_self_attn_out_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_1_self_attn_out_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(93745024)))]; + 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(94924736)))]; + 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, 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(94926336)))]; + 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(94927936)))]; + 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_13_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 = const()[name = tensor("text_encoder_text_model_encoder_layers_1_mlp_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(94929536)))]; + 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(99648192)))]; + 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, x = input_29_cast_fp16)[name = tensor("linear_10_cast_fp16")]; + tensor var_240_to_fp16 = const()[name = tensor("op_240_to_fp16"), val = tensor(0x1.b3cp+0)]; + tensor var_241_cast_fp16 = mul(x = linear_10_cast_fp16, y = var_240_to_fp16)[name = tensor("op_241_cast_fp16")]; + tensor var_242_cast_fp16 = sigmoid(x = var_241_cast_fp16)[name = tensor("op_242_cast_fp16")]; + tensor input_33_cast_fp16 = mul(x = linear_10_cast_fp16, y = var_242_cast_fp16)[name = tensor("input_33_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_1_mlp_fc2_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_1_mlp_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(99654400)))]; + 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(104373056)))]; + 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, 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(104374656)))]; + 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(104376256)))]; + 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_13_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 = const()[name = tensor("text_encoder_text_model_encoder_layers_2_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(104377856)))]; + 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(105557568)))]; + 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, x = hidden_states_13_cast_fp16)[name = tensor("linear_12_cast_fp16")]; + tensor var_267_to_fp16 = const()[name = tensor("op_267_to_fp16"), val = tensor(0x1p-3)]; + tensor tensor_17_cast_fp16 = mul(x = linear_12_cast_fp16, y = var_267_to_fp16)[name = tensor("tensor_17_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_2_self_attn_k_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_2_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(105559168)))]; + 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(106738880)))]; + 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, x = hidden_states_13_cast_fp16)[name = tensor("linear_13_cast_fp16")]; + tensor var_272 = const()[name = tensor("op_272"), val = tensor([1, -1, 12, 64])]; + tensor var_273_cast_fp16 = reshape(shape = var_272, x = linear_13_cast_fp16)[name = tensor("op_273_cast_fp16")]; + tensor var_274_perm_0 = const()[name = tensor("op_274_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor text_encoder_text_model_encoder_layers_2_self_attn_v_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_2_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(106740480)))]; + 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(107920192)))]; + 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, x = hidden_states_13_cast_fp16)[name = tensor("linear_14_cast_fp16")]; + tensor var_279 = const()[name = tensor("op_279"), val = tensor([1, -1, 12, 64])]; + tensor var_280_cast_fp16 = reshape(shape = var_279, x = linear_14_cast_fp16)[name = tensor("op_280_cast_fp16")]; + tensor var_281_perm_0 = const()[name = tensor("op_281_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor var_288 = const()[name = tensor("op_288"), val = tensor([1, 77, 12, 64])]; + tensor var_289_cast_fp16 = reshape(shape = var_288, x = tensor_17_cast_fp16)[name = tensor("op_289_cast_fp16")]; + tensor var_290_perm_0 = const()[name = tensor("op_290_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor var_292 = const()[name = tensor("op_292"), val = tensor([12, -1, 64])]; + tensor transpose_38 = transpose(perm = var_290_perm_0, x = var_289_cast_fp16)[name = tensor("transpose_38")]; + tensor query_states_5_cast_fp16 = reshape(shape = var_292, x = transpose_38)[name = tensor("query_states_5_cast_fp16")]; + tensor var_294 = const()[name = tensor("op_294"), val = tensor([12, -1, 64])]; + tensor transpose_40 = transpose(perm = var_274_perm_0, x = var_273_cast_fp16)[name = tensor("transpose_40")]; + tensor key_states_11_cast_fp16 = reshape(shape = var_294, x = transpose_40)[name = tensor("key_states_11_cast_fp16")]; + tensor var_296 = const()[name = tensor("op_296"), val = tensor([12, -1, 64])]; + tensor transpose_39 = transpose(perm = var_281_perm_0, x = var_280_cast_fp16)[name = tensor("transpose_39")]; + tensor value_states_11_cast_fp16 = reshape(shape = var_296, x = transpose_39)[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_301 = const()[name = tensor("op_301"), val = tensor([1, 12, 77, 77])]; + tensor var_302_cast_fp16 = reshape(shape = var_301, x = attn_weights_13_cast_fp16)[name = tensor("op_302_cast_fp16")]; + tensor attn_weights_15_cast_fp16 = add(x = var_302_cast_fp16, y = causal_attention_mask_to_fp16)[name = tensor("attn_weights_15_cast_fp16")]; + tensor var_307 = const()[name = tensor("op_307"), val = tensor([12, 77, 77])]; + tensor input_37_cast_fp16 = reshape(shape = var_307, 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_312 = const()[name = tensor("op_312"), val = tensor([1, 12, 77, 64])]; + tensor attn_output_15_cast_fp16 = reshape(shape = var_312, 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_315 = const()[name = tensor("op_315"), val = tensor([1, 77, 768])]; + tensor transpose_37 = transpose(perm = attn_output_17_perm_0, x = attn_output_15_cast_fp16)[name = tensor("transpose_37")]; + tensor input_41_cast_fp16 = reshape(shape = var_315, x = transpose_37)[name = tensor("input_41_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_2_self_attn_out_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_2_self_attn_out_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(107921792)))]; + 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(109101504)))]; + 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, 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(109103104)))]; + 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(109104704)))]; + 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_13_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 = const()[name = tensor("text_encoder_text_model_encoder_layers_2_mlp_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(109106304)))]; + 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(113824960)))]; + 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, x = input_45_cast_fp16)[name = tensor("linear_16_cast_fp16")]; + tensor var_330_to_fp16 = const()[name = tensor("op_330_to_fp16"), val = tensor(0x1.b3cp+0)]; + tensor var_331_cast_fp16 = mul(x = linear_16_cast_fp16, y = var_330_to_fp16)[name = tensor("op_331_cast_fp16")]; + tensor var_332_cast_fp16 = sigmoid(x = var_331_cast_fp16)[name = tensor("op_332_cast_fp16")]; + tensor input_49_cast_fp16 = mul(x = linear_16_cast_fp16, y = var_332_cast_fp16)[name = tensor("input_49_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_2_mlp_fc2_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_2_mlp_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(113831168)))]; + 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(118549824)))]; + 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, 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(118551424)))]; + 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(118553024)))]; + 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_13_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 = const()[name = tensor("text_encoder_text_model_encoder_layers_3_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(118554624)))]; + 