| program(1.0) | |
| [buildInfo = dict<tensor<string, []>, tensor<string, []>>({{"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<ios16>(tensor<fp32, [1, 77]> input_ids) { | |
| tensor<int32, []> var_5 = const()[name = tensor<string, []>("op_5"), val = tensor<int32, []>(-1)]; | |
| tensor<bool, []> var_6 = const()[name = tensor<string, []>("op_6"), val = tensor<bool, []>(false)]; | |
| tensor<string, []> cast_1_dtype_0 = const()[name = tensor<string, []>("cast_1_dtype_0"), val = tensor<string, []>("int32")]; | |
| tensor<int32, []> inputs_embeds_axis_0 = const()[name = tensor<string, []>("inputs_embeds_axis_0"), val = tensor<int32, []>(0)]; | |
| tensor<int32, []> inputs_embeds_batch_dims_0 = const()[name = tensor<string, []>("inputs_embeds_batch_dims_0"), val = tensor<int32, []>(0)]; | |
| tensor<fp16, [49408, 768]> text_encoder_text_model_embeddings_token_embedding_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_embeddings_token_embedding_weight_to_fp16"), val = tensor<fp16, [49408, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(64)))]; | |
| tensor<int32, [1, 77]> cast_127 = cast(dtype = cast_1_dtype_0, x = input_ids)[name = tensor<string, []>("cast_127")]; | |
| tensor<fp16, [1, 77, 768]> 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<string, []>("inputs_embeds_cast_fp16")]; | |
| tensor<fp16, [1, 77, 768]> position_embeddings_to_fp16 = const()[name = tensor<string, []>("position_embeddings_to_fp16"), val = tensor<fp16, [1, 77, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(75890816)))]; | |
| tensor<fp16, [1, 77, 768]> input_3_cast_fp16 = add(x = inputs_embeds_cast_fp16, y = position_embeddings_to_fp16)[name = tensor<string, []>("input_3_cast_fp16")]; | |
| tensor<int32, [1]> hidden_states_1_axes_0 = const()[name = tensor<string, []>("hidden_states_1_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_0_layer_norm1_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_0_layer_norm1_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(76009152)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_0_layer_norm1_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_0_layer_norm1_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(76010752)))]; | |
| tensor<fp16, []> var_13_to_fp16 = const()[name = tensor<string, []>("op_13_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; | |
| tensor<fp16, [1, 77, 768]> 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<string, []>("hidden_states_1_cast_fp16")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_0_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_0_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(76012352)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_0_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_0_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(77192064)))]; | |
| tensor<fp16, [1, 77, 768]> 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<string, []>("linear_0_cast_fp16")]; | |
| tensor<fp16, []> var_87_to_fp16 = const()[name = tensor<string, []>("op_87_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; | |
| tensor<fp16, [1, 77, 768]> tensor_5_cast_fp16 = mul(x = linear_0_cast_fp16, y = var_87_to_fp16)[name = tensor<string, []>("tensor_5_cast_fp16")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_0_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_0_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(77193664)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_0_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_0_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(78373376)))]; | |
| tensor<fp16, [1, 77, 768]> 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<string, []>("linear_1_cast_fp16")]; | |
| tensor<int32, [4]> var_92 = const()[name = tensor<string, []>("op_92"), val = tensor<int32, [4]>([1, -1, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_93_cast_fp16 = reshape(shape = var_92, x = linear_1_cast_fp16)[name = tensor<string, []>("op_93_cast_fp16")]; | |
| tensor<int32, [4]> var_94_perm_0 = const()[name = tensor<string, []>("op_94_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_0_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_0_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(78374976)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_0_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_0_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(79554688)))]; | |
| tensor<fp16, [1, 77, 768]> 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<string, []>("linear_2_cast_fp16")]; | |
| tensor<int32, [4]> var_99 = const()[name = tensor<string, []>("op_99"), val = tensor<int32, [4]>([1, -1, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_100_cast_fp16 = reshape(shape = var_99, x = linear_2_cast_fp16)[name = tensor<string, []>("op_100_cast_fp16")]; | |
| tensor<int32, [4]> var_101_perm_0 = const()[name = tensor<string, []>("op_101_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [4]> var_108 = const()[name = tensor<string, []>("op_108"), val = tensor<int32, [4]>([1, 77, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_109_cast_fp16 = reshape(shape = var_108, x = tensor_5_cast_fp16)[name = tensor<string, []>("op_109_cast_fp16")]; | |
| tensor<int32, [4]> var_110_perm_0 = const()[name = tensor<string, []>("op_110_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> var_112 = const()[name = tensor<string, []>("op_112"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> transpose_46 = transpose(perm = var_110_perm_0, x = var_109_cast_fp16)[name = tensor<string, []>("transpose_46")]; | |
| tensor<fp16, [12, 77, 64]> query_states_1_cast_fp16 = reshape(shape = var_112, x = transpose_46)[name = tensor<string, []>("query_states_1_cast_fp16")]; | |
| tensor<int32, [3]> var_114 = const()[name = tensor<string, []>("op_114"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> transpose_48 = transpose(perm = var_94_perm_0, x = var_93_cast_fp16)[name = tensor<string, []>("transpose_48")]; | |
| tensor<fp16, [12, 77, 64]> key_states_3_cast_fp16 = reshape(shape = var_114, x = transpose_48)[name = tensor<string, []>("key_states_3_cast_fp16")]; | |
| tensor<int32, [3]> var_116 = const()[name = tensor<string, []>("op_116"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> transpose_47 = transpose(perm = var_101_perm_0, x = var_100_cast_fp16)[name = tensor<string, []>("transpose_47")]; | |
| tensor<fp16, [12, 77, 64]> value_states_3_cast_fp16 = reshape(shape = var_116, x = transpose_47)[name = tensor<string, []>("value_states_3_cast_fp16")]; | |
| tensor<bool, []> attn_weights_1_transpose_x_1 = const()[name = tensor<string, []>("attn_weights_1_transpose_x_1"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> attn_weights_1_transpose_y_1 = const()[name = tensor<string, []>("attn_weights_1_transpose_y_1"), val = tensor<bool, []>(true)]; | |
| tensor<fp16, [12, 77, 77]> 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<string, []>("attn_weights_1_cast_fp16")]; | |
| tensor<int32, [4]> var_121 = const()[name = tensor<string, []>("op_121"), val = tensor<int32, [4]>([1, 12, 77, 77])]; | |
| tensor<fp16, [1, 12, 77, 77]> var_122_cast_fp16 = reshape(shape = var_121, x = attn_weights_1_cast_fp16)[name = tensor<string, []>("op_122_cast_fp16")]; | |
| tensor<fp16, [1, 1, 77, 77]> causal_attention_mask_to_fp16 = const()[name = tensor<string, []>("causal_attention_mask_to_fp16"), val = tensor<fp16, [1, 1, 77, 77]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(79556288)))]; | |
| tensor<fp16, [1, 12, 77, 77]> attn_weights_3_cast_fp16 = add(x = var_122_cast_fp16, y = causal_attention_mask_to_fp16)[name = tensor<string, []>("attn_weights_3_cast_fp16")]; | |
| tensor<int32, [3]> var_127 = const()[name = tensor<string, []>("op_127"), val = tensor<int32, [3]>([12, 77, 77])]; | |
| tensor<fp16, [12, 77, 77]> input_5_cast_fp16 = reshape(shape = var_127, x = attn_weights_3_cast_fp16)[name = tensor<string, []>("input_5_cast_fp16")]; | |
| tensor<fp16, [12, 77, 77]> input_7_cast_fp16 = softmax(axis = var_5, x = input_5_cast_fp16)[name = tensor<string, []>("input_7_cast_fp16")]; | |
| tensor<bool, []> attn_output_1_transpose_x_0 = const()[name = tensor<string, []>("attn_output_1_transpose_x_0"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> attn_output_1_transpose_y_0 = const()[name = tensor<string, []>("attn_output_1_transpose_y_0"), val = tensor<bool, []>(false)]; | |
| tensor<fp16, [12, 77, 64]> 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<string, []>("attn_output_1_cast_fp16")]; | |
| tensor<int32, [4]> var_132 = const()[name = tensor<string, []>("op_132"), val = tensor<int32, [4]>([1, 12, 77, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> attn_output_3_cast_fp16 = reshape(shape = var_132, x = attn_output_1_cast_fp16)[name = tensor<string, []>("attn_output_3_cast_fp16")]; | |
| tensor<int32, [4]> attn_output_5_perm_0 = const()[name = tensor<string, []>("attn_output_5_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> var_135 = const()[name = tensor<string, []>("op_135"), val = tensor<int32, [3]>([1, 77, 768])]; | |
| tensor<fp16, [1, 77, 12, 64]> transpose_45 = transpose(perm = attn_output_5_perm_0, x = attn_output_3_cast_fp16)[name = tensor<string, []>("transpose_45")]; | |
| tensor<fp16, [1, 77, 768]> input_9_cast_fp16 = reshape(shape = var_135, x = transpose_45)[name = tensor<string, []>("input_9_cast_fp16")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_0_self_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_0_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(79568256)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_0_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_0_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(80747968)))]; | |
| tensor<fp16, [1, 77, 768]> 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<string, []>("linear_3_cast_fp16")]; | |
| tensor<fp16, [1, 77, 768]> input_11_cast_fp16 = add(x = input_3_cast_fp16, y = linear_3_cast_fp16)[name = tensor<string, []>("input_11_cast_fp16")]; | |
| tensor<int32, [1]> input_13_axes_0 = const()[name = tensor<string, []>("input_13_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_0_layer_norm2_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_0_layer_norm2_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(80749568)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_0_layer_norm2_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_0_layer_norm2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(80751168)))]; | |
| tensor<fp16, [1, 77, 768]> 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<string, []>("input_13_cast_fp16")]; | |
| tensor<fp16, [3072, 768]> text_encoder_text_model_encoder_layers_0_mlp_fc1_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_0_mlp_fc1_weight_to_fp16"), val = tensor<fp16, [3072, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(80752768)))]; | |
| tensor<fp16, [3072]> text_encoder_text_model_encoder_layers_0_mlp_fc1_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_0_mlp_fc1_bias_to_fp16"), val = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(85471424)))]; | |
| tensor<fp16, [1, 77, 3072]> 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<string, []>("linear_4_cast_fp16")]; | |
| tensor<fp16, []> var_150_to_fp16 = const()[name = tensor<string, []>("op_150_to_fp16"), val = tensor<fp16, []>(0x1.b3cp+0)]; | |
| tensor<fp16, [1, 77, 3072]> var_151_cast_fp16 = mul(x = linear_4_cast_fp16, y = var_150_to_fp16)[name = tensor<string, []>("op_151_cast_fp16")]; | |
| tensor<fp16, [1, 77, 3072]> var_152_cast_fp16 = sigmoid(x = var_151_cast_fp16)[name = tensor<string, []>("op_152_cast_fp16")]; | |
| tensor<fp16, [1, 77, 3072]> input_17_cast_fp16 = mul(x = linear_4_cast_fp16, y = var_152_cast_fp16)[name = tensor<string, []>("input_17_cast_fp16")]; | |
| tensor<fp16, [768, 3072]> text_encoder_text_model_encoder_layers_0_mlp_fc2_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_0_mlp_fc2_weight_to_fp16"), val = tensor<fp16, [768, 3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(85477632)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_0_mlp_fc2_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_0_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(90196288)))]; | |
| tensor<fp16, [1, 77, 768]> 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<string, []>("linear_5_cast_fp16")]; | |
| tensor<fp16, [1, 77, 768]> input_19_cast_fp16 = add(x = input_11_cast_fp16, y = linear_5_cast_fp16)[name = tensor<string, []>("input_19_cast_fp16")]; | |
| tensor<int32, [1]> hidden_states_7_axes_0 = const()[name = tensor<string, []>("hidden_states_7_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_1_layer_norm1_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_1_layer_norm1_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(90197888)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_1_layer_norm1_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_1_layer_norm1_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(90199488)))]; | |
| tensor<fp16, [1, 77, 768]> 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<string, []>("hidden_states_7_cast_fp16")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_1_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_1_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(90201088)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_1_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_1_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(91380800)))]; | |
| tensor<fp16, [1, 77, 768]> 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<string, []>("linear_6_cast_fp16")]; | |
| tensor<fp16, []> var_177_to_fp16 = const()[name = tensor<string, []>("op_177_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; | |
| tensor<fp16, [1, 77, 768]> tensor_11_cast_fp16 = mul(x = linear_6_cast_fp16, y = var_177_to_fp16)[name = tensor<string, []>("tensor_11_cast_fp16")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_1_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_1_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(91382400)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_1_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_1_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(92562112)))]; | |
| tensor<fp16, [1, 77, 768]> 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<string, []>("linear_7_cast_fp16")]; | |
| tensor<int32, [4]> var_182 = const()[name = tensor<string, []>("op_182"), val = tensor<int32, [4]>([1, -1, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_183_cast_fp16 = reshape(shape = var_182, x = linear_7_cast_fp16)[name = tensor<string, []>("op_183_cast_fp16")]; | |
| tensor<int32, [4]> var_184_perm_0 = const()[name = tensor<string, []>("op_184_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_1_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_1_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(92563712)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_1_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_1_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(93743424)))]; | |
| tensor<fp16, [1, 77, 768]> 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<string, []>("linear_8_cast_fp16")]; | |
| tensor<int32, [4]> var_189 = const()[name = tensor<string, []>("op_189"), val = tensor<int32, [4]>([1, -1, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_190_cast_fp16 = reshape(shape = var_189, x = linear_8_cast_fp16)[name = tensor<string, []>("op_190_cast_fp16")]; | |
| tensor<int32, [4]> var_191_perm_0 = const()[name = tensor<string, []>("op_191_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [4]> var_198 = const()[name = tensor<string, []>("op_198"), val = tensor<int32, [4]>([1, 77, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_199_cast_fp16 = reshape(shape = var_198, x = tensor_11_cast_fp16)[name = tensor<string, []>("op_199_cast_fp16")]; | |
| tensor<int32, [4]> var_200_perm_0 = const()[name = tensor<string, []>("op_200_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> var_202 = const()[name = tensor<string, []>("op_202"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> transpose_42 = transpose(perm = var_200_perm_0, x = var_199_cast_fp16)[name = tensor<string, []>("transpose_42")]; | |
