| program(1.0) |
| [buildInfo = dict<tensor<string, []>, tensor<string, []>>({{"coremlc-component-MIL", "3510.2.1"}, {"coremlc-version", "3500.32.1"}, {"coremltools-component-torch", "2.5.0"}, {"coremltools-source-dialect", "TorchScript"}, {"coremltools-version", "9.0"}})] |
| { |
| func main<ios16>(tensor<int32, [1]> cache_length, tensor<fp16, [1, 448]> decoder_key_padding_mask, tensor<fp16, [1, 768, 1, 1500]> encoder_output_embeds, tensor<int32, [1]> input_ids, tensor<fp16, [1, 9216, 1, 448]> key_cache, tensor<fp16, [1, 448]> kv_cache_update_mask, tensor<fp16, [1, 9216, 1, 448]> value_cache) { |
| tensor<int32, []> var_40_axis_0 = const()[name = tensor<string, []>("op_40_axis_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> var_40_batch_dims_0 = const()[name = tensor<string, []>("op_40_batch_dims_0"), val = tensor<int32, []>(0)]; |
| tensor<fp16, [51865, 768]> embed_tokens_weight_to_fp16 = const()[name = tensor<string, []>("embed_tokens_weight_to_fp16"), val = tensor<fp16, [51865, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(64)))]; |
| tensor<fp16, [1, 768]> var_40_cast_fp16 = gather(axis = var_40_axis_0, batch_dims = var_40_batch_dims_0, indices = input_ids, x = embed_tokens_weight_to_fp16)[name = tensor<string, []>("op_40_cast_fp16")]; |
| tensor<int32, []> var_44_axis_0 = const()[name = tensor<string, []>("op_44_axis_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> var_44_batch_dims_0 = const()[name = tensor<string, []>("op_44_batch_dims_0"), val = tensor<int32, []>(0)]; |
| tensor<fp16, [448, 768]> embed_positions_weight_to_fp16 = const()[name = tensor<string, []>("embed_positions_weight_to_fp16"), val = tensor<fp16, [448, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(79664768)))]; |
| tensor<fp16, [1, 768]> var_44_cast_fp16 = gather(axis = var_44_axis_0, batch_dims = var_44_batch_dims_0, indices = cache_length, x = embed_positions_weight_to_fp16)[name = tensor<string, []>("op_44_cast_fp16")]; |
| tensor<fp16, [1, 768]> hidden_states_1_cast_fp16 = add(x = var_40_cast_fp16, y = var_44_cast_fp16)[name = tensor<string, []>("hidden_states_1_cast_fp16")]; |
| tensor<int32, [1]> var_58_axes_0 = const()[name = tensor<string, []>("op_58_axes_0"), val = tensor<int32, [1]>([2])]; |
| tensor<fp16, [1, 768, 1]> var_58_cast_fp16 = expand_dims(axes = var_58_axes_0, x = hidden_states_1_cast_fp16)[name = tensor<string, []>("op_58_cast_fp16")]; |
| tensor<int32, [1]> inputs_1_axes_0 = const()[name = tensor<string, []>("inputs_1_axes_0"), val = tensor<int32, [1]>([3])]; |
| tensor<fp16, [1, 768, 1, 1]> inputs_1_cast_fp16 = expand_dims(axes = inputs_1_axes_0, x = var_58_cast_fp16)[name = tensor<string, []>("inputs_1_cast_fp16")]; |
| tensor<int32, [12]> tile_0 = const()[name = tensor<string, []>("tile_0"), val = tensor<int32, [12]>([768, 768, 768, 768, 768, 768, 768, 768, 768, 768, 768, 768])]; |
| tensor<int32, []> var_63_axis_0 = const()[name = tensor<string, []>("op_63_axis_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [1, 768, 1, 448]> var_63_cast_fp16_0, tensor<fp16, [1, 768, 1, 448]> var_63_cast_fp16_1, tensor<fp16, [1, 768, 1, 448]> var_63_cast_fp16_2, tensor<fp16, [1, 768, 1, 448]> var_63_cast_fp16_3, tensor<fp16, [1, 768, 1, 448]> var_63_cast_fp16_4, tensor<fp16, [1, 768, 1, 448]> var_63_cast_fp16_5, tensor<fp16, [1, 768, 1, 448]> var_63_cast_fp16_6, tensor<fp16, [1, 768, 1, 448]> var_63_cast_fp16_7, tensor<fp16, [1, 768, 1, 448]> var_63_cast_fp16_8, tensor<fp16, [1, 768, 1, 448]> var_63_cast_fp16_9, tensor<fp16, [1, 768, 1, 448]> var_63_cast_fp16_10, tensor<fp16, [1, 768, 1, 448]> var_63_cast_fp16_11 = split(axis = var_63_axis_0, split_sizes = tile_0, x = key_cache)[name = tensor<string, []>("op_63_cast_fp16")]; |
| tensor<int32, [12]> tile_1 = const()[name = tensor<string, []>("tile_1"), val = tensor<int32, [12]>([768, 768, 768, 768, 768, 768, 768, 768, 768, 768, 768, 768])]; |
| tensor<int32, []> var_78_axis_0 = const()[name = tensor<string, []>("op_78_axis_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [1, 768, 1, 448]> var_78_cast_fp16_0, tensor<fp16, [1, 768, 1, 448]> var_78_cast_fp16_1, tensor<fp16, [1, 768, 1, 448]> var_78_cast_fp16_2, tensor<fp16, [1, 768, 1, 448]> var_78_cast_fp16_3, tensor<fp16, [1, 768, 1, 448]> var_78_cast_fp16_4, tensor<fp16, [1, 768, 1, 448]> var_78_cast_fp16_5, tensor<fp16, [1, 768, 1, 448]> var_78_cast_fp16_6, tensor<fp16, [1, 768, 1, 448]> var_78_cast_fp16_7, tensor<fp16, [1, 768, 1, 448]> var_78_cast_fp16_8, tensor<fp16, [1, 768, 1, 448]> var_78_cast_fp16_9, tensor<fp16, [1, 768, 1, 448]> var_78_cast_fp16_10, tensor<fp16, [1, 768, 1, 448]> var_78_cast_fp16_11 = split(axis = var_78_axis_0, split_sizes = tile_1, x = value_cache)[name = tensor<string, []>("op_78_cast_fp16")]; |
| tensor<int32, []> var_96 = const()[name = tensor<string, []>("op_96"), val = tensor<int32, []>(3)]; |
| tensor<int32, [1]> out_1_axes_0 = const()[name = tensor<string, []>("out_1_axes_0"), val = tensor<int32, [1]>([1])]; |
| tensor<fp16, []> var_121_to_fp16 = const()[name = tensor<string, []>("op_121_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, 768, 1, 1]> out_1_cast_fp16 = layer_norm(axes = out_1_axes_0, epsilon = var_121_to_fp16, x = inputs_1_cast_fp16)[name = tensor<string, []>("out_1_cast_fp16")]; |
| tensor<fp16, [768]> obj_1_mean_0_to_fp16 = const()[name = tensor<string, []>("obj_1_mean_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(80352960)))]; |
| tensor<fp16, [768]> obj_1_variance_0_to_fp16 = const()[name = tensor<string, []>("obj_1_variance_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(80354560)))]; |
| tensor<fp16, [768]> obj_1_gamma_0_to_fp16 = const()[name = tensor<string, []>("obj_1_gamma_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(80356160)))]; |
| tensor<fp16, [768]> obj_1_beta_0_to_fp16 = const()[name = tensor<string, []>("obj_1_beta_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(80357760)))]; |
| tensor<fp16, []> obj_1_epsilon_0_to_fp16 = const()[name = tensor<string, []>("obj_1_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, 768, 1, 1]> obj_1_cast_fp16 = batch_norm(beta = obj_1_beta_0_to_fp16, epsilon = obj_1_epsilon_0_to_fp16, gamma = obj_1_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_1_cast_fp16)[name = tensor<string, []>("obj_1_cast_fp16")]; |
| tensor<string, []> query_1_pad_type_0 = const()[name = tensor<string, []>("query_1_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> query_1_strides_0 = const()[name = tensor<string, []>("query_1_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> query_1_pad_0 = const()[name = tensor<string, []>("query_1_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> query_1_dilations_0 = const()[name = tensor<string, []>("query_1_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> query_1_groups_0 = const()[name = tensor<string, []>("query_1_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 768, 1, 1]> layers_0_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_0_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [768, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(80359360)))]; |
| tensor<fp16, [768]> layers_0_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("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, []>(81539072)))]; |
| tensor<fp16, [1, 768, 1, 1]> query_1_cast_fp16 = conv(bias = layers_0_self_attn_q_proj_bias_to_fp16, dilations = query_1_dilations_0, groups = query_1_groups_0, pad = query_1_pad_0, pad_type = query_1_pad_type_0, strides = query_1_strides_0, weight = layers_0_self_attn_q_proj_weight_to_fp16, x = obj_1_cast_fp16)[name = tensor<string, []>("query_1_cast_fp16")]; |
| tensor<string, []> current_key_1_pad_type_0 = const()[name = tensor<string, []>("current_key_1_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> current_key_1_strides_0 = const()[name = tensor<string, []>("current_key_1_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> current_key_1_pad_0 = const()[name = tensor<string, []>("current_key_1_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> current_key_1_dilations_0 = const()[name = tensor<string, []>("current_key_1_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> current_key_1_groups_0 = const()[name = tensor<string, []>("current_key_1_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 768, 1, 1]> layers_0_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_0_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [768, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(81540672)))]; |
| tensor<fp16, [1, 768, 1, 1]> current_key_1_cast_fp16 = conv(dilations = current_key_1_dilations_0, groups = current_key_1_groups_0, pad = current_key_1_pad_0, pad_type = current_key_1_pad_type_0, strides = current_key_1_strides_0, weight = layers_0_self_attn_k_proj_weight_to_fp16, x = obj_1_cast_fp16)[name = tensor<string, []>("current_key_1_cast_fp16")]; |
| tensor<string, []> current_value_1_pad_type_0 = const()[name = tensor<string, []>("current_value_1_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> current_value_1_strides_0 = const()[name = tensor<string, []>("current_value_1_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> current_value_1_pad_0 = const()[name = tensor<string, []>("current_value_1_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> current_value_1_dilations_0 = const()[name = tensor<string, []>("current_value_1_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> current_value_1_groups_0 = const()[name = tensor<string, []>("current_value_1_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 768, 1, 1]> layers_0_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_0_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [768, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(82720384)))]; |
| tensor<fp16, [768]> layers_0_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("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, []>(83900096)))]; |
| tensor<fp16, [1, 768, 1, 1]> current_value_1_cast_fp16 = conv(bias = layers_0_self_attn_v_proj_bias_to_fp16, dilations = current_value_1_dilations_0, groups = current_value_1_groups_0, pad = current_value_1_pad_0, pad_type = current_value_1_pad_type_0, strides = current_value_1_strides_0, weight = layers_0_self_attn_v_proj_weight_to_fp16, x = obj_1_cast_fp16)[name = tensor<string, []>("current_value_1_cast_fp16")]; |
| tensor<int32, [1]> var_156_axes_0 = const()[name = tensor<string, []>("op_156_axes_0"), val = tensor<int32, [1]>([1])]; |
| tensor<fp16, [1, 1, 448]> var_156_cast_fp16 = expand_dims(axes = var_156_axes_0, x = kv_cache_update_mask)[name = tensor<string, []>("op_156_cast_fp16")]; |
| tensor<int32, [1]> var_157_axes_0 = const()[name = tensor<string, []>("op_157_axes_0"), val = tensor<int32, [1]>([2])]; |
| tensor<fp16, [1, 1, 1, 448]> var_157_cast_fp16 = expand_dims(axes = var_157_axes_0, x = var_156_cast_fp16)[name = tensor<string, []>("op_157_cast_fp16")]; |
| tensor<fp16, []> var_97_to_fp16 = const()[name = tensor<string, []>("op_97_to_fp16"), val = tensor<fp16, []>(0x1p+0)]; |
| tensor<fp16, [1, 1, 1, 448]> var_159_cast_fp16 = sub(x = var_97_to_fp16, y = var_157_cast_fp16)[name = tensor<string, []>("op_159_cast_fp16")]; |
| tensor<fp16, [1, 768, 1, 448]> var_160_cast_fp16 = mul(x = var_63_cast_fp16_0, y = var_159_cast_fp16)[name = tensor<string, []>("op_160_cast_fp16")]; |
| tensor<fp16, [1, 768, 1, 448]> var_161_cast_fp16 = mul(x = current_key_1_cast_fp16, y = var_157_cast_fp16)[name = tensor<string, []>("op_161_cast_fp16")]; |
| tensor<fp16, [1, 768, 1, 448]> key_1_cast_fp16 = add(x = var_160_cast_fp16, y = var_161_cast_fp16)[name = tensor<string, []>("key_1_cast_fp16")]; |
| tensor<fp16, [1, 768, 1, 448]> var_164_cast_fp16 = mul(x = var_78_cast_fp16_0, y = var_159_cast_fp16)[name = tensor<string, []>("op_164_cast_fp16")]; |
| tensor<fp16, [1, 768, 1, 448]> var_165_cast_fp16 = mul(x = current_value_1_cast_fp16, y = var_157_cast_fp16)[name = tensor<string, []>("op_165_cast_fp16")]; |
| tensor<fp16, [1, 768, 1, 448]> value_1_cast_fp16 = add(x = var_164_cast_fp16, y = var_165_cast_fp16)[name = tensor<string, []>("value_1_cast_fp16")]; |
| tensor<int32, [4]> var_169 = const()[name = tensor<string, []>("op_169"), val = tensor<int32, [4]>([1, 12, 64, 1])]; |
| tensor<fp16, [1, 12, 64, 1]> mh_q_1_cast_fp16 = reshape(shape = var_169, x = query_1_cast_fp16)[name = tensor<string, []>("mh_q_1_cast_fp16")]; |
| tensor<fp16, []> var_171_to_fp16 = const()[name = tensor<string, []>("op_171_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; |
| tensor<fp16, [1, 12, 64, 1]> var_172_cast_fp16 = mul(x = mh_q_1_cast_fp16, y = var_171_to_fp16)[name = tensor<string, []>("op_172_cast_fp16")]; |
| tensor<int32, [4]> var_175 = const()[name = tensor<string, []>("op_175"), val = tensor<int32, [4]>([1, 12, 64, 448])]; |
| tensor<fp16, [1, 12, 64, 448]> var_176_cast_fp16 = reshape(shape = var_175, x = key_1_cast_fp16)[name = tensor<string, []>("op_176_cast_fp16")]; |
| tensor<bool, []> mh_w_1_transpose_x_0 = const()[name = tensor<string, []>("mh_w_1_transpose_x_0"), val = tensor<bool, []>(true)]; |
| tensor<bool, []> mh_w_1_transpose_y_0 = const()[name = tensor<string, []>("mh_w_1_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [1, 12, 1, 448]> mh_w_1_cast_fp16 = matmul(transpose_x = mh_w_1_transpose_x_0, transpose_y = mh_w_1_transpose_y_0, x = var_172_cast_fp16, y = var_176_cast_fp16)[name = tensor<string, []>("mh_w_1_cast_fp16")]; |
| tensor<int32, [1]> var_180_axes_0 = const()[name = tensor<string, []>("op_180_axes_0"), val = tensor<int32, [1]>([1])]; |
| tensor<fp16, [1, 1, 448]> var_180_cast_fp16 = expand_dims(axes = var_180_axes_0, x = decoder_key_padding_mask)[name = tensor<string, []>("op_180_cast_fp16")]; |
| tensor<int32, [1]> var_181_axes_0 = const()[name = tensor<string, []>("op_181_axes_0"), val = tensor<int32, [1]>([2])]; |
| tensor<fp16, [1, 1, 1, 448]> var_181_cast_fp16 = expand_dims(axes = var_181_axes_0, x = var_180_cast_fp16)[name = tensor<string, []>("op_181_cast_fp16")]; |
| tensor<fp16, [1, 12, 1, 448]> mh_w_3_cast_fp16 = add(x = mh_w_1_cast_fp16, y = var_181_cast_fp16)[name = tensor<string, []>("mh_w_3_cast_fp16")]; |
| tensor<fp16, [1, 12, 1, 448]> var_184_cast_fp16 = softmax(axis = var_96, x = mh_w_3_cast_fp16)[name = tensor<string, []>("op_184_cast_fp16")]; |
| tensor<int32, [4]> var_185 = const()[name = tensor<string, []>("op_185"), val = tensor<int32, [4]>([1, 12, 64, 448])]; |
| tensor<fp16, [1, 12, 64, 448]> var_186_cast_fp16 = reshape(shape = var_185, x = value_1_cast_fp16)[name = tensor<string, []>("op_186_cast_fp16")]; |
| tensor<bool, []> attn_1_transpose_x_0 = const()[name = tensor<string, []>("attn_1_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_1_transpose_y_0 = const()[name = tensor<string, []>("attn_1_transpose_y_0"), val = tensor<bool, []>(true)]; |
| tensor<fp16, [1, 12, 64, 1]> attn_1_cast_fp16 = matmul(transpose_x = attn_1_transpose_x_0, transpose_y = attn_1_transpose_y_0, x = var_186_cast_fp16, y = var_184_cast_fp16)[name = tensor<string, []>("attn_1_cast_fp16")]; |
| tensor<int32, [4]> var_189 = const()[name = tensor<string, []>("op_189"), val = tensor<int32, [4]>([1, 768, 1, 1])]; |
| tensor<fp16, [1, 768, 1, 1]> input_1_cast_fp16 = reshape(shape = var_189, x = attn_1_cast_fp16)[name = tensor<string, []>("input_1_cast_fp16")]; |
| tensor<string, []> obj_7_pad_type_0 = const()[name = tensor<string, []>("obj_7_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> obj_7_strides_0 = const()[name = tensor<string, []>("obj_7_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> obj_7_pad_0 = const()[name = tensor<string, []>("obj_7_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> obj_7_dilations_0 = const()[name = tensor<string, []>("obj_7_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> obj_7_groups_0 = const()[name = tensor<string, []>("obj_7_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 768, 1, 1]> layers_0_self_attn_o_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_0_self_attn_o_proj_weight_to_fp16"), val = tensor<fp16, [768, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(83901696)))]; |
| tensor<fp16, [768]> layers_0_self_attn_o_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_0_self_attn_o_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(85081408)))]; |
| tensor<fp16, [1, 768, 1, 1]> obj_7_cast_fp16 = conv(bias = layers_0_self_attn_o_proj_bias_to_fp16, dilations = obj_7_dilations_0, groups = obj_7_groups_0, pad = obj_7_pad_0, pad_type = obj_7_pad_type_0, strides = obj_7_strides_0, weight = layers_0_self_attn_o_proj_weight_to_fp16, x = input_1_cast_fp16)[name = tensor<string, []>("obj_7_cast_fp16")]; |
| tensor<fp16, [1, 768, 1, 1]> inputs_3_cast_fp16 = add(x = inputs_1_cast_fp16, y = obj_7_cast_fp16)[name = tensor<string, []>("inputs_3_cast_fp16")]; |
| tensor<int32, [1]> out_3_axes_0 = const()[name = tensor<string, []>("out_3_axes_0"), val = tensor<int32, [1]>([1])]; |
| tensor<fp16, []> var_211_to_fp16 = const()[name = tensor<string, []>("op_211_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, 768, 1, 1]> out_3_cast_fp16 = layer_norm(axes = out_3_axes_0, epsilon = var_211_to_fp16, x = inputs_3_cast_fp16)[name = tensor<string, []>("out_3_cast_fp16")]; |
| tensor<fp16, [768]> obj_9_gamma_0_to_fp16 = const()[name = tensor<string, []>("obj_9_gamma_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(85083008)))]; |
| tensor<fp16, [768]> obj_9_beta_0_to_fp16 = const()[name = tensor<string, []>("obj_9_beta_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(85084608)))]; |
| tensor<fp16, []> obj_9_epsilon_0_to_fp16 = const()[name = tensor<string, []>("obj_9_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, 768, 1, 1]> obj_9_cast_fp16 = batch_norm(beta = obj_9_beta_0_to_fp16, epsilon = obj_9_epsilon_0_to_fp16, gamma = obj_9_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_3_cast_fp16)[name = tensor<string, []>("obj_9_cast_fp16")]; |