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(119734336)))]; + 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, x = hidden_states_19_cast_fp16)[name = tensor("linear_18_cast_fp16")]; + tensor var_357_to_fp16 = const()[name = tensor("op_357_to_fp16"), val = tensor(0x1p-3)]; + tensor tensor_23_cast_fp16 = mul(x = linear_18_cast_fp16, y = var_357_to_fp16)[name = tensor("tensor_23_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_3_self_attn_k_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_3_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(119735936)))]; + 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(120915648)))]; + 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, x = hidden_states_19_cast_fp16)[name = tensor("linear_19_cast_fp16")]; + tensor var_362 = const()[name = tensor("op_362"), val = tensor([1, -1, 12, 64])]; + tensor var_363_cast_fp16 = reshape(shape = var_362, x = linear_19_cast_fp16)[name = tensor("op_363_cast_fp16")]; + tensor var_364_perm_0 = const()[name = tensor("op_364_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor text_encoder_text_model_encoder_layers_3_self_attn_v_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_3_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(120917248)))]; + 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(122096960)))]; + 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, x = hidden_states_19_cast_fp16)[name = tensor("linear_20_cast_fp16")]; + tensor var_369 = const()[name = tensor("op_369"), val = tensor([1, -1, 12, 64])]; + tensor var_370_cast_fp16 = reshape(shape = var_369, x = linear_20_cast_fp16)[name = tensor("op_370_cast_fp16")]; + tensor var_371_perm_0 = const()[name = tensor("op_371_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor var_378 = const()[name = tensor("op_378"), val = tensor([1, 77, 12, 64])]; + tensor var_379_cast_fp16 = reshape(shape = var_378, x = tensor_23_cast_fp16)[name = tensor("op_379_cast_fp16")]; + tensor var_380_perm_0 = const()[name = tensor("op_380_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor var_382 = const()[name = tensor("op_382"), val = tensor([12, -1, 64])]; + tensor transpose_34 = transpose(perm = var_380_perm_0, x = var_379_cast_fp16)[name = tensor("transpose_34")]; + tensor query_states_7_cast_fp16 = reshape(shape = var_382, x = transpose_34)[name = tensor("query_states_7_cast_fp16")]; + tensor var_384 = const()[name = tensor("op_384"), val = tensor([12, -1, 64])]; + tensor transpose_36 = transpose(perm = var_364_perm_0, x = var_363_cast_fp16)[name = tensor("transpose_36")]; + tensor key_states_15_cast_fp16 = reshape(shape = var_384, x = transpose_36)[name = tensor("key_states_15_cast_fp16")]; + tensor var_386 = const()[name = tensor("op_386"), val = tensor([12, -1, 64])]; + tensor transpose_35 = transpose(perm = var_371_perm_0, x = var_370_cast_fp16)[name = tensor("transpose_35")]; + tensor value_states_15_cast_fp16 = reshape(shape = var_386, x = transpose_35)[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_391 = const()[name = tensor("op_391"), val = tensor([1, 12, 77, 77])]; + tensor var_392_cast_fp16 = reshape(shape = var_391, x = attn_weights_19_cast_fp16)[name = tensor("op_392_cast_fp16")]; + tensor attn_weights_21_cast_fp16 = add(x = var_392_cast_fp16, y = causal_attention_mask_to_fp16)[name = tensor("attn_weights_21_cast_fp16")]; + tensor var_397 = const()[name = tensor("op_397"), val = tensor([12, 77, 77])]; + tensor input_53_cast_fp16 = reshape(shape = var_397, 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_402 = const()[name = tensor("op_402"), val = tensor([1, 12, 77, 64])]; + tensor attn_output_21_cast_fp16 = reshape(shape = var_402, 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_405 = const()[name = tensor("op_405"), val = tensor([1, 77, 768])]; + tensor transpose_33 = transpose(perm = attn_output_23_perm_0, x = attn_output_21_cast_fp16)[name = tensor("transpose_33")]; + tensor input_57_cast_fp16 = reshape(shape = var_405, x = transpose_33)[name = tensor("input_57_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_3_self_attn_out_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_3_self_attn_out_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(122098560)))]; + 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(123278272)))]; + 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, 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(123279872)))]; + 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(123281472)))]; + 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_13_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 = const()[name = tensor("text_encoder_text_model_encoder_layers_3_mlp_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(123283072)))]; + 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(128001728)))]; + 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, x = input_61_cast_fp16)[name = tensor("linear_22_cast_fp16")]; + tensor var_420_to_fp16 = const()[name = tensor("op_420_to_fp16"), val = tensor(0x1.b3cp+0)]; + tensor var_421_cast_fp16 = mul(x = linear_22_cast_fp16, y = var_420_to_fp16)[name = tensor("op_421_cast_fp16")]; + tensor var_422_cast_fp16 = sigmoid(x = var_421_cast_fp16)[name = tensor("op_422_cast_fp16")]; + tensor input_65_cast_fp16 = mul(x = linear_22_cast_fp16, y = var_422_cast_fp16)[name = tensor("input_65_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_3_mlp_fc2_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_3_mlp_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(128007936)))]; + 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(132726592)))]; + 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, 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(132728192)))]; + 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(132729792)))]; + 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_13_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 = const()[name = tensor("text_encoder_text_model_encoder_layers_4_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(132731392)))]; + 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(133911104)))]; + 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, x = hidden_states_25_cast_fp16)[name = tensor("linear_24_cast_fp16")]; + tensor var_447_to_fp16 = const()[name = tensor("op_447_to_fp16"), val = tensor(0x1p-3)]; + tensor tensor_29_cast_fp16 = mul(x = linear_24_cast_fp16, y = var_447_to_fp16)[name = tensor("tensor_29_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_4_self_attn_k_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_4_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(133912704)))]; + 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(135092416)))]; + 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, x = hidden_states_25_cast_fp16)[name = tensor("linear_25_cast_fp16")]; + tensor var_452 = const()[name = tensor("op_452"), val = tensor([1, -1, 12, 64])]; + tensor var_453_cast_fp16 = reshape(shape = var_452, x = linear_25_cast_fp16)[name = tensor("op_453_cast_fp16")]; + tensor var_454_perm_0 = const()[name = tensor("op_454_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor text_encoder_text_model_encoder_layers_4_self_attn_v_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_4_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(135094016)))]; + 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(136273728)))]; + 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, x = hidden_states_25_cast_fp16)[name = tensor("linear_26_cast_fp16")]; + tensor var_459 = const()[name = tensor("op_459"), val = tensor([1, -1, 12, 64])]; + tensor var_460_cast_fp16 = reshape(shape = var_459, x = linear_26_cast_fp16)[name = tensor("op_460_cast_fp16")]; + tensor var_461_perm_0 = const()[name = tensor("op_461_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor var_468 = const()[name = tensor("op_468"), val = tensor([1, 77, 12, 64])]; + tensor