| tensor<fp16, [12, 77, 64]> query_states_3_cast_fp16 = reshape(shape = var_202, x = transpose_42)[name = tensor<string, []>("query_states_3_cast_fp16")]; | |
| tensor<int32, [3]> var_204 = const()[name = tensor<string, []>("op_204"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> transpose_44 = transpose(perm = var_184_perm_0, x = var_183_cast_fp16)[name = tensor<string, []>("transpose_44")]; | |
| tensor<fp16, [12, 77, 64]> key_states_7_cast_fp16 = reshape(shape = var_204, x = transpose_44)[name = tensor<string, []>("key_states_7_cast_fp16")]; | |
| tensor<int32, [3]> var_206 = const()[name = tensor<string, []>("op_206"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> transpose_43 = transpose(perm = var_191_perm_0, x = var_190_cast_fp16)[name = tensor<string, []>("transpose_43")]; | |
| tensor<fp16, [12, 77, 64]> value_states_7_cast_fp16 = reshape(shape = var_206, x = transpose_43)[name = tensor<string, []>("value_states_7_cast_fp16")]; | |
| tensor<bool, []> attn_weights_7_transpose_x_1 = const()[name = tensor<string, []>("attn_weights_7_transpose_x_1"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> attn_weights_7_transpose_y_1 = const()[name = tensor<string, []>("attn_weights_7_transpose_y_1"), val = tensor<bool, []>(true)]; | |
| tensor<fp16, [12, 77, 77]> 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<string, []>("attn_weights_7_cast_fp16")]; | |
| tensor<int32, [4]> var_211 = const()[name = tensor<string, []>("op_211"), val = tensor<int32, [4]>([1, 12, 77, 77])]; | |
| tensor<fp16, [1, 12, 77, 77]> var_212_cast_fp16 = reshape(shape = var_211, x = attn_weights_7_cast_fp16)[name = tensor<string, []>("op_212_cast_fp16")]; | |
| tensor<fp16, [1, 12, 77, 77]> attn_weights_9_cast_fp16 = add(x = var_212_cast_fp16, y = causal_attention_mask_to_fp16)[name = tensor<string, []>("attn_weights_9_cast_fp16")]; | |
| tensor<int32, [3]> var_217 = const()[name = tensor<string, []>("op_217"), val = tensor<int32, [3]>([12, 77, 77])]; | |
| tensor<fp16, [12, 77, 77]> input_21_cast_fp16 = reshape(shape = var_217, x = attn_weights_9_cast_fp16)[name = tensor<string, []>("input_21_cast_fp16")]; | |
| tensor<fp16, [12, 77, 77]> input_23_cast_fp16 = softmax(axis = var_5, x = input_21_cast_fp16)[name = tensor<string, []>("input_23_cast_fp16")]; | |
| tensor<bool, []> attn_output_7_transpose_x_0 = const()[name = tensor<string, []>("attn_output_7_transpose_x_0"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> attn_output_7_transpose_y_0 = const()[name = tensor<string, []>("attn_output_7_transpose_y_0"), val = tensor<bool, []>(false)]; | |
| tensor<fp16, [12, 77, 64]> 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<string, []>("attn_output_7_cast_fp16")]; | |
| tensor<int32, [4]> var_222 = const()[name = tensor<string, []>("op_222"), val = tensor<int32, [4]>([1, 12, 77, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> attn_output_9_cast_fp16 = reshape(shape = var_222, x = attn_output_7_cast_fp16)[name = tensor<string, []>("attn_output_9_cast_fp16")]; | |
| tensor<int32, [4]> attn_output_11_perm_0 = const()[name = tensor<string, []>("attn_output_11_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> var_225 = const()[name = tensor<string, []>("op_225"), val = tensor<int32, [3]>([1, 77, 768])]; | |
| tensor<fp16, [1, 77, 12, 64]> transpose_41 = transpose(perm = attn_output_11_perm_0, x = attn_output_9_cast_fp16)[name = tensor<string, []>("transpose_41")]; | |
| tensor<fp16, [1, 77, 768]> input_25_cast_fp16 = reshape(shape = var_225, x = transpose_41)[name = tensor<string, []>("input_25_cast_fp16")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_1_self_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_1_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(93745024)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_1_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_1_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(94924736)))]; | |
| tensor<fp16, [1, 77, 768]> 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<string, []>("linear_9_cast_fp16")]; | |
| tensor<fp16, [1, 77, 768]> input_27_cast_fp16 = add(x = input_19_cast_fp16, y = linear_9_cast_fp16)[name = tensor<string, []>("input_27_cast_fp16")]; | |
| tensor<int32, [1]> input_29_axes_0 = const()[name = tensor<string, []>("input_29_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_1_layer_norm2_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_1_layer_norm2_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(94926336)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_1_layer_norm2_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_1_layer_norm2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(94927936)))]; | |
| tensor<fp16, [1, 77, 768]> 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<string, []>("input_29_cast_fp16")]; | |
| tensor<fp16, [3072, 768]> text_encoder_text_model_encoder_layers_1_mlp_fc1_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_1_mlp_fc1_weight_to_fp16"), val = tensor<fp16, [3072, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(94929536)))]; | |
| tensor<fp16, [3072]> text_encoder_text_model_encoder_layers_1_mlp_fc1_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_1_mlp_fc1_bias_to_fp16"), val = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(99648192)))]; | |
| tensor<fp16, [1, 77, 3072]> 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<string, []>("linear_10_cast_fp16")]; | |
| tensor<fp16, []> var_240_to_fp16 = const()[name = tensor<string, []>("op_240_to_fp16"), val = tensor<fp16, []>(0x1.b3cp+0)]; | |
| tensor<fp16, [1, 77, 3072]> var_241_cast_fp16 = mul(x = linear_10_cast_fp16, y = var_240_to_fp16)[name = tensor<string, []>("op_241_cast_fp16")]; | |
| tensor<fp16, [1, 77, 3072]> var_242_cast_fp16 = sigmoid(x = var_241_cast_fp16)[name = tensor<string, []>("op_242_cast_fp16")]; | |
| tensor<fp16, [1, 77, 3072]> input_33_cast_fp16 = mul(x = linear_10_cast_fp16, y = var_242_cast_fp16)[name = tensor<string, []>("input_33_cast_fp16")]; | |
| tensor<fp16, [768, 3072]> text_encoder_text_model_encoder_layers_1_mlp_fc2_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_1_mlp_fc2_weight_to_fp16"), val = tensor<fp16, [768, 3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(99654400)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_1_mlp_fc2_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_1_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(104373056)))]; | |
| tensor<fp16, [1, 77, 768]> 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<string, []>("linear_11_cast_fp16")]; | |
| tensor<fp16, [1, 77, 768]> input_35_cast_fp16 = add(x = input_27_cast_fp16, y = linear_11_cast_fp16)[name = tensor<string, []>("input_35_cast_fp16")]; | |
| tensor<int32, [1]> hidden_states_13_axes_0 = const()[name = tensor<string, []>("hidden_states_13_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_2_layer_norm1_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_2_layer_norm1_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(104374656)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_2_layer_norm1_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_2_layer_norm1_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(104376256)))]; | |
| tensor<fp16, [1, 77, 768]> 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<string, []>("hidden_states_13_cast_fp16")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_2_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_2_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(104377856)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_2_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_2_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(105557568)))]; | |
| tensor<fp16, [1, 77, 768]> 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<string, []>("linear_12_cast_fp16")]; | |
| tensor<fp16, []> var_267_to_fp16 = const()[name = tensor<string, []>("op_267_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; | |
| tensor<fp16, [1, 77, 768]> tensor_17_cast_fp16 = mul(x = linear_12_cast_fp16, y = var_267_to_fp16)[name = tensor<string, []>("tensor_17_cast_fp16")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_2_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_2_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(105559168)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_2_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_2_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(106738880)))]; | |
| tensor<fp16, [1, 77, 768]> 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<string, []>("linear_13_cast_fp16")]; | |
| tensor<int32, [4]> var_272 = const()[name = tensor<string, []>("op_272"), val = tensor<int32, [4]>([1, -1, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_273_cast_fp16 = reshape(shape = var_272, x = linear_13_cast_fp16)[name = tensor<string, []>("op_273_cast_fp16")]; | |
| tensor<int32, [4]> var_274_perm_0 = const()[name = tensor<string, []>("op_274_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_2_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_2_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(106740480)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_2_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_2_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(107920192)))]; | |
| tensor<fp16, [1, 77, 768]> 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<string, []>("linear_14_cast_fp16")]; | |
| tensor<int32, [4]> var_279 = const()[name = tensor<string, []>("op_279"), val = tensor<int32, [4]>([1, -1, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_280_cast_fp16 = reshape(shape = var_279, x = linear_14_cast_fp16)[name = tensor<string, []>("op_280_cast_fp16")]; | |
| tensor<int32, [4]> var_281_perm_0 = const()[name = tensor<string, []>("op_281_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [4]> var_288 = const()[name = tensor<string, []>("op_288"), val = tensor<int32, [4]>([1, 77, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_289_cast_fp16 = reshape(shape = var_288, x = tensor_17_cast_fp16)[name = tensor<string, []>("op_289_cast_fp16")]; | |
| tensor<int32, [4]> var_290_perm_0 = const()[name = tensor<string, []>("op_290_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> var_292 = const()[name = tensor<string, []>("op_292"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> transpose_38 = transpose(perm = var_290_perm_0, x = var_289_cast_fp16)[name = tensor<string, []>("transpose_38")]; | |
| tensor<fp16, [12, 77, 64]> query_states_5_cast_fp16 = reshape(shape = var_292, x = transpose_38)[name = tensor<string, []>("query_states_5_cast_fp16")]; | |
| tensor<int32, [3]> var_294 = const()[name = tensor<string, []>("op_294"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> transpose_40 = transpose(perm = var_274_perm_0, x = var_273_cast_fp16)[name = tensor<string, []>("transpose_40")]; | |
| tensor<fp16, [12, 77, 64]> key_states_11_cast_fp16 = reshape(shape = var_294, x = transpose_40)[name = tensor<string, []>("key_states_11_cast_fp16")]; | |
| tensor<int32, [3]> var_296 = const()[name = tensor<string, []>("op_296"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> transpose_39 = transpose(perm = var_281_perm_0, x = var_280_cast_fp16)[name = tensor<string, []>("transpose_39")]; | |
| tensor<fp16, [12, 77, 64]> value_states_11_cast_fp16 = reshape(shape = var_296, x = transpose_39)[name = tensor<string, []>("value_states_11_cast_fp16")]; | |
| tensor<bool, []> attn_weights_13_transpose_x_1 = const()[name = tensor<string, []>("attn_weights_13_transpose_x_1"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> attn_weights_13_transpose_y_1 = const()[name = tensor<string, []>("attn_weights_13_transpose_y_1"), val = tensor<bool, []>(true)]; | |
| tensor<fp16, [12, 77, 77]> 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<string, []>("attn_weights_13_cast_fp16")]; | |
| tensor<int32, [4]> var_301 = const()[name = tensor<string, []>("op_301"), val = tensor<int32, [4]>([1, 12, 77, 77])]; | |
| tensor<fp16, [1, 12, 77, 77]> var_302_cast_fp16 = reshape(shape = var_301, x = attn_weights_13_cast_fp16)[name = tensor<string, []>("op_302_cast_fp16")]; | |
| tensor<fp16, [1, 12, 77, 77]> attn_weights_15_cast_fp16 = add(x = var_302_cast_fp16, y = causal_attention_mask_to_fp16)[name = tensor<string, []>("attn_weights_15_cast_fp16")]; | |
| tensor<int32, [3]> var_307 = const()[name = tensor<string, []>("op_307"), val = tensor<int32, [3]>([12, 77, 77])]; | |
| tensor<fp16, [12, 77, 77]> input_37_cast_fp16 = reshape(shape = var_307, x = attn_weights_15_cast_fp16)[name = tensor<string, []>("input_37_cast_fp16")]; | |
| tensor<fp16, [12, 77, 77]> input_39_cast_fp16 = softmax(axis = var_5, x = input_37_cast_fp16)[name = tensor<string, []>("input_39_cast_fp16")]; | |
| tensor<bool, []> attn_output_13_transpose_x_0 = const()[name = tensor<string, []>("attn_output_13_transpose_x_0"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> attn_output_13_transpose_y_0 = const()[name = tensor<string, []>("attn_output_13_transpose_y_0"), val = tensor<bool, []>(false)]; | |
| tensor<fp16, [12, 77, 64]> 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<string, []>("attn_output_13_cast_fp16")]; | |
| tensor<int32, [4]> var_312 = const()[name = tensor<string, []>("op_312"), val = tensor<int32, [4]>([1, 12, 77, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> attn_output_15_cast_fp16 = reshape(shape = var_312, x = attn_output_13_cast_fp16)[name = tensor<string, []>("attn_output_15_cast_fp16")]; | |
| tensor<int32, [4]> attn_output_17_perm_0 = const()[name = tensor<string, []>("attn_output_17_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> var_315 = const()[name = tensor<string, []>("op_315"), val = tensor<int32, [3]>([1, 77, 768])]; | |
| tensor<fp16, [1, 77, 12, 64]> transpose_37 = transpose(perm = attn_output_17_perm_0, x = attn_output_15_cast_fp16)[name = tensor<string, []>("transpose_37")]; | |
| tensor<fp16, [1, 77, 768]> input_41_cast_fp16 = reshape(shape = var_315, x = transpose_37)[name = tensor<string, []>("input_41_cast_fp16")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_2_self_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_2_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(107921792)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_2_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_2_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(109101504)))]; | |
| tensor<fp16, [1, 77, 768]> 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<string, []>("linear_15_cast_fp16")]; | |
| tensor<fp16, [1, 77, 768]> input_43_cast_fp16 = add(x = input_35_cast_fp16, y = linear_15_cast_fp16)[name = tensor<string, []>("input_43_cast_fp16")]; | |
| tensor<int32, [1]> input_45_axes_0 = const()[name = tensor<string, []>("input_45_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_2_layer_norm2_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_2_layer_norm2_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(109103104)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_2_layer_norm2_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_2_layer_norm2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(109104704)))]; | |
| tensor<fp16, [1, 77, 768]> 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<string, []>("input_45_cast_fp16")]; | |
| tensor<fp16, [3072, 768]> text_encoder_text_model_encoder_layers_2_mlp_fc1_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_2_mlp_fc1_weight_to_fp16"), val = tensor<fp16, [3072, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(109106304)))]; | |
| tensor<fp16, [3072]> text_encoder_text_model_encoder_layers_2_mlp_fc1_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_2_mlp_fc1_bias_to_fp16"), val = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(113824960)))]; | |
| tensor<fp16, [1, 77, 3072]> 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<string, []>("linear_16_cast_fp16")]; | |
| tensor<fp16, []> var_330_to_fp16 = const()[name = tensor<string, []>("op_330_to_fp16"), val = tensor<fp16, []>(0x1.b3cp+0)]; | |
| tensor<fp16, [1, 77, 3072]> var_331_cast_fp16 = mul(x = linear_16_cast_fp16, y = var_330_to_fp16)[name = tensor<string, []>("op_331_cast_fp16")]; | |
| tensor<fp16, [1, 77, 3072]> var_332_cast_fp16 = sigmoid(x = var_331_cast_fp16)[name = tensor<string, []>("op_332_cast_fp16")]; | |