| tensor<string, []> query_3_pad_type_0 = const()[name = tensor<string, []>("query_3_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> query_3_strides_0 = const()[name = tensor<string, []>("query_3_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> query_3_pad_0 = const()[name = tensor<string, []>("query_3_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> query_3_dilations_0 = const()[name = tensor<string, []>("query_3_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> query_3_groups_0 = const()[name = tensor<string, []>("query_3_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 768, 1, 1]> layers_0_encoder_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_0_encoder_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [768, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(85086208)))]; |
| tensor<fp16, [768]> layers_0_encoder_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_0_encoder_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(86265920)))]; |
| tensor<fp16, [1, 768, 1, 1]> query_3_cast_fp16 = conv(bias = layers_0_encoder_attn_q_proj_bias_to_fp16, dilations = query_3_dilations_0, groups = query_3_groups_0, pad = query_3_pad_0, pad_type = query_3_pad_type_0, strides = query_3_strides_0, weight = layers_0_encoder_attn_q_proj_weight_to_fp16, x = obj_9_cast_fp16)[name = tensor<string, []>("query_3_cast_fp16")]; |
| tensor<string, []> key_3_pad_type_0 = const()[name = tensor<string, []>("key_3_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> key_3_strides_0 = const()[name = tensor<string, []>("key_3_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> key_3_pad_0 = const()[name = tensor<string, []>("key_3_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> key_3_dilations_0 = const()[name = tensor<string, []>("key_3_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> key_3_groups_0 = const()[name = tensor<string, []>("key_3_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 768, 1, 1]> layers_0_encoder_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_0_encoder_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [768, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(86267520)))]; |
| tensor<fp16, [1, 768, 1, 1500]> key_3_cast_fp16 = conv(dilations = key_3_dilations_0, groups = key_3_groups_0, pad = key_3_pad_0, pad_type = key_3_pad_type_0, strides = key_3_strides_0, weight = layers_0_encoder_attn_k_proj_weight_to_fp16, x = encoder_output_embeds)[name = tensor<string, []>("key_3_cast_fp16")]; |
| tensor<string, []> value_3_pad_type_0 = const()[name = tensor<string, []>("value_3_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> value_3_strides_0 = const()[name = tensor<string, []>("value_3_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> value_3_pad_0 = const()[name = tensor<string, []>("value_3_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> value_3_dilations_0 = const()[name = tensor<string, []>("value_3_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> value_3_groups_0 = const()[name = tensor<string, []>("value_3_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 768, 1, 1]> layers_0_encoder_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_0_encoder_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [768, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(87447232)))]; |
| tensor<fp16, [768]> layers_0_encoder_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_0_encoder_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(88626944)))]; |
| tensor<fp16, [1, 768, 1, 1500]> value_3_cast_fp16 = conv(bias = layers_0_encoder_attn_v_proj_bias_to_fp16, dilations = value_3_dilations_0, groups = value_3_groups_0, pad = value_3_pad_0, pad_type = value_3_pad_type_0, strides = value_3_strides_0, weight = layers_0_encoder_attn_v_proj_weight_to_fp16, x = encoder_output_embeds)[name = tensor<string, []>("value_3_cast_fp16")]; |
| tensor<int32, [4]> var_247 = const()[name = tensor<string, []>("op_247"), val = tensor<int32, [4]>([1, 12, 64, 1])]; |
| tensor<fp16, [1, 12, 64, 1]> mh_q_3_cast_fp16 = reshape(shape = var_247, x = query_3_cast_fp16)[name = tensor<string, []>("mh_q_3_cast_fp16")]; |
| tensor<fp16, []> var_249_to_fp16 = const()[name = tensor<string, []>("op_249_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; |
| tensor<fp16, [1, 12, 64, 1]> var_250_cast_fp16 = mul(x = mh_q_3_cast_fp16, y = var_249_to_fp16)[name = tensor<string, []>("op_250_cast_fp16")]; |
| tensor<int32, [4]> var_253 = const()[name = tensor<string, []>("op_253"), val = tensor<int32, [4]>([1, 12, 64, 1500])]; |
| tensor<fp16, [1, 12, 64, 1500]> var_254_cast_fp16 = reshape(shape = var_253, x = key_3_cast_fp16)[name = tensor<string, []>("op_254_cast_fp16")]; |
| tensor<bool, []> mh_w_5_transpose_x_0 = const()[name = tensor<string, []>("mh_w_5_transpose_x_0"), val = tensor<bool, []>(true)]; |
| tensor<bool, []> mh_w_5_transpose_y_0 = const()[name = tensor<string, []>("mh_w_5_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [1, 12, 1, 1500]> mh_w_5_cast_fp16 = matmul(transpose_x = mh_w_5_transpose_x_0, transpose_y = mh_w_5_transpose_y_0, x = var_250_cast_fp16, y = var_254_cast_fp16)[name = tensor<string, []>("mh_w_5_cast_fp16")]; |
| tensor<fp16, [1, 12, 1, 1500]> obj_13_cast_fp16 = softmax(axis = var_96, x = mh_w_5_cast_fp16)[name = tensor<string, []>("obj_13_cast_fp16")]; |
| tensor<int32, [4]> var_258 = const()[name = tensor<string, []>("op_258"), val = tensor<int32, [4]>([1, 12, 64, 1500])]; |
| tensor<fp16, [1, 12, 64, 1500]> var_259_cast_fp16 = reshape(shape = var_258, x = value_3_cast_fp16)[name = tensor<string, []>("op_259_cast_fp16")]; |
| tensor<bool, []> attn_3_transpose_x_0 = const()[name = tensor<string, []>("attn_3_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_3_transpose_y_0 = const()[name = tensor<string, []>("attn_3_transpose_y_0"), val = tensor<bool, []>(true)]; |
| tensor<fp16, [1, 12, 64, 1]> attn_3_cast_fp16 = matmul(transpose_x = attn_3_transpose_x_0, transpose_y = attn_3_transpose_y_0, x = var_259_cast_fp16, y = obj_13_cast_fp16)[name = tensor<string, []>("attn_3_cast_fp16")]; |
| tensor<int32, [4]> var_262 = const()[name = tensor<string, []>("op_262"), val = tensor<int32, [4]>([1, 768, 1, 1])]; |
| tensor<fp16, [1, 768, 1, 1]> input_3_cast_fp16 = reshape(shape = var_262, x = attn_3_cast_fp16)[name = tensor<string, []>("input_3_cast_fp16")]; |
| tensor<string, []> obj_11_pad_type_0 = const()[name = tensor<string, []>("obj_11_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> obj_11_strides_0 = const()[name = tensor<string, []>("obj_11_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> obj_11_pad_0 = const()[name = tensor<string, []>("obj_11_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> obj_11_dilations_0 = const()[name = tensor<string, []>("obj_11_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> obj_11_groups_0 = const()[name = tensor<string, []>("obj_11_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 768, 1, 1]> layers_0_encoder_attn_o_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_0_encoder_attn_o_proj_weight_to_fp16"), val = tensor<fp16, [768, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(88628544)))]; |
| tensor<fp16, [768]> layers_0_encoder_attn_o_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_0_encoder_attn_o_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(89808256)))]; |
| tensor<fp16, [1, 768, 1, 1]> obj_11_cast_fp16 = conv(bias = layers_0_encoder_attn_o_proj_bias_to_fp16, dilations = obj_11_dilations_0, groups = obj_11_groups_0, pad = obj_11_pad_0, pad_type = obj_11_pad_type_0, strides = obj_11_strides_0, weight = layers_0_encoder_attn_o_proj_weight_to_fp16, x = input_3_cast_fp16)[name = tensor<string, []>("obj_11_cast_fp16")]; |
| tensor<fp16, [1, 768, 1, 1]> inputs_5_cast_fp16 = add(x = inputs_3_cast_fp16, y = obj_11_cast_fp16)[name = tensor<string, []>("inputs_5_cast_fp16")]; |
| tensor<int32, [1]> out_5_axes_0 = const()[name = tensor<string, []>("out_5_axes_0"), val = tensor<int32, [1]>([1])]; |
| tensor<fp16, []> var_280_to_fp16 = const()[name = tensor<string, []>("op_280_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, 768, 1, 1]> out_5_cast_fp16 = layer_norm(axes = out_5_axes_0, epsilon = var_280_to_fp16, x = inputs_5_cast_fp16)[name = tensor<string, []>("out_5_cast_fp16")]; |
| tensor<fp16, [768]> input_5_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_5_gamma_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(89809856)))]; |
| tensor<fp16, [768]> input_5_beta_0_to_fp16 = const()[name = tensor<string, []>("input_5_beta_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(89811456)))]; |
| tensor<fp16, []> input_5_epsilon_0_to_fp16 = const()[name = tensor<string, []>("input_5_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, 768, 1, 1]> input_5_cast_fp16 = batch_norm(beta = input_5_beta_0_to_fp16, epsilon = input_5_epsilon_0_to_fp16, gamma = input_5_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_5_cast_fp16)[name = tensor<string, []>("input_5_cast_fp16")]; |
| tensor<string, []> input_7_pad_type_0 = const()[name = tensor<string, []>("input_7_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> input_7_strides_0 = const()[name = tensor<string, []>("input_7_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> input_7_pad_0 = const()[name = tensor<string, []>("input_7_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> input_7_dilations_0 = const()[name = tensor<string, []>("input_7_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> input_7_groups_0 = const()[name = tensor<string, []>("input_7_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [3072, 768, 1, 1]> layers_0_fc1_weight_to_fp16 = const()[name = tensor<string, []>("layers_0_fc1_weight_to_fp16"), val = tensor<fp16, [3072, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(89813056)))]; |
| tensor<fp16, [3072]> layers_0_fc1_bias_to_fp16 = const()[name = tensor<string, []>("layers_0_fc1_bias_to_fp16"), val = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(94531712)))]; |
| tensor<fp16, [1, 3072, 1, 1]> input_7_cast_fp16 = conv(bias = layers_0_fc1_bias_to_fp16, dilations = input_7_dilations_0, groups = input_7_groups_0, pad = input_7_pad_0, pad_type = input_7_pad_type_0, strides = input_7_strides_0, weight = layers_0_fc1_weight_to_fp16, x = input_5_cast_fp16)[name = tensor<string, []>("input_7_cast_fp16")]; |
| tensor<string, []> input_9_mode_0 = const()[name = tensor<string, []>("input_9_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 3072, 1, 1]> input_9_cast_fp16 = gelu(mode = input_9_mode_0, x = input_7_cast_fp16)[name = tensor<string, []>("input_9_cast_fp16")]; |
| tensor<string, []> hidden_states_3_pad_type_0 = const()[name = tensor<string, []>("hidden_states_3_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> hidden_states_3_strides_0 = const()[name = tensor<string, []>("hidden_states_3_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> hidden_states_3_pad_0 = const()[name = tensor<string, []>("hidden_states_3_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> hidden_states_3_dilations_0 = const()[name = tensor<string, []>("hidden_states_3_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> hidden_states_3_groups_0 = const()[name = tensor<string, []>("hidden_states_3_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 3072, 1, 1]> layers_0_fc2_weight_to_fp16 = const()[name = tensor<string, []>("layers_0_fc2_weight_to_fp16"), val = tensor<fp16, [768, 3072, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(94537920)))]; |
| tensor<fp16, [768]> layers_0_fc2_bias_to_fp16 = const()[name = tensor<string, []>("layers_0_fc2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(99256576)))]; |
| tensor<fp16, [1, 768, 1, 1]> hidden_states_3_cast_fp16 = conv(bias = layers_0_fc2_bias_to_fp16, dilations = hidden_states_3_dilations_0, groups = hidden_states_3_groups_0, pad = hidden_states_3_pad_0, pad_type = hidden_states_3_pad_type_0, strides = hidden_states_3_strides_0, weight = layers_0_fc2_weight_to_fp16, x = input_9_cast_fp16)[name = tensor<string, []>("hidden_states_3_cast_fp16")]; |
| tensor<fp16, [1, 768, 1, 1]> inputs_7_cast_fp16 = add(x = inputs_5_cast_fp16, y = hidden_states_3_cast_fp16)[name = tensor<string, []>("inputs_7_cast_fp16")]; |
| tensor<int32, []> var_315 = const()[name = tensor<string, []>("op_315"), val = tensor<int32, []>(3)]; |
| tensor<int32, [1]> out_7_axes_0 = const()[name = tensor<string, []>("out_7_axes_0"), val = tensor<int32, [1]>([1])]; |
| tensor<fp16, []> var_340_to_fp16 = const()[name = tensor<string, []>("op_340_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, 768, 1, 1]> out_7_cast_fp16 = layer_norm(axes = out_7_axes_0, epsilon = var_340_to_fp16, x = inputs_7_cast_fp16)[name = tensor<string, []>("out_7_cast_fp16")]; |
| tensor<fp16, [768]> obj_15_gamma_0_to_fp16 = const()[name = tensor<string, []>("obj_15_gamma_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(99258176)))]; |
| tensor<fp16, [768]> obj_15_beta_0_to_fp16 = const()[name = tensor<string, []>("obj_15_beta_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(99259776)))]; |
| tensor<fp16, []> obj_15_epsilon_0_to_fp16 = const()[name = tensor<string, []>("obj_15_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, 768, 1, 1]> obj_15_cast_fp16 = batch_norm(beta = obj_15_beta_0_to_fp16, epsilon = obj_15_epsilon_0_to_fp16, gamma = obj_15_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_7_cast_fp16)[name = tensor<string, []>("obj_15_cast_fp16")]; |
| tensor<string, []> query_5_pad_type_0 = const()[name = tensor<string, []>("query_5_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> query_5_strides_0 = const()[name = tensor<string, []>("query_5_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> query_5_pad_0 = const()[name = tensor<string, []>("query_5_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> query_5_dilations_0 = const()[name = tensor<string, []>("query_5_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> query_5_groups_0 = const()[name = tensor<string, []>("query_5_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 768, 1, 1]> layers_1_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_1_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [768, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(99261376)))]; |
| tensor<fp16, [768]> layers_1_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("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, []>(100441088)))]; |
| tensor<fp16, [1, 768, 1, 1]> query_5_cast_fp16 = conv(bias = layers_1_self_attn_q_proj_bias_to_fp16, dilations = query_5_dilations_0, groups = query_5_groups_0, pad = query_5_pad_0, pad_type = query_5_pad_type_0, strides = query_5_strides_0, weight = layers_1_self_attn_q_proj_weight_to_fp16, x = obj_15_cast_fp16)[name = tensor<string, []>("query_5_cast_fp16")]; |
| tensor<string, []> current_key_3_pad_type_0 = const()[name = tensor<string, []>("current_key_3_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> current_key_3_strides_0 = const()[name = tensor<string, []>("current_key_3_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> current_key_3_pad_0 = const()[name = tensor<string, []>("current_key_3_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> current_key_3_dilations_0 = const()[name = tensor<string, []>("current_key_3_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> current_key_3_groups_0 = const()[name = tensor<string, []>("current_key_3_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 768, 1, 1]> layers_1_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_1_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [768, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(100442688)))]; |
| tensor<fp16, [1, 768, 1, 1]> current_key_3_cast_fp16 = conv(dilations = current_key_3_dilations_0, groups = current_key_3_groups_0, pad = current_key_3_pad_0, pad_type = current_key_3_pad_type_0, strides = current_key_3_strides_0, weight = layers_1_self_attn_k_proj_weight_to_fp16, x = obj_15_cast_fp16)[name = tensor<string, []>("current_key_3_cast_fp16")]; |
| tensor<string, []> current_value_3_pad_type_0 = const()[name = tensor<string, []>("current_value_3_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> current_value_3_strides_0 = const()[name = tensor<string, []>("current_value_3_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> current_value_3_pad_0 = const()[name = tensor<string, []>("current_value_3_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> current_value_3_dilations_0 = const()[name = tensor<string, []>("current_value_3_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> current_value_3_groups_0 = const()[name = tensor<string, []>("current_value_3_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 768, 1, 1]> layers_1_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_1_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [768, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(101622400)))]; |
| tensor<fp16, [768]> layers_1_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("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, []>(102802112)))]; |
| tensor<fp16, [1, 768, 1, 1]> current_value_3_cast_fp16 = conv(bias = layers_1_self_attn_v_proj_bias_to_fp16, dilations = current_value_3_dilations_0, groups = current_value_3_groups_0, pad = current_value_3_pad_0, pad_type = current_value_3_pad_type_0, strides = current_value_3_strides_0, weight = layers_1_self_attn_v_proj_weight_to_fp16, x = obj_15_cast_fp16)[name = tensor<string, []>("current_value_3_cast_fp16")]; |
| tensor<fp16, [1, 768, 1, 448]> var_379_cast_fp16 = mul(x = var_63_cast_fp16_1, y = var_159_cast_fp16)[name = tensor<string, []>("op_379_cast_fp16")]; |
| tensor<fp16, [1, 768, 1, 448]> var_380_cast_fp16 = mul(x = current_key_3_cast_fp16, y = var_157_cast_fp16)[name = tensor<string, []>("op_380_cast_fp16")]; |
| tensor<fp16, [1, 768, 1, 448]> key_5_cast_fp16 = add(x = var_379_cast_fp16, y = var_380_cast_fp16)[name = tensor<string, []>("key_5_cast_fp16")]; |
| tensor<fp16, [1, 768, 1, 448]> var_383_cast_fp16 = mul(x = var_78_cast_fp16_1, y = var_159_cast_fp16)[name = tensor<string, []>("op_383_cast_fp16")]; |
| tensor<fp16, [1, 768, 1, 448]> var_384_cast_fp16 = mul(x = current_value_3_cast_fp16, y = var_157_cast_fp16)[name = tensor<string, []>("op_384_cast_fp16")]; |