var_469_cast_fp16 = reshape(shape = var_468, x = tensor_29_cast_fp16)[name = tensor("op_469_cast_fp16")]; + tensor var_470_perm_0 = const()[name = tensor("op_470_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor var_472 = const()[name = tensor("op_472"), val = tensor([12, -1, 64])]; + tensor transpose_30 = transpose(perm = var_470_perm_0, x = var_469_cast_fp16)[name = tensor("transpose_30")]; + tensor query_states_9_cast_fp16 = reshape(shape = var_472, x = transpose_30)[name = tensor("query_states_9_cast_fp16")]; + tensor var_474 = const()[name = tensor("op_474"), val = tensor([12, -1, 64])]; + tensor transpose_32 = transpose(perm = var_454_perm_0, x = var_453_cast_fp16)[name = tensor("transpose_32")]; + tensor key_states_19_cast_fp16 = reshape(shape = var_474, x = transpose_32)[name = tensor("key_states_19_cast_fp16")]; + tensor var_476 = const()[name = tensor("op_476"), val = tensor([12, -1, 64])]; + tensor transpose_31 = transpose(perm = var_461_perm_0, x = var_460_cast_fp16)[name = tensor("transpose_31")]; + tensor value_states_19_cast_fp16 = reshape(shape = var_476, x = transpose_31)[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_481 = const()[name = tensor("op_481"), val = tensor([1, 12, 77, 77])]; + tensor var_482_cast_fp16 = reshape(shape = var_481, x = attn_weights_25_cast_fp16)[name = tensor("op_482_cast_fp16")]; + tensor attn_weights_27_cast_fp16 = add(x = var_482_cast_fp16, y = causal_attention_mask_to_fp16)[name = tensor("attn_weights_27_cast_fp16")]; + tensor var_487 = const()[name = tensor("op_487"), val = tensor([12, 77, 77])]; + tensor input_69_cast_fp16 = reshape(shape = var_487, 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_492 = const()[name = tensor("op_492"), val = tensor([1, 12, 77, 64])]; + tensor attn_output_27_cast_fp16 = reshape(shape = var_492, 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_495 = const()[name = tensor("op_495"), val = tensor([1, 77, 768])]; + tensor transpose_29 = transpose(perm = attn_output_29_perm_0, x = attn_output_27_cast_fp16)[name = tensor("transpose_29")]; + tensor input_73_cast_fp16 = reshape(shape = var_495, x = transpose_29)[name = tensor("input_73_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_4_self_attn_out_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_4_self_attn_out_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(136275328)))]; + 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(137455040)))]; + 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, 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(137456640)))]; + 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(137458240)))]; + 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_13_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 = const()[name = tensor("text_encoder_text_model_encoder_layers_4_mlp_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(137459840)))]; + 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(142178496)))]; + 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, x = input_77_cast_fp16)[name = tensor("linear_28_cast_fp16")]; + tensor var_510_to_fp16 = const()[name = tensor("op_510_to_fp16"), val = tensor(0x1.b3cp+0)]; + tensor var_511_cast_fp16 = mul(x = linear_28_cast_fp16, y = var_510_to_fp16)[name = tensor("op_511_cast_fp16")]; + tensor var_512_cast_fp16 = sigmoid(x = var_511_cast_fp16)[name = tensor("op_512_cast_fp16")]; + tensor input_81_cast_fp16 = mul(x = linear_28_cast_fp16, y = var_512_cast_fp16)[name = tensor("input_81_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_4_mlp_fc2_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_4_mlp_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(142184704)))]; + 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(146903360)))]; + 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, 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(146904960)))]; + 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(146906560)))]; + 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_13_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 = const()[name = tensor("text_encoder_text_model_encoder_layers_5_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(146908160)))]; + 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(148087872)))]; + 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, x = hidden_states_31_cast_fp16)[name = tensor("linear_30_cast_fp16")]; + tensor var_537_to_fp16 = const()[name = tensor("op_537_to_fp16"), val = tensor(0x1p-3)]; + tensor tensor_35_cast_fp16 = mul(x = linear_30_cast_fp16, y = var_537_to_fp16)[name = tensor("tensor_35_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_5_self_attn_k_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_5_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(148089472)))]; + 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(149269184)))]; + 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, x = hidden_states_31_cast_fp16)[name = tensor("linear_31_cast_fp16")]; + tensor var_542 = const()[name = tensor("op_542"), val = tensor([1, -1, 12, 64])]; + tensor var_543_cast_fp16 = reshape(shape = var_542, x = linear_31_cast_fp16)[name = tensor("op_543_cast_fp16")]; + tensor var_544_perm_0 = const()[name = tensor("op_544_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor text_encoder_text_model_encoder_layers_5_self_attn_v_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_5_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(149270784)))]; + 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(150450496)))]; + 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, x = hidden_states_31_cast_fp16)[name = tensor("linear_32_cast_fp16")]; + tensor var_549 = const()[name = tensor("op_549"), val = tensor([1, -1, 12, 64])]; + tensor var_550_cast_fp16 = reshape(shape = var_549, x = linear_32_cast_fp16)[name = tensor("op_550_cast_fp16")]; + tensor var_551_perm_0 = const()[name = tensor("op_551_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor var_558 = const()[name = tensor("op_558"), val = tensor([1, 77, 12, 64])]; + tensor var_559_cast_fp16 = reshape(shape = var_558, x = tensor_35_cast_fp16)[name = tensor("op_559_cast_fp16")]; + tensor var_560_perm_0 = const()[name = tensor("op_560_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor var_562 = const()[name = tensor("op_562"), val = tensor([12, -1, 64])]; + tensor transpose_26 = transpose(perm = var_560_perm_0, x = var_559_cast_fp16)[name = tensor("transpose_26")]; + tensor query_states_11_cast_fp16 = reshape(shape = var_562, x = transpose_26)[name = tensor("query_states_11_cast_fp16")]; + tensor var_564 = const()[name = tensor("op_564"), val = tensor([12, -1, 64])]; + tensor transpose_28 = transpose(perm = var_544_perm_0, x = var_543_cast_fp16)[name = tensor("transpose_28")]; + tensor key_states_23_cast_fp16 = reshape(shape = var_564, x = transpose_28)[name = tensor("key_states_23_cast_fp16")]; + tensor var_566 = const()[name = tensor("op_566"), val = tensor([12, -1, 64])]; + tensor transpose_27 = transpose(perm = var_551_perm_0, x = var_550_cast_fp16)[name = tensor("transpose_27")]; + tensor value_states_23_cast_fp16 = reshape(shape = var_566, x = transpose_27)[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_571 = const()[name = tensor("op_571"), val = tensor([1, 12, 77, 77])]; + tensor var_572_cast_fp16 = reshape(shape = var_571, x = attn_weights_31_cast_fp16)[name = tensor("op_572_cast_fp16")]; + tensor attn_weights_33_cast_fp16 = add(x = var_572_cast_fp16, y = causal_attention_mask_to_fp16)[name = tensor("attn_weights_33_cast_fp16")]; + tensor var_577 = const()[name = tensor("op_577"), val = tensor([12, 77, 77])]; + tensor input_85_cast_fp16 = reshape(shape = var_577, 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_582 = const()[name = tensor("op_582"), val = tensor([1, 12, 77, 64])]; + tensor attn_output_33_cast_fp16 = reshape(shape = var_582, 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_585 = const()[name = tensor("op_585"), val = tensor([1, 