| tensor<fp16, [1, 77, 3072]> input_49_cast_fp16 = mul(x = linear_16_cast_fp16, y = var_332_cast_fp16)[name = tensor<string, []>("input_49_cast_fp16")]; | |
| tensor<fp16, [768, 3072]> text_encoder_text_model_encoder_layers_2_mlp_fc2_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_2_mlp_fc2_weight_to_fp16"), val = tensor<fp16, [768, 3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(113831168)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_2_mlp_fc2_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_2_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(118549824)))]; | |
| tensor<fp16, [1, 77, 768]> 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<string, []>("linear_17_cast_fp16")]; | |
| tensor<fp16, [1, 77, 768]> input_51_cast_fp16 = add(x = input_43_cast_fp16, y = linear_17_cast_fp16)[name = tensor<string, []>("input_51_cast_fp16")]; | |
| tensor<int32, [1]> hidden_states_19_axes_0 = const()[name = tensor<string, []>("hidden_states_19_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_3_layer_norm1_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_3_layer_norm1_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(118551424)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_3_layer_norm1_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_3_layer_norm1_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(118553024)))]; | |
| tensor<fp16, [1, 77, 768]> 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<string, []>("hidden_states_19_cast_fp16")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_3_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_3_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(118554624)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_3_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_3_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(119734336)))]; | |
| tensor<fp16, [1, 77, 768]> 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<string, []>("linear_18_cast_fp16")]; | |
| tensor<fp16, []> var_357_to_fp16 = const()[name = tensor<string, []>("op_357_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; | |
| tensor<fp16, [1, 77, 768]> tensor_23_cast_fp16 = mul(x = linear_18_cast_fp16, y = var_357_to_fp16)[name = tensor<string, []>("tensor_23_cast_fp16")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_3_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_3_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(119735936)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_3_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_3_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(120915648)))]; | |
| tensor<fp16, [1, 77, 768]> 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<string, []>("linear_19_cast_fp16")]; | |
| tensor<int32, [4]> var_362 = const()[name = tensor<string, []>("op_362"), val = tensor<int32, [4]>([1, -1, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_363_cast_fp16 = reshape(shape = var_362, x = linear_19_cast_fp16)[name = tensor<string, []>("op_363_cast_fp16")]; | |
| tensor<int32, [4]> var_364_perm_0 = const()[name = tensor<string, []>("op_364_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_3_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_3_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(120917248)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_3_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_3_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(122096960)))]; | |
| tensor<fp16, [1, 77, 768]> 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<string, []>("linear_20_cast_fp16")]; | |
| tensor<int32, [4]> var_369 = const()[name = tensor<string, []>("op_369"), val = tensor<int32, [4]>([1, -1, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_370_cast_fp16 = reshape(shape = var_369, x = linear_20_cast_fp16)[name = tensor<string, []>("op_370_cast_fp16")]; | |
| tensor<int32, [4]> var_371_perm_0 = const()[name = tensor<string, []>("op_371_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [4]> var_378 = const()[name = tensor<string, []>("op_378"), val = tensor<int32, [4]>([1, 77, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_379_cast_fp16 = reshape(shape = var_378, x = tensor_23_cast_fp16)[name = tensor<string, []>("op_379_cast_fp16")]; | |
| tensor<int32, [4]> var_380_perm_0 = const()[name = tensor<string, []>("op_380_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> var_382 = const()[name = tensor<string, []>("op_382"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> transpose_34 = transpose(perm = var_380_perm_0, x = var_379_cast_fp16)[name = tensor<string, []>("transpose_34")]; | |
| tensor<fp16, [12, 77, 64]> query_states_7_cast_fp16 = reshape(shape = var_382, x = transpose_34)[name = tensor<string, []>("query_states_7_cast_fp16")]; | |
| tensor<int32, [3]> var_384 = const()[name = tensor<string, []>("op_384"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> transpose_36 = transpose(perm = var_364_perm_0, x = var_363_cast_fp16)[name = tensor<string, []>("transpose_36")]; | |
| tensor<fp16, [12, 77, 64]> key_states_15_cast_fp16 = reshape(shape = var_384, x = transpose_36)[name = tensor<string, []>("key_states_15_cast_fp16")]; | |
| tensor<int32, [3]> var_386 = const()[name = tensor<string, []>("op_386"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> transpose_35 = transpose(perm = var_371_perm_0, x = var_370_cast_fp16)[name = tensor<string, []>("transpose_35")]; | |
| tensor<fp16, [12, 77, 64]> value_states_15_cast_fp16 = reshape(shape = var_386, x = transpose_35)[name = tensor<string, []>("value_states_15_cast_fp16")]; | |
| tensor<bool, []> attn_weights_19_transpose_x_1 = const()[name = tensor<string, []>("attn_weights_19_transpose_x_1"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> attn_weights_19_transpose_y_1 = const()[name = tensor<string, []>("attn_weights_19_transpose_y_1"), val = tensor<bool, []>(true)]; | |
| tensor<fp16, [12, 77, 77]> 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<string, []>("attn_weights_19_cast_fp16")]; | |
| tensor<int32, [4]> var_391 = const()[name = tensor<string, []>("op_391"), val = tensor<int32, [4]>([1, 12, 77, 77])]; | |
| tensor<fp16, [1, 12, 77, 77]> var_392_cast_fp16 = reshape(shape = var_391, x = attn_weights_19_cast_fp16)[name = tensor<string, []>("op_392_cast_fp16")]; | |
| tensor<fp16, [1, 12, 77, 77]> attn_weights_21_cast_fp16 = add(x = var_392_cast_fp16, y = causal_attention_mask_to_fp16)[name = tensor<string, []>("attn_weights_21_cast_fp16")]; | |
| tensor<int32, [3]> var_397 = const()[name = tensor<string, []>("op_397"), val = tensor<int32, [3]>([12, 77, 77])]; | |
| tensor<fp16, [12, 77, 77]> input_53_cast_fp16 = reshape(shape = var_397, x = attn_weights_21_cast_fp16)[name = tensor<string, []>("input_53_cast_fp16")]; | |
| tensor<fp16, [12, 77, 77]> input_55_cast_fp16 = softmax(axis = var_5, x = input_53_cast_fp16)[name = tensor<string, []>("input_55_cast_fp16")]; | |
| tensor<bool, []> attn_output_19_transpose_x_0 = const()[name = tensor<string, []>("attn_output_19_transpose_x_0"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> attn_output_19_transpose_y_0 = const()[name = tensor<string, []>("attn_output_19_transpose_y_0"), val = tensor<bool, []>(false)]; | |
| tensor<fp16, [12, 77, 64]> 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<string, []>("attn_output_19_cast_fp16")]; | |
| tensor<int32, [4]> var_402 = const()[name = tensor<string, []>("op_402"), val = tensor<int32, [4]>([1, 12, 77, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> attn_output_21_cast_fp16 = reshape(shape = var_402, x = attn_output_19_cast_fp16)[name = tensor<string, []>("attn_output_21_cast_fp16")]; | |
| tensor<int32, [4]> attn_output_23_perm_0 = const()[name = tensor<string, []>("attn_output_23_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> var_405 = const()[name = tensor<string, []>("op_405"), val = tensor<int32, [3]>([1, 77, 768])]; | |
| tensor<fp16, [1, 77, 12, 64]> transpose_33 = transpose(perm = attn_output_23_perm_0, x = attn_output_21_cast_fp16)[name = tensor<string, []>("transpose_33")]; | |
| tensor<fp16, [1, 77, 768]> input_57_cast_fp16 = reshape(shape = var_405, x = transpose_33)[name = tensor<string, []>("input_57_cast_fp16")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_3_self_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_3_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(122098560)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_3_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_3_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(123278272)))]; | |
| tensor<fp16, [1, 77, 768]> 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<string, []>("linear_21_cast_fp16")]; | |
| tensor<fp16, [1, 77, 768]> input_59_cast_fp16 = add(x = input_51_cast_fp16, y = linear_21_cast_fp16)[name = tensor<string, []>("input_59_cast_fp16")]; | |
| tensor<int32, [1]> input_61_axes_0 = const()[name = tensor<string, []>("input_61_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_3_layer_norm2_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_3_layer_norm2_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(123279872)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_3_layer_norm2_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_3_layer_norm2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(123281472)))]; | |
| tensor<fp16, [1, 77, 768]> 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<string, []>("input_61_cast_fp16")]; | |
| tensor<fp16, [3072, 768]> text_encoder_text_model_encoder_layers_3_mlp_fc1_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_3_mlp_fc1_weight_to_fp16"), val = tensor<fp16, [3072, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(123283072)))]; | |
| tensor<fp16, [3072]> text_encoder_text_model_encoder_layers_3_mlp_fc1_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_3_mlp_fc1_bias_to_fp16"), val = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(128001728)))]; | |
| tensor<fp16, [1, 77, 3072]> 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<string, []>("linear_22_cast_fp16")]; | |
| tensor<fp16, []> var_420_to_fp16 = const()[name = tensor<string, []>("op_420_to_fp16"), val = tensor<fp16, []>(0x1.b3cp+0)]; | |
| tensor<fp16, [1, 77, 3072]> var_421_cast_fp16 = mul(x = linear_22_cast_fp16, y = var_420_to_fp16)[name = tensor<string, []>("op_421_cast_fp16")]; | |
| tensor<fp16, [1, 77, 3072]> var_422_cast_fp16 = sigmoid(x = var_421_cast_fp16)[name = tensor<string, []>("op_422_cast_fp16")]; | |
| tensor<fp16, [1, 77, 3072]> input_65_cast_fp16 = mul(x = linear_22_cast_fp16, y = var_422_cast_fp16)[name = tensor<string, []>("input_65_cast_fp16")]; | |
| tensor<fp16, [768, 3072]> text_encoder_text_model_encoder_layers_3_mlp_fc2_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_3_mlp_fc2_weight_to_fp16"), val = tensor<fp16, [768, 3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(128007936)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_3_mlp_fc2_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_3_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(132726592)))]; | |
| tensor<fp16, [1, 77, 768]> 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<string, []>("linear_23_cast_fp16")]; | |
| tensor<fp16, [1, 77, 768]> input_67_cast_fp16 = add(x = input_59_cast_fp16, y = linear_23_cast_fp16)[name = tensor<string, []>("input_67_cast_fp16")]; | |
| tensor<int32, [1]> hidden_states_25_axes_0 = const()[name = tensor<string, []>("hidden_states_25_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_4_layer_norm1_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_4_layer_norm1_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(132728192)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_4_layer_norm1_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_4_layer_norm1_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(132729792)))]; | |
| tensor<fp16, [1, 77, 768]> 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<string, []>("hidden_states_25_cast_fp16")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_4_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_4_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(132731392)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_4_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_4_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(133911104)))]; | |
| tensor<fp16, [1, 77, 768]> 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<string, []>("linear_24_cast_fp16")]; | |
| tensor<fp16, []> var_447_to_fp16 = const()[name = tensor<string, []>("op_447_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; | |
| tensor<fp16, [1, 77, 768]> tensor_29_cast_fp16 = mul(x = linear_24_cast_fp16, y = var_447_to_fp16)[name = tensor<string, []>("tensor_29_cast_fp16")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_4_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_4_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(133912704)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_4_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_4_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(135092416)))]; | |
| tensor<fp16, [1, 77, 768]> 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<string, []>("linear_25_cast_fp16")]; | |
| tensor<int32, [4]> var_452 = const()[name = tensor<string, []>("op_452"), val = tensor<int32, [4]>([1, -1, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_453_cast_fp16 = reshape(shape = var_452, x = linear_25_cast_fp16)[name = tensor<string, []>("op_453_cast_fp16")]; | |
| tensor<int32, [4]> var_454_perm_0 = const()[name = tensor<string, []>("op_454_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_4_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_4_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(135094016)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_4_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_4_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(136273728)))]; | |
| tensor<fp16, [1, 77, 768]> 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<string, []>("linear_26_cast_fp16")]; | |
| tensor<int32, [4]> var_459 = const()[name = tensor<string, []>("op_459"), val = tensor<int32, [4]>([1, -1, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_460_cast_fp16 = reshape(shape = var_459, x = linear_26_cast_fp16)[name = tensor<string, []>("op_460_cast_fp16")]; | |
| tensor<int32, [4]> var_461_perm_0 = const()[name = tensor<string, []>("op_461_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [4]> var_468 = const()[name = tensor<string, []>("op_468"), val = tensor<int32, [4]>([1, 77, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_469_cast_fp16 = reshape(shape = var_468, x = tensor_29_cast_fp16)[name = tensor<string, []>("op_469_cast_fp16")]; | |
| tensor<int32, [4]> var_470_perm_0 = const()[name = tensor<string, []>("op_470_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> var_472 = const()[name = tensor<string, []>("op_472"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> transpose_30 = transpose(perm = var_470_perm_0, x = var_469_cast_fp16)[name = tensor<string, []>("transpose_30")]; | |
| tensor<fp16, [12, 77, 64]> query_states_9_cast_fp16 = reshape(shape = var_472, x = transpose_30)[name = tensor<string, []>("query_states_9_cast_fp16")]; | |
| tensor<int32, [3]> var_474 = const()[name = tensor<string, []>("op_474"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> transpose_32 = transpose(perm = var_454_perm_0, x = var_453_cast_fp16)[name = tensor<string, []>("transpose_32")]; | |
| tensor<fp16, [12, 77, 64]> key_states_19_cast_fp16 = reshape(shape = var_474, x = transpose_32)[name = tensor<string, []>("key_states_19_cast_fp16")]; | |
| tensor<int32, [3]> var_476 = const()[name = tensor<string, []>("op_476"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> transpose_31 = transpose(perm = var_461_perm_0, x = var_460_cast_fp16)[name = tensor<string, []>("transpose_31")]; | |
| tensor<fp16, [12, 77, 64]> value_states_19_cast_fp16 = reshape(shape = var_476, x = transpose_31)[name = tensor<string, []>("value_states_19_cast_fp16")]; | |
| tensor<bool, []> attn_weights_25_transpose_x_1 = const()[name = tensor<string, []>("attn_weights_25_transpose_x_1"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> attn_weights_25_transpose_y_1 = const()[name = tensor<string, []>("attn_weights_25_transpose_y_1"), val = tensor<bool, []>(true)]; | |