| tensor<fp16, [1, 768, 1, 448]> value_5_cast_fp16 = add(x = var_383_cast_fp16, y = var_384_cast_fp16)[name = tensor<string, []>("value_5_cast_fp16")]; |
| tensor<int32, [4]> var_388 = const()[name = tensor<string, []>("op_388"), val = tensor<int32, [4]>([1, 12, 64, 1])]; |
| tensor<fp16, [1, 12, 64, 1]> mh_q_5_cast_fp16 = reshape(shape = var_388, x = query_5_cast_fp16)[name = tensor<string, []>("mh_q_5_cast_fp16")]; |
| tensor<fp16, []> var_390_to_fp16 = const()[name = tensor<string, []>("op_390_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; |
| tensor<fp16, [1, 12, 64, 1]> var_391_cast_fp16 = mul(x = mh_q_5_cast_fp16, y = var_390_to_fp16)[name = tensor<string, []>("op_391_cast_fp16")]; |
| tensor<int32, [4]> var_394 = const()[name = tensor<string, []>("op_394"), val = tensor<int32, [4]>([1, 12, 64, 448])]; |
| tensor<fp16, [1, 12, 64, 448]> var_395_cast_fp16 = reshape(shape = var_394, x = key_5_cast_fp16)[name = tensor<string, []>("op_395_cast_fp16")]; |
| tensor<bool, []> mh_w_7_transpose_x_0 = const()[name = tensor<string, []>("mh_w_7_transpose_x_0"), val = tensor<bool, []>(true)]; |
| tensor<bool, []> mh_w_7_transpose_y_0 = const()[name = tensor<string, []>("mh_w_7_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [1, 12, 1, 448]> mh_w_7_cast_fp16 = matmul(transpose_x = mh_w_7_transpose_x_0, transpose_y = mh_w_7_transpose_y_0, x = var_391_cast_fp16, y = var_395_cast_fp16)[name = tensor<string, []>("mh_w_7_cast_fp16")]; |
| tensor<fp16, [1, 12, 1, 448]> mh_w_9_cast_fp16 = add(x = mh_w_7_cast_fp16, y = var_181_cast_fp16)[name = tensor<string, []>("mh_w_9_cast_fp16")]; |
| tensor<fp16, [1, 12, 1, 448]> var_403_cast_fp16 = softmax(axis = var_315, x = mh_w_9_cast_fp16)[name = tensor<string, []>("op_403_cast_fp16")]; |
| tensor<int32, [4]> var_404 = const()[name = tensor<string, []>("op_404"), val = tensor<int32, [4]>([1, 12, 64, 448])]; |
| tensor<fp16, [1, 12, 64, 448]> var_405_cast_fp16 = reshape(shape = var_404, x = value_5_cast_fp16)[name = tensor<string, []>("op_405_cast_fp16")]; |
| tensor<bool, []> attn_5_transpose_x_0 = const()[name = tensor<string, []>("attn_5_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_5_transpose_y_0 = const()[name = tensor<string, []>("attn_5_transpose_y_0"), val = tensor<bool, []>(true)]; |
| tensor<fp16, [1, 12, 64, 1]> attn_5_cast_fp16 = matmul(transpose_x = attn_5_transpose_x_0, transpose_y = attn_5_transpose_y_0, x = var_405_cast_fp16, y = var_403_cast_fp16)[name = tensor<string, []>("attn_5_cast_fp16")]; |
| tensor<int32, [4]> var_408 = const()[name = tensor<string, []>("op_408"), val = tensor<int32, [4]>([1, 768, 1, 1])]; |
| tensor<fp16, [1, 768, 1, 1]> input_11_cast_fp16 = reshape(shape = var_408, x = attn_5_cast_fp16)[name = tensor<string, []>("input_11_cast_fp16")]; |
| tensor<string, []> obj_21_pad_type_0 = const()[name = tensor<string, []>("obj_21_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> obj_21_strides_0 = const()[name = tensor<string, []>("obj_21_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> obj_21_pad_0 = const()[name = tensor<string, []>("obj_21_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> obj_21_dilations_0 = const()[name = tensor<string, []>("obj_21_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> obj_21_groups_0 = const()[name = tensor<string, []>("obj_21_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 768, 1, 1]> layers_1_self_attn_o_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_1_self_attn_o_proj_weight_to_fp16"), val = tensor<fp16, [768, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(102803712)))]; |
| tensor<fp16, [768]> layers_1_self_attn_o_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_1_self_attn_o_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(103983424)))]; |
| tensor<fp16, [1, 768, 1, 1]> obj_21_cast_fp16 = conv(bias = layers_1_self_attn_o_proj_bias_to_fp16, dilations = obj_21_dilations_0, groups = obj_21_groups_0, pad = obj_21_pad_0, pad_type = obj_21_pad_type_0, strides = obj_21_strides_0, weight = layers_1_self_attn_o_proj_weight_to_fp16, x = input_11_cast_fp16)[name = tensor<string, []>("obj_21_cast_fp16")]; |
| tensor<fp16, [1, 768, 1, 1]> inputs_9_cast_fp16 = add(x = inputs_7_cast_fp16, y = obj_21_cast_fp16)[name = tensor<string, []>("inputs_9_cast_fp16")]; |
| tensor<int32, [1]> out_9_axes_0 = const()[name = tensor<string, []>("out_9_axes_0"), val = tensor<int32, [1]>([1])]; |
| tensor<fp16, []> var_430_to_fp16 = const()[name = tensor<string, []>("op_430_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, 768, 1, 1]> out_9_cast_fp16 = layer_norm(axes = out_9_axes_0, epsilon = var_430_to_fp16, x = inputs_9_cast_fp16)[name = tensor<string, []>("out_9_cast_fp16")]; |
| tensor<fp16, [768]> obj_23_gamma_0_to_fp16 = const()[name = tensor<string, []>("obj_23_gamma_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(103985024)))]; |
| tensor<fp16, [768]> obj_23_beta_0_to_fp16 = const()[name = tensor<string, []>("obj_23_beta_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(103986624)))]; |
| tensor<fp16, []> obj_23_epsilon_0_to_fp16 = const()[name = tensor<string, []>("obj_23_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, 768, 1, 1]> obj_23_cast_fp16 = batch_norm(beta = obj_23_beta_0_to_fp16, epsilon = obj_23_epsilon_0_to_fp16, gamma = obj_23_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_9_cast_fp16)[name = tensor<string, []>("obj_23_cast_fp16")]; |
| tensor<string, []> query_7_pad_type_0 = const()[name = tensor<string, []>("query_7_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> query_7_strides_0 = const()[name = tensor<string, []>("query_7_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> query_7_pad_0 = const()[name = tensor<string, []>("query_7_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> query_7_dilations_0 = const()[name = tensor<string, []>("query_7_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> query_7_groups_0 = const()[name = tensor<string, []>("query_7_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 768, 1, 1]> layers_1_encoder_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_1_encoder_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [768, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(103988224)))]; |
| tensor<fp16, [768]> layers_1_encoder_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_1_encoder_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(105167936)))]; |
| tensor<fp16, [1, 768, 1, 1]> query_7_cast_fp16 = conv(bias = layers_1_encoder_attn_q_proj_bias_to_fp16, dilations = query_7_dilations_0, groups = query_7_groups_0, pad = query_7_pad_0, pad_type = query_7_pad_type_0, strides = query_7_strides_0, weight = layers_1_encoder_attn_q_proj_weight_to_fp16, x = obj_23_cast_fp16)[name = tensor<string, []>("query_7_cast_fp16")]; |
| tensor<string, []> key_7_pad_type_0 = const()[name = tensor<string, []>("key_7_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> key_7_strides_0 = const()[name = tensor<string, []>("key_7_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> key_7_pad_0 = const()[name = tensor<string, []>("key_7_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> key_7_dilations_0 = const()[name = tensor<string, []>("key_7_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> key_7_groups_0 = const()[name = tensor<string, []>("key_7_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 768, 1, 1]> layers_1_encoder_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_1_encoder_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [768, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(105169536)))]; |
| tensor<fp16, [1, 768, 1, 1500]> key_7_cast_fp16 = conv(dilations = key_7_dilations_0, groups = key_7_groups_0, pad = key_7_pad_0, pad_type = key_7_pad_type_0, strides = key_7_strides_0, weight = layers_1_encoder_attn_k_proj_weight_to_fp16, x = encoder_output_embeds)[name = tensor<string, []>("key_7_cast_fp16")]; |
| tensor<string, []> value_7_pad_type_0 = const()[name = tensor<string, []>("value_7_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> value_7_strides_0 = const()[name = tensor<string, []>("value_7_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> value_7_pad_0 = const()[name = tensor<string, []>("value_7_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> value_7_dilations_0 = const()[name = tensor<string, []>("value_7_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> value_7_groups_0 = const()[name = tensor<string, []>("value_7_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 768, 1, 1]> layers_1_encoder_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_1_encoder_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [768, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(106349248)))]; |
| tensor<fp16, [768]> layers_1_encoder_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_1_encoder_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(107528960)))]; |
| tensor<fp16, [1, 768, 1, 1500]> value_7_cast_fp16 = conv(bias = layers_1_encoder_attn_v_proj_bias_to_fp16, dilations = value_7_dilations_0, groups = value_7_groups_0, pad = value_7_pad_0, pad_type = value_7_pad_type_0, strides = value_7_strides_0, weight = layers_1_encoder_attn_v_proj_weight_to_fp16, x = encoder_output_embeds)[name = tensor<string, []>("value_7_cast_fp16")]; |
| tensor<int32, [4]> var_466 = const()[name = tensor<string, []>("op_466"), val = tensor<int32, [4]>([1, 12, 64, 1])]; |
| tensor<fp16, [1, 12, 64, 1]> mh_q_7_cast_fp16 = reshape(shape = var_466, x = query_7_cast_fp16)[name = tensor<string, []>("mh_q_7_cast_fp16")]; |
| tensor<fp16, []> var_468_to_fp16 = const()[name = tensor<string, []>("op_468_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; |
| tensor<fp16, [1, 12, 64, 1]> var_469_cast_fp16 = mul(x = mh_q_7_cast_fp16, y = var_468_to_fp16)[name = tensor<string, []>("op_469_cast_fp16")]; |
| tensor<int32, [4]> var_472 = const()[name = tensor<string, []>("op_472"), val = tensor<int32, [4]>([1, 12, 64, 1500])]; |
| tensor<fp16, [1, 12, 64, 1500]> var_473_cast_fp16 = reshape(shape = var_472, x = key_7_cast_fp16)[name = tensor<string, []>("op_473_cast_fp16")]; |
| tensor<bool, []> mh_w_11_transpose_x_0 = const()[name = tensor<string, []>("mh_w_11_transpose_x_0"), val = tensor<bool, []>(true)]; |
| tensor<bool, []> mh_w_11_transpose_y_0 = const()[name = tensor<string, []>("mh_w_11_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [1, 12, 1, 1500]> mh_w_11_cast_fp16 = matmul(transpose_x = mh_w_11_transpose_x_0, transpose_y = mh_w_11_transpose_y_0, x = var_469_cast_fp16, y = var_473_cast_fp16)[name = tensor<string, []>("mh_w_11_cast_fp16")]; |
| tensor<fp16, [1, 12, 1, 1500]> obj_27_cast_fp16 = softmax(axis = var_315, x = mh_w_11_cast_fp16)[name = tensor<string, []>("obj_27_cast_fp16")]; |
| tensor<int32, [4]> var_477 = const()[name = tensor<string, []>("op_477"), val = tensor<int32, [4]>([1, 12, 64, 1500])]; |
| tensor<fp16, [1, 12, 64, 1500]> var_478_cast_fp16 = reshape(shape = var_477, x = value_7_cast_fp16)[name = tensor<string, []>("op_478_cast_fp16")]; |
| tensor<bool, []> attn_7_transpose_x_0 = const()[name = tensor<string, []>("attn_7_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_7_transpose_y_0 = const()[name = tensor<string, []>("attn_7_transpose_y_0"), val = tensor<bool, []>(true)]; |
| tensor<fp16, [1, 12, 64, 1]> attn_7_cast_fp16 = matmul(transpose_x = attn_7_transpose_x_0, transpose_y = attn_7_transpose_y_0, x = var_478_cast_fp16, y = obj_27_cast_fp16)[name = tensor<string, []>("attn_7_cast_fp16")]; |
| tensor<int32, [4]> var_481 = const()[name = tensor<string, []>("op_481"), val = tensor<int32, [4]>([1, 768, 1, 1])]; |
| tensor<fp16, [1, 768, 1, 1]> input_13_cast_fp16 = reshape(shape = var_481, x = attn_7_cast_fp16)[name = tensor<string, []>("input_13_cast_fp16")]; |
| tensor<string, []> obj_25_pad_type_0 = const()[name = tensor<string, []>("obj_25_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> obj_25_strides_0 = const()[name = tensor<string, []>("obj_25_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> obj_25_pad_0 = const()[name = tensor<string, []>("obj_25_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> obj_25_dilations_0 = const()[name = tensor<string, []>("obj_25_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> obj_25_groups_0 = const()[name = tensor<string, []>("obj_25_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 768, 1, 1]> layers_1_encoder_attn_o_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_1_encoder_attn_o_proj_weight_to_fp16"), val = tensor<fp16, [768, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(107530560)))]; |
| tensor<fp16, [768]> layers_1_encoder_attn_o_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_1_encoder_attn_o_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(108710272)))]; |
| tensor<fp16, [1, 768, 1, 1]> obj_25_cast_fp16 = conv(bias = layers_1_encoder_attn_o_proj_bias_to_fp16, dilations = obj_25_dilations_0, groups = obj_25_groups_0, pad = obj_25_pad_0, pad_type = obj_25_pad_type_0, strides = obj_25_strides_0, weight = layers_1_encoder_attn_o_proj_weight_to_fp16, x = input_13_cast_fp16)[name = tensor<string, []>("obj_25_cast_fp16")]; |
| tensor<fp16, [1, 768, 1, 1]> inputs_11_cast_fp16 = add(x = inputs_9_cast_fp16, y = obj_25_cast_fp16)[name = tensor<string, []>("inputs_11_cast_fp16")]; |
| tensor<int32, [1]> out_11_axes_0 = const()[name = tensor<string, []>("out_11_axes_0"), val = tensor<int32, [1]>([1])]; |
| tensor<fp16, []> var_499_to_fp16 = const()[name = tensor<string, []>("op_499_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, 768, 1, 1]> out_11_cast_fp16 = layer_norm(axes = out_11_axes_0, epsilon = var_499_to_fp16, x = inputs_11_cast_fp16)[name = tensor<string, []>("out_11_cast_fp16")]; |
| tensor<fp16, [768]> input_15_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_15_gamma_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(108711872)))]; |
| tensor<fp16, [768]> input_15_beta_0_to_fp16 = const()[name = tensor<string, []>("input_15_beta_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(108713472)))]; |
| tensor<fp16, []> input_15_epsilon_0_to_fp16 = const()[name = tensor<string, []>("input_15_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, 768, 1, 1]> input_15_cast_fp16 = batch_norm(beta = input_15_beta_0_to_fp16, epsilon = input_15_epsilon_0_to_fp16, gamma = input_15_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_11_cast_fp16)[name = tensor<string, []>("input_15_cast_fp16")]; |
| tensor<string, []> input_17_pad_type_0 = const()[name = tensor<string, []>("input_17_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> input_17_strides_0 = const()[name = tensor<string, []>("input_17_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> input_17_pad_0 = const()[name = tensor<string, []>("input_17_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> input_17_dilations_0 = const()[name = tensor<string, []>("input_17_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> input_17_groups_0 = const()[name = tensor<string, []>("input_17_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [3072, 768, 1, 1]> layers_1_fc1_weight_to_fp16 = const()[name = tensor<string, []>("layers_1_fc1_weight_to_fp16"), val = tensor<fp16, [3072, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(108715072)))]; |
| tensor<fp16, [3072]> layers_1_fc1_bias_to_fp16 = const()[name = tensor<string, []>("layers_1_fc1_bias_to_fp16"), val = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(113433728)))]; |
| tensor<fp16, [1, 3072, 1, 1]> input_17_cast_fp16 = conv(bias = layers_1_fc1_bias_to_fp16, dilations = input_17_dilations_0, groups = input_17_groups_0, pad = input_17_pad_0, pad_type = input_17_pad_type_0, strides = input_17_strides_0, weight = layers_1_fc1_weight_to_fp16, x = input_15_cast_fp16)[name = tensor<string, []>("input_17_cast_fp16")]; |
| tensor<string, []> input_19_mode_0 = const()[name = tensor<string, []>("input_19_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 3072, 1, 1]> input_19_cast_fp16 = gelu(mode = input_19_mode_0, x = input_17_cast_fp16)[name = tensor<string, []>("input_19_cast_fp16")]; |
| tensor<string, []> hidden_states_5_pad_type_0 = const()[name = tensor<string, []>("hidden_states_5_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> hidden_states_5_strides_0 = const()[name = tensor<string, []>("hidden_states_5_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> hidden_states_5_pad_0 = const()[name = tensor<string, []>("hidden_states_5_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> hidden_states_5_dilations_0 = const()[name = tensor<string, []>("hidden_states_5_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> hidden_states_5_groups_0 = const()[name = tensor<string, []>("hidden_states_5_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 3072, 1, 1]> layers_1_fc2_weight_to_fp16 = const()[name = tensor<string, []>("layers_1_fc2_weight_to_fp16"), val = tensor<fp16, [768, 3072, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(113439936)))]; |
| tensor<fp16, [768]> layers_1_fc2_bias_to_fp16 = const()[name = tensor<string, []>("layers_1_fc2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(118158592)))]; |
| tensor<fp16, [1, 768, 1, 1]> hidden_states_5_cast_fp16 = conv(bias = layers_1_fc2_bias_to_fp16, dilations = hidden_states_5_dilations_0, groups = hidden_states_5_groups_0, pad = hidden_states_5_pad_0, pad_type = hidden_states_5_pad_type_0, strides = hidden_states_5_strides_0, weight = layers_1_fc2_weight_to_fp16, x = input_19_cast_fp16)[name = tensor<string, []>("hidden_states_5_cast_fp16")]; |
| tensor<fp16, [1, 768, 1, 1]> inputs_13_cast_fp16 = add(x = inputs_11_cast_fp16, y = hidden_states_5_cast_fp16)[name = tensor<string, []>("inputs_13_cast_fp16")]; |
| tensor<int32, []> var_534 = const()[name = tensor<string, []>("op_534"), val = tensor<int32, []>(3)]; |
| tensor<int32, [1]> out_13_axes_0 = const()[name = tensor<string, []>("out_13_axes_0"), val = tensor<int32, [1]>([1])]; |