77, 768])]; + tensor transpose_25 = transpose(perm = attn_output_35_perm_0, x = attn_output_33_cast_fp16)[name = tensor("transpose_25")]; + tensor input_89_cast_fp16 = reshape(shape = var_585, x = transpose_25)[name = tensor("input_89_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_5_self_attn_out_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_5_self_attn_out_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(150452096)))]; + 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(151631808)))]; + 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, 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(151633408)))]; + 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(151635008)))]; + 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_13_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 = const()[name = tensor("text_encoder_text_model_encoder_layers_5_mlp_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(151636608)))]; + 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(156355264)))]; + 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, x = input_93_cast_fp16)[name = tensor("linear_34_cast_fp16")]; + tensor var_600_to_fp16 = const()[name = tensor("op_600_to_fp16"), val = tensor(0x1.b3cp+0)]; + tensor var_601_cast_fp16 = mul(x = linear_34_cast_fp16, y = var_600_to_fp16)[name = tensor("op_601_cast_fp16")]; + tensor var_602_cast_fp16 = sigmoid(x = var_601_cast_fp16)[name = tensor("op_602_cast_fp16")]; + tensor input_97_cast_fp16 = mul(x = linear_34_cast_fp16, y = var_602_cast_fp16)[name = tensor("input_97_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_5_mlp_fc2_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_5_mlp_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(156361472)))]; + 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(161080128)))]; + 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, 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(161081728)))]; + 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(161083328)))]; + 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_13_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 = const()[name = tensor("text_encoder_text_model_encoder_layers_6_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(161084928)))]; + 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(162264640)))]; + 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, x = hidden_states_37_cast_fp16)[name = tensor("linear_36_cast_fp16")]; + tensor var_627_to_fp16 = const()[name = tensor("op_627_to_fp16"), val = tensor(0x1p-3)]; + tensor tensor_41_cast_fp16 = mul(x = linear_36_cast_fp16, y = var_627_to_fp16)[name = tensor("tensor_41_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_6_self_attn_k_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_6_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(162266240)))]; + 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(163445952)))]; + 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, x = hidden_states_37_cast_fp16)[name = tensor("linear_37_cast_fp16")]; + tensor var_632 = const()[name = tensor("op_632"), val = tensor([1, -1, 12, 64])]; + tensor var_633_cast_fp16 = reshape(shape = var_632, x = linear_37_cast_fp16)[name = tensor("op_633_cast_fp16")]; + tensor var_634_perm_0 = const()[name = tensor("op_634_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor text_encoder_text_model_encoder_layers_6_self_attn_v_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_6_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(163447552)))]; + 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(164627264)))]; + 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, x = hidden_states_37_cast_fp16)[name = tensor("linear_38_cast_fp16")]; + tensor var_639 = const()[name = tensor("op_639"), val = tensor([1, -1, 12, 64])]; + tensor var_640_cast_fp16 = reshape(shape = var_639, x = linear_38_cast_fp16)[name = tensor("op_640_cast_fp16")]; + tensor var_641_perm_0 = const()[name = tensor("op_641_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor var_648 = const()[name = tensor("op_648"), val = tensor([1, 77, 12, 64])]; + tensor var_649_cast_fp16 = reshape(shape = var_648, x = tensor_41_cast_fp16)[name = tensor("op_649_cast_fp16")]; + tensor var_650_perm_0 = const()[name = tensor("op_650_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor var_652 = const()[name = tensor("op_652"), val = tensor([12, -1, 64])]; + tensor transpose_22 = transpose(perm = var_650_perm_0, x = var_649_cast_fp16)[name = tensor("transpose_22")]; + tensor query_states_13_cast_fp16 = reshape(shape = var_652, x = transpose_22)[name = tensor("query_states_13_cast_fp16")]; + tensor var_654 = const()[name = tensor("op_654"), val = tensor([12, -1, 64])]; + tensor transpose_24 = transpose(perm = var_634_perm_0, x = var_633_cast_fp16)[name = tensor("transpose_24")]; + tensor key_states_27_cast_fp16 = reshape(shape = var_654, x = transpose_24)[name = tensor("key_states_27_cast_fp16")]; + tensor var_656 = const()[name = tensor("op_656"), val = tensor([12, -1, 64])]; + tensor transpose_23 = transpose(perm = var_641_perm_0, x = var_640_cast_fp16)[name = tensor("transpose_23")]; + tensor value_states_27_cast_fp16 = reshape(shape = var_656, x = transpose_23)[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_661 = const()[name = tensor("op_661"), val = tensor([1, 12, 77, 77])]; + tensor var_662_cast_fp16 = reshape(shape = var_661, x = attn_weights_37_cast_fp16)[name = tensor("op_662_cast_fp16")]; + tensor attn_weights_39_cast_fp16 = add(x = var_662_cast_fp16, y = causal_attention_mask_to_fp16)[name = tensor("attn_weights_39_cast_fp16")]; + tensor var_667 = const()[name = tensor("op_667"), val = tensor([12, 77, 77])]; + tensor input_101_cast_fp16 = reshape(shape = var_667, 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_672 = const()[name = tensor("op_672"), val = tensor([1, 12, 77, 64])]; + tensor attn_output_39_cast_fp16 = reshape(shape = var_672, 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_675 = const()[name = tensor("op_675"), val = tensor([1, 77, 768])]; + tensor transpose_21 = transpose(perm = attn_output_41_perm_0, x = attn_output_39_cast_fp16)[name = tensor("transpose_21")]; + tensor input_105_cast_fp16 = reshape(shape = var_675, x = transpose_21)[name = tensor("input_105_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_6_self_attn_out_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_6_self_attn_out_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(164628864)))]; + 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(165808576)))]; + 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, 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(165810176)))]; + 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(165811776)))]; + 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_13_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 = const()[name = tensor("text_encoder_text_model_encoder_layers_6_mlp_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(165813376)))]; + 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(170532032)))]; + 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, x = input_109_cast_fp16)[name = tensor("linear_40_cast_fp16")]; + tensor var_690_to_fp16 = const()[name = tensor("op_690_to_fp16"), val = tensor(0x1.b3cp+0)]; + tensor var_691_cast_fp16 = mul(x = linear_40_cast_fp16, y = var_690_to_fp16)[name = tensor("op_691_cast_fp16")]; + tensor var_692_cast_fp16 = sigmoid(x = var_691_cast_fp16)[name = tensor("op_692_cast_fp16")]; + tensor input_113_cast_fp16 = mul(x = linear_40_cast_fp16, y = var_692_cast_fp16)[name = tensor("input_113_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_6_mlp_fc2_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_6_mlp_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(170538240)))]; + 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(175256896)))]; + 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, 