| tensor<fp16, [12, 77, 77]> 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<string, []>("attn_weights_25_cast_fp16")]; | |
| tensor<int32, [4]> var_481 = const()[name = tensor<string, []>("op_481"), val = tensor<int32, [4]>([1, 12, 77, 77])]; | |
| tensor<fp16, [1, 12, 77, 77]> var_482_cast_fp16 = reshape(shape = var_481, x = attn_weights_25_cast_fp16)[name = tensor<string, []>("op_482_cast_fp16")]; | |
| tensor<fp16, [1, 12, 77, 77]> attn_weights_27_cast_fp16 = add(x = var_482_cast_fp16, y = causal_attention_mask_to_fp16)[name = tensor<string, []>("attn_weights_27_cast_fp16")]; | |
| tensor<int32, [3]> var_487 = const()[name = tensor<string, []>("op_487"), val = tensor<int32, [3]>([12, 77, 77])]; | |
| tensor<fp16, [12, 77, 77]> input_69_cast_fp16 = reshape(shape = var_487, x = attn_weights_27_cast_fp16)[name = tensor<string, []>("input_69_cast_fp16")]; | |
| tensor<fp16, [12, 77, 77]> input_71_cast_fp16 = softmax(axis = var_5, x = input_69_cast_fp16)[name = tensor<string, []>("input_71_cast_fp16")]; | |
| tensor<bool, []> attn_output_25_transpose_x_0 = const()[name = tensor<string, []>("attn_output_25_transpose_x_0"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> attn_output_25_transpose_y_0 = const()[name = tensor<string, []>("attn_output_25_transpose_y_0"), val = tensor<bool, []>(false)]; | |
| tensor<fp16, [12, 77, 64]> 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<string, []>("attn_output_25_cast_fp16")]; | |
| tensor<int32, [4]> var_492 = const()[name = tensor<string, []>("op_492"), val = tensor<int32, [4]>([1, 12, 77, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> attn_output_27_cast_fp16 = reshape(shape = var_492, x = attn_output_25_cast_fp16)[name = tensor<string, []>("attn_output_27_cast_fp16")]; | |
| tensor<int32, [4]> attn_output_29_perm_0 = const()[name = tensor<string, []>("attn_output_29_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> var_495 = const()[name = tensor<string, []>("op_495"), val = tensor<int32, [3]>([1, 77, 768])]; | |
| tensor<fp16, [1, 77, 12, 64]> transpose_29 = transpose(perm = attn_output_29_perm_0, x = attn_output_27_cast_fp16)[name = tensor<string, []>("transpose_29")]; | |
| tensor<fp16, [1, 77, 768]> input_73_cast_fp16 = reshape(shape = var_495, x = transpose_29)[name = tensor<string, []>("input_73_cast_fp16")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_4_self_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_4_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(136275328)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_4_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_4_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(137455040)))]; | |
| tensor<fp16, [1, 77, 768]> 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<string, []>("linear_27_cast_fp16")]; | |
| tensor<fp16, [1, 77, 768]> input_75_cast_fp16 = add(x = input_67_cast_fp16, y = linear_27_cast_fp16)[name = tensor<string, []>("input_75_cast_fp16")]; | |
| tensor<int32, [1]> input_77_axes_0 = const()[name = tensor<string, []>("input_77_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_4_layer_norm2_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_4_layer_norm2_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(137456640)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_4_layer_norm2_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_4_layer_norm2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(137458240)))]; | |
| tensor<fp16, [1, 77, 768]> 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<string, []>("input_77_cast_fp16")]; | |
| tensor<fp16, [3072, 768]> text_encoder_text_model_encoder_layers_4_mlp_fc1_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_4_mlp_fc1_weight_to_fp16"), val = tensor<fp16, [3072, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(137459840)))]; | |
| tensor<fp16, [3072]> text_encoder_text_model_encoder_layers_4_mlp_fc1_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_4_mlp_fc1_bias_to_fp16"), val = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(142178496)))]; | |
| tensor<fp16, [1, 77, 3072]> 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<string, []>("linear_28_cast_fp16")]; | |
| tensor<fp16, []> var_510_to_fp16 = const()[name = tensor<string, []>("op_510_to_fp16"), val = tensor<fp16, []>(0x1.b3cp+0)]; | |
| tensor<fp16, [1, 77, 3072]> var_511_cast_fp16 = mul(x = linear_28_cast_fp16, y = var_510_to_fp16)[name = tensor<string, []>("op_511_cast_fp16")]; | |
| tensor<fp16, [1, 77, 3072]> var_512_cast_fp16 = sigmoid(x = var_511_cast_fp16)[name = tensor<string, []>("op_512_cast_fp16")]; | |
| tensor<fp16, [1, 77, 3072]> input_81_cast_fp16 = mul(x = linear_28_cast_fp16, y = var_512_cast_fp16)[name = tensor<string, []>("input_81_cast_fp16")]; | |
| tensor<fp16, [768, 3072]> text_encoder_text_model_encoder_layers_4_mlp_fc2_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_4_mlp_fc2_weight_to_fp16"), val = tensor<fp16, [768, 3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(142184704)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_4_mlp_fc2_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_4_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(146903360)))]; | |
| tensor<fp16, [1, 77, 768]> 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<string, []>("linear_29_cast_fp16")]; | |
| tensor<fp16, [1, 77, 768]> input_83_cast_fp16 = add(x = input_75_cast_fp16, y = linear_29_cast_fp16)[name = tensor<string, []>("input_83_cast_fp16")]; | |
| tensor<int32, [1]> hidden_states_31_axes_0 = const()[name = tensor<string, []>("hidden_states_31_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_5_layer_norm1_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_5_layer_norm1_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(146904960)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_5_layer_norm1_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_5_layer_norm1_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(146906560)))]; | |
| tensor<fp16, [1, 77, 768]> 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<string, []>("hidden_states_31_cast_fp16")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_5_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_5_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(146908160)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_5_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_5_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(148087872)))]; | |
| tensor<fp16, [1, 77, 768]> 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<string, []>("linear_30_cast_fp16")]; | |
| tensor<fp16, []> var_537_to_fp16 = const()[name = tensor<string, []>("op_537_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; | |
| tensor<fp16, [1, 77, 768]> tensor_35_cast_fp16 = mul(x = linear_30_cast_fp16, y = var_537_to_fp16)[name = tensor<string, []>("tensor_35_cast_fp16")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_5_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_5_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(148089472)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_5_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_5_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(149269184)))]; | |
| tensor<fp16, [1, 77, 768]> 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<string, []>("linear_31_cast_fp16")]; | |
| tensor<int32, [4]> var_542 = const()[name = tensor<string, []>("op_542"), val = tensor<int32, [4]>([1, -1, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_543_cast_fp16 = reshape(shape = var_542, x = linear_31_cast_fp16)[name = tensor<string, []>("op_543_cast_fp16")]; | |
| tensor<int32, [4]> var_544_perm_0 = const()[name = tensor<string, []>("op_544_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_5_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_5_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(149270784)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_5_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_5_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(150450496)))]; | |
| tensor<fp16, [1, 77, 768]> 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<string, []>("linear_32_cast_fp16")]; | |
| tensor<int32, [4]> var_549 = const()[name = tensor<string, []>("op_549"), val = tensor<int32, [4]>([1, -1, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_550_cast_fp16 = reshape(shape = var_549, x = linear_32_cast_fp16)[name = tensor<string, []>("op_550_cast_fp16")]; | |
| tensor<int32, [4]> var_551_perm_0 = const()[name = tensor<string, []>("op_551_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [4]> var_558 = const()[name = tensor<string, []>("op_558"), val = tensor<int32, [4]>([1, 77, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_559_cast_fp16 = reshape(shape = var_558, x = tensor_35_cast_fp16)[name = tensor<string, []>("op_559_cast_fp16")]; | |
| tensor<int32, [4]> var_560_perm_0 = const()[name = tensor<string, []>("op_560_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> var_562 = const()[name = tensor<string, []>("op_562"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> transpose_26 = transpose(perm = var_560_perm_0, x = var_559_cast_fp16)[name = tensor<string, []>("transpose_26")]; | |
| tensor<fp16, [12, 77, 64]> query_states_11_cast_fp16 = reshape(shape = var_562, x = transpose_26)[name = tensor<string, []>("query_states_11_cast_fp16")]; | |
| tensor<int32, [3]> var_564 = const()[name = tensor<string, []>("op_564"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> transpose_28 = transpose(perm = var_544_perm_0, x = var_543_cast_fp16)[name = tensor<string, []>("transpose_28")]; | |
| tensor<fp16, [12, 77, 64]> key_states_23_cast_fp16 = reshape(shape = var_564, x = transpose_28)[name = tensor<string, []>("key_states_23_cast_fp16")]; | |
| tensor<int32, [3]> var_566 = const()[name = tensor<string, []>("op_566"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> transpose_27 = transpose(perm = var_551_perm_0, x = var_550_cast_fp16)[name = tensor<string, []>("transpose_27")]; | |
| tensor<fp16, [12, 77, 64]> value_states_23_cast_fp16 = reshape(shape = var_566, x = transpose_27)[name = tensor<string, []>("value_states_23_cast_fp16")]; | |
| tensor<bool, []> attn_weights_31_transpose_x_1 = const()[name = tensor<string, []>("attn_weights_31_transpose_x_1"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> attn_weights_31_transpose_y_1 = const()[name = tensor<string, []>("attn_weights_31_transpose_y_1"), val = tensor<bool, []>(true)]; | |
| tensor<fp16, [12, 77, 77]> 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<string, []>("attn_weights_31_cast_fp16")]; | |
| tensor<int32, [4]> var_571 = const()[name = tensor<string, []>("op_571"), val = tensor<int32, [4]>([1, 12, 77, 77])]; | |
| tensor<fp16, [1, 12, 77, 77]> var_572_cast_fp16 = reshape(shape = var_571, x = attn_weights_31_cast_fp16)[name = tensor<string, []>("op_572_cast_fp16")]; | |
| tensor<fp16, [1, 12, 77, 77]> attn_weights_33_cast_fp16 = add(x = var_572_cast_fp16, y = causal_attention_mask_to_fp16)[name = tensor<string, []>("attn_weights_33_cast_fp16")]; | |
| tensor<int32, [3]> var_577 = const()[name = tensor<string, []>("op_577"), val = tensor<int32, [3]>([12, 77, 77])]; | |
| tensor<fp16, [12, 77, 77]> input_85_cast_fp16 = reshape(shape = var_577, x = attn_weights_33_cast_fp16)[name = tensor<string, []>("input_85_cast_fp16")]; | |
| tensor<fp16, [12, 77, 77]> input_87_cast_fp16 = softmax(axis = var_5, x = input_85_cast_fp16)[name = tensor<string, []>("input_87_cast_fp16")]; | |
| tensor<bool, []> attn_output_31_transpose_x_0 = const()[name = tensor<string, []>("attn_output_31_transpose_x_0"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> attn_output_31_transpose_y_0 = const()[name = tensor<string, []>("attn_output_31_transpose_y_0"), val = tensor<bool, []>(false)]; | |
| tensor<fp16, [12, 77, 64]> 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<string, []>("attn_output_31_cast_fp16")]; | |
| tensor<int32, [4]> var_582 = const()[name = tensor<string, []>("op_582"), val = tensor<int32, [4]>([1, 12, 77, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> attn_output_33_cast_fp16 = reshape(shape = var_582, x = attn_output_31_cast_fp16)[name = tensor<string, []>("attn_output_33_cast_fp16")]; | |
| tensor<int32, [4]> attn_output_35_perm_0 = const()[name = tensor<string, []>("attn_output_35_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> var_585 = const()[name = tensor<string, []>("op_585"), val = tensor<int32, [3]>([1, 77, 768])]; | |
| tensor<fp16, [1, 77, 12, 64]> transpose_25 = transpose(perm = attn_output_35_perm_0, x = attn_output_33_cast_fp16)[name = tensor<string, []>("transpose_25")]; | |
| tensor<fp16, [1, 77, 768]> input_89_cast_fp16 = reshape(shape = var_585, x = transpose_25)[name = tensor<string, []>("input_89_cast_fp16")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_5_self_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_5_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(150452096)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_5_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_5_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(151631808)))]; | |
| tensor<fp16, [1, 77, 768]> 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<string, []>("linear_33_cast_fp16")]; | |
| tensor<fp16, [1, 77, 768]> input_91_cast_fp16 = add(x = input_83_cast_fp16, y = linear_33_cast_fp16)[name = tensor<string, []>("input_91_cast_fp16")]; | |
| tensor<int32, [1]> input_93_axes_0 = const()[name = tensor<string, []>("input_93_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_5_layer_norm2_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_5_layer_norm2_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(151633408)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_5_layer_norm2_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_5_layer_norm2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(151635008)))]; | |
| tensor<fp16, [1, 77, 768]> 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<string, []>("input_93_cast_fp16")]; | |
| tensor<fp16, [3072, 768]> text_encoder_text_model_encoder_layers_5_mlp_fc1_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_5_mlp_fc1_weight_to_fp16"), val = tensor<fp16, [3072, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(151636608)))]; | |
| tensor<fp16, [3072]> text_encoder_text_model_encoder_layers_5_mlp_fc1_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_5_mlp_fc1_bias_to_fp16"), val = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(156355264)))]; | |
| tensor<fp16, [1, 77, 3072]> 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<string, []>("linear_34_cast_fp16")]; | |
| tensor<fp16, []> var_600_to_fp16 = const()[name = tensor<string, []>("op_600_to_fp16"), val = tensor<fp16, []>(0x1.b3cp+0)]; | |
| tensor<fp16, [1, 77, 3072]> var_601_cast_fp16 = mul(x = linear_34_cast_fp16, y = var_600_to_fp16)[name = tensor<string, []>("op_601_cast_fp16")]; | |
| tensor<fp16, [1, 77, 3072]> var_602_cast_fp16 = sigmoid(x = var_601_cast_fp16)[name = tensor<string, []>("op_602_cast_fp16")]; | |
| tensor<fp16, [1, 77, 3072]> input_97_cast_fp16 = mul(x = linear_34_cast_fp16, y = var_602_cast_fp16)[name = tensor<string, []>("input_97_cast_fp16")]; | |
| tensor<fp16, [768, 3072]> text_encoder_text_model_encoder_layers_5_mlp_fc2_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_5_mlp_fc2_weight_to_fp16"), val = tensor<fp16, [768, 3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(156361472)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_5_mlp_fc2_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_5_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(161080128)))]; | |
| tensor<fp16, [1, 77, 768]> 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<string, []>("linear_35_cast_fp16")]; | |
| tensor<fp16, [1, 77, 768]> input_99_cast_fp16 = add(x = input_91_cast_fp16, y = linear_35_cast_fp16)[name = tensor<string, []>("input_99_cast_fp16")]; | |