| tensor<fp16, []> var_559_to_fp16 = const()[name = tensor<string, []>("op_559_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, 768, 1, 1]> out_13_cast_fp16 = layer_norm(axes = out_13_axes_0, epsilon = var_559_to_fp16, x = inputs_13_cast_fp16)[name = tensor<string, []>("out_13_cast_fp16")]; |
| tensor<fp16, [768]> obj_29_gamma_0_to_fp16 = const()[name = tensor<string, []>("obj_29_gamma_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(118160192)))]; |
| tensor<fp16, [768]> obj_29_beta_0_to_fp16 = const()[name = tensor<string, []>("obj_29_beta_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(118161792)))]; |
| tensor<fp16, []> obj_29_epsilon_0_to_fp16 = const()[name = tensor<string, []>("obj_29_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, 768, 1, 1]> obj_29_cast_fp16 = batch_norm(beta = obj_29_beta_0_to_fp16, epsilon = obj_29_epsilon_0_to_fp16, gamma = obj_29_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_13_cast_fp16)[name = tensor<string, []>("obj_29_cast_fp16")]; |
| tensor<string, []> query_9_pad_type_0 = const()[name = tensor<string, []>("query_9_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> query_9_strides_0 = const()[name = tensor<string, []>("query_9_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> query_9_pad_0 = const()[name = tensor<string, []>("query_9_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> query_9_dilations_0 = const()[name = tensor<string, []>("query_9_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> query_9_groups_0 = const()[name = tensor<string, []>("query_9_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 768, 1, 1]> layers_2_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_2_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [768, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(118163392)))]; |
| tensor<fp16, [768]> layers_2_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("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, []>(119343104)))]; |
| tensor<fp16, [1, 768, 1, 1]> query_9_cast_fp16 = conv(bias = layers_2_self_attn_q_proj_bias_to_fp16, dilations = query_9_dilations_0, groups = query_9_groups_0, pad = query_9_pad_0, pad_type = query_9_pad_type_0, strides = query_9_strides_0, weight = layers_2_self_attn_q_proj_weight_to_fp16, x = obj_29_cast_fp16)[name = tensor<string, []>("query_9_cast_fp16")]; |
| tensor<string, []> current_key_5_pad_type_0 = const()[name = tensor<string, []>("current_key_5_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> current_key_5_strides_0 = const()[name = tensor<string, []>("current_key_5_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> current_key_5_pad_0 = const()[name = tensor<string, []>("current_key_5_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> current_key_5_dilations_0 = const()[name = tensor<string, []>("current_key_5_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> current_key_5_groups_0 = const()[name = tensor<string, []>("current_key_5_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 768, 1, 1]> layers_2_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_2_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [768, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(119344704)))]; |
| tensor<fp16, [1, 768, 1, 1]> current_key_5_cast_fp16 = conv(dilations = current_key_5_dilations_0, groups = current_key_5_groups_0, pad = current_key_5_pad_0, pad_type = current_key_5_pad_type_0, strides = current_key_5_strides_0, weight = layers_2_self_attn_k_proj_weight_to_fp16, x = obj_29_cast_fp16)[name = tensor<string, []>("current_key_5_cast_fp16")]; |
| tensor<string, []> current_value_5_pad_type_0 = const()[name = tensor<string, []>("current_value_5_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> current_value_5_strides_0 = const()[name = tensor<string, []>("current_value_5_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> current_value_5_pad_0 = const()[name = tensor<string, []>("current_value_5_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> current_value_5_dilations_0 = const()[name = tensor<string, []>("current_value_5_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> current_value_5_groups_0 = const()[name = tensor<string, []>("current_value_5_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 768, 1, 1]> layers_2_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_2_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [768, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(120524416)))]; |
| tensor<fp16, [768]> layers_2_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("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, []>(121704128)))]; |
| tensor<fp16, [1, 768, 1, 1]> current_value_5_cast_fp16 = conv(bias = layers_2_self_attn_v_proj_bias_to_fp16, dilations = current_value_5_dilations_0, groups = current_value_5_groups_0, pad = current_value_5_pad_0, pad_type = current_value_5_pad_type_0, strides = current_value_5_strides_0, weight = layers_2_self_attn_v_proj_weight_to_fp16, x = obj_29_cast_fp16)[name = tensor<string, []>("current_value_5_cast_fp16")]; |
| tensor<fp16, [1, 768, 1, 448]> var_598_cast_fp16 = mul(x = var_63_cast_fp16_2, y = var_159_cast_fp16)[name = tensor<string, []>("op_598_cast_fp16")]; |
| tensor<fp16, [1, 768, 1, 448]> var_599_cast_fp16 = mul(x = current_key_5_cast_fp16, y = var_157_cast_fp16)[name = tensor<string, []>("op_599_cast_fp16")]; |
| tensor<fp16, [1, 768, 1, 448]> key_9_cast_fp16 = add(x = var_598_cast_fp16, y = var_599_cast_fp16)[name = tensor<string, []>("key_9_cast_fp16")]; |
| tensor<fp16, [1, 768, 1, 448]> var_602_cast_fp16 = mul(x = var_78_cast_fp16_2, y = var_159_cast_fp16)[name = tensor<string, []>("op_602_cast_fp16")]; |
| tensor<fp16, [1, 768, 1, 448]> var_603_cast_fp16 = mul(x = current_value_5_cast_fp16, y = var_157_cast_fp16)[name = tensor<string, []>("op_603_cast_fp16")]; |
| tensor<fp16, [1, 768, 1, 448]> value_9_cast_fp16 = add(x = var_602_cast_fp16, y = var_603_cast_fp16)[name = tensor<string, []>("value_9_cast_fp16")]; |
| tensor<int32, [4]> var_607 = const()[name = tensor<string, []>("op_607"), val = tensor<int32, [4]>([1, 12, 64, 1])]; |
| tensor<fp16, [1, 12, 64, 1]> mh_q_9_cast_fp16 = reshape(shape = var_607, x = query_9_cast_fp16)[name = tensor<string, []>("mh_q_9_cast_fp16")]; |
| tensor<fp16, []> var_609_to_fp16 = const()[name = tensor<string, []>("op_609_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; |
| tensor<fp16, [1, 12, 64, 1]> var_610_cast_fp16 = mul(x = mh_q_9_cast_fp16, y = var_609_to_fp16)[name = tensor<string, []>("op_610_cast_fp16")]; |
| tensor<int32, [4]> var_613 = const()[name = tensor<string, []>("op_613"), val = tensor<int32, [4]>([1, 12, 64, 448])]; |
| tensor<fp16, [1, 12, 64, 448]> var_614_cast_fp16 = reshape(shape = var_613, x = key_9_cast_fp16)[name = tensor<string, []>("op_614_cast_fp16")]; |
| tensor<bool, []> mh_w_13_transpose_x_0 = const()[name = tensor<string, []>("mh_w_13_transpose_x_0"), val = tensor<bool, []>(true)]; |
| tensor<bool, []> mh_w_13_transpose_y_0 = const()[name = tensor<string, []>("mh_w_13_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [1, 12, 1, 448]> mh_w_13_cast_fp16 = matmul(transpose_x = mh_w_13_transpose_x_0, transpose_y = mh_w_13_transpose_y_0, x = var_610_cast_fp16, y = var_614_cast_fp16)[name = tensor<string, []>("mh_w_13_cast_fp16")]; |
| tensor<fp16, [1, 12, 1, 448]> mh_w_15_cast_fp16 = add(x = mh_w_13_cast_fp16, y = var_181_cast_fp16)[name = tensor<string, []>("mh_w_15_cast_fp16")]; |
| tensor<fp16, [1, 12, 1, 448]> var_622_cast_fp16 = softmax(axis = var_534, x = mh_w_15_cast_fp16)[name = tensor<string, []>("op_622_cast_fp16")]; |
| tensor<int32, [4]> var_623 = const()[name = tensor<string, []>("op_623"), val = tensor<int32, [4]>([1, 12, 64, 448])]; |
| tensor<fp16, [1, 12, 64, 448]> var_624_cast_fp16 = reshape(shape = var_623, x = value_9_cast_fp16)[name = tensor<string, []>("op_624_cast_fp16")]; |
| tensor<bool, []> attn_9_transpose_x_0 = const()[name = tensor<string, []>("attn_9_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_9_transpose_y_0 = const()[name = tensor<string, []>("attn_9_transpose_y_0"), val = tensor<bool, []>(true)]; |
| tensor<fp16, [1, 12, 64, 1]> attn_9_cast_fp16 = matmul(transpose_x = attn_9_transpose_x_0, transpose_y = attn_9_transpose_y_0, x = var_624_cast_fp16, y = var_622_cast_fp16)[name = tensor<string, []>("attn_9_cast_fp16")]; |
| tensor<int32, [4]> var_627 = const()[name = tensor<string, []>("op_627"), val = tensor<int32, [4]>([1, 768, 1, 1])]; |
| tensor<fp16, [1, 768, 1, 1]> input_21_cast_fp16 = reshape(shape = var_627, x = attn_9_cast_fp16)[name = tensor<string, []>("input_21_cast_fp16")]; |
| tensor<string, []> obj_35_pad_type_0 = const()[name = tensor<string, []>("obj_35_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> obj_35_strides_0 = const()[name = tensor<string, []>("obj_35_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> obj_35_pad_0 = const()[name = tensor<string, []>("obj_35_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> obj_35_dilations_0 = const()[name = tensor<string, []>("obj_35_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> obj_35_groups_0 = const()[name = tensor<string, []>("obj_35_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 768, 1, 1]> layers_2_self_attn_o_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_2_self_attn_o_proj_weight_to_fp16"), val = tensor<fp16, [768, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(121705728)))]; |
| tensor<fp16, [768]> layers_2_self_attn_o_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_2_self_attn_o_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(122885440)))]; |
| tensor<fp16, [1, 768, 1, 1]> obj_35_cast_fp16 = conv(bias = layers_2_self_attn_o_proj_bias_to_fp16, dilations = obj_35_dilations_0, groups = obj_35_groups_0, pad = obj_35_pad_0, pad_type = obj_35_pad_type_0, strides = obj_35_strides_0, weight = layers_2_self_attn_o_proj_weight_to_fp16, x = input_21_cast_fp16)[name = tensor<string, []>("obj_35_cast_fp16")]; |
| tensor<fp16, [1, 768, 1, 1]> inputs_15_cast_fp16 = add(x = inputs_13_cast_fp16, y = obj_35_cast_fp16)[name = tensor<string, []>("inputs_15_cast_fp16")]; |
| tensor<int32, [1]> out_15_axes_0 = const()[name = tensor<string, []>("out_15_axes_0"), val = tensor<int32, [1]>([1])]; |
| tensor<fp16, []> var_649_to_fp16 = const()[name = tensor<string, []>("op_649_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, 768, 1, 1]> out_15_cast_fp16 = layer_norm(axes = out_15_axes_0, epsilon = var_649_to_fp16, x = inputs_15_cast_fp16)[name = tensor<string, []>("out_15_cast_fp16")]; |
| tensor<fp16, [768]> obj_37_gamma_0_to_fp16 = const()[name = tensor<string, []>("obj_37_gamma_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(122887040)))]; |
| tensor<fp16, [768]> obj_37_beta_0_to_fp16 = const()[name = tensor<string, []>("obj_37_beta_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(122888640)))]; |
| tensor<fp16, []> obj_37_epsilon_0_to_fp16 = const()[name = tensor<string, []>("obj_37_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, 768, 1, 1]> obj_37_cast_fp16 = batch_norm(beta = obj_37_beta_0_to_fp16, epsilon = obj_37_epsilon_0_to_fp16, gamma = obj_37_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_15_cast_fp16)[name = tensor<string, []>("obj_37_cast_fp16")]; |
| tensor<string, []> query_11_pad_type_0 = const()[name = tensor<string, []>("query_11_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> query_11_strides_0 = const()[name = tensor<string, []>("query_11_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> query_11_pad_0 = const()[name = tensor<string, []>("query_11_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> query_11_dilations_0 = const()[name = tensor<string, []>("query_11_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> query_11_groups_0 = const()[name = tensor<string, []>("query_11_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 768, 1, 1]> layers_2_encoder_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_2_encoder_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [768, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(122890240)))]; |
| tensor<fp16, [768]> layers_2_encoder_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_2_encoder_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(124069952)))]; |
| tensor<fp16, [1, 768, 1, 1]> query_11_cast_fp16 = conv(bias = layers_2_encoder_attn_q_proj_bias_to_fp16, dilations = query_11_dilations_0, groups = query_11_groups_0, pad = query_11_pad_0, pad_type = query_11_pad_type_0, strides = query_11_strides_0, weight = layers_2_encoder_attn_q_proj_weight_to_fp16, x = obj_37_cast_fp16)[name = tensor<string, []>("query_11_cast_fp16")]; |
| tensor<string, []> key_11_pad_type_0 = const()[name = tensor<string, []>("key_11_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> key_11_strides_0 = const()[name = tensor<string, []>("key_11_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> key_11_pad_0 = const()[name = tensor<string, []>("key_11_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> key_11_dilations_0 = const()[name = tensor<string, []>("key_11_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> key_11_groups_0 = const()[name = tensor<string, []>("key_11_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 768, 1, 1]> layers_2_encoder_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_2_encoder_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [768, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(124071552)))]; |
| tensor<fp16, [1, 768, 1, 1500]> key_11_cast_fp16 = conv(dilations = key_11_dilations_0, groups = key_11_groups_0, pad = key_11_pad_0, pad_type = key_11_pad_type_0, strides = key_11_strides_0, weight = layers_2_encoder_attn_k_proj_weight_to_fp16, x = encoder_output_embeds)[name = tensor<string, []>("key_11_cast_fp16")]; |
| tensor<string, []> value_11_pad_type_0 = const()[name = tensor<string, []>("value_11_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> value_11_strides_0 = const()[name = tensor<string, []>("value_11_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> value_11_pad_0 = const()[name = tensor<string, []>("value_11_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> value_11_dilations_0 = const()[name = tensor<string, []>("value_11_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> value_11_groups_0 = const()[name = tensor<string, []>("value_11_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 768, 1, 1]> layers_2_encoder_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_2_encoder_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [768, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(125251264)))]; |
| tensor<fp16, [768]> layers_2_encoder_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_2_encoder_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(126430976)))]; |
| tensor<fp16, [1, 768, 1, 1500]> value_11_cast_fp16 = conv(bias = layers_2_encoder_attn_v_proj_bias_to_fp16, dilations = value_11_dilations_0, groups = value_11_groups_0, pad = value_11_pad_0, pad_type = value_11_pad_type_0, strides = value_11_strides_0, weight = layers_2_encoder_attn_v_proj_weight_to_fp16, x = encoder_output_embeds)[name = tensor<string, []>("value_11_cast_fp16")]; |
| tensor<int32, [4]> var_685 = const()[name = tensor<string, []>("op_685"), val = tensor<int32, [4]>([1, 12, 64, 1])]; |
| tensor<fp16, [1, 12, 64, 1]> mh_q_11_cast_fp16 = reshape(shape = var_685, x = query_11_cast_fp16)[name = tensor<string, []>("mh_q_11_cast_fp16")]; |
| tensor<fp16, []> var_687_to_fp16 = const()[name = tensor<string, []>("op_687_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; |
| tensor<fp16, [1, 12, 64, 1]> var_688_cast_fp16 = mul(x = mh_q_11_cast_fp16, y = var_687_to_fp16)[name = tensor<string, []>("op_688_cast_fp16")]; |
| tensor<int32, [4]> var_691 = const()[name = tensor<string, []>("op_691"), val = tensor<int32, [4]>([1, 12, 64, 1500])]; |
| tensor<fp16, [1, 12, 64, 1500]> var_692_cast_fp16 = reshape(shape = var_691, x = key_11_cast_fp16)[name = tensor<string, []>("op_692_cast_fp16")]; |
| tensor<bool, []> mh_w_17_transpose_x_0 = const()[name = tensor<string, []>("mh_w_17_transpose_x_0"), val = tensor<bool, []>(true)]; |
| tensor<bool, []> mh_w_17_transpose_y_0 = const()[name = tensor<string, []>("mh_w_17_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [1, 12, 1, 1500]> mh_w_17_cast_fp16 = matmul(transpose_x = mh_w_17_transpose_x_0, transpose_y = mh_w_17_transpose_y_0, x = var_688_cast_fp16, y = var_692_cast_fp16)[name = tensor<string, []>("mh_w_17_cast_fp16")]; |
| tensor<fp16, [1, 12, 1, 1500]> obj_41_cast_fp16 = softmax(axis = var_534, x = mh_w_17_cast_fp16)[name = tensor<string, []>("obj_41_cast_fp16")]; |
| tensor<int32, [4]> var_696 = const()[name = tensor<string, []>("op_696"), val = tensor<int32, [4]>([1, 12, 64, 1500])]; |
| tensor<fp16, [1, 12, 64, 1500]> var_697_cast_fp16 = reshape(shape = var_696, x = value_11_cast_fp16)[name = tensor<string, []>("op_697_cast_fp16")]; |
| tensor<bool, []> attn_11_transpose_x_0 = const()[name = tensor<string, []>("attn_11_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_11_transpose_y_0 = const()[name = tensor<string, []>("attn_11_transpose_y_0"), val = tensor<bool, []>(true)]; |
| tensor<fp16, [1, 12, 64, 1]> attn_11_cast_fp16 = matmul(transpose_x = attn_11_transpose_x_0, transpose_y = attn_11_transpose_y_0, x = var_697_cast_fp16, y = obj_41_cast_fp16)[name = tensor<string, []>("attn_11_cast_fp16")]; |
| tensor<int32, [4]> var_700 = const()[name = tensor<string, []>("op_700"), val = tensor<int32, [4]>([1, 768, 1, 1])]; |
| tensor<fp16, [1, 768, 1, 1]> input_23_cast_fp16 = reshape(shape = var_700, x = attn_11_cast_fp16)[name = tensor<string, []>("input_23_cast_fp16")]; |