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(175258496)))]; + 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(175260096)))]; + 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_13_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 = const()[name = tensor("text_encoder_text_model_encoder_layers_7_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(175261696)))]; + 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(176441408)))]; + 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, x = hidden_states_43_cast_fp16)[name = tensor("linear_42_cast_fp16")]; + tensor var_717_to_fp16 = const()[name = tensor("op_717_to_fp16"), val = tensor(0x1p-3)]; + tensor tensor_47_cast_fp16 = mul(x = linear_42_cast_fp16, y = var_717_to_fp16)[name = tensor("tensor_47_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_7_self_attn_k_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_7_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(176443008)))]; + 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(177622720)))]; + 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, x = hidden_states_43_cast_fp16)[name = tensor("linear_43_cast_fp16")]; + tensor var_722 = const()[name = tensor("op_722"), val = tensor([1, -1, 12, 64])]; + tensor var_723_cast_fp16 = reshape(shape = var_722, x = linear_43_cast_fp16)[name = tensor("op_723_cast_fp16")]; + tensor var_724_perm_0 = const()[name = tensor("op_724_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor text_encoder_text_model_encoder_layers_7_self_attn_v_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_7_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(177624320)))]; + 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(178804032)))]; + 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, x = hidden_states_43_cast_fp16)[name = tensor("linear_44_cast_fp16")]; + tensor var_729 = const()[name = tensor("op_729"), val = tensor([1, -1, 12, 64])]; + tensor var_730_cast_fp16 = reshape(shape = var_729, x = linear_44_cast_fp16)[name = tensor("op_730_cast_fp16")]; + tensor var_731_perm_0 = const()[name = tensor("op_731_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor var_738 = const()[name = tensor("op_738"), val = tensor([1, 77, 12, 64])]; + tensor var_739_cast_fp16 = reshape(shape = var_738, x = tensor_47_cast_fp16)[name = tensor("op_739_cast_fp16")]; + tensor var_740_perm_0 = const()[name = tensor("op_740_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor var_742 = const()[name = tensor("op_742"), val = tensor([12, -1, 64])]; + tensor transpose_18 = transpose(perm = var_740_perm_0, x = var_739_cast_fp16)[name = tensor("transpose_18")]; + tensor query_states_15_cast_fp16 = reshape(shape = var_742, x = transpose_18)[name = tensor("query_states_15_cast_fp16")]; + tensor var_744 = const()[name = tensor("op_744"), val = tensor([12, -1, 64])]; + tensor transpose_20 = transpose(perm = var_724_perm_0, x = var_723_cast_fp16)[name = tensor("transpose_20")]; + tensor key_states_31_cast_fp16 = reshape(shape = var_744, x = transpose_20)[name = tensor("key_states_31_cast_fp16")]; + tensor var_746 = const()[name = tensor("op_746"), val = tensor([12, -1, 64])]; + tensor transpose_19 = transpose(perm = var_731_perm_0, x = var_730_cast_fp16)[name = tensor("transpose_19")]; + tensor value_states_31_cast_fp16 = reshape(shape = var_746, x = transpose_19)[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_751 = const()[name = tensor("op_751"), val = tensor([1, 12, 77, 77])]; + tensor var_752_cast_fp16 = reshape(shape = var_751, x = attn_weights_43_cast_fp16)[name = tensor("op_752_cast_fp16")]; + tensor attn_weights_45_cast_fp16 = add(x = var_752_cast_fp16, y = causal_attention_mask_to_fp16)[name = tensor("attn_weights_45_cast_fp16")]; + tensor var_757 = const()[name = tensor("op_757"), val = tensor([12, 77, 77])]; + tensor input_117_cast_fp16 = reshape(shape = var_757, 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_762 = const()[name = tensor("op_762"), val = tensor([1, 12, 77, 64])]; + tensor attn_output_45_cast_fp16 = reshape(shape = var_762, 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_765 = const()[name = tensor("op_765"), val = tensor([1, 77, 768])]; + tensor transpose_17 = transpose(perm = attn_output_47_perm_0, x = attn_output_45_cast_fp16)[name = tensor("transpose_17")]; + tensor input_121_cast_fp16 = reshape(shape = var_765, x = transpose_17)[name = tensor("input_121_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_7_self_attn_out_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_7_self_attn_out_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(178805632)))]; + 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(179985344)))]; + 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, 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(179986944)))]; + 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(179988544)))]; + 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_13_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 = const()[name = tensor("text_encoder_text_model_encoder_layers_7_mlp_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(179990144)))]; + 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(184708800)))]; + 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, x = input_125_cast_fp16)[name = tensor("linear_46_cast_fp16")]; + tensor var_780_to_fp16 = const()[name = tensor("op_780_to_fp16"), val = tensor(0x1.b3cp+0)]; + tensor var_781_cast_fp16 = mul(x = linear_46_cast_fp16, y = var_780_to_fp16)[name = tensor("op_781_cast_fp16")]; + tensor var_782_cast_fp16 = sigmoid(x = var_781_cast_fp16)[name = tensor("op_782_cast_fp16")]; + tensor input_129_cast_fp16 = mul(x = linear_46_cast_fp16, y = var_782_cast_fp16)[name = tensor("input_129_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_7_mlp_fc2_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_7_mlp_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(184715008)))]; + 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(189433664)))]; + 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, 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(189435264)))]; + 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(189436864)))]; + 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_13_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 = const()[name = tensor("text_encoder_text_model_encoder_layers_8_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(189438464)))]; + 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(190618176)))]; + 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, x = hidden_states_49_cast_fp16)[name = tensor("linear_48_cast_fp16")]; + tensor var_807_to_fp16 = const()[name = tensor("op_807_to_fp16"), val = tensor(0x1p-3)]; + tensor tensor_53_cast_fp16 = mul(x = linear_48_cast_fp16, y = var_807_to_fp16)[name = tensor("tensor_53_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_8_self_attn_k_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_8_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(190619776)))]; + 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(191799488)))]; + 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, x = hidden_states_49_cast_fp16)[name = tensor("linear_49_cast_fp16")]; + tensor var_812 = const()[name = tensor("op_812"), val = tensor([1, -1, 12, 64])]; + tensor var_813_cast_fp16 = reshape(shape = var_812, x = linear_49_cast_fp16)[name = tensor("op_813_cast_fp16")]; + tensor var_814_perm_0 = const()[name = tensor("op_814_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor text_encoder_text_model_encoder_layers_8_self_attn_v_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_8_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(191801088)))]; + 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(192980800)))]; + 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, x = hidden_states_49_cast_fp16)[name = tensor("linear_50_cast_fp16")]; + tensor var_819 = const()[name = tensor("op_819"), val = tensor([1, -1, 12, 