| tensor<int32, [1]> hidden_states_37_axes_0 = const()[name = tensor<string, []>("hidden_states_37_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_6_layer_norm1_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_6_layer_norm1_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(161081728)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_6_layer_norm1_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_6_layer_norm1_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(161083328)))]; | |
| tensor<fp16, [1, 77, 768]> 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<string, []>("hidden_states_37_cast_fp16")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_6_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_6_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(161084928)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_6_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_6_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(162264640)))]; | |
| tensor<fp16, [1, 77, 768]> 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<string, []>("linear_36_cast_fp16")]; | |
| tensor<fp16, []> var_627_to_fp16 = const()[name = tensor<string, []>("op_627_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; | |
| tensor<fp16, [1, 77, 768]> tensor_41_cast_fp16 = mul(x = linear_36_cast_fp16, y = var_627_to_fp16)[name = tensor<string, []>("tensor_41_cast_fp16")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_6_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_6_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(162266240)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_6_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_6_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(163445952)))]; | |
| tensor<fp16, [1, 77, 768]> 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<string, []>("linear_37_cast_fp16")]; | |
| tensor<int32, [4]> var_632 = const()[name = tensor<string, []>("op_632"), val = tensor<int32, [4]>([1, -1, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_633_cast_fp16 = reshape(shape = var_632, x = linear_37_cast_fp16)[name = tensor<string, []>("op_633_cast_fp16")]; | |
| tensor<int32, [4]> var_634_perm_0 = const()[name = tensor<string, []>("op_634_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_6_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_6_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(163447552)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_6_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_6_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(164627264)))]; | |
| tensor<fp16, [1, 77, 768]> 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<string, []>("linear_38_cast_fp16")]; | |
| tensor<int32, [4]> var_639 = const()[name = tensor<string, []>("op_639"), val = tensor<int32, [4]>([1, -1, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_640_cast_fp16 = reshape(shape = var_639, x = linear_38_cast_fp16)[name = tensor<string, []>("op_640_cast_fp16")]; | |
| tensor<int32, [4]> var_641_perm_0 = const()[name = tensor<string, []>("op_641_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [4]> var_648 = const()[name = tensor<string, []>("op_648"), val = tensor<int32, [4]>([1, 77, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_649_cast_fp16 = reshape(shape = var_648, x = tensor_41_cast_fp16)[name = tensor<string, []>("op_649_cast_fp16")]; | |
| tensor<int32, [4]> var_650_perm_0 = const()[name = tensor<string, []>("op_650_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> var_652 = const()[name = tensor<string, []>("op_652"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> transpose_22 = transpose(perm = var_650_perm_0, x = var_649_cast_fp16)[name = tensor<string, []>("transpose_22")]; | |
| tensor<fp16, [12, 77, 64]> query_states_13_cast_fp16 = reshape(shape = var_652, x = transpose_22)[name = tensor<string, []>("query_states_13_cast_fp16")]; | |
| tensor<int32, [3]> var_654 = const()[name = tensor<string, []>("op_654"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> transpose_24 = transpose(perm = var_634_perm_0, x = var_633_cast_fp16)[name = tensor<string, []>("transpose_24")]; | |
| tensor<fp16, [12, 77, 64]> key_states_27_cast_fp16 = reshape(shape = var_654, x = transpose_24)[name = tensor<string, []>("key_states_27_cast_fp16")]; | |
| tensor<int32, [3]> var_656 = const()[name = tensor<string, []>("op_656"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> transpose_23 = transpose(perm = var_641_perm_0, x = var_640_cast_fp16)[name = tensor<string, []>("transpose_23")]; | |
| tensor<fp16, [12, 77, 64]> value_states_27_cast_fp16 = reshape(shape = var_656, x = transpose_23)[name = tensor<string, []>("value_states_27_cast_fp16")]; | |
| tensor<bool, []> attn_weights_37_transpose_x_1 = const()[name = tensor<string, []>("attn_weights_37_transpose_x_1"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> attn_weights_37_transpose_y_1 = const()[name = tensor<string, []>("attn_weights_37_transpose_y_1"), val = tensor<bool, []>(true)]; | |
| tensor<fp16, [12, 77, 77]> 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<string, []>("attn_weights_37_cast_fp16")]; | |
| tensor<int32, [4]> var_661 = const()[name = tensor<string, []>("op_661"), val = tensor<int32, [4]>([1, 12, 77, 77])]; | |
| tensor<fp16, [1, 12, 77, 77]> var_662_cast_fp16 = reshape(shape = var_661, x = attn_weights_37_cast_fp16)[name = tensor<string, []>("op_662_cast_fp16")]; | |
| tensor<fp16, [1, 12, 77, 77]> attn_weights_39_cast_fp16 = add(x = var_662_cast_fp16, y = causal_attention_mask_to_fp16)[name = tensor<string, []>("attn_weights_39_cast_fp16")]; | |
| tensor<int32, [3]> var_667 = const()[name = tensor<string, []>("op_667"), val = tensor<int32, [3]>([12, 77, 77])]; | |
| tensor<fp16, [12, 77, 77]> input_101_cast_fp16 = reshape(shape = var_667, x = attn_weights_39_cast_fp16)[name = tensor<string, []>("input_101_cast_fp16")]; | |
| tensor<fp16, [12, 77, 77]> input_103_cast_fp16 = softmax(axis = var_5, x = input_101_cast_fp16)[name = tensor<string, []>("input_103_cast_fp16")]; | |
| tensor<bool, []> attn_output_37_transpose_x_0 = const()[name = tensor<string, []>("attn_output_37_transpose_x_0"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> attn_output_37_transpose_y_0 = const()[name = tensor<string, []>("attn_output_37_transpose_y_0"), val = tensor<bool, []>(false)]; | |
| tensor<fp16, [12, 77, 64]> 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<string, []>("attn_output_37_cast_fp16")]; | |
| tensor<int32, [4]> var_672 = const()[name = tensor<string, []>("op_672"), val = tensor<int32, [4]>([1, 12, 77, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> attn_output_39_cast_fp16 = reshape(shape = var_672, x = attn_output_37_cast_fp16)[name = tensor<string, []>("attn_output_39_cast_fp16")]; | |
| tensor<int32, [4]> attn_output_41_perm_0 = const()[name = tensor<string, []>("attn_output_41_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> var_675 = const()[name = tensor<string, []>("op_675"), val = tensor<int32, [3]>([1, 77, 768])]; | |
| tensor<fp16, [1, 77, 12, 64]> transpose_21 = transpose(perm = attn_output_41_perm_0, x = attn_output_39_cast_fp16)[name = tensor<string, []>("transpose_21")]; | |
| tensor<fp16, [1, 77, 768]> input_105_cast_fp16 = reshape(shape = var_675, x = transpose_21)[name = tensor<string, []>("input_105_cast_fp16")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_6_self_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_6_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(164628864)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_6_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_6_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(165808576)))]; | |
| tensor<fp16, [1, 77, 768]> 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<string, []>("linear_39_cast_fp16")]; | |
| tensor<fp16, [1, 77, 768]> input_107_cast_fp16 = add(x = input_99_cast_fp16, y = linear_39_cast_fp16)[name = tensor<string, []>("input_107_cast_fp16")]; | |
| tensor<int32, [1]> input_109_axes_0 = const()[name = tensor<string, []>("input_109_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_6_layer_norm2_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_6_layer_norm2_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(165810176)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_6_layer_norm2_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_6_layer_norm2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(165811776)))]; | |
| tensor<fp16, [1, 77, 768]> 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<string, []>("input_109_cast_fp16")]; | |
| tensor<fp16, [3072, 768]> text_encoder_text_model_encoder_layers_6_mlp_fc1_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_6_mlp_fc1_weight_to_fp16"), val = tensor<fp16, [3072, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(165813376)))]; | |
| tensor<fp16, [3072]> text_encoder_text_model_encoder_layers_6_mlp_fc1_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_6_mlp_fc1_bias_to_fp16"), val = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(170532032)))]; | |
| tensor<fp16, [1, 77, 3072]> 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<string, []>("linear_40_cast_fp16")]; | |
| tensor<fp16, []> var_690_to_fp16 = const()[name = tensor<string, []>("op_690_to_fp16"), val = tensor<fp16, []>(0x1.b3cp+0)]; | |
| tensor<fp16, [1, 77, 3072]> var_691_cast_fp16 = mul(x = linear_40_cast_fp16, y = var_690_to_fp16)[name = tensor<string, []>("op_691_cast_fp16")]; | |
| tensor<fp16, [1, 77, 3072]> var_692_cast_fp16 = sigmoid(x = var_691_cast_fp16)[name = tensor<string, []>("op_692_cast_fp16")]; | |
| tensor<fp16, [1, 77, 3072]> input_113_cast_fp16 = mul(x = linear_40_cast_fp16, y = var_692_cast_fp16)[name = tensor<string, []>("input_113_cast_fp16")]; | |
| tensor<fp16, [768, 3072]> text_encoder_text_model_encoder_layers_6_mlp_fc2_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_6_mlp_fc2_weight_to_fp16"), val = tensor<fp16, [768, 3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(170538240)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_6_mlp_fc2_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_6_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(175256896)))]; | |
| tensor<fp16, [1, 77, 768]> 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<string, []>("linear_41_cast_fp16")]; | |
| tensor<fp16, [1, 77, 768]> input_115_cast_fp16 = add(x = input_107_cast_fp16, y = linear_41_cast_fp16)[name = tensor<string, []>("input_115_cast_fp16")]; | |
| tensor<int32, [1]> hidden_states_43_axes_0 = const()[name = tensor<string, []>("hidden_states_43_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_7_layer_norm1_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_7_layer_norm1_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(175258496)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_7_layer_norm1_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_7_layer_norm1_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(175260096)))]; | |
| tensor<fp16, [1, 77, 768]> 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<string, []>("hidden_states_43_cast_fp16")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_7_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_7_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(175261696)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_7_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_7_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(176441408)))]; | |
| tensor<fp16, [1, 77, 768]> 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<string, []>("linear_42_cast_fp16")]; | |
| tensor<fp16, []> var_717_to_fp16 = const()[name = tensor<string, []>("op_717_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; | |
| tensor<fp16, [1, 77, 768]> tensor_47_cast_fp16 = mul(x = linear_42_cast_fp16, y = var_717_to_fp16)[name = tensor<string, []>("tensor_47_cast_fp16")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_7_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_7_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(176443008)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_7_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_7_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(177622720)))]; | |
| tensor<fp16, [1, 77, 768]> 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<string, []>("linear_43_cast_fp16")]; | |
| tensor<int32, [4]> var_722 = const()[name = tensor<string, []>("op_722"), val = tensor<int32, [4]>([1, -1, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_723_cast_fp16 = reshape(shape = var_722, x = linear_43_cast_fp16)[name = tensor<string, []>("op_723_cast_fp16")]; | |
| tensor<int32, [4]> var_724_perm_0 = const()[name = tensor<string, []>("op_724_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_7_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_7_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(177624320)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_7_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_7_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(178804032)))]; | |
| tensor<fp16, [1, 77, 768]> 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<string, []>("linear_44_cast_fp16")]; | |
| tensor<int32, [4]> var_729 = const()[name = tensor<string, []>("op_729"), val = tensor<int32, [4]>([1, -1, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_730_cast_fp16 = reshape(shape = var_729, x = linear_44_cast_fp16)[name = tensor<string, []>("op_730_cast_fp16")]; | |
| tensor<int32, [4]> var_731_perm_0 = const()[name = tensor<string, []>("op_731_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [4]> var_738 = const()[name = tensor<string, []>("op_738"), val = tensor<int32, [4]>([1, 77, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_739_cast_fp16 = reshape(shape = var_738, x = tensor_47_cast_fp16)[name = tensor<string, []>("op_739_cast_fp16")]; | |
| tensor<int32, [4]> var_740_perm_0 = const()[name = tensor<string, []>("op_740_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> var_742 = const()[name = tensor<string, []>("op_742"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> transpose_18 = transpose(perm = var_740_perm_0, x = var_739_cast_fp16)[name = tensor<string, []>("transpose_18")]; | |
| tensor<fp16, [12, 77, 64]> query_states_15_cast_fp16 = reshape(shape = var_742, x = transpose_18)[name = tensor<string, []>("query_states_15_cast_fp16")]; | |
| tensor<int32, [3]> var_744 = const()[name = tensor<string, []>("op_744"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> transpose_20 = transpose(perm = var_724_perm_0, x = var_723_cast_fp16)[name = tensor<string, []>("transpose_20")]; | |
| tensor<fp16, [12, 77, 64]> key_states_31_cast_fp16 = reshape(shape = var_744, x = transpose_20)[name = tensor<string, []>("key_states_31_cast_fp16")]; | |
| tensor<int32, [3]> var_746 = const()[name = tensor<string, []>("op_746"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> transpose_19 = transpose(perm = var_731_perm_0, x = var_730_cast_fp16)[name = tensor<string, []>("transpose_19")]; | |
| tensor<fp16, [12, 77, 64]> value_states_31_cast_fp16 = reshape(shape = var_746, x = transpose_19)[name = tensor<string, []>("value_states_31_cast_fp16")]; | |
| tensor<bool, []> attn_weights_43_transpose_x_1 = const()[name = tensor<string, []>("attn_weights_43_transpose_x_1"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> attn_weights_43_transpose_y_1 = const()[name = tensor<string, []>("attn_weights_43_transpose_y_1"), val = tensor<bool, []>(true)]; | |
| tensor<fp16, [12, 77, 77]> 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<string, []>("attn_weights_43_cast_fp16")]; | |
| tensor<int32, [4]> var_751 = const()[name = tensor<string, []>("op_751"), val = tensor<int32, [4]>([1, 12, 77, 77])]; | |
| tensor<fp16, [1, 12, 77, 77]> var_752_cast_fp16 = reshape(shape = var_751, x = attn_weights_43_cast_fp16)[name = tensor<string, []>("op_752_cast_fp16")]; | |
| tensor<fp16, [1, 12, 77, 77]> attn_weights_45_cast_fp16 = add(x = var_752_cast_fp16, y = causal_attention_mask_to_fp16)[name = tensor<string, []>("attn_weights_45_cast_fp16")]; | |
| tensor<int32, [3]> var_757 = const()[name = tensor<string, []>("op_757"), val = tensor<int32, [3]>([12, 77, 77])]; | |
| tensor<fp16, [12, 77, 77]> input_117_cast_fp16 = reshape(shape = var_757, x = attn_weights_45_cast_fp16)[name = tensor<string, []>("input_117_cast_fp16")]; | |
| tensor<fp16, [12, 77, 77]> input_119_cast_fp16 = softmax(axis = var_5, x = input_117_cast_fp16)[name = tensor<string, []>("input_119_cast_fp16")]; | |