| tensor<string, []> obj_39_pad_type_0 = const()[name = tensor<string, []>("obj_39_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> obj_39_strides_0 = const()[name = tensor<string, []>("obj_39_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> obj_39_pad_0 = const()[name = tensor<string, []>("obj_39_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> obj_39_dilations_0 = const()[name = tensor<string, []>("obj_39_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> obj_39_groups_0 = const()[name = tensor<string, []>("obj_39_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 768, 1, 1]> layers_2_encoder_attn_o_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_2_encoder_attn_o_proj_weight_to_fp16"), val = tensor<fp16, [768, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(126432576)))]; |
| tensor<fp16, [768]> layers_2_encoder_attn_o_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_2_encoder_attn_o_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(127612288)))]; |
| tensor<fp16, [1, 768, 1, 1]> obj_39_cast_fp16 = conv(bias = layers_2_encoder_attn_o_proj_bias_to_fp16, dilations = obj_39_dilations_0, groups = obj_39_groups_0, pad = obj_39_pad_0, pad_type = obj_39_pad_type_0, strides = obj_39_strides_0, weight = layers_2_encoder_attn_o_proj_weight_to_fp16, x = input_23_cast_fp16)[name = tensor<string, []>("obj_39_cast_fp16")]; |
| tensor<fp16, [1, 768, 1, 1]> inputs_17_cast_fp16 = add(x = inputs_15_cast_fp16, y = obj_39_cast_fp16)[name = tensor<string, []>("inputs_17_cast_fp16")]; |
| tensor<int32, [1]> out_17_axes_0 = const()[name = tensor<string, []>("out_17_axes_0"), val = tensor<int32, [1]>([1])]; |
| tensor<fp16, []> var_718_to_fp16 = const()[name = tensor<string, []>("op_718_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, 768, 1, 1]> out_17_cast_fp16 = layer_norm(axes = out_17_axes_0, epsilon = var_718_to_fp16, x = inputs_17_cast_fp16)[name = tensor<string, []>("out_17_cast_fp16")]; |
| tensor<fp16, [768]> input_25_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_25_gamma_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(127613888)))]; |
| tensor<fp16, [768]> input_25_beta_0_to_fp16 = const()[name = tensor<string, []>("input_25_beta_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(127615488)))]; |
| tensor<fp16, []> input_25_epsilon_0_to_fp16 = const()[name = tensor<string, []>("input_25_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, 768, 1, 1]> input_25_cast_fp16 = batch_norm(beta = input_25_beta_0_to_fp16, epsilon = input_25_epsilon_0_to_fp16, gamma = input_25_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_17_cast_fp16)[name = tensor<string, []>("input_25_cast_fp16")]; |
| tensor<string, []> input_27_pad_type_0 = const()[name = tensor<string, []>("input_27_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> input_27_strides_0 = const()[name = tensor<string, []>("input_27_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> input_27_pad_0 = const()[name = tensor<string, []>("input_27_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> input_27_dilations_0 = const()[name = tensor<string, []>("input_27_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> input_27_groups_0 = const()[name = tensor<string, []>("input_27_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [3072, 768, 1, 1]> layers_2_fc1_weight_to_fp16 = const()[name = tensor<string, []>("layers_2_fc1_weight_to_fp16"), val = tensor<fp16, [3072, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(127617088)))]; |
| tensor<fp16, [3072]> layers_2_fc1_bias_to_fp16 = const()[name = tensor<string, []>("layers_2_fc1_bias_to_fp16"), val = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(132335744)))]; |
| tensor<fp16, [1, 3072, 1, 1]> input_27_cast_fp16 = conv(bias = layers_2_fc1_bias_to_fp16, dilations = input_27_dilations_0, groups = input_27_groups_0, pad = input_27_pad_0, pad_type = input_27_pad_type_0, strides = input_27_strides_0, weight = layers_2_fc1_weight_to_fp16, x = input_25_cast_fp16)[name = tensor<string, []>("input_27_cast_fp16")]; |
| tensor<string, []> input_29_mode_0 = const()[name = tensor<string, []>("input_29_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 3072, 1, 1]> input_29_cast_fp16 = gelu(mode = input_29_mode_0, x = input_27_cast_fp16)[name = tensor<string, []>("input_29_cast_fp16")]; |
| tensor<string, []> hidden_states_7_pad_type_0 = const()[name = tensor<string, []>("hidden_states_7_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> hidden_states_7_strides_0 = const()[name = tensor<string, []>("hidden_states_7_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> hidden_states_7_pad_0 = const()[name = tensor<string, []>("hidden_states_7_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> hidden_states_7_dilations_0 = const()[name = tensor<string, []>("hidden_states_7_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> hidden_states_7_groups_0 = const()[name = tensor<string, []>("hidden_states_7_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 3072, 1, 1]> layers_2_fc2_weight_to_fp16 = const()[name = tensor<string, []>("layers_2_fc2_weight_to_fp16"), val = tensor<fp16, [768, 3072, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(132341952)))]; |
| tensor<fp16, [768]> layers_2_fc2_bias_to_fp16 = const()[name = tensor<string, []>("layers_2_fc2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(137060608)))]; |
| tensor<fp16, [1, 768, 1, 1]> hidden_states_7_cast_fp16 = conv(bias = layers_2_fc2_bias_to_fp16, dilations = hidden_states_7_dilations_0, groups = hidden_states_7_groups_0, pad = hidden_states_7_pad_0, pad_type = hidden_states_7_pad_type_0, strides = hidden_states_7_strides_0, weight = layers_2_fc2_weight_to_fp16, x = input_29_cast_fp16)[name = tensor<string, []>("hidden_states_7_cast_fp16")]; |
| tensor<fp16, [1, 768, 1, 1]> inputs_19_cast_fp16 = add(x = inputs_17_cast_fp16, y = hidden_states_7_cast_fp16)[name = tensor<string, []>("inputs_19_cast_fp16")]; |
| tensor<int32, []> var_753 = const()[name = tensor<string, []>("op_753"), val = tensor<int32, []>(3)]; |
| tensor<int32, [1]> out_19_axes_0 = const()[name = tensor<string, []>("out_19_axes_0"), val = tensor<int32, [1]>([1])]; |
| tensor<fp16, []> var_778_to_fp16 = const()[name = tensor<string, []>("op_778_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, 768, 1, 1]> out_19_cast_fp16 = layer_norm(axes = out_19_axes_0, epsilon = var_778_to_fp16, x = inputs_19_cast_fp16)[name = tensor<string, []>("out_19_cast_fp16")]; |
| tensor<fp16, [768]> obj_43_gamma_0_to_fp16 = const()[name = tensor<string, []>("obj_43_gamma_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(137062208)))]; |
| tensor<fp16, [768]> obj_43_beta_0_to_fp16 = const()[name = tensor<string, []>("obj_43_beta_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(137063808)))]; |
| tensor<fp16, []> obj_43_epsilon_0_to_fp16 = const()[name = tensor<string, []>("obj_43_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, 768, 1, 1]> obj_43_cast_fp16 = batch_norm(beta = obj_43_beta_0_to_fp16, epsilon = obj_43_epsilon_0_to_fp16, gamma = obj_43_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_19_cast_fp16)[name = tensor<string, []>("obj_43_cast_fp16")]; |
| tensor<string, []> query_13_pad_type_0 = const()[name = tensor<string, []>("query_13_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> query_13_strides_0 = const()[name = tensor<string, []>("query_13_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> query_13_pad_0 = const()[name = tensor<string, []>("query_13_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> query_13_dilations_0 = const()[name = tensor<string, []>("query_13_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> query_13_groups_0 = const()[name = tensor<string, []>("query_13_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 768, 1, 1]> layers_3_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_3_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [768, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(137065408)))]; |
| tensor<fp16, [768]> layers_3_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("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, []>(138245120)))]; |
| tensor<fp16, [1, 768, 1, 1]> query_13_cast_fp16 = conv(bias = layers_3_self_attn_q_proj_bias_to_fp16, dilations = query_13_dilations_0, groups = query_13_groups_0, pad = query_13_pad_0, pad_type = query_13_pad_type_0, strides = query_13_strides_0, weight = layers_3_self_attn_q_proj_weight_to_fp16, x = obj_43_cast_fp16)[name = tensor<string, []>("query_13_cast_fp16")]; |
| tensor<string, []> current_key_7_pad_type_0 = const()[name = tensor<string, []>("current_key_7_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> current_key_7_strides_0 = const()[name = tensor<string, []>("current_key_7_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> current_key_7_pad_0 = const()[name = tensor<string, []>("current_key_7_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> current_key_7_dilations_0 = const()[name = tensor<string, []>("current_key_7_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> current_key_7_groups_0 = const()[name = tensor<string, []>("current_key_7_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 768, 1, 1]> layers_3_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_3_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [768, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(138246720)))]; |
| tensor<fp16, [1, 768, 1, 1]> current_key_7_cast_fp16 = conv(dilations = current_key_7_dilations_0, groups = current_key_7_groups_0, pad = current_key_7_pad_0, pad_type = current_key_7_pad_type_0, strides = current_key_7_strides_0, weight = layers_3_self_attn_k_proj_weight_to_fp16, x = obj_43_cast_fp16)[name = tensor<string, []>("current_key_7_cast_fp16")]; |
| tensor<string, []> current_value_7_pad_type_0 = const()[name = tensor<string, []>("current_value_7_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> current_value_7_strides_0 = const()[name = tensor<string, []>("current_value_7_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> current_value_7_pad_0 = const()[name = tensor<string, []>("current_value_7_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> current_value_7_dilations_0 = const()[name = tensor<string, []>("current_value_7_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> current_value_7_groups_0 = const()[name = tensor<string, []>("current_value_7_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 768, 1, 1]> layers_3_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_3_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [768, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(139426432)))]; |
| tensor<fp16, [768]> layers_3_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("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, []>(140606144)))]; |
| tensor<fp16, [1, 768, 1, 1]> current_value_7_cast_fp16 = conv(bias = layers_3_self_attn_v_proj_bias_to_fp16, dilations = current_value_7_dilations_0, groups = current_value_7_groups_0, pad = current_value_7_pad_0, pad_type = current_value_7_pad_type_0, strides = current_value_7_strides_0, weight = layers_3_self_attn_v_proj_weight_to_fp16, x = obj_43_cast_fp16)[name = tensor<string, []>("current_value_7_cast_fp16")]; |
| tensor<fp16, [1, 768, 1, 448]> var_817_cast_fp16 = mul(x = var_63_cast_fp16_3, y = var_159_cast_fp16)[name = tensor<string, []>("op_817_cast_fp16")]; |
| tensor<fp16, [1, 768, 1, 448]> var_818_cast_fp16 = mul(x = current_key_7_cast_fp16, y = var_157_cast_fp16)[name = tensor<string, []>("op_818_cast_fp16")]; |
| tensor<fp16, [1, 768, 1, 448]> key_13_cast_fp16 = add(x = var_817_cast_fp16, y = var_818_cast_fp16)[name = tensor<string, []>("key_13_cast_fp16")]; |
| tensor<fp16, [1, 768, 1, 448]> var_821_cast_fp16 = mul(x = var_78_cast_fp16_3, y = var_159_cast_fp16)[name = tensor<string, []>("op_821_cast_fp16")]; |
| tensor<fp16, [1, 768, 1, 448]> var_822_cast_fp16 = mul(x = current_value_7_cast_fp16, y = var_157_cast_fp16)[name = tensor<string, []>("op_822_cast_fp16")]; |
| tensor<fp16, [1, 768, 1, 448]> value_13_cast_fp16 = add(x = var_821_cast_fp16, y = var_822_cast_fp16)[name = tensor<string, []>("value_13_cast_fp16")]; |
| tensor<int32, [4]> var_826 = const()[name = tensor<string, []>("op_826"), val = tensor<int32, [4]>([1, 12, 64, 1])]; |
| tensor<fp16, [1, 12, 64, 1]> mh_q_13_cast_fp16 = reshape(shape = var_826, x = query_13_cast_fp16)[name = tensor<string, []>("mh_q_13_cast_fp16")]; |
| tensor<fp16, []> var_828_to_fp16 = const()[name = tensor<string, []>("op_828_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; |
| tensor<fp16, [1, 12, 64, 1]> var_829_cast_fp16 = mul(x = mh_q_13_cast_fp16, y = var_828_to_fp16)[name = tensor<string, []>("op_829_cast_fp16")]; |
| tensor<int32, [4]> var_832 = const()[name = tensor<string, []>("op_832"), val = tensor<int32, [4]>([1, 12, 64, 448])]; |
| tensor<fp16, [1, 12, 64, 448]> var_833_cast_fp16 = reshape(shape = var_832, x = key_13_cast_fp16)[name = tensor<string, []>("op_833_cast_fp16")]; |
| tensor<bool, []> mh_w_19_transpose_x_0 = const()[name = tensor<string, []>("mh_w_19_transpose_x_0"), val = tensor<bool, []>(true)]; |
| tensor<bool, []> mh_w_19_transpose_y_0 = const()[name = tensor<string, []>("mh_w_19_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [1, 12, 1, 448]> mh_w_19_cast_fp16 = matmul(transpose_x = mh_w_19_transpose_x_0, transpose_y = mh_w_19_transpose_y_0, x = var_829_cast_fp16, y = var_833_cast_fp16)[name = tensor<string, []>("mh_w_19_cast_fp16")]; |
| tensor<fp16, [1, 12, 1, 448]> mh_w_21_cast_fp16 = add(x = mh_w_19_cast_fp16, y = var_181_cast_fp16)[name = tensor<string, []>("mh_w_21_cast_fp16")]; |
| tensor<fp16, [1, 12, 1, 448]> var_841_cast_fp16 = softmax(axis = var_753, x = mh_w_21_cast_fp16)[name = tensor<string, []>("op_841_cast_fp16")]; |
| tensor<int32, [4]> var_842 = const()[name = tensor<string, []>("op_842"), val = tensor<int32, [4]>([1, 12, 64, 448])]; |
| tensor<fp16, [1, 12, 64, 448]> var_843_cast_fp16 = reshape(shape = var_842, x = value_13_cast_fp16)[name = tensor<string, []>("op_843_cast_fp16")]; |
| tensor<bool, []> attn_13_transpose_x_0 = const()[name = tensor<string, []>("attn_13_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_13_transpose_y_0 = const()[name = tensor<string, []>("attn_13_transpose_y_0"), val = tensor<bool, []>(true)]; |
| tensor<fp16, [1, 12, 64, 1]> attn_13_cast_fp16 = matmul(transpose_x = attn_13_transpose_x_0, transpose_y = attn_13_transpose_y_0, x = var_843_cast_fp16, y = var_841_cast_fp16)[name = tensor<string, []>("attn_13_cast_fp16")]; |
| tensor<int32, [4]> var_846 = const()[name = tensor<string, []>("op_846"), val = tensor<int32, [4]>([1, 768, 1, 1])]; |
| tensor<fp16, [1, 768, 1, 1]> input_31_cast_fp16 = reshape(shape = var_846, x = attn_13_cast_fp16)[name = tensor<string, []>("input_31_cast_fp16")]; |
| tensor<string, []> obj_49_pad_type_0 = const()[name = tensor<string, []>("obj_49_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> obj_49_strides_0 = const()[name = tensor<string, []>("obj_49_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> obj_49_pad_0 = const()[name = tensor<string, []>("obj_49_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> obj_49_dilations_0 = const()[name = tensor<string, []>("obj_49_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> obj_49_groups_0 = const()[name = tensor<string, []>("obj_49_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 768, 1, 1]> layers_3_self_attn_o_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_3_self_attn_o_proj_weight_to_fp16"), val = tensor<fp16, [768, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(140607744)))]; |
| tensor<fp16, [768]> layers_3_self_attn_o_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_3_self_attn_o_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(141787456)))]; |
| tensor<fp16, [1, 768, 1, 1]> obj_49_cast_fp16 = conv(bias = layers_3_self_attn_o_proj_bias_to_fp16, dilations = obj_49_dilations_0, groups = obj_49_groups_0, pad = obj_49_pad_0, pad_type = obj_49_pad_type_0, strides = obj_49_strides_0, weight = layers_3_self_attn_o_proj_weight_to_fp16, x = input_31_cast_fp16)[name = tensor<string, []>("obj_49_cast_fp16")]; |
| tensor<fp16, [1, 768, 1, 1]> inputs_21_cast_fp16 = add(x = inputs_19_cast_fp16, y = obj_49_cast_fp16)[name = tensor<string, []>("inputs_21_cast_fp16")]; |
| tensor<int32, [1]> out_21_axes_0 = const()[name = tensor<string, []>("out_21_axes_0"), val = tensor<int32, [1]>([1])]; |
| tensor<fp16, []> var_868_to_fp16 = const()[name = tensor<string, []>("op_868_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, 768, 1, 1]> out_21_cast_fp16 = layer_norm(axes = out_21_axes_0, epsilon = var_868_to_fp16, x = inputs_21_cast_fp16)[name = tensor<string, []>("out_21_cast_fp16")]; |
| tensor<fp16, [768]> obj_51_gamma_0_to_fp16 = const()[name = tensor<string, []>("obj_51_gamma_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(141789056)))]; |
| tensor<fp16, [768]> obj_51_beta_0_to_fp16 = const()[name = tensor<string, []>("obj_51_beta_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(141790656)))]; |
| tensor<fp16, []> obj_51_epsilon_0_to_fp16 = const()[name = tensor<string, []>("obj_51_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, 768, 1, 1]> obj_51_cast_fp16 = batch_norm(beta = obj_51_beta_0_to_fp16, epsilon = obj_51_epsilon_0_to_fp16, gamma = obj_51_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_21_cast_fp16)[name = tensor<string, []>("obj_51_cast_fp16")]; |
| tensor<string, []> query_15_pad_type_0 = const()[name = tensor<string, []>("query_15_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> query_15_strides_0 = const()[name = tensor<string, []>("query_15_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> query_15_pad_0 = const()[name = tensor<string, []>("query_15_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> query_15_dilations_0 = const()[name = tensor<string, []>("query_15_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> query_15_groups_0 = const()[name = tensor<string, []>("query_15_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 768, 1, 1]> layers_3_encoder_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_3_encoder_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [768, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(141792256)))]; |
| tensor<fp16, [768]> layers_3_encoder_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_3_encoder_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(142971968)))]; |