64])]; + tensor var_820_cast_fp16 = reshape(shape = var_819, x = linear_50_cast_fp16)[name = tensor("op_820_cast_fp16")]; + tensor var_821_perm_0 = const()[name = tensor("op_821_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor var_828 = const()[name = tensor("op_828"), val = tensor([1, 77, 12, 64])]; + tensor var_829_cast_fp16 = reshape(shape = var_828, x = tensor_53_cast_fp16)[name = tensor("op_829_cast_fp16")]; + tensor var_830_perm_0 = const()[name = tensor("op_830_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor var_832 = const()[name = tensor("op_832"), val = tensor([12, -1, 64])]; + tensor transpose_14 = transpose(perm = var_830_perm_0, x = var_829_cast_fp16)[name = tensor("transpose_14")]; + tensor query_states_17_cast_fp16 = reshape(shape = var_832, x = transpose_14)[name = tensor("query_states_17_cast_fp16")]; + tensor var_834 = const()[name = tensor("op_834"), val = tensor([12, -1, 64])]; + tensor transpose_16 = transpose(perm = var_814_perm_0, x = var_813_cast_fp16)[name = tensor("transpose_16")]; + tensor key_states_35_cast_fp16 = reshape(shape = var_834, x = transpose_16)[name = tensor("key_states_35_cast_fp16")]; + tensor var_836 = const()[name = tensor("op_836"), val = tensor([12, -1, 64])]; + tensor transpose_15 = transpose(perm = var_821_perm_0, x = var_820_cast_fp16)[name = tensor("transpose_15")]; + tensor value_states_35_cast_fp16 = reshape(shape = var_836, x = transpose_15)[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_841 = const()[name = tensor("op_841"), val = tensor([1, 12, 77, 77])]; + tensor var_842_cast_fp16 = reshape(shape = var_841, x = attn_weights_49_cast_fp16)[name = tensor("op_842_cast_fp16")]; + tensor attn_weights_51_cast_fp16 = add(x = var_842_cast_fp16, y = causal_attention_mask_to_fp16)[name = tensor("attn_weights_51_cast_fp16")]; + tensor var_847 = const()[name = tensor("op_847"), val = tensor([12, 77, 77])]; + tensor input_133_cast_fp16 = reshape(shape = var_847, 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_852 = const()[name = tensor("op_852"), val = tensor([1, 12, 77, 64])]; + tensor attn_output_51_cast_fp16 = reshape(shape = var_852, 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_855 = const()[name = tensor("op_855"), val = tensor([1, 77, 768])]; + tensor transpose_13 = transpose(perm = attn_output_53_perm_0, x = attn_output_51_cast_fp16)[name = tensor("transpose_13")]; + tensor input_137_cast_fp16 = reshape(shape = var_855, x = transpose_13)[name = tensor("input_137_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_8_self_attn_out_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_8_self_attn_out_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(192982400)))]; + 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(194162112)))]; + 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, 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(194163712)))]; + 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(194165312)))]; + 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_13_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 = const()[name = tensor("text_encoder_text_model_encoder_layers_8_mlp_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(194166912)))]; + 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(198885568)))]; + 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, x = input_141_cast_fp16)[name = tensor("linear_52_cast_fp16")]; + tensor var_870_to_fp16 = const()[name = tensor("op_870_to_fp16"), val = tensor(0x1.b3cp+0)]; + tensor var_871_cast_fp16 = mul(x = linear_52_cast_fp16, y = var_870_to_fp16)[name = tensor("op_871_cast_fp16")]; + tensor var_872_cast_fp16 = sigmoid(x = var_871_cast_fp16)[name = tensor("op_872_cast_fp16")]; + tensor input_145_cast_fp16 = mul(x = linear_52_cast_fp16, y = var_872_cast_fp16)[name = tensor("input_145_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_8_mlp_fc2_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_8_mlp_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(198891776)))]; + 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(203610432)))]; + 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, 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(203612032)))]; + 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(203613632)))]; + 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_13_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 = const()[name = tensor("text_encoder_text_model_encoder_layers_9_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(203615232)))]; + 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(204794944)))]; + 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, x = hidden_states_55_cast_fp16)[name = tensor("linear_54_cast_fp16")]; + tensor var_897_to_fp16 = const()[name = tensor("op_897_to_fp16"), val = tensor(0x1p-3)]; + tensor tensor_59_cast_fp16 = mul(x = linear_54_cast_fp16, y = var_897_to_fp16)[name = tensor("tensor_59_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_9_self_attn_k_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_9_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(204796544)))]; + 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(205976256)))]; + 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, x = hidden_states_55_cast_fp16)[name = tensor("linear_55_cast_fp16")]; + tensor var_902 = const()[name = tensor("op_902"), val = tensor([1, -1, 12, 64])]; + tensor var_903_cast_fp16 = reshape(shape = var_902, x = linear_55_cast_fp16)[name = tensor("op_903_cast_fp16")]; + tensor var_904_perm_0 = const()[name = tensor("op_904_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor text_encoder_text_model_encoder_layers_9_self_attn_v_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_9_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(205977856)))]; + 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(207157568)))]; + 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, x = hidden_states_55_cast_fp16)[name = tensor("linear_56_cast_fp16")]; + tensor var_909 = const()[name = tensor("op_909"), val = tensor([1, -1, 12, 64])]; + tensor var_910_cast_fp16 = reshape(shape = var_909, x = linear_56_cast_fp16)[name = tensor("op_910_cast_fp16")]; + tensor var_911_perm_0 = const()[name = tensor("op_911_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor var_918 = const()[name = tensor("op_918"), val = tensor([1, 77, 12, 64])]; + tensor var_919_cast_fp16 = reshape(shape = var_918, x = tensor_59_cast_fp16)[name = tensor("op_919_cast_fp16")]; + tensor var_920_perm_0 = const()[name = tensor("op_920_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor var_922 = const()[name = tensor("op_922"), val = tensor([12, -1, 64])]; + tensor transpose_10 = transpose(perm = var_920_perm_0, x = var_919_cast_fp16)[name = tensor("transpose_10")]; + tensor query_states_19_cast_fp16 = reshape(shape = var_922, x = transpose_10)[name = tensor("query_states_19_cast_fp16")]; + tensor var_924 = const()[name = tensor("op_924"), val = tensor([12, -1, 64])]; + tensor transpose_12 = transpose(perm = var_904_perm_0, x = var_903_cast_fp16)[name = tensor("transpose_12")]; + tensor key_states_39_cast_fp16 = reshape(shape = var_924, x = transpose_12)[name = tensor("key_states_39_cast_fp16")]; + tensor var_926 = const()[name = tensor("op_926"), val = tensor([12, -1, 64])]; + tensor transpose_11 = transpose(perm = var_911_perm_0, x = var_910_cast_fp16)[name = tensor("transpose_11")]; + tensor value_states_39_cast_fp16 = reshape(shape = var_926, x = transpose_11)[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_931 = const()[name = tensor("op_931"), val = tensor([1, 12, 77, 77])]; + tensor var_932_cast_fp16 = reshape(shape = var_931, x = attn_weights_55_cast_fp16)[name = tensor("op_932_cast_fp16")]; + tensor attn_weights_57_cast_fp16 = add(x = var_932_cast_fp16, y = causal_attention_mask_to_fp16)[name = tensor("attn_weights_57_cast_fp16")]; + tensor var_937 = const()[name = tensor("op_937"), val = tensor([12, 77, 77])]; + tensor