| tensor<bool, []> attn_output_43_transpose_x_0 = const()[name = tensor<string, []>("attn_output_43_transpose_x_0"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> attn_output_43_transpose_y_0 = const()[name = tensor<string, []>("attn_output_43_transpose_y_0"), val = tensor<bool, []>(false)]; | |
| tensor<fp16, [12, 77, 64]> 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<string, []>("attn_output_43_cast_fp16")]; | |
| tensor<int32, [4]> var_762 = const()[name = tensor<string, []>("op_762"), val = tensor<int32, [4]>([1, 12, 77, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> attn_output_45_cast_fp16 = reshape(shape = var_762, x = attn_output_43_cast_fp16)[name = tensor<string, []>("attn_output_45_cast_fp16")]; | |
| tensor<int32, [4]> attn_output_47_perm_0 = const()[name = tensor<string, []>("attn_output_47_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> var_765 = const()[name = tensor<string, []>("op_765"), val = tensor<int32, [3]>([1, 77, 768])]; | |
| tensor<fp16, [1, 77, 12, 64]> transpose_17 = transpose(perm = attn_output_47_perm_0, x = attn_output_45_cast_fp16)[name = tensor<string, []>("transpose_17")]; | |
| tensor<fp16, [1, 77, 768]> input_121_cast_fp16 = reshape(shape = var_765, x = transpose_17)[name = tensor<string, []>("input_121_cast_fp16")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_7_self_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_7_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(178805632)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_7_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_7_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(179985344)))]; | |
| tensor<fp16, [1, 77, 768]> 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<string, []>("linear_45_cast_fp16")]; | |
| tensor<fp16, [1, 77, 768]> input_123_cast_fp16 = add(x = input_115_cast_fp16, y = linear_45_cast_fp16)[name = tensor<string, []>("input_123_cast_fp16")]; | |
| tensor<int32, [1]> input_125_axes_0 = const()[name = tensor<string, []>("input_125_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_7_layer_norm2_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_7_layer_norm2_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(179986944)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_7_layer_norm2_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_7_layer_norm2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(179988544)))]; | |
| tensor<fp16, [1, 77, 768]> 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<string, []>("input_125_cast_fp16")]; | |
| tensor<fp16, [3072, 768]> text_encoder_text_model_encoder_layers_7_mlp_fc1_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_7_mlp_fc1_weight_to_fp16"), val = tensor<fp16, [3072, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(179990144)))]; | |
| tensor<fp16, [3072]> text_encoder_text_model_encoder_layers_7_mlp_fc1_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_7_mlp_fc1_bias_to_fp16"), val = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(184708800)))]; | |
| tensor<fp16, [1, 77, 3072]> 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<string, []>("linear_46_cast_fp16")]; | |
| tensor<fp16, []> var_780_to_fp16 = const()[name = tensor<string, []>("op_780_to_fp16"), val = tensor<fp16, []>(0x1.b3cp+0)]; | |
| tensor<fp16, [1, 77, 3072]> var_781_cast_fp16 = mul(x = linear_46_cast_fp16, y = var_780_to_fp16)[name = tensor<string, []>("op_781_cast_fp16")]; | |
| tensor<fp16, [1, 77, 3072]> var_782_cast_fp16 = sigmoid(x = var_781_cast_fp16)[name = tensor<string, []>("op_782_cast_fp16")]; | |
| tensor<fp16, [1, 77, 3072]> input_129_cast_fp16 = mul(x = linear_46_cast_fp16, y = var_782_cast_fp16)[name = tensor<string, []>("input_129_cast_fp16")]; | |
| tensor<fp16, [768, 3072]> text_encoder_text_model_encoder_layers_7_mlp_fc2_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_7_mlp_fc2_weight_to_fp16"), val = tensor<fp16, [768, 3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(184715008)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_7_mlp_fc2_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_7_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(189433664)))]; | |
| tensor<fp16, [1, 77, 768]> 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<string, []>("linear_47_cast_fp16")]; | |
| tensor<fp16, [1, 77, 768]> input_131_cast_fp16 = add(x = input_123_cast_fp16, y = linear_47_cast_fp16)[name = tensor<string, []>("input_131_cast_fp16")]; | |
| tensor<int32, [1]> hidden_states_49_axes_0 = const()[name = tensor<string, []>("hidden_states_49_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_8_layer_norm1_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_8_layer_norm1_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(189435264)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_8_layer_norm1_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_8_layer_norm1_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(189436864)))]; | |
| tensor<fp16, [1, 77, 768]> 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<string, []>("hidden_states_49_cast_fp16")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_8_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_8_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(189438464)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_8_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_8_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(190618176)))]; | |
| tensor<fp16, [1, 77, 768]> 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<string, []>("linear_48_cast_fp16")]; | |
| tensor<fp16, []> var_807_to_fp16 = const()[name = tensor<string, []>("op_807_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; | |
| tensor<fp16, [1, 77, 768]> tensor_53_cast_fp16 = mul(x = linear_48_cast_fp16, y = var_807_to_fp16)[name = tensor<string, []>("tensor_53_cast_fp16")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_8_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_8_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(190619776)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_8_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_8_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(191799488)))]; | |
| tensor<fp16, [1, 77, 768]> 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<string, []>("linear_49_cast_fp16")]; | |
| tensor<int32, [4]> var_812 = const()[name = tensor<string, []>("op_812"), val = tensor<int32, [4]>([1, -1, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_813_cast_fp16 = reshape(shape = var_812, x = linear_49_cast_fp16)[name = tensor<string, []>("op_813_cast_fp16")]; | |
| tensor<int32, [4]> var_814_perm_0 = const()[name = tensor<string, []>("op_814_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_8_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_8_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(191801088)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_8_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_8_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(192980800)))]; | |
| tensor<fp16, [1, 77, 768]> 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<string, []>("linear_50_cast_fp16")]; | |
| tensor<int32, [4]> var_819 = const()[name = tensor<string, []>("op_819"), val = tensor<int32, [4]>([1, -1, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_820_cast_fp16 = reshape(shape = var_819, x = linear_50_cast_fp16)[name = tensor<string, []>("op_820_cast_fp16")]; | |
| tensor<int32, [4]> var_821_perm_0 = const()[name = tensor<string, []>("op_821_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [4]> var_828 = const()[name = tensor<string, []>("op_828"), val = tensor<int32, [4]>([1, 77, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_829_cast_fp16 = reshape(shape = var_828, x = tensor_53_cast_fp16)[name = tensor<string, []>("op_829_cast_fp16")]; | |
| tensor<int32, [4]> var_830_perm_0 = const()[name = tensor<string, []>("op_830_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> var_832 = const()[name = tensor<string, []>("op_832"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> transpose_14 = transpose(perm = var_830_perm_0, x = var_829_cast_fp16)[name = tensor<string, []>("transpose_14")]; | |
| tensor<fp16, [12, 77, 64]> query_states_17_cast_fp16 = reshape(shape = var_832, x = transpose_14)[name = tensor<string, []>("query_states_17_cast_fp16")]; | |
| tensor<int32, [3]> var_834 = const()[name = tensor<string, []>("op_834"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> transpose_16 = transpose(perm = var_814_perm_0, x = var_813_cast_fp16)[name = tensor<string, []>("transpose_16")]; | |
| tensor<fp16, [12, 77, 64]> key_states_35_cast_fp16 = reshape(shape = var_834, x = transpose_16)[name = tensor<string, []>("key_states_35_cast_fp16")]; | |
| tensor<int32, [3]> var_836 = const()[name = tensor<string, []>("op_836"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> transpose_15 = transpose(perm = var_821_perm_0, x = var_820_cast_fp16)[name = tensor<string, []>("transpose_15")]; | |
| tensor<fp16, [12, 77, 64]> value_states_35_cast_fp16 = reshape(shape = var_836, x = transpose_15)[name = tensor<string, []>("value_states_35_cast_fp16")]; | |
| tensor<bool, []> attn_weights_49_transpose_x_1 = const()[name = tensor<string, []>("attn_weights_49_transpose_x_1"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> attn_weights_49_transpose_y_1 = const()[name = tensor<string, []>("attn_weights_49_transpose_y_1"), val = tensor<bool, []>(true)]; | |
| tensor<fp16, [12, 77, 77]> 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<string, []>("attn_weights_49_cast_fp16")]; | |
| tensor<int32, [4]> var_841 = const()[name = tensor<string, []>("op_841"), val = tensor<int32, [4]>([1, 12, 77, 77])]; | |
| tensor<fp16, [1, 12, 77, 77]> var_842_cast_fp16 = reshape(shape = var_841, x = attn_weights_49_cast_fp16)[name = tensor<string, []>("op_842_cast_fp16")]; | |
| tensor<fp16, [1, 12, 77, 77]> attn_weights_51_cast_fp16 = add(x = var_842_cast_fp16, y = causal_attention_mask_to_fp16)[name = tensor<string, []>("attn_weights_51_cast_fp16")]; | |
| tensor<int32, [3]> var_847 = const()[name = tensor<string, []>("op_847"), val = tensor<int32, [3]>([12, 77, 77])]; | |
| tensor<fp16, [12, 77, 77]> input_133_cast_fp16 = reshape(shape = var_847, x = attn_weights_51_cast_fp16)[name = tensor<string, []>("input_133_cast_fp16")]; | |
| tensor<fp16, [12, 77, 77]> input_135_cast_fp16 = softmax(axis = var_5, x = input_133_cast_fp16)[name = tensor<string, []>("input_135_cast_fp16")]; | |
| tensor<bool, []> attn_output_49_transpose_x_0 = const()[name = tensor<string, []>("attn_output_49_transpose_x_0"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> attn_output_49_transpose_y_0 = const()[name = tensor<string, []>("attn_output_49_transpose_y_0"), val = tensor<bool, []>(false)]; | |
| tensor<fp16, [12, 77, 64]> 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<string, []>("attn_output_49_cast_fp16")]; | |
| tensor<int32, [4]> var_852 = const()[name = tensor<string, []>("op_852"), val = tensor<int32, [4]>([1, 12, 77, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> attn_output_51_cast_fp16 = reshape(shape = var_852, x = attn_output_49_cast_fp16)[name = tensor<string, []>("attn_output_51_cast_fp16")]; | |
| tensor<int32, [4]> attn_output_53_perm_0 = const()[name = tensor<string, []>("attn_output_53_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> var_855 = const()[name = tensor<string, []>("op_855"), val = tensor<int32, [3]>([1, 77, 768])]; | |
| tensor<fp16, [1, 77, 12, 64]> transpose_13 = transpose(perm = attn_output_53_perm_0, x = attn_output_51_cast_fp16)[name = tensor<string, []>("transpose_13")]; | |
| tensor<fp16, [1, 77, 768]> input_137_cast_fp16 = reshape(shape = var_855, x = transpose_13)[name = tensor<string, []>("input_137_cast_fp16")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_8_self_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_8_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(192982400)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_8_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_8_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(194162112)))]; | |
| tensor<fp16, [1, 77, 768]> 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<string, []>("linear_51_cast_fp16")]; | |
| tensor<fp16, [1, 77, 768]> input_139_cast_fp16 = add(x = input_131_cast_fp16, y = linear_51_cast_fp16)[name = tensor<string, []>("input_139_cast_fp16")]; | |
| tensor<int32, [1]> input_141_axes_0 = const()[name = tensor<string, []>("input_141_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_8_layer_norm2_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_8_layer_norm2_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(194163712)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_8_layer_norm2_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_8_layer_norm2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(194165312)))]; | |
| tensor<fp16, [1, 77, 768]> 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<string, []>("input_141_cast_fp16")]; | |
| tensor<fp16, [3072, 768]> text_encoder_text_model_encoder_layers_8_mlp_fc1_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_8_mlp_fc1_weight_to_fp16"), val = tensor<fp16, [3072, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(194166912)))]; | |
| tensor<fp16, [3072]> text_encoder_text_model_encoder_layers_8_mlp_fc1_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_8_mlp_fc1_bias_to_fp16"), val = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(198885568)))]; | |
| tensor<fp16, [1, 77, 3072]> 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<string, []>("linear_52_cast_fp16")]; | |
| tensor<fp16, []> var_870_to_fp16 = const()[name = tensor<string, []>("op_870_to_fp16"), val = tensor<fp16, []>(0x1.b3cp+0)]; | |
| tensor<fp16, [1, 77, 3072]> var_871_cast_fp16 = mul(x = linear_52_cast_fp16, y = var_870_to_fp16)[name = tensor<string, []>("op_871_cast_fp16")]; | |
| tensor<fp16, [1, 77, 3072]> var_872_cast_fp16 = sigmoid(x = var_871_cast_fp16)[name = tensor<string, []>("op_872_cast_fp16")]; | |
| tensor<fp16, [1, 77, 3072]> input_145_cast_fp16 = mul(x = linear_52_cast_fp16, y = var_872_cast_fp16)[name = tensor<string, []>("input_145_cast_fp16")]; | |
| tensor<fp16, [768, 3072]> text_encoder_text_model_encoder_layers_8_mlp_fc2_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_8_mlp_fc2_weight_to_fp16"), val = tensor<fp16, [768, 3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(198891776)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_8_mlp_fc2_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_8_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(203610432)))]; | |
| tensor<fp16, [1, 77, 768]> 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<string, []>("linear_53_cast_fp16")]; | |
| tensor<fp16, [1, 77, 768]> input_147_cast_fp16 = add(x = input_139_cast_fp16, y = linear_53_cast_fp16)[name = tensor<string, []>("input_147_cast_fp16")]; | |
| tensor<int32, [1]> hidden_states_55_axes_0 = const()[name = tensor<string, []>("hidden_states_55_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_9_layer_norm1_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_9_layer_norm1_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(203612032)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_9_layer_norm1_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_9_layer_norm1_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(203613632)))]; | |
| tensor<fp16, [1, 77, 768]> 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<string, []>("hidden_states_55_cast_fp16")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_9_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_9_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(203615232)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_9_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_9_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(204794944)))]; | |
| tensor<fp16, [1, 77, 768]> 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<string, []>("linear_54_cast_fp16")]; | |
| tensor<fp16, []> var_897_to_fp16 = const()[name = tensor<string, []>("op_897_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; | |