| tensor<fp16, [1, 768, 1, 1]> query_15_cast_fp16 = conv(bias = layers_3_encoder_attn_q_proj_bias_to_fp16, dilations = query_15_dilations_0, groups = query_15_groups_0, pad = query_15_pad_0, pad_type = query_15_pad_type_0, strides = query_15_strides_0, weight = layers_3_encoder_attn_q_proj_weight_to_fp16, x = obj_51_cast_fp16)[name = tensor<string, []>("query_15_cast_fp16")]; |
| tensor<string, []> key_15_pad_type_0 = const()[name = tensor<string, []>("key_15_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> key_15_strides_0 = const()[name = tensor<string, []>("key_15_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> key_15_pad_0 = const()[name = tensor<string, []>("key_15_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> key_15_dilations_0 = const()[name = tensor<string, []>("key_15_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> key_15_groups_0 = const()[name = tensor<string, []>("key_15_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 768, 1, 1]> layers_3_encoder_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_3_encoder_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [768, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(142973568)))]; |
| tensor<fp16, [1, 768, 1, 1500]> key_15_cast_fp16 = conv(dilations = key_15_dilations_0, groups = key_15_groups_0, pad = key_15_pad_0, pad_type = key_15_pad_type_0, strides = key_15_strides_0, weight = layers_3_encoder_attn_k_proj_weight_to_fp16, x = encoder_output_embeds)[name = tensor<string, []>("key_15_cast_fp16")]; |
| tensor<string, []> value_15_pad_type_0 = const()[name = tensor<string, []>("value_15_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> value_15_strides_0 = const()[name = tensor<string, []>("value_15_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> value_15_pad_0 = const()[name = tensor<string, []>("value_15_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> value_15_dilations_0 = const()[name = tensor<string, []>("value_15_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> value_15_groups_0 = const()[name = tensor<string, []>("value_15_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 768, 1, 1]> layers_3_encoder_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_3_encoder_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [768, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(144153280)))]; |
| tensor<fp16, [768]> layers_3_encoder_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_3_encoder_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(145332992)))]; |
| tensor<fp16, [1, 768, 1, 1500]> value_15_cast_fp16 = conv(bias = layers_3_encoder_attn_v_proj_bias_to_fp16, dilations = value_15_dilations_0, groups = value_15_groups_0, pad = value_15_pad_0, pad_type = value_15_pad_type_0, strides = value_15_strides_0, weight = layers_3_encoder_attn_v_proj_weight_to_fp16, x = encoder_output_embeds)[name = tensor<string, []>("value_15_cast_fp16")]; |
| tensor<int32, [4]> var_904 = const()[name = tensor<string, []>("op_904"), val = tensor<int32, [4]>([1, 12, 64, 1])]; |
| tensor<fp16, [1, 12, 64, 1]> mh_q_15_cast_fp16 = reshape(shape = var_904, x = query_15_cast_fp16)[name = tensor<string, []>("mh_q_15_cast_fp16")]; |
| tensor<fp16, []> var_906_to_fp16 = const()[name = tensor<string, []>("op_906_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; |
| tensor<fp16, [1, 12, 64, 1]> var_907_cast_fp16 = mul(x = mh_q_15_cast_fp16, y = var_906_to_fp16)[name = tensor<string, []>("op_907_cast_fp16")]; |
| tensor<int32, [4]> var_910 = const()[name = tensor<string, []>("op_910"), val = tensor<int32, [4]>([1, 12, 64, 1500])]; |
| tensor<fp16, [1, 12, 64, 1500]> var_911_cast_fp16 = reshape(shape = var_910, x = key_15_cast_fp16)[name = tensor<string, []>("op_911_cast_fp16")]; |
| tensor<bool, []> mh_w_23_transpose_x_0 = const()[name = tensor<string, []>("mh_w_23_transpose_x_0"), val = tensor<bool, []>(true)]; |
| tensor<bool, []> mh_w_23_transpose_y_0 = const()[name = tensor<string, []>("mh_w_23_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [1, 12, 1, 1500]> mh_w_23_cast_fp16 = matmul(transpose_x = mh_w_23_transpose_x_0, transpose_y = mh_w_23_transpose_y_0, x = var_907_cast_fp16, y = var_911_cast_fp16)[name = tensor<string, []>("mh_w_23_cast_fp16")]; |
| tensor<fp16, [1, 12, 1, 1500]> obj_55_cast_fp16 = softmax(axis = var_753, x = mh_w_23_cast_fp16)[name = tensor<string, []>("obj_55_cast_fp16")]; |
| tensor<int32, [4]> var_915 = const()[name = tensor<string, []>("op_915"), val = tensor<int32, [4]>([1, 12, 64, 1500])]; |
| tensor<fp16, [1, 12, 64, 1500]> var_916_cast_fp16 = reshape(shape = var_915, x = value_15_cast_fp16)[name = tensor<string, []>("op_916_cast_fp16")]; |
| tensor<bool, []> attn_15_transpose_x_0 = const()[name = tensor<string, []>("attn_15_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_15_transpose_y_0 = const()[name = tensor<string, []>("attn_15_transpose_y_0"), val = tensor<bool, []>(true)]; |
| tensor<fp16, [1, 12, 64, 1]> attn_15_cast_fp16 = matmul(transpose_x = attn_15_transpose_x_0, transpose_y = attn_15_transpose_y_0, x = var_916_cast_fp16, y = obj_55_cast_fp16)[name = tensor<string, []>("attn_15_cast_fp16")]; |
| tensor<int32, [4]> var_919 = const()[name = tensor<string, []>("op_919"), val = tensor<int32, [4]>([1, 768, 1, 1])]; |
| tensor<fp16, [1, 768, 1, 1]> input_33_cast_fp16 = reshape(shape = var_919, x = attn_15_cast_fp16)[name = tensor<string, []>("input_33_cast_fp16")]; |
| tensor<string, []> obj_53_pad_type_0 = const()[name = tensor<string, []>("obj_53_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> obj_53_strides_0 = const()[name = tensor<string, []>("obj_53_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> obj_53_pad_0 = const()[name = tensor<string, []>("obj_53_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> obj_53_dilations_0 = const()[name = tensor<string, []>("obj_53_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> obj_53_groups_0 = const()[name = tensor<string, []>("obj_53_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 768, 1, 1]> layers_3_encoder_attn_o_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_3_encoder_attn_o_proj_weight_to_fp16"), val = tensor<fp16, [768, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(145334592)))]; |
| tensor<fp16, [768]> layers_3_encoder_attn_o_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_3_encoder_attn_o_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(146514304)))]; |
| tensor<fp16, [1, 768, 1, 1]> obj_53_cast_fp16 = conv(bias = layers_3_encoder_attn_o_proj_bias_to_fp16, dilations = obj_53_dilations_0, groups = obj_53_groups_0, pad = obj_53_pad_0, pad_type = obj_53_pad_type_0, strides = obj_53_strides_0, weight = layers_3_encoder_attn_o_proj_weight_to_fp16, x = input_33_cast_fp16)[name = tensor<string, []>("obj_53_cast_fp16")]; |
| tensor<fp16, [1, 768, 1, 1]> inputs_23_cast_fp16 = add(x = inputs_21_cast_fp16, y = obj_53_cast_fp16)[name = tensor<string, []>("inputs_23_cast_fp16")]; |
| tensor<int32, [1]> out_23_axes_0 = const()[name = tensor<string, []>("out_23_axes_0"), val = tensor<int32, [1]>([1])]; |
| tensor<fp16, []> var_937_to_fp16 = const()[name = tensor<string, []>("op_937_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, 768, 1, 1]> out_23_cast_fp16 = layer_norm(axes = out_23_axes_0, epsilon = var_937_to_fp16, x = inputs_23_cast_fp16)[name = tensor<string, []>("out_23_cast_fp16")]; |
| tensor<fp16, [768]> input_35_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_35_gamma_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(146515904)))]; |
| tensor<fp16, [768]> input_35_beta_0_to_fp16 = const()[name = tensor<string, []>("input_35_beta_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(146517504)))]; |
| tensor<fp16, []> input_35_epsilon_0_to_fp16 = const()[name = tensor<string, []>("input_35_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, 768, 1, 1]> input_35_cast_fp16 = batch_norm(beta = input_35_beta_0_to_fp16, epsilon = input_35_epsilon_0_to_fp16, gamma = input_35_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_23_cast_fp16)[name = tensor<string, []>("input_35_cast_fp16")]; |
| tensor<string, []> input_37_pad_type_0 = const()[name = tensor<string, []>("input_37_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> input_37_strides_0 = const()[name = tensor<string, []>("input_37_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> input_37_pad_0 = const()[name = tensor<string, []>("input_37_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> input_37_dilations_0 = const()[name = tensor<string, []>("input_37_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> input_37_groups_0 = const()[name = tensor<string, []>("input_37_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [3072, 768, 1, 1]> layers_3_fc1_weight_to_fp16 = const()[name = tensor<string, []>("layers_3_fc1_weight_to_fp16"), val = tensor<fp16, [3072, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(146519104)))]; |
| tensor<fp16, [3072]> layers_3_fc1_bias_to_fp16 = const()[name = tensor<string, []>("layers_3_fc1_bias_to_fp16"), val = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(151237760)))]; |
| tensor<fp16, [1, 3072, 1, 1]> input_37_cast_fp16 = conv(bias = layers_3_fc1_bias_to_fp16, dilations = input_37_dilations_0, groups = input_37_groups_0, pad = input_37_pad_0, pad_type = input_37_pad_type_0, strides = input_37_strides_0, weight = layers_3_fc1_weight_to_fp16, x = input_35_cast_fp16)[name = tensor<string, []>("input_37_cast_fp16")]; |
| tensor<string, []> input_39_mode_0 = const()[name = tensor<string, []>("input_39_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 3072, 1, 1]> input_39_cast_fp16 = gelu(mode = input_39_mode_0, x = input_37_cast_fp16)[name = tensor<string, []>("input_39_cast_fp16")]; |
| tensor<string, []> hidden_states_9_pad_type_0 = const()[name = tensor<string, []>("hidden_states_9_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> hidden_states_9_strides_0 = const()[name = tensor<string, []>("hidden_states_9_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> hidden_states_9_pad_0 = const()[name = tensor<string, []>("hidden_states_9_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> hidden_states_9_dilations_0 = const()[name = tensor<string, []>("hidden_states_9_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> hidden_states_9_groups_0 = const()[name = tensor<string, []>("hidden_states_9_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 3072, 1, 1]> layers_3_fc2_weight_to_fp16 = const()[name = tensor<string, []>("layers_3_fc2_weight_to_fp16"), val = tensor<fp16, [768, 3072, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(151243968)))]; |
| tensor<fp16, [768]> layers_3_fc2_bias_to_fp16 = const()[name = tensor<string, []>("layers_3_fc2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(155962624)))]; |
| tensor<fp16, [1, 768, 1, 1]> hidden_states_9_cast_fp16 = conv(bias = layers_3_fc2_bias_to_fp16, dilations = hidden_states_9_dilations_0, groups = hidden_states_9_groups_0, pad = hidden_states_9_pad_0, pad_type = hidden_states_9_pad_type_0, strides = hidden_states_9_strides_0, weight = layers_3_fc2_weight_to_fp16, x = input_39_cast_fp16)[name = tensor<string, []>("hidden_states_9_cast_fp16")]; |
| tensor<fp16, [1, 768, 1, 1]> inputs_25_cast_fp16 = add(x = inputs_23_cast_fp16, y = hidden_states_9_cast_fp16)[name = tensor<string, []>("inputs_25_cast_fp16")]; |
| tensor<int32, []> var_972 = const()[name = tensor<string, []>("op_972"), val = tensor<int32, []>(3)]; |
| tensor<int32, [1]> out_25_axes_0 = const()[name = tensor<string, []>("out_25_axes_0"), val = tensor<int32, [1]>([1])]; |
| tensor<fp16, []> var_997_to_fp16 = const()[name = tensor<string, []>("op_997_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, 768, 1, 1]> out_25_cast_fp16 = layer_norm(axes = out_25_axes_0, epsilon = var_997_to_fp16, x = inputs_25_cast_fp16)[name = tensor<string, []>("out_25_cast_fp16")]; |
| tensor<fp16, [768]> obj_57_gamma_0_to_fp16 = const()[name = tensor<string, []>("obj_57_gamma_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(155964224)))]; |
| tensor<fp16, [768]> obj_57_beta_0_to_fp16 = const()[name = tensor<string, []>("obj_57_beta_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(155965824)))]; |
| tensor<fp16, []> obj_57_epsilon_0_to_fp16 = const()[name = tensor<string, []>("obj_57_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, 768, 1, 1]> obj_57_cast_fp16 = batch_norm(beta = obj_57_beta_0_to_fp16, epsilon = obj_57_epsilon_0_to_fp16, gamma = obj_57_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_25_cast_fp16)[name = tensor<string, []>("obj_57_cast_fp16")]; |
| tensor<string, []> query_17_pad_type_0 = const()[name = tensor<string, []>("query_17_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> query_17_strides_0 = const()[name = tensor<string, []>("query_17_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> query_17_pad_0 = const()[name = tensor<string, []>("query_17_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> query_17_dilations_0 = const()[name = tensor<string, []>("query_17_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> query_17_groups_0 = const()[name = tensor<string, []>("query_17_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 768, 1, 1]> layers_4_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_4_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [768, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(155967424)))]; |
| tensor<fp16, [768]> layers_4_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("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, []>(157147136)))]; |
| tensor<fp16, [1, 768, 1, 1]> query_17_cast_fp16 = conv(bias = layers_4_self_attn_q_proj_bias_to_fp16, dilations = query_17_dilations_0, groups = query_17_groups_0, pad = query_17_pad_0, pad_type = query_17_pad_type_0, strides = query_17_strides_0, weight = layers_4_self_attn_q_proj_weight_to_fp16, x = obj_57_cast_fp16)[name = tensor<string, []>("query_17_cast_fp16")]; |
| tensor<string, []> current_key_9_pad_type_0 = const()[name = tensor<string, []>("current_key_9_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> current_key_9_strides_0 = const()[name = tensor<string, []>("current_key_9_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> current_key_9_pad_0 = const()[name = tensor<string, []>("current_key_9_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> current_key_9_dilations_0 = const()[name = tensor<string, []>("current_key_9_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> current_key_9_groups_0 = const()[name = tensor<string, []>("current_key_9_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 768, 1, 1]> layers_4_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_4_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [768, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(157148736)))]; |
| tensor<fp16, [1, 768, 1, 1]> current_key_9_cast_fp16 = conv(dilations = current_key_9_dilations_0, groups = current_key_9_groups_0, pad = current_key_9_pad_0, pad_type = current_key_9_pad_type_0, strides = current_key_9_strides_0, weight = layers_4_self_attn_k_proj_weight_to_fp16, x = obj_57_cast_fp16)[name = tensor<string, []>("current_key_9_cast_fp16")]; |
| tensor<string, []> current_value_9_pad_type_0 = const()[name = tensor<string, []>("current_value_9_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> current_value_9_strides_0 = const()[name = tensor<string, []>("current_value_9_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> current_value_9_pad_0 = const()[name = tensor<string, []>("current_value_9_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> current_value_9_dilations_0 = const()[name = tensor<string, []>("current_value_9_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> current_value_9_groups_0 = const()[name = tensor<string, []>("current_value_9_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 768, 1, 1]> layers_4_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_4_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [768, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(158328448)))]; |
| tensor<fp16, [768]> layers_4_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("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, []>(159508160)))]; |
| tensor<fp16, [1, 768, 1, 1]> current_value_9_cast_fp16 = conv(bias = layers_4_self_attn_v_proj_bias_to_fp16, dilations = current_value_9_dilations_0, groups = current_value_9_groups_0, pad = current_value_9_pad_0, pad_type = current_value_9_pad_type_0, strides = current_value_9_strides_0, weight = layers_4_self_attn_v_proj_weight_to_fp16, x = obj_57_cast_fp16)[name = tensor<string, []>("current_value_9_cast_fp16")]; |
| tensor<fp16, [1, 768, 1, 448]> var_1036_cast_fp16 = mul(x = var_63_cast_fp16_4, y = var_159_cast_fp16)[name = tensor<string, []>("op_1036_cast_fp16")]; |
| tensor<fp16, [1, 768, 1, 448]> var_1037_cast_fp16 = mul(x = current_key_9_cast_fp16, y = var_157_cast_fp16)[name = tensor<string, []>("op_1037_cast_fp16")]; |
| tensor<fp16, [1, 768, 1, 448]> key_17_cast_fp16 = add(x = var_1036_cast_fp16, y = var_1037_cast_fp16)[name = tensor<string, []>("key_17_cast_fp16")]; |
| tensor<fp16, [1, 768, 1, 448]> var_1040_cast_fp16 = mul(x = var_78_cast_fp16_4, y = var_159_cast_fp16)[name = tensor<string, []>("op_1040_cast_fp16")]; |
| tensor<fp16, [1, 768, 1, 448]> var_1041_cast_fp16 = mul(x = current_value_9_cast_fp16, y = var_157_cast_fp16)[name = tensor<string, []>("op_1041_cast_fp16")]; |
| tensor<fp16, [1, 768, 1, 448]> value_17_cast_fp16 = add(x = var_1040_cast_fp16, y = var_1041_cast_fp16)[name = tensor<string, []>("value_17_cast_fp16")]; |
| tensor<int32, [4]> var_1045 = const()[name = tensor<string, []>("op_1045"), val = tensor<int32, [4]>([1, 12, 64, 1])]; |
| tensor<fp16, [1, 12, 64, 1]> mh_q_17_cast_fp16 = reshape(shape = var_1045, x = query_17_cast_fp16)[name = tensor<string, []>("mh_q_17_cast_fp16")]; |
| tensor<fp16, []> var_1047_to_fp16 = const()[name = tensor<string, []>("op_1047_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; |