input_149_cast_fp16 = reshape(shape = var_937, 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_942 = const()[name = tensor("op_942"), val = tensor([1, 12, 77, 64])]; + tensor attn_output_57_cast_fp16 = reshape(shape = var_942, 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_945 = const()[name = tensor("op_945"), val = tensor([1, 77, 768])]; + tensor transpose_9 = transpose(perm = attn_output_59_perm_0, x = attn_output_57_cast_fp16)[name = tensor("transpose_9")]; + tensor input_153_cast_fp16 = reshape(shape = var_945, x = transpose_9)[name = tensor("input_153_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_9_self_attn_out_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_9_self_attn_out_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(207159168)))]; + 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(208338880)))]; + 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, 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(208340480)))]; + 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(208342080)))]; + 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_13_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 = const()[name = tensor("text_encoder_text_model_encoder_layers_9_mlp_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(208343680)))]; + 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(213062336)))]; + 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, x = input_157_cast_fp16)[name = tensor("linear_58_cast_fp16")]; + tensor var_960_to_fp16 = const()[name = tensor("op_960_to_fp16"), val = tensor(0x1.b3cp+0)]; + tensor var_961_cast_fp16 = mul(x = linear_58_cast_fp16, y = var_960_to_fp16)[name = tensor("op_961_cast_fp16")]; + tensor var_962_cast_fp16 = sigmoid(x = var_961_cast_fp16)[name = tensor("op_962_cast_fp16")]; + tensor input_161_cast_fp16 = mul(x = linear_58_cast_fp16, y = var_962_cast_fp16)[name = tensor("input_161_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_9_mlp_fc2_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_9_mlp_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(213068544)))]; + 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(217787200)))]; + 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, 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(217788800)))]; + 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(217790400)))]; + 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_13_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 = const()[name = tensor("text_encoder_text_model_encoder_layers_10_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(217792000)))]; + 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(218971712)))]; + 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, x = hidden_states_61_cast_fp16)[name = tensor("linear_60_cast_fp16")]; + tensor var_987_to_fp16 = const()[name = tensor("op_987_to_fp16"), val = tensor(0x1p-3)]; + tensor tensor_65_cast_fp16 = mul(x = linear_60_cast_fp16, y = var_987_to_fp16)[name = tensor("tensor_65_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_10_self_attn_k_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_10_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(218973312)))]; + 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(220153024)))]; + 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, x = hidden_states_61_cast_fp16)[name = tensor("linear_61_cast_fp16")]; + tensor var_992 = const()[name = tensor("op_992"), val = tensor([1, -1, 12, 64])]; + tensor var_993_cast_fp16 = reshape(shape = var_992, x = linear_61_cast_fp16)[name = tensor("op_993_cast_fp16")]; + tensor var_994_perm_0 = const()[name = tensor("op_994_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor text_encoder_text_model_encoder_layers_10_self_attn_v_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_10_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(220154624)))]; + 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(221334336)))]; + 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, x = hidden_states_61_cast_fp16)[name = tensor("linear_62_cast_fp16")]; + tensor var_999 = const()[name = tensor("op_999"), val = tensor([1, -1, 12, 64])]; + tensor var_1000_cast_fp16 = reshape(shape = var_999, x = linear_62_cast_fp16)[name = tensor("op_1000_cast_fp16")]; + tensor var_1001_perm_0 = const()[name = tensor("op_1001_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor var_1008 = const()[name = tensor("op_1008"), val = tensor([1, 77, 12, 64])]; + tensor var_1009_cast_fp16 = reshape(shape = var_1008, x = tensor_65_cast_fp16)[name = tensor("op_1009_cast_fp16")]; + tensor var_1010_perm_0 = const()[name = tensor("op_1010_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor var_1012 = const()[name = tensor("op_1012"), val = tensor([12, -1, 64])]; + tensor transpose_6 = transpose(perm = var_1010_perm_0, x = var_1009_cast_fp16)[name = tensor("transpose_6")]; + tensor query_states_21_cast_fp16 = reshape(shape = var_1012, x = transpose_6)[name = tensor("query_states_21_cast_fp16")]; + tensor var_1014 = const()[name = tensor("op_1014"), val = tensor([12, -1, 64])]; + tensor transpose_8 = transpose(perm = var_994_perm_0, x = var_993_cast_fp16)[name = tensor("transpose_8")]; + tensor key_states_43_cast_fp16 = reshape(shape = var_1014, x = transpose_8)[name = tensor("key_states_43_cast_fp16")]; + tensor var_1016 = const()[name = tensor("op_1016"), val = tensor([12, -1, 64])]; + tensor transpose_7 = transpose(perm = var_1001_perm_0, x = var_1000_cast_fp16)[name = tensor("transpose_7")]; + tensor value_states_43_cast_fp16 = reshape(shape = var_1016, x = transpose_7)[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_1021 = const()[name = tensor("op_1021"), val = tensor([1, 12, 77, 77])]; + tensor var_1022_cast_fp16 = reshape(shape = var_1021, x = attn_weights_61_cast_fp16)[name = tensor("op_1022_cast_fp16")]; + tensor attn_weights_63_cast_fp16 = add(x = var_1022_cast_fp16, y = causal_attention_mask_to_fp16)[name = tensor("attn_weights_63_cast_fp16")]; + tensor var_1027 = const()[name = tensor("op_1027"), val = tensor([12, 77, 77])]; + tensor input_165_cast_fp16 = reshape(shape = var_1027, 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_1032 = const()[name = tensor("op_1032"), val = tensor([1, 12, 77, 64])]; + tensor attn_output_63_cast_fp16 = reshape(shape = var_1032, 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_1035 = const()[name = tensor("op_1035"), val = tensor([1, 77, 768])]; + tensor transpose_5 = transpose(perm = attn_output_65_perm_0, x = attn_output_63_cast_fp16)[name = tensor("transpose_5")]; + tensor input_169_cast_fp16 = reshape(shape = var_1035, x = transpose_5)[name = tensor("input_169_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_10_self_attn_out_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_10_self_attn_out_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(221335936)))]; + 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(222515648)))]; + 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, 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(222517248)))]; + 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(222518848)))]; + 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_13_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 = const()[name = tensor("text_encoder_text_model_encoder_layers_10_mlp_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(222520448)))]; + 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(227239104)))]; + 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, x = input_173_cast_fp16)[name = tensor("linear_64_cast_fp16")]; + tensor var_1050_to_fp16 = const()[name = tensor("op_1050_to_fp16"), val = tensor(0x1.b3cp+0)]; + tensor var_1051_cast_fp16 = mul(x = linear_64_cast_fp16, y = var_1050_to_fp16)[name = tensor("op_1051_cast_fp16")]; + tensor var_1052_cast_fp16 = sigmoid(x = var_1051_cast_fp16)[name = tensor("op_1052_cast_fp16")]; + tensor input_177_cast_fp16 = mul(x = linear_64_cast_fp16, y = var_1052_cast_fp16)[name = tensor("input_177_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_10_mlp_fc2_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_10_mlp_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(227245312)))]; + 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(231963968)))]; + 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, 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 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(231965568)))]; + 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(231967168)))]; + 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_13_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 = const()[name = tensor("text_encoder_text_model_encoder_layers_11_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(231968768)))]; + 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(233148480)))]; + 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, x = hidden_states_67_cast_fp16)[name = tensor("linear_66_cast_fp16")]; + tensor var_1077_to_fp16 = const()[name = tensor("op_1077_to_fp16"), val = tensor(0x1p-3)]; + tensor tensor_cast_fp16 = mul(x = linear_66_cast_fp16, y = var_1077_to_fp16)[name = tensor("tensor_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_11_self_attn_k_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_11_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(233150080)))]; + 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(234329792)))]; + 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, x = hidden_states_67_cast_fp16)[name = tensor("linear_67_cast_fp16")]; + tensor var_1082 = const()[name = tensor("op_1082"), val = tensor([1, -1, 12, 64])]; + tensor var_1083_cast_fp16 = reshape(shape = var_1082, x = linear_67_cast_fp16)[name = tensor("op_1083_cast_fp16")]; + tensor var_1084_perm_0 = const()[name = tensor("op_1084_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor text_encoder_text_model_encoder_layers_11_self_attn_v_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_11_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(234331392)))]; + 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(235511104)))]; + 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, x = hidden_states_67_cast_fp16)[name = tensor("linear_68_cast_fp16")]; + tensor var_1089 = const()[name = tensor("op_1089"), val = tensor([1, -1, 12, 64])]; + tensor var_1090_cast_fp16 = reshape(shape = var_1089, x = linear_68_cast_fp16)[name = tensor("op_1090_cast_fp16")]; + tensor var_1091_perm_0 = const()[name = tensor("op_1091_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor var_1098 = const()[name = tensor("op_1098"), val = tensor([1, 77, 12, 64])]; + tensor var_1099_cast_fp16 = reshape(shape = var_1098, x = tensor_cast_fp16)[name = tensor("op_1099_cast_fp16")]; + tensor var_1100_perm_0 = const()[name = tensor("op_1100_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor var_1102 = const()[name = tensor("op_1102"), val = tensor([12, -1, 64])]; + tensor transpose_2 = transpose(perm = var_1100_perm_0, x = var_1099_cast_fp16)[name = tensor("transpose_2")]; + tensor query_states_cast_fp16 = reshape(shape = var_1102, x = transpose_2)[name = tensor("query_states_cast_fp16")]; + tensor var_1104 = const()[name = tensor("op_1104"), val = tensor([12, -1, 64])]; + tensor transpose_4 = transpose(perm = var_1084_perm_0, x = var_1083_cast_fp16)[name = tensor("transpose_4")]; + tensor key_states_cast_fp16 = reshape(shape = var_1104, x = transpose_4)[name = tensor("key_states_cast_fp16")]; + tensor var_1106 = const()[name = tensor("op_1106"), val = tensor([12, -1, 64])]; + tensor transpose_3 = transpose(perm = var_1091_perm_0, x = var_1090_cast_fp16)[name = tensor("transpose_3")]; + tensor value_states_cast_fp16 = reshape(shape = var_1106, x = transpose_3)[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_1111 = const()[name = tensor("op_1111"), val = tensor([1, 12, 77, 77])]; + tensor var_1112_cast_fp16 = reshape(shape = var_1111, x = attn_weights_67_cast_fp16)[name = tensor("op_1112_cast_fp16")]; + tensor attn_weights_69_cast_fp16 = add(x = var_1112_cast_fp16, y = causal_attention_mask_to_fp16)[name = tensor("attn_weights_69_cast_fp16")]; + tensor var_1117 = const()[name = tensor("op_1117"), val = tensor([12, 77, 77])]; + tensor input_181_cast_fp16 = reshape(shape = var_1117, 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_1122 = const()[name = tensor("op_1122"), val = tensor([1, 12, 77, 64])]; + tensor attn_output_69_cast_fp16 = reshape(shape = var_1122, 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_1125 = const()[name = tensor("op_1125"), val = tensor([1, 77, 768])]; + tensor transpose_1 = transpose(perm = attn_output_perm_0, x = attn_output_69_cast_fp16)[name = tensor("transpose_1")]; + tensor input_185_cast_fp16 = reshape(shape = var_1125, x = transpose_1)[name = tensor("input_185_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_11_self_attn_out_proj_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_11_self_attn_out_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(235512704)))]; + 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(236692416)))]; + 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, 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(236694016)))]; + 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(236695616)))]; + 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_13_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 = const()[name = tensor("text_encoder_text_model_encoder_layers_11_mlp_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(236697216)))]; + 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(241415872)))]; + 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, x = input_189_cast_fp16)[name = tensor("linear_70_cast_fp16")]; + tensor var_1140_to_fp16 = const()[name = tensor("op_1140_to_fp16"), val = tensor(0x1.b3cp+0)]; + tensor var_1141_cast_fp16 = mul(x = linear_70_cast_fp16, y = var_1140_to_fp16)[name = tensor("op_1141_cast_fp16")]; + tensor var_1142_cast_fp16 = sigmoid(x = var_1141_cast_fp16)[name = tensor("op_1142_cast_fp16")]; + tensor input_193_cast_fp16 = mul(x = linear_70_cast_fp16, y = var_1142_cast_fp16)[name = tensor("input_193_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_11_mlp_fc2_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_11_mlp_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(241422080)))]; + 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(246140736)))]; + 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, 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(246142336)))]; + 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(246143936)))]; + 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_13_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 last_hidden_state_cast_fp16_to_fp32_dtype_0 = const()[name = tensor("last_hidden_state_cast_fp16_to_fp32_dtype_0"), val = tensor("fp32")]; + tensor var_1153 = const()[name = tensor("op_1153"), val = tensor([0])]; + tensor var_1155 = reduce_argmax(axis = var_5, keep_dims = var_6, x = cast_127)[name = tensor("op_1155")]; + 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_1153, var_1155))[name = tensor("stack_0")]; + tensor var_1157_transpose_batch_dims_0 = const()[name = tensor("op_1157_transpose_batch_dims_0"), val = tensor(0)]; + tensor var_1157_transpose_cast_fp16 = gather_nd(batch_dims = var_1157_transpose_batch_dims_0, indices = stack_0, x = last_hidden_state_cast_fp16)[name = tensor("op_1157_transpose_cast_fp16")]; + tensor var_1157_cast_fp16_to_fp32_dtype_0 = const()[name = tensor("op_1157_cast_fp16_to_fp32_dtype_0"), val = tensor("fp32")]; + tensor pooled_outputs = cast(dtype = var_1157_cast_fp16_to_fp32_dtype_0, x = var_1157_transpose_cast_fp16)[name = tensor("cast_125")]; + tensor last_hidden_state = cast(dtype = last_hidden_state_cast_fp16_to_fp32_dtype_0, x = last_hidden_state_cast_fp16)[name = tensor("cast_126")]; + } -> (last_hidden_state, pooled_outputs); +} \ No newline at end of file