| tensor<fp16, [1, 77, 768]> tensor_59_cast_fp16 = mul(x = linear_54_cast_fp16, y = var_897_to_fp16)[name = tensor<string, []>("tensor_59_cast_fp16")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_9_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_9_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(204796544)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_9_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_9_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(205976256)))]; | |
| tensor<fp16, [1, 77, 768]> 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<string, []>("linear_55_cast_fp16")]; | |
| tensor<int32, [4]> var_902 = const()[name = tensor<string, []>("op_902"), val = tensor<int32, [4]>([1, -1, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_903_cast_fp16 = reshape(shape = var_902, x = linear_55_cast_fp16)[name = tensor<string, []>("op_903_cast_fp16")]; | |
| tensor<int32, [4]> var_904_perm_0 = const()[name = tensor<string, []>("op_904_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_9_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_9_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(205977856)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_9_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_9_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(207157568)))]; | |
| tensor<fp16, [1, 77, 768]> 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<string, []>("linear_56_cast_fp16")]; | |
| tensor<int32, [4]> var_909 = const()[name = tensor<string, []>("op_909"), val = tensor<int32, [4]>([1, -1, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_910_cast_fp16 = reshape(shape = var_909, x = linear_56_cast_fp16)[name = tensor<string, []>("op_910_cast_fp16")]; | |
| tensor<int32, [4]> var_911_perm_0 = const()[name = tensor<string, []>("op_911_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [4]> var_918 = const()[name = tensor<string, []>("op_918"), val = tensor<int32, [4]>([1, 77, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_919_cast_fp16 = reshape(shape = var_918, x = tensor_59_cast_fp16)[name = tensor<string, []>("op_919_cast_fp16")]; | |
| tensor<int32, [4]> var_920_perm_0 = const()[name = tensor<string, []>("op_920_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> var_922 = const()[name = tensor<string, []>("op_922"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> transpose_10 = transpose(perm = var_920_perm_0, x = var_919_cast_fp16)[name = tensor<string, []>("transpose_10")]; | |
| tensor<fp16, [12, 77, 64]> query_states_19_cast_fp16 = reshape(shape = var_922, x = transpose_10)[name = tensor<string, []>("query_states_19_cast_fp16")]; | |
| tensor<int32, [3]> var_924 = const()[name = tensor<string, []>("op_924"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> transpose_12 = transpose(perm = var_904_perm_0, x = var_903_cast_fp16)[name = tensor<string, []>("transpose_12")]; | |
| tensor<fp16, [12, 77, 64]> key_states_39_cast_fp16 = reshape(shape = var_924, x = transpose_12)[name = tensor<string, []>("key_states_39_cast_fp16")]; | |
| tensor<int32, [3]> var_926 = const()[name = tensor<string, []>("op_926"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> transpose_11 = transpose(perm = var_911_perm_0, x = var_910_cast_fp16)[name = tensor<string, []>("transpose_11")]; | |
| tensor<fp16, [12, 77, 64]> value_states_39_cast_fp16 = reshape(shape = var_926, x = transpose_11)[name = tensor<string, []>("value_states_39_cast_fp16")]; | |
| tensor<bool, []> attn_weights_55_transpose_x_1 = const()[name = tensor<string, []>("attn_weights_55_transpose_x_1"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> attn_weights_55_transpose_y_1 = const()[name = tensor<string, []>("attn_weights_55_transpose_y_1"), val = tensor<bool, []>(true)]; | |
| tensor<fp16, [12, 77, 77]> 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<string, []>("attn_weights_55_cast_fp16")]; | |
| tensor<int32, [4]> var_931 = const()[name = tensor<string, []>("op_931"), val = tensor<int32, [4]>([1, 12, 77, 77])]; | |
| tensor<fp16, [1, 12, 77, 77]> var_932_cast_fp16 = reshape(shape = var_931, x = attn_weights_55_cast_fp16)[name = tensor<string, []>("op_932_cast_fp16")]; | |
| tensor<fp16, [1, 12, 77, 77]> attn_weights_57_cast_fp16 = add(x = var_932_cast_fp16, y = causal_attention_mask_to_fp16)[name = tensor<string, []>("attn_weights_57_cast_fp16")]; | |
| tensor<int32, [3]> var_937 = const()[name = tensor<string, []>("op_937"), val = tensor<int32, [3]>([12, 77, 77])]; | |
| tensor<fp16, [12, 77, 77]> input_149_cast_fp16 = reshape(shape = var_937, x = attn_weights_57_cast_fp16)[name = tensor<string, []>("input_149_cast_fp16")]; | |
| tensor<fp16, [12, 77, 77]> input_151_cast_fp16 = softmax(axis = var_5, x = input_149_cast_fp16)[name = tensor<string, []>("input_151_cast_fp16")]; | |
| tensor<bool, []> attn_output_55_transpose_x_0 = const()[name = tensor<string, []>("attn_output_55_transpose_x_0"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> attn_output_55_transpose_y_0 = const()[name = tensor<string, []>("attn_output_55_transpose_y_0"), val = tensor<bool, []>(false)]; | |
| tensor<fp16, [12, 77, 64]> 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<string, []>("attn_output_55_cast_fp16")]; | |
| tensor<int32, [4]> var_942 = const()[name = tensor<string, []>("op_942"), val = tensor<int32, [4]>([1, 12, 77, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> attn_output_57_cast_fp16 = reshape(shape = var_942, x = attn_output_55_cast_fp16)[name = tensor<string, []>("attn_output_57_cast_fp16")]; | |
| tensor<int32, [4]> attn_output_59_perm_0 = const()[name = tensor<string, []>("attn_output_59_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> var_945 = const()[name = tensor<string, []>("op_945"), val = tensor<int32, [3]>([1, 77, 768])]; | |
| tensor<fp16, [1, 77, 12, 64]> transpose_9 = transpose(perm = attn_output_59_perm_0, x = attn_output_57_cast_fp16)[name = tensor<string, []>("transpose_9")]; | |
| tensor<fp16, [1, 77, 768]> input_153_cast_fp16 = reshape(shape = var_945, x = transpose_9)[name = tensor<string, []>("input_153_cast_fp16")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_9_self_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_9_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(207159168)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_9_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_9_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(208338880)))]; | |
| tensor<fp16, [1, 77, 768]> 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<string, []>("linear_57_cast_fp16")]; | |
| tensor<fp16, [1, 77, 768]> input_155_cast_fp16 = add(x = input_147_cast_fp16, y = linear_57_cast_fp16)[name = tensor<string, []>("input_155_cast_fp16")]; | |
| tensor<int32, [1]> input_157_axes_0 = const()[name = tensor<string, []>("input_157_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_9_layer_norm2_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_9_layer_norm2_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(208340480)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_9_layer_norm2_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_9_layer_norm2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(208342080)))]; | |
| tensor<fp16, [1, 77, 768]> 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<string, []>("input_157_cast_fp16")]; | |
| tensor<fp16, [3072, 768]> text_encoder_text_model_encoder_layers_9_mlp_fc1_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_9_mlp_fc1_weight_to_fp16"), val = tensor<fp16, [3072, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(208343680)))]; | |
| tensor<fp16, [3072]> text_encoder_text_model_encoder_layers_9_mlp_fc1_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_9_mlp_fc1_bias_to_fp16"), val = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(213062336)))]; | |
| tensor<fp16, [1, 77, 3072]> 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<string, []>("linear_58_cast_fp16")]; | |
| tensor<fp16, []> var_960_to_fp16 = const()[name = tensor<string, []>("op_960_to_fp16"), val = tensor<fp16, []>(0x1.b3cp+0)]; | |
| tensor<fp16, [1, 77, 3072]> var_961_cast_fp16 = mul(x = linear_58_cast_fp16, y = var_960_to_fp16)[name = tensor<string, []>("op_961_cast_fp16")]; | |
| tensor<fp16, [1, 77, 3072]> var_962_cast_fp16 = sigmoid(x = var_961_cast_fp16)[name = tensor<string, []>("op_962_cast_fp16")]; | |
| tensor<fp16, [1, 77, 3072]> input_161_cast_fp16 = mul(x = linear_58_cast_fp16, y = var_962_cast_fp16)[name = tensor<string, []>("input_161_cast_fp16")]; | |
| tensor<fp16, [768, 3072]> text_encoder_text_model_encoder_layers_9_mlp_fc2_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_9_mlp_fc2_weight_to_fp16"), val = tensor<fp16, [768, 3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(213068544)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_9_mlp_fc2_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_9_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(217787200)))]; | |
| tensor<fp16, [1, 77, 768]> 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<string, []>("linear_59_cast_fp16")]; | |
| tensor<fp16, [1, 77, 768]> input_163_cast_fp16 = add(x = input_155_cast_fp16, y = linear_59_cast_fp16)[name = tensor<string, []>("input_163_cast_fp16")]; | |
| tensor<int32, [1]> hidden_states_61_axes_0 = const()[name = tensor<string, []>("hidden_states_61_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_10_layer_norm1_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_10_layer_norm1_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(217788800)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_10_layer_norm1_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_10_layer_norm1_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(217790400)))]; | |
| tensor<fp16, [1, 77, 768]> 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<string, []>("hidden_states_61_cast_fp16")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_10_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_10_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(217792000)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_10_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_10_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(218971712)))]; | |
| tensor<fp16, [1, 77, 768]> 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<string, []>("linear_60_cast_fp16")]; | |
| tensor<fp16, []> var_987_to_fp16 = const()[name = tensor<string, []>("op_987_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; | |
| tensor<fp16, [1, 77, 768]> tensor_65_cast_fp16 = mul(x = linear_60_cast_fp16, y = var_987_to_fp16)[name = tensor<string, []>("tensor_65_cast_fp16")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_10_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_10_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(218973312)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_10_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_10_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(220153024)))]; | |
| tensor<fp16, [1, 77, 768]> 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<string, []>("linear_61_cast_fp16")]; | |
| tensor<int32, [4]> var_992 = const()[name = tensor<string, []>("op_992"), val = tensor<int32, [4]>([1, -1, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_993_cast_fp16 = reshape(shape = var_992, x = linear_61_cast_fp16)[name = tensor<string, []>("op_993_cast_fp16")]; | |
| tensor<int32, [4]> var_994_perm_0 = const()[name = tensor<string, []>("op_994_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_10_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_10_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(220154624)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_10_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_10_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(221334336)))]; | |
| tensor<fp16, [1, 77, 768]> 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<string, []>("linear_62_cast_fp16")]; | |
| tensor<int32, [4]> var_999 = const()[name = tensor<string, []>("op_999"), val = tensor<int32, [4]>([1, -1, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_1000_cast_fp16 = reshape(shape = var_999, x = linear_62_cast_fp16)[name = tensor<string, []>("op_1000_cast_fp16")]; | |
| tensor<int32, [4]> var_1001_perm_0 = const()[name = tensor<string, []>("op_1001_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [4]> var_1008 = const()[name = tensor<string, []>("op_1008"), val = tensor<int32, [4]>([1, 77, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_1009_cast_fp16 = reshape(shape = var_1008, x = tensor_65_cast_fp16)[name = tensor<string, []>("op_1009_cast_fp16")]; | |
| tensor<int32, [4]> var_1010_perm_0 = const()[name = tensor<string, []>("op_1010_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> var_1012 = const()[name = tensor<string, []>("op_1012"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> transpose_6 = transpose(perm = var_1010_perm_0, x = var_1009_cast_fp16)[name = tensor<string, []>("transpose_6")]; | |
| tensor<fp16, [12, 77, 64]> query_states_21_cast_fp16 = reshape(shape = var_1012, x = transpose_6)[name = tensor<string, []>("query_states_21_cast_fp16")]; | |
| tensor<int32, [3]> var_1014 = const()[name = tensor<string, []>("op_1014"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> transpose_8 = transpose(perm = var_994_perm_0, x = var_993_cast_fp16)[name = tensor<string, []>("transpose_8")]; | |
| tensor<fp16, [12, 77, 64]> key_states_43_cast_fp16 = reshape(shape = var_1014, x = transpose_8)[name = tensor<string, []>("key_states_43_cast_fp16")]; | |
| tensor<int32, [3]> var_1016 = const()[name = tensor<string, []>("op_1016"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> transpose_7 = transpose(perm = var_1001_perm_0, x = var_1000_cast_fp16)[name = tensor<string, []>("transpose_7")]; | |
| tensor<fp16, [12, 77, 64]> value_states_43_cast_fp16 = reshape(shape = var_1016, x = transpose_7)[name = tensor<string, []>("value_states_43_cast_fp16")]; | |
| tensor<bool, []> attn_weights_61_transpose_x_1 = const()[name = tensor<string, []>("attn_weights_61_transpose_x_1"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> attn_weights_61_transpose_y_1 = const()[name = tensor<string, []>("attn_weights_61_transpose_y_1"), val = tensor<bool, []>(true)]; | |
| tensor<fp16, [12, 77, 77]> 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<string, []>("attn_weights_61_cast_fp16")]; | |
| tensor<int32, [4]> var_1021 = const()[name = tensor<string, []>("op_1021"), val = tensor<int32, [4]>([1, 12, 77, 77])]; | |
| tensor<fp16, [1, 12, 77, 77]> var_1022_cast_fp16 = reshape(shape = var_1021, x = attn_weights_61_cast_fp16)[name = tensor<string, []>("op_1022_cast_fp16")]; | |
| tensor<fp16, [1, 12, 77, 77]> attn_weights_63_cast_fp16 = add(x = var_1022_cast_fp16, y = causal_attention_mask_to_fp16)[name = tensor<string, []>("attn_weights_63_cast_fp16")]; | |
| tensor<int32, [3]> var_1027 = const()[name = tensor<string, []>("op_1027"), val = tensor<int32, [3]>([12, 77, 77])]; | |
| tensor<fp16, [12, 77, 77]> input_165_cast_fp16 = reshape(shape = var_1027, x = attn_weights_63_cast_fp16)[name = tensor<string, []>("input_165_cast_fp16")]; | |
| tensor<fp16, [12, 77, 77]> input_167_cast_fp16 = softmax(axis = var_5, x = input_165_cast_fp16)[name = tensor<string, []>("input_167_cast_fp16")]; | |
| tensor<bool, []> attn_output_61_transpose_x_0 = const()[name = tensor<string, []>("attn_output_61_transpose_x_0"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> attn_output_61_transpose_y_0 = const()[name = tensor<string, []>("attn_output_61_transpose_y_0"), val = tensor<bool, []>(false)]; | |
| tensor<fp16, [12, 77, 64]> 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<string, []>("attn_output_61_cast_fp16")]; | |
| tensor<int32, [4]> var_1032 = const()[name = tensor<string, []>("op_1032"), val = tensor<int32, [4]>([1, 12, 77, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> attn_output_63_cast_fp16 = reshape(shape = var_1032, x = attn_output_61_cast_fp16)[name = tensor<string, []>("attn_output_63_cast_fp16")]; | |
| tensor<int32, [4]> attn_output_65_perm_0 = const()[name = tensor<string, []>("attn_output_65_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> var_1035 = const()[name = tensor<string, []>("op_1035"), val = tensor<int32, [3]>([1, 77, 768])]; | |