| tensor<fp16, [1, 12, 64, 1]> var_1048_cast_fp16 = mul(x = mh_q_17_cast_fp16, y = var_1047_to_fp16)[name = tensor<string, []>("op_1048_cast_fp16")]; |
| tensor<int32, [4]> var_1051 = const()[name = tensor<string, []>("op_1051"), val = tensor<int32, [4]>([1, 12, 64, 448])]; |
| tensor<fp16, [1, 12, 64, 448]> var_1052_cast_fp16 = reshape(shape = var_1051, x = key_17_cast_fp16)[name = tensor<string, []>("op_1052_cast_fp16")]; |
| tensor<bool, []> mh_w_25_transpose_x_0 = const()[name = tensor<string, []>("mh_w_25_transpose_x_0"), val = tensor<bool, []>(true)]; |
| tensor<bool, []> mh_w_25_transpose_y_0 = const()[name = tensor<string, []>("mh_w_25_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [1, 12, 1, 448]> mh_w_25_cast_fp16 = matmul(transpose_x = mh_w_25_transpose_x_0, transpose_y = mh_w_25_transpose_y_0, x = var_1048_cast_fp16, y = var_1052_cast_fp16)[name = tensor<string, []>("mh_w_25_cast_fp16")]; |
| tensor<fp16, [1, 12, 1, 448]> mh_w_27_cast_fp16 = add(x = mh_w_25_cast_fp16, y = var_181_cast_fp16)[name = tensor<string, []>("mh_w_27_cast_fp16")]; |
| tensor<fp16, [1, 12, 1, 448]> var_1060_cast_fp16 = softmax(axis = var_972, x = mh_w_27_cast_fp16)[name = tensor<string, []>("op_1060_cast_fp16")]; |
| tensor<int32, [4]> var_1061 = const()[name = tensor<string, []>("op_1061"), val = tensor<int32, [4]>([1, 12, 64, 448])]; |
| tensor<fp16, [1, 12, 64, 448]> var_1062_cast_fp16 = reshape(shape = var_1061, x = value_17_cast_fp16)[name = tensor<string, []>("op_1062_cast_fp16")]; |
| tensor<bool, []> attn_17_transpose_x_0 = const()[name = tensor<string, []>("attn_17_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_17_transpose_y_0 = const()[name = tensor<string, []>("attn_17_transpose_y_0"), val = tensor<bool, []>(true)]; |
| tensor<fp16, [1, 12, 64, 1]> attn_17_cast_fp16 = matmul(transpose_x = attn_17_transpose_x_0, transpose_y = attn_17_transpose_y_0, x = var_1062_cast_fp16, y = var_1060_cast_fp16)[name = tensor<string, []>("attn_17_cast_fp16")]; |
| tensor<int32, [4]> var_1065 = const()[name = tensor<string, []>("op_1065"), val = tensor<int32, [4]>([1, 768, 1, 1])]; |
| tensor<fp16, [1, 768, 1, 1]> input_41_cast_fp16 = reshape(shape = var_1065, x = attn_17_cast_fp16)[name = tensor<string, []>("input_41_cast_fp16")]; |
| tensor<string, []> obj_63_pad_type_0 = const()[name = tensor<string, []>("obj_63_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> obj_63_strides_0 = const()[name = tensor<string, []>("obj_63_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> obj_63_pad_0 = const()[name = tensor<string, []>("obj_63_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> obj_63_dilations_0 = const()[name = tensor<string, []>("obj_63_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> obj_63_groups_0 = const()[name = tensor<string, []>("obj_63_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 768, 1, 1]> layers_4_self_attn_o_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_4_self_attn_o_proj_weight_to_fp16"), val = tensor<fp16, [768, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(159509760)))]; |
| tensor<fp16, [768]> layers_4_self_attn_o_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_4_self_attn_o_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(160689472)))]; |
| tensor<fp16, [1, 768, 1, 1]> obj_63_cast_fp16 = conv(bias = layers_4_self_attn_o_proj_bias_to_fp16, dilations = obj_63_dilations_0, groups = obj_63_groups_0, pad = obj_63_pad_0, pad_type = obj_63_pad_type_0, strides = obj_63_strides_0, weight = layers_4_self_attn_o_proj_weight_to_fp16, x = input_41_cast_fp16)[name = tensor<string, []>("obj_63_cast_fp16")]; |
| tensor<fp16, [1, 768, 1, 1]> inputs_27_cast_fp16 = add(x = inputs_25_cast_fp16, y = obj_63_cast_fp16)[name = tensor<string, []>("inputs_27_cast_fp16")]; |
| tensor<int32, [1]> out_27_axes_0 = const()[name = tensor<string, []>("out_27_axes_0"), val = tensor<int32, [1]>([1])]; |
| tensor<fp16, []> var_1087_to_fp16 = const()[name = tensor<string, []>("op_1087_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, 768, 1, 1]> out_27_cast_fp16 = layer_norm(axes = out_27_axes_0, epsilon = var_1087_to_fp16, x = inputs_27_cast_fp16)[name = tensor<string, []>("out_27_cast_fp16")]; |
| tensor<fp16, [768]> obj_65_gamma_0_to_fp16 = const()[name = tensor<string, []>("obj_65_gamma_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(160691072)))]; |
| tensor<fp16, [768]> obj_65_beta_0_to_fp16 = const()[name = tensor<string, []>("obj_65_beta_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(160692672)))]; |
| tensor<fp16, []> obj_65_epsilon_0_to_fp16 = const()[name = tensor<string, []>("obj_65_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, 768, 1, 1]> obj_65_cast_fp16 = batch_norm(beta = obj_65_beta_0_to_fp16, epsilon = obj_65_epsilon_0_to_fp16, gamma = obj_65_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_27_cast_fp16)[name = tensor<string, []>("obj_65_cast_fp16")]; |
| tensor<string, []> query_19_pad_type_0 = const()[name = tensor<string, []>("query_19_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> query_19_strides_0 = const()[name = tensor<string, []>("query_19_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> query_19_pad_0 = const()[name = tensor<string, []>("query_19_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> query_19_dilations_0 = const()[name = tensor<string, []>("query_19_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> query_19_groups_0 = const()[name = tensor<string, []>("query_19_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 768, 1, 1]> layers_4_encoder_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_4_encoder_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [768, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(160694272)))]; |
| tensor<fp16, [768]> layers_4_encoder_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_4_encoder_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(161873984)))]; |
| tensor<fp16, [1, 768, 1, 1]> query_19_cast_fp16 = conv(bias = layers_4_encoder_attn_q_proj_bias_to_fp16, dilations = query_19_dilations_0, groups = query_19_groups_0, pad = query_19_pad_0, pad_type = query_19_pad_type_0, strides = query_19_strides_0, weight = layers_4_encoder_attn_q_proj_weight_to_fp16, x = obj_65_cast_fp16)[name = tensor<string, []>("query_19_cast_fp16")]; |
| tensor<string, []> key_19_pad_type_0 = const()[name = tensor<string, []>("key_19_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> key_19_strides_0 = const()[name = tensor<string, []>("key_19_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> key_19_pad_0 = const()[name = tensor<string, []>("key_19_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> key_19_dilations_0 = const()[name = tensor<string, []>("key_19_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> key_19_groups_0 = const()[name = tensor<string, []>("key_19_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 768, 1, 1]> layers_4_encoder_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_4_encoder_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [768, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(161875584)))]; |
| tensor<fp16, [1, 768, 1, 1500]> key_19_cast_fp16 = conv(dilations = key_19_dilations_0, groups = key_19_groups_0, pad = key_19_pad_0, pad_type = key_19_pad_type_0, strides = key_19_strides_0, weight = layers_4_encoder_attn_k_proj_weight_to_fp16, x = encoder_output_embeds)[name = tensor<string, []>("key_19_cast_fp16")]; |
| tensor<string, []> value_19_pad_type_0 = const()[name = tensor<string, []>("value_19_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> value_19_strides_0 = const()[name = tensor<string, []>("value_19_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> value_19_pad_0 = const()[name = tensor<string, []>("value_19_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> value_19_dilations_0 = const()[name = tensor<string, []>("value_19_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> value_19_groups_0 = const()[name = tensor<string, []>("value_19_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 768, 1, 1]> layers_4_encoder_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_4_encoder_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [768, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(163055296)))]; |
| tensor<fp16, [768]> layers_4_encoder_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_4_encoder_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(164235008)))]; |
| tensor<fp16, [1, 768, 1, 1500]> value_19_cast_fp16 = conv(bias = layers_4_encoder_attn_v_proj_bias_to_fp16, dilations = value_19_dilations_0, groups = value_19_groups_0, pad = value_19_pad_0, pad_type = value_19_pad_type_0, strides = value_19_strides_0, weight = layers_4_encoder_attn_v_proj_weight_to_fp16, x = encoder_output_embeds)[name = tensor<string, []>("value_19_cast_fp16")]; |
| tensor<int32, [4]> var_1123 = const()[name = tensor<string, []>("op_1123"), val = tensor<int32, [4]>([1, 12, 64, 1])]; |
| tensor<fp16, [1, 12, 64, 1]> mh_q_19_cast_fp16 = reshape(shape = var_1123, x = query_19_cast_fp16)[name = tensor<string, []>("mh_q_19_cast_fp16")]; |
| tensor<fp16, []> var_1125_to_fp16 = const()[name = tensor<string, []>("op_1125_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; |
| tensor<fp16, [1, 12, 64, 1]> var_1126_cast_fp16 = mul(x = mh_q_19_cast_fp16, y = var_1125_to_fp16)[name = tensor<string, []>("op_1126_cast_fp16")]; |
| tensor<int32, [4]> var_1129 = const()[name = tensor<string, []>("op_1129"), val = tensor<int32, [4]>([1, 12, 64, 1500])]; |
| tensor<fp16, [1, 12, 64, 1500]> var_1130_cast_fp16 = reshape(shape = var_1129, x = key_19_cast_fp16)[name = tensor<string, []>("op_1130_cast_fp16")]; |
| tensor<bool, []> mh_w_29_transpose_x_0 = const()[name = tensor<string, []>("mh_w_29_transpose_x_0"), val = tensor<bool, []>(true)]; |
| tensor<bool, []> mh_w_29_transpose_y_0 = const()[name = tensor<string, []>("mh_w_29_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [1, 12, 1, 1500]> mh_w_29_cast_fp16 = matmul(transpose_x = mh_w_29_transpose_x_0, transpose_y = mh_w_29_transpose_y_0, x = var_1126_cast_fp16, y = var_1130_cast_fp16)[name = tensor<string, []>("mh_w_29_cast_fp16")]; |
| tensor<fp16, [1, 12, 1, 1500]> obj_69_cast_fp16 = softmax(axis = var_972, x = mh_w_29_cast_fp16)[name = tensor<string, []>("obj_69_cast_fp16")]; |
| tensor<int32, [4]> var_1134 = const()[name = tensor<string, []>("op_1134"), val = tensor<int32, [4]>([1, 12, 64, 1500])]; |
| tensor<fp16, [1, 12, 64, 1500]> var_1135_cast_fp16 = reshape(shape = var_1134, x = value_19_cast_fp16)[name = tensor<string, []>("op_1135_cast_fp16")]; |
| tensor<bool, []> attn_19_transpose_x_0 = const()[name = tensor<string, []>("attn_19_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_19_transpose_y_0 = const()[name = tensor<string, []>("attn_19_transpose_y_0"), val = tensor<bool, []>(true)]; |
| tensor<fp16, [1, 12, 64, 1]> attn_19_cast_fp16 = matmul(transpose_x = attn_19_transpose_x_0, transpose_y = attn_19_transpose_y_0, x = var_1135_cast_fp16, y = obj_69_cast_fp16)[name = tensor<string, []>("attn_19_cast_fp16")]; |
| tensor<int32, [4]> var_1138 = const()[name = tensor<string, []>("op_1138"), val = tensor<int32, [4]>([1, 768, 1, 1])]; |
| tensor<fp16, [1, 768, 1, 1]> input_43_cast_fp16 = reshape(shape = var_1138, x = attn_19_cast_fp16)[name = tensor<string, []>("input_43_cast_fp16")]; |
| tensor<string, []> obj_67_pad_type_0 = const()[name = tensor<string, []>("obj_67_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> obj_67_strides_0 = const()[name = tensor<string, []>("obj_67_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> obj_67_pad_0 = const()[name = tensor<string, []>("obj_67_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> obj_67_dilations_0 = const()[name = tensor<string, []>("obj_67_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> obj_67_groups_0 = const()[name = tensor<string, []>("obj_67_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 768, 1, 1]> layers_4_encoder_attn_o_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_4_encoder_attn_o_proj_weight_to_fp16"), val = tensor<fp16, [768, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(164236608)))]; |
| tensor<fp16, [768]> layers_4_encoder_attn_o_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_4_encoder_attn_o_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(165416320)))]; |
| tensor<fp16, [1, 768, 1, 1]> obj_67_cast_fp16 = conv(bias = layers_4_encoder_attn_o_proj_bias_to_fp16, dilations = obj_67_dilations_0, groups = obj_67_groups_0, pad = obj_67_pad_0, pad_type = obj_67_pad_type_0, strides = obj_67_strides_0, weight = layers_4_encoder_attn_o_proj_weight_to_fp16, x = input_43_cast_fp16)[name = tensor<string, []>("obj_67_cast_fp16")]; |
| tensor<fp16, [1, 768, 1, 1]> inputs_29_cast_fp16 = add(x = inputs_27_cast_fp16, y = obj_67_cast_fp16)[name = tensor<string, []>("inputs_29_cast_fp16")]; |
| tensor<int32, [1]> out_29_axes_0 = const()[name = tensor<string, []>("out_29_axes_0"), val = tensor<int32, [1]>([1])]; |
| tensor<fp16, []> var_1156_to_fp16 = const()[name = tensor<string, []>("op_1156_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, 768, 1, 1]> out_29_cast_fp16 = layer_norm(axes = out_29_axes_0, epsilon = var_1156_to_fp16, x = inputs_29_cast_fp16)[name = tensor<string, []>("out_29_cast_fp16")]; |
| tensor<fp16, [768]> input_45_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_45_gamma_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(165417920)))]; |
| tensor<fp16, [768]> input_45_beta_0_to_fp16 = const()[name = tensor<string, []>("input_45_beta_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(165419520)))]; |
| tensor<fp16, []> input_45_epsilon_0_to_fp16 = const()[name = tensor<string, []>("input_45_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, 768, 1, 1]> input_45_cast_fp16 = batch_norm(beta = input_45_beta_0_to_fp16, epsilon = input_45_epsilon_0_to_fp16, gamma = input_45_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_29_cast_fp16)[name = tensor<string, []>("input_45_cast_fp16")]; |
| tensor<string, []> input_47_pad_type_0 = const()[name = tensor<string, []>("input_47_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> input_47_strides_0 = const()[name = tensor<string, []>("input_47_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> input_47_pad_0 = const()[name = tensor<string, []>("input_47_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> input_47_dilations_0 = const()[name = tensor<string, []>("input_47_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> input_47_groups_0 = const()[name = tensor<string, []>("input_47_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [3072, 768, 1, 1]> layers_4_fc1_weight_to_fp16 = const()[name = tensor<string, []>("layers_4_fc1_weight_to_fp16"), val = tensor<fp16, [3072, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(165421120)))]; |
| tensor<fp16, [3072]> layers_4_fc1_bias_to_fp16 = const()[name = tensor<string, []>("layers_4_fc1_bias_to_fp16"), val = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(170139776)))]; |
| tensor<fp16, [1, 3072, 1, 1]> input_47_cast_fp16 = conv(bias = layers_4_fc1_bias_to_fp16, dilations = input_47_dilations_0, groups = input_47_groups_0, pad = input_47_pad_0, pad_type = input_47_pad_type_0, strides = input_47_strides_0, weight = layers_4_fc1_weight_to_fp16, x = input_45_cast_fp16)[name = tensor<string, []>("input_47_cast_fp16")]; |
| tensor<string, []> input_49_mode_0 = const()[name = tensor<string, []>("input_49_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 3072, 1, 1]> input_49_cast_fp16 = gelu(mode = input_49_mode_0, x = input_47_cast_fp16)[name = tensor<string, []>("input_49_cast_fp16")]; |
| tensor<string, []> hidden_states_11_pad_type_0 = const()[name = tensor<string, []>("hidden_states_11_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> hidden_states_11_strides_0 = const()[name = tensor<string, []>("hidden_states_11_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> hidden_states_11_pad_0 = const()[name = tensor<string, []>("hidden_states_11_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> hidden_states_11_dilations_0 = const()[name = tensor<string, []>("hidden_states_11_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> hidden_states_11_groups_0 = const()[name = tensor<string, []>("hidden_states_11_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 3072, 1, 1]> layers_4_fc2_weight_to_fp16 = const()[name = tensor<string, []>("layers_4_fc2_weight_to_fp16"), val = tensor<fp16, [768, 3072, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(170145984)))]; |
| tensor<fp16, [768]> layers_4_fc2_bias_to_fp16 = const()[name = tensor<string, []>("layers_4_fc2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(174864640)))]; |
| tensor<fp16, [1, 768, 1, 1]> hidden_states_11_cast_fp16 = conv(bias = layers_4_fc2_bias_to_fp16, dilations = hidden_states_11_dilations_0, groups = hidden_states_11_groups_0, pad = hidden_states_11_pad_0, pad_type = hidden_states_11_pad_type_0, strides = hidden_states_11_strides_0, weight = layers_4_fc2_weight_to_fp16, x = input_49_cast_fp16)[name = tensor<string, []>("hidden_states_11_cast_fp16")]; |
| tensor<fp16, [1, 768, 1, 1]> inputs_31_cast_fp16 = add(x = inputs_29_cast_fp16, y = hidden_states_11_cast_fp16)[name = tensor<string, []>("inputs_31_cast_fp16")]; |
| tensor<int32, []> var_1191 = const()[name = tensor<string, []>("op_1191"), val = tensor<int32, []>(3)]; |
| tensor<int32, [1]> out_31_axes_0 = const()[name = tensor<string, []>("out_31_axes_0"), val = tensor<int32, [1]>([1])]; |
| tensor<fp16, []> var_1216_to_fp16 = const()[name = tensor<string, []>("op_1216_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, 768, 1, 1]> out_31_cast_fp16 = layer_norm(axes = out_31_axes_0, epsilon = var_1216_to_fp16, x = inputs_31_cast_fp16)[name = tensor<string, []>("out_31_cast_fp16")]; |
| tensor<fp16, [768]> obj_71_gamma_0_to_fp16 = const()[name = tensor<string, []>("obj_71_gamma_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(174866240)))]; |
| tensor<fp16, [768]> obj_71_beta_0_to_fp16 = const()[name = tensor<string, []>("obj_71_beta_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(174867840)))]; |
| tensor<fp16, []> obj_71_epsilon_0_to_fp16 = const()[name = tensor<string, []>("obj_71_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, 768, 1, 1]> obj_71_cast_fp16 = batch_norm(beta = obj_71_beta_0_to_fp16, epsilon = obj_71_epsilon_0_to_fp16, gamma = obj_71_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_31_cast_fp16)[name = tensor<string, []>("obj_71_cast_fp16")]; |
| tensor<string, []> query_21_pad_type_0 = const()[name = tensor<string, []>("query_21_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> query_21_strides_0 = const()[name = tensor<string, []>("query_21_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> query_21_pad_0 = const()[name = tensor<string, []>("query_21_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> query_21_dilations_0 = const()[name = tensor<string, []>("query_21_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> query_21_groups_0 = const()[name = tensor<string, []>("query_21_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 768, 1, 1]> layers_5_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_5_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [768, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(174869440)))]; |
| tensor<fp16, [768]> layers_5_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("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, []>(176049152)))]; |
| tensor<fp16, [1, 768, 1, 1]> query_21_cast_fp16 = conv(bias = layers_5_self_attn_q_proj_bias_to_fp16, dilations = query_21_dilations_0, groups = query_21_groups_0, pad = query_21_pad_0, pad_type = query_21_pad_type_0, strides = query_21_strides_0, weight = layers_5_self_attn_q_proj_weight_to_fp16, x = obj_71_cast_fp16)[name = tensor<string, []>("query_21_cast_fp16")]; |
| tensor<string, []> current_key_11_pad_type_0 = const()[name = tensor<string, []>("current_key_11_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> current_key_11_strides_0 = const()[name = tensor<string, []>("current_key_11_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> current_key_11_pad_0 = const()[name = tensor<string, []>("current_key_11_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> current_key_11_dilations_0 = const()[name = tensor<string, []>("current_key_11_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> current_key_11_groups_0 = const()[name = tensor<string, []>("current_key_11_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 768, 1, 1]> layers_5_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_5_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [768, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(176050752)))]; |
| tensor<fp16, [1, 768, 1, 1]> current_key_11_cast_fp16 = conv(dilations = current_key_11_dilations_0, groups = current_key_11_groups_0, pad = current_key_11_pad_0, pad_type = current_key_11_pad_type_0, strides = current_key_11_strides_0, weight = layers_5_self_attn_k_proj_weight_to_fp16, x = obj_71_cast_fp16)[name = tensor<string, []>("current_key_11_cast_fp16")]; |
| tensor<string, []> current_value_11_pad_type_0 = const()[name = tensor<string, []>("current_value_11_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> current_value_11_strides_0 = const()[name = tensor<string, []>("current_value_11_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> current_value_11_pad_0 = const()[name = tensor<string, []>("current_value_11_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> current_value_11_dilations_0 = const()[name = tensor<string, []>("current_value_11_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> current_value_11_groups_0 = const()[name = tensor<string, []>("current_value_11_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 768, 1, 1]> layers_5_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_5_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [768, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(177230464)))]; |
| tensor<fp16, [768]> layers_5_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("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, []>(178410176)))]; |
| tensor<fp16, [1, 768, 1, 1]> current_value_11_cast_fp16 = conv(bias = layers_5_self_attn_v_proj_bias_to_fp16, dilations = current_value_11_dilations_0, groups = current_value_11_groups_0, pad = current_value_11_pad_0, pad_type = current_value_11_pad_type_0, strides = current_value_11_strides_0, weight = layers_5_self_attn_v_proj_weight_to_fp16, x = obj_71_cast_fp16)[name = tensor<string, []>("current_value_11_cast_fp16")]; |
| tensor<fp16, [1, 768, 1, 448]> var_1255_cast_fp16 = mul(x = var_63_cast_fp16_5, y = var_159_cast_fp16)[name = tensor<string, []>("op_1255_cast_fp16")]; |
| tensor<fp16, [1, 768, 1, 448]> var_1256_cast_fp16 = mul(x = current_key_11_cast_fp16, y = var_157_cast_fp16)[name = tensor<string, []>("op_1256_cast_fp16")]; |
| tensor<fp16, [1, 768, 1, 448]> key_21_cast_fp16 = add(x = var_1255_cast_fp16, y = var_1256_cast_fp16)[name = tensor<string, []>("key_21_cast_fp16")]; |
| tensor<fp16, [1, 768, 1, 448]> var_1259_cast_fp16 = mul(x = var_78_cast_fp16_5, y = var_159_cast_fp16)[name = tensor<string, []>("op_1259_cast_fp16")]; |
| tensor<fp16, [1, 768, 1, 448]> var_1260_cast_fp16 = mul(x = current_value_11_cast_fp16, y = var_157_cast_fp16)[name = tensor<string, []>("op_1260_cast_fp16")]; |
| tensor<fp16, [1, 768, 1, 448]> value_21_cast_fp16 = add(x = var_1259_cast_fp16, y = var_1260_cast_fp16)[name = tensor<string, []>("value_21_cast_fp16")]; |
| tensor<int32, [4]> var_1264 = const()[name = tensor<string, []>("op_1264"), val = tensor<int32, [4]>([1, 12, 64, 1])]; |
| tensor<fp16, [1, 12, 64, 1]> mh_q_21_cast_fp16 = reshape(shape = var_1264, x = query_21_cast_fp16)[name = tensor<string, []>("mh_q_21_cast_fp16")]; |
| tensor<fp16, []> var_1266_to_fp16 = const()[name = tensor<string, []>("op_1266_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; |
| tensor<fp16, [1, 12, 64, 1]> var_1267_cast_fp16 = mul(x = mh_q_21_cast_fp16, y = var_1266_to_fp16)[name = tensor<string, []>("op_1267_cast_fp16")]; |
| tensor<int32, [4]> var_1270 = const()[name = tensor<string, []>("op_1270"), val = tensor<int32, [4]>([1, 12, 64, 448])]; |
| tensor<fp16, [1, 12, 64, 448]> var_1271_cast_fp16 = reshape(shape = var_1270, x = key_21_cast_fp16)[name = tensor<string, []>("op_1271_cast_fp16")]; |
| tensor<bool, []> mh_w_31_transpose_x_0 = const()[name = tensor<string, []>("mh_w_31_transpose_x_0"), val = tensor<bool, []>(true)]; |
| tensor<bool, []> mh_w_31_transpose_y_0 = const()[name = tensor<string, []>("mh_w_31_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [1, 12, 1, 448]> mh_w_31_cast_fp16 = matmul(transpose_x = mh_w_31_transpose_x_0, transpose_y = mh_w_31_transpose_y_0, x = var_1267_cast_fp16, y = var_1271_cast_fp16)[name = tensor<string, []>("mh_w_31_cast_fp16")]; |
| tensor<fp16, [1, 12, 1, 448]> mh_w_33_cast_fp16 = add(x = mh_w_31_cast_fp16, y = var_181_cast_fp16)[name = tensor<string, []>("mh_w_33_cast_fp16")]; |
| tensor<fp16, [1, 12, 1, 448]> var_1279_cast_fp16 = softmax(axis = var_1191, x = mh_w_33_cast_fp16)[name = tensor<string, []>("op_1279_cast_fp16")]; |
| tensor<int32, [4]> var_1280 = const()[name = tensor<string, []>("op_1280"), val = tensor<int32, [4]>([1, 12, 64, 448])]; |
| tensor<fp16, [1, 12, 64, 448]> var_1281_cast_fp16 = reshape(shape = var_1280, x = value_21_cast_fp16)[name = tensor<string, []>("op_1281_cast_fp16")]; |
| tensor<bool, []> attn_21_transpose_x_0 = const()[name = tensor<string, []>("attn_21_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_21_transpose_y_0 = const()[name = tensor<string, []>("attn_21_transpose_y_0"), val = tensor<bool, []>(true)]; |
| tensor<fp16, [1, 12, 64, 1]> attn_21_cast_fp16 = matmul(transpose_x = attn_21_transpose_x_0, transpose_y = attn_21_transpose_y_0, x = var_1281_cast_fp16, y = var_1279_cast_fp16)[name = tensor<string, []>("attn_21_cast_fp16")]; |
| tensor<int32, [4]> var_1284 = const()[name = tensor<string, []>("op_1284"), val = tensor<int32, [4]>([1, 768, 1, 1])]; |
| tensor<fp16, [1, 768, 1, 1]> input_51_cast_fp16 = reshape(shape = var_1284, x = attn_21_cast_fp16)[name = tensor<string, []>("input_51_cast_fp16")]; |
| tensor<string, []> obj_77_pad_type_0 = const()[name = tensor<string, []>("obj_77_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> obj_77_strides_0 = const()[name = tensor<string, []>("obj_77_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> obj_77_pad_0 = const()[name = tensor<string, []>("obj_77_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> obj_77_dilations_0 = const()[name = tensor<string, []>("obj_77_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> obj_77_groups_0 = const()[name = tensor<string, []>("obj_77_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 768, 1, 1]> layers_5_self_attn_o_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_5_self_attn_o_proj_weight_to_fp16"), val = tensor<fp16, [768, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(178411776)))]; |
| tensor<fp16, [768]> layers_5_self_attn_o_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_5_self_attn_o_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(179591488)))]; |
| tensor<fp16, [1, 768, 1, 1]> obj_77_cast_fp16 = conv(bias = layers_5_self_attn_o_proj_bias_to_fp16, dilations = obj_77_dilations_0, groups = obj_77_groups_0, pad = obj_77_pad_0, pad_type = obj_77_pad_type_0, strides = obj_77_strides_0, weight = layers_5_self_attn_o_proj_weight_to_fp16, x = input_51_cast_fp16)[name = tensor<string, []>("obj_77_cast_fp16")]; |
| tensor<fp16, [1, 768, 1, 1]> inputs_33_cast_fp16 = add(x = inputs_31_cast_fp16, y = obj_77_cast_fp16)[name = tensor<string, []>("inputs_33_cast_fp16")]; |
| tensor<int32, [1]> out_33_axes_0 = const()[name = tensor<string, []>("out_33_axes_0"), val = tensor<int32, [1]>([1])]; |
| tensor<fp16, []> var_1306_to_fp16 = const()[name = tensor<string, []>("op_1306_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, 768, 1, 1]> out_33_cast_fp16 = layer_norm(axes = out_33_axes_0, epsilon = var_1306_to_fp16, x = inputs_33_cast_fp16)[name = tensor<string, []>("out_33_cast_fp16")]; |
| tensor<fp16, [768]> obj_79_gamma_0_to_fp16 = const()[name = tensor<string, []>("obj_79_gamma_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(179593088)))]; |
| tensor<fp16, [768]> obj_79_beta_0_to_fp16 = const()[name = tensor<string, []>("obj_79_beta_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(179594688)))]; |
| tensor<fp16, []> obj_79_epsilon_0_to_fp16 = const()[name = tensor<string, []>("obj_79_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, 768, 1, 1]> obj_79_cast_fp16 = batch_norm(beta = obj_79_beta_0_to_fp16, epsilon = obj_79_epsilon_0_to_fp16, gamma = obj_79_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_33_cast_fp16)[name = tensor<string, []>("obj_79_cast_fp16")]; |
| tensor<string, []> query_23_pad_type_0 = const()[name = tensor<string, []>("query_23_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> query_23_strides_0 = const()[name = tensor<string, []>("query_23_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> query_23_pad_0 = const()[name = tensor<string, []>("query_23_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> query_23_dilations_0 = const()[name = tensor<string, []>("query_23_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> query_23_groups_0 = const()[name = tensor<string, []>("query_23_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 768, 1, 1]> layers_5_encoder_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_5_encoder_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [768, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(179596288)))]; |
| tensor<fp16, [768]> layers_5_encoder_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_5_encoder_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(180776000)))]; |
| tensor<fp16, [1, 768, 1, 1]> query_23_cast_fp16 = conv(bias = layers_5_encoder_attn_q_proj_bias_to_fp16, dilations = query_23_dilations_0, groups = query_23_groups_0, pad = query_23_pad_0, pad_type = query_23_pad_type_0, strides = query_23_strides_0, weight = layers_5_encoder_attn_q_proj_weight_to_fp16, x = obj_79_cast_fp16)[name = tensor<string, []>("query_23_cast_fp16")]; |
| tensor<string, []> key_23_pad_type_0 = const()[name = tensor<string, []>("key_23_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> key_23_strides_0 = const()[name = tensor<string, []>("key_23_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> key_23_pad_0 = const()[name = tensor<string, []>("key_23_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> key_23_dilations_0 = const()[name = tensor<string, []>("key_23_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> key_23_groups_0 = const()[name = tensor<string, []>("key_23_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 768, 1, 1]> layers_5_encoder_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_5_encoder_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [768, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(180777600)))]; |
| tensor<fp16, [1, 768, 1, 1500]> key_23_cast_fp16 = conv(dilations = key_23_dilations_0, groups = key_23_groups_0, pad = key_23_pad_0, pad_type = key_23_pad_type_0, strides = key_23_strides_0, weight = layers_5_encoder_attn_k_proj_weight_to_fp16, x = encoder_output_embeds)[name = tensor<string, []>("key_23_cast_fp16")]; |
| tensor<string, []> value_23_pad_type_0 = const()[name = tensor<string, []>("value_23_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> value_23_strides_0 = const()[name = tensor<string, []>("value_23_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> value_23_pad_0 = const()[name = tensor<string, []>("value_23_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> value_23_dilations_0 = const()[name = tensor<string, []>("value_23_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> value_23_groups_0 = const()[name = tensor<string, []>("value_23_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 768, 1, 1]> layers_5_encoder_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_5_encoder_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [768, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(181957312)))]; |
| tensor<fp16, [768]> layers_5_encoder_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_5_encoder_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(183137024)))]; |
| tensor<fp16, [1, 768, 1, 1500]> value_23_cast_fp16 = conv(bias = layers_5_encoder_attn_v_proj_bias_to_fp16, dilations = value_23_dilations_0, groups = value_23_groups_0, pad = value_23_pad_0, pad_type = value_23_pad_type_0, strides = value_23_strides_0, weight = layers_5_encoder_attn_v_proj_weight_to_fp16, x = encoder_output_embeds)[name = tensor<string, []>("value_23_cast_fp16")]; |
| tensor<int32, [4]> var_1342 = const()[name = tensor<string, []>("op_1342"), val = tensor<int32, [4]>([1, 12, 64, 1])]; |
| tensor<fp16, [1, 12, 64, 1]> mh_q_23_cast_fp16 = reshape(shape = var_1342, x = query_23_cast_fp16)[name = tensor<string, []>("mh_q_23_cast_fp16")]; |
| tensor<fp16, []> var_1344_to_fp16 = const()[name = tensor<string, []>("op_1344_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; |
| tensor<fp16, [1, 12, 64, 1]> var_1345_cast_fp16 = mul(x = mh_q_23_cast_fp16, y = var_1344_to_fp16)[name = tensor<string, []>("op_1345_cast_fp16")]; |
| tensor<int32, [4]> var_1348 = const()[name = tensor<string, []>("op_1348"), val = tensor<int32, [4]>([1, 12, 64, 1500])]; |
| tensor<fp16, [1, 12, 64, 1500]> var_1349_cast_fp16 = reshape(shape = var_1348, x = key_23_cast_fp16)[name = tensor<string, []>("op_1349_cast_fp16")]; |
| tensor<bool, []> mh_w_35_transpose_x_0 = const()[name = tensor<string, []>("mh_w_35_transpose_x_0"), val = tensor<bool, []>(true)]; |
| tensor<bool, []> mh_w_35_transpose_y_0 = const()[name = tensor<string, []>("mh_w_35_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [1, 12, 1, 1500]> mh_w_35_cast_fp16 = matmul(transpose_x = mh_w_35_transpose_x_0, transpose_y = mh_w_35_transpose_y_0, x = var_1345_cast_fp16, y = var_1349_cast_fp16)[name = tensor<string, []>("mh_w_35_cast_fp16")]; |
| tensor<fp16, [1, 12, 1, 1500]> obj_83_cast_fp16 = softmax(axis = var_1191, x = mh_w_35_cast_fp16)[name = tensor<string, []>("obj_83_cast_fp16")]; |
| tensor<int32, [4]> var_1353 = const()[name = tensor<string, []>("op_1353"), val = tensor<int32, [4]>([1, 12, 64, 1500])]; |
| tensor<fp16, [1, 12, 64, 1500]> var_1354_cast_fp16 = reshape(shape = var_1353, x = value_23_cast_fp16)[name = tensor<string, []>("op_1354_cast_fp16")]; |
| tensor<bool, []> attn_23_transpose_x_0 = const()[name = tensor<string, []>("attn_23_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_23_transpose_y_0 = const()[name = tensor<string, []>("attn_23_transpose_y_0"), val = tensor<bool, []>(true)]; |
| tensor<fp16, [1, 12, 64, 1]> attn_23_cast_fp16 = matmul(transpose_x = attn_23_transpose_x_0, transpose_y = attn_23_transpose_y_0, x = var_1354_cast_fp16, y = obj_83_cast_fp16)[name = tensor<string, []>("attn_23_cast_fp16")]; |
| tensor<int32, [4]> var_1357 = const()[name = tensor<string, []>("op_1357"), val = tensor<int32, [4]>([1, 768, 1, 1])]; |
| tensor<fp16, [1, 768, 1, 1]> input_53_cast_fp16 = reshape(shape = var_1357, x = attn_23_cast_fp16)[name = tensor<string, []>("input_53_cast_fp16")]; |
| tensor<string, []> obj_81_pad_type_0 = const()[name = tensor<string, []>("obj_81_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> obj_81_strides_0 = const()[name = tensor<string, []>("obj_81_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> obj_81_pad_0 = const()[name = tensor<string, []>("obj_81_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> obj_81_dilations_0 = const()[name = tensor<string, []>("obj_81_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> obj_81_groups_0 = const()[name = tensor<string, []>("obj_81_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 768, 1, 1]> layers_5_encoder_attn_o_