| tensor<fp16, [1, 77, 12, 64]> transpose_5 = transpose(perm = attn_output_65_perm_0, x = attn_output_63_cast_fp16)[name = tensor<string, []>("transpose_5")]; | |
| tensor<fp16, [1, 77, 768]> input_169_cast_fp16 = reshape(shape = var_1035, x = transpose_5)[name = tensor<string, []>("input_169_cast_fp16")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_10_self_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_10_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(221335936)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_10_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_10_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(222515648)))]; | |
| tensor<fp16, [1, 77, 768]> 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<string, []>("linear_63_cast_fp16")]; | |
| tensor<fp16, [1, 77, 768]> input_171_cast_fp16 = add(x = input_163_cast_fp16, y = linear_63_cast_fp16)[name = tensor<string, []>("input_171_cast_fp16")]; | |
| tensor<int32, [1]> input_173_axes_0 = const()[name = tensor<string, []>("input_173_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_10_layer_norm2_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_10_layer_norm2_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(222517248)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_10_layer_norm2_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_10_layer_norm2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(222518848)))]; | |
| tensor<fp16, [1, 77, 768]> 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<string, []>("input_173_cast_fp16")]; | |
| tensor<fp16, [3072, 768]> text_encoder_text_model_encoder_layers_10_mlp_fc1_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_10_mlp_fc1_weight_to_fp16"), val = tensor<fp16, [3072, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(222520448)))]; | |
| tensor<fp16, [3072]> text_encoder_text_model_encoder_layers_10_mlp_fc1_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_10_mlp_fc1_bias_to_fp16"), val = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(227239104)))]; | |
| tensor<fp16, [1, 77, 3072]> 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<string, []>("linear_64_cast_fp16")]; | |
| tensor<fp16, []> var_1050_to_fp16 = const()[name = tensor<string, []>("op_1050_to_fp16"), val = tensor<fp16, []>(0x1.b3cp+0)]; | |
| tensor<fp16, [1, 77, 3072]> var_1051_cast_fp16 = mul(x = linear_64_cast_fp16, y = var_1050_to_fp16)[name = tensor<string, []>("op_1051_cast_fp16")]; | |
| tensor<fp16, [1, 77, 3072]> var_1052_cast_fp16 = sigmoid(x = var_1051_cast_fp16)[name = tensor<string, []>("op_1052_cast_fp16")]; | |
| tensor<fp16, [1, 77, 3072]> input_177_cast_fp16 = mul(x = linear_64_cast_fp16, y = var_1052_cast_fp16)[name = tensor<string, []>("input_177_cast_fp16")]; | |
| tensor<fp16, [768, 3072]> text_encoder_text_model_encoder_layers_10_mlp_fc2_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_10_mlp_fc2_weight_to_fp16"), val = tensor<fp16, [768, 3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(227245312)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_10_mlp_fc2_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_10_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(231963968)))]; | |
| tensor<fp16, [1, 77, 768]> 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<string, []>("linear_65_cast_fp16")]; | |
| tensor<fp16, [1, 77, 768]> input_179_cast_fp16 = add(x = input_171_cast_fp16, y = linear_65_cast_fp16)[name = tensor<string, []>("input_179_cast_fp16")]; | |
| tensor<int32, [1]> hidden_states_67_axes_0 = const()[name = tensor<string, []>("hidden_states_67_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_11_layer_norm1_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_11_layer_norm1_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(231965568)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_11_layer_norm1_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_11_layer_norm1_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(231967168)))]; | |
| tensor<fp16, [1, 77, 768]> 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<string, []>("hidden_states_67_cast_fp16")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_11_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_11_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(231968768)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_11_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_11_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(233148480)))]; | |
| tensor<fp16, [1, 77, 768]> 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<string, []>("linear_66_cast_fp16")]; | |
| tensor<fp16, []> var_1077_to_fp16 = const()[name = tensor<string, []>("op_1077_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; | |
| tensor<fp16, [1, 77, 768]> tensor_cast_fp16 = mul(x = linear_66_cast_fp16, y = var_1077_to_fp16)[name = tensor<string, []>("tensor_cast_fp16")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_11_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_11_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(233150080)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_11_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_11_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(234329792)))]; | |
| tensor<fp16, [1, 77, 768]> 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<string, []>("linear_67_cast_fp16")]; | |
| tensor<int32, [4]> var_1082 = const()[name = tensor<string, []>("op_1082"), val = tensor<int32, [4]>([1, -1, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_1083_cast_fp16 = reshape(shape = var_1082, x = linear_67_cast_fp16)[name = tensor<string, []>("op_1083_cast_fp16")]; | |
| tensor<int32, [4]> var_1084_perm_0 = const()[name = tensor<string, []>("op_1084_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_11_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_11_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(234331392)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_11_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_11_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(235511104)))]; | |
| tensor<fp16, [1, 77, 768]> 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<string, []>("linear_68_cast_fp16")]; | |
| tensor<int32, [4]> var_1089 = const()[name = tensor<string, []>("op_1089"), val = tensor<int32, [4]>([1, -1, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_1090_cast_fp16 = reshape(shape = var_1089, x = linear_68_cast_fp16)[name = tensor<string, []>("op_1090_cast_fp16")]; | |
| tensor<int32, [4]> var_1091_perm_0 = const()[name = tensor<string, []>("op_1091_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [4]> var_1098 = const()[name = tensor<string, []>("op_1098"), val = tensor<int32, [4]>([1, 77, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_1099_cast_fp16 = reshape(shape = var_1098, x = tensor_cast_fp16)[name = tensor<string, []>("op_1099_cast_fp16")]; | |
| tensor<int32, [4]> var_1100_perm_0 = const()[name = tensor<string, []>("op_1100_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> var_1102 = const()[name = tensor<string, []>("op_1102"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> transpose_2 = transpose(perm = var_1100_perm_0, x = var_1099_cast_fp16)[name = tensor<string, []>("transpose_2")]; | |
| tensor<fp16, [12, 77, 64]> query_states_cast_fp16 = reshape(shape = var_1102, x = transpose_2)[name = tensor<string, []>("query_states_cast_fp16")]; | |
| tensor<int32, [3]> var_1104 = const()[name = tensor<string, []>("op_1104"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> transpose_4 = transpose(perm = var_1084_perm_0, x = var_1083_cast_fp16)[name = tensor<string, []>("transpose_4")]; | |
| tensor<fp16, [12, 77, 64]> key_states_cast_fp16 = reshape(shape = var_1104, x = transpose_4)[name = tensor<string, []>("key_states_cast_fp16")]; | |
| tensor<int32, [3]> var_1106 = const()[name = tensor<string, []>("op_1106"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> transpose_3 = transpose(perm = var_1091_perm_0, x = var_1090_cast_fp16)[name = tensor<string, []>("transpose_3")]; | |
| tensor<fp16, [12, 77, 64]> value_states_cast_fp16 = reshape(shape = var_1106, x = transpose_3)[name = tensor<string, []>("value_states_cast_fp16")]; | |
| tensor<bool, []> attn_weights_67_transpose_x_1 = const()[name = tensor<string, []>("attn_weights_67_transpose_x_1"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> attn_weights_67_transpose_y_1 = const()[name = tensor<string, []>("attn_weights_67_transpose_y_1"), val = tensor<bool, []>(true)]; | |
| tensor<fp16, [12, 77, 77]> 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<string, []>("attn_weights_67_cast_fp16")]; | |
| tensor<int32, [4]> var_1111 = const()[name = tensor<string, []>("op_1111"), val = tensor<int32, [4]>([1, 12, 77, 77])]; | |
| tensor<fp16, [1, 12, 77, 77]> var_1112_cast_fp16 = reshape(shape = var_1111, x = attn_weights_67_cast_fp16)[name = tensor<string, []>("op_1112_cast_fp16")]; | |
| tensor<fp16, [1, 12, 77, 77]> attn_weights_69_cast_fp16 = add(x = var_1112_cast_fp16, y = causal_attention_mask_to_fp16)[name = tensor<string, []>("attn_weights_69_cast_fp16")]; | |
| tensor<int32, [3]> var_1117 = const()[name = tensor<string, []>("op_1117"), val = tensor<int32, [3]>([12, 77, 77])]; | |
| tensor<fp16, [12, 77, 77]> input_181_cast_fp16 = reshape(shape = var_1117, x = attn_weights_69_cast_fp16)[name = tensor<string, []>("input_181_cast_fp16")]; | |
| tensor<fp16, [12, 77, 77]> input_183_cast_fp16 = softmax(axis = var_5, x = input_181_cast_fp16)[name = tensor<string, []>("input_183_cast_fp16")]; | |
| tensor<bool, []> attn_output_67_transpose_x_0 = const()[name = tensor<string, []>("attn_output_67_transpose_x_0"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> attn_output_67_transpose_y_0 = const()[name = tensor<string, []>("attn_output_67_transpose_y_0"), val = tensor<bool, []>(false)]; | |
| tensor<fp16, [12, 77, 64]> 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<string, []>("attn_output_67_cast_fp16")]; | |
| tensor<int32, [4]> var_1122 = const()[name = tensor<string, []>("op_1122"), val = tensor<int32, [4]>([1, 12, 77, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> attn_output_69_cast_fp16 = reshape(shape = var_1122, x = attn_output_67_cast_fp16)[name = tensor<string, []>("attn_output_69_cast_fp16")]; | |
| tensor<int32, [4]> attn_output_perm_0 = const()[name = tensor<string, []>("attn_output_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> var_1125 = const()[name = tensor<string, []>("op_1125"), val = tensor<int32, [3]>([1, 77, 768])]; | |
| tensor<fp16, [1, 77, 12, 64]> transpose_1 = transpose(perm = attn_output_perm_0, x = attn_output_69_cast_fp16)[name = tensor<string, []>("transpose_1")]; | |
| tensor<fp16, [1, 77, 768]> input_185_cast_fp16 = reshape(shape = var_1125, x = transpose_1)[name = tensor<string, []>("input_185_cast_fp16")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_11_self_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_11_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(235512704)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_11_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_11_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(236692416)))]; | |
| tensor<fp16, [1, 77, 768]> 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<string, []>("linear_69_cast_fp16")]; | |
| tensor<fp16, [1, 77, 768]> input_187_cast_fp16 = add(x = input_179_cast_fp16, y = linear_69_cast_fp16)[name = tensor<string, []>("input_187_cast_fp16")]; | |
| tensor<int32, [1]> input_189_axes_0 = const()[name = tensor<string, []>("input_189_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_11_layer_norm2_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_11_layer_norm2_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(236694016)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_11_layer_norm2_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_11_layer_norm2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(236695616)))]; | |
| tensor<fp16, [1, 77, 768]> 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<string, []>("input_189_cast_fp16")]; | |
| tensor<fp16, [3072, 768]> text_encoder_text_model_encoder_layers_11_mlp_fc1_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_11_mlp_fc1_weight_to_fp16"), val = tensor<fp16, [3072, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(236697216)))]; | |
| tensor<fp16, [3072]> text_encoder_text_model_encoder_layers_11_mlp_fc1_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_11_mlp_fc1_bias_to_fp16"), val = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(241415872)))]; | |
| tensor<fp16, [1, 77, 3072]> 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<string, []>("linear_70_cast_fp16")]; | |
| tensor<fp16, []> var_1140_to_fp16 = const()[name = tensor<string, []>("op_1140_to_fp16"), val = tensor<fp16, []>(0x1.b3cp+0)]; | |
| tensor<fp16, [1, 77, 3072]> var_1141_cast_fp16 = mul(x = linear_70_cast_fp16, y = var_1140_to_fp16)[name = tensor<string, []>("op_1141_cast_fp16")]; | |
| tensor<fp16, [1, 77, 3072]> var_1142_cast_fp16 = sigmoid(x = var_1141_cast_fp16)[name = tensor<string, []>("op_1142_cast_fp16")]; | |
| tensor<fp16, [1, 77, 3072]> input_193_cast_fp16 = mul(x = linear_70_cast_fp16, y = var_1142_cast_fp16)[name = tensor<string, []>("input_193_cast_fp16")]; | |
| tensor<fp16, [768, 3072]> text_encoder_text_model_encoder_layers_11_mlp_fc2_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_11_mlp_fc2_weight_to_fp16"), val = tensor<fp16, [768, 3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(241422080)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_11_mlp_fc2_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_11_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(246140736)))]; | |
| tensor<fp16, [1, 77, 768]> 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<string, []>("linear_71_cast_fp16")]; | |
| tensor<fp16, [1, 77, 768]> input_cast_fp16 = add(x = input_187_cast_fp16, y = linear_71_cast_fp16)[name = tensor<string, []>("input_cast_fp16")]; | |
| tensor<int32, [1]> last_hidden_state_axes_0 = const()[name = tensor<string, []>("last_hidden_state_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> text_encoder_text_model_final_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_final_layer_norm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(246142336)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_final_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_final_layer_norm_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(246143936)))]; | |
| tensor<fp16, [1, 77, 768]> 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<string, []>("last_hidden_state_cast_fp16")]; | |
| tensor<string, []> last_hidden_state_cast_fp16_to_fp32_dtype_0 = const()[name = tensor<string, []>("last_hidden_state_cast_fp16_to_fp32_dtype_0"), val = tensor<string, []>("fp32")]; | |
| tensor<int32, [1]> var_1153 = const()[name = tensor<string, []>("op_1153"), val = tensor<int32, [1]>([0])]; | |
| tensor<int32, [1]> var_1155 = reduce_argmax(axis = var_5, keep_dims = var_6, x = cast_127)[name = tensor<string, []>("op_1155")]; | |
| tensor<int32, []> stack_0_axis_0 = const()[name = tensor<string, []>("stack_0_axis_0"), val = tensor<int32, []>(1)]; | |
| tensor<int32, [1, 2]> stack_0 = stack(axis = stack_0_axis_0, values = (var_1153, var_1155))[name = tensor<string, []>("stack_0")]; | |
| tensor<int32, []> var_1157_transpose_batch_dims_0 = const()[name = tensor<string, []>("op_1157_transpose_batch_dims_0"), val = tensor<int32, []>(0)]; | |
| tensor<fp16, [1, 768]> 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<string, []>("op_1157_transpose_cast_fp16")]; | |
| tensor<string, []> var_1157_cast_fp16_to_fp32_dtype_0 = const()[name = tensor<string, []>("op_1157_cast_fp16_to_fp32_dtype_0"), val = tensor<string, []>("fp32")]; | |
| tensor<fp32, [1, 768]> pooled_outputs = cast(dtype = var_1157_cast_fp16_to_fp32_dtype_0, x = var_1157_transpose_cast_fp16)[name = tensor<string, []>("cast_125")]; | |
| tensor<fp32, [1, 77, 768]> last_hidden_state = cast(dtype = last_hidden_state_cast_fp16_to_fp32_dtype_0, x = last_hidden_state_cast_fp16)[name = tensor<string, []>("cast_126")]; | |
| } -> (last_hidden_state, pooled_outputs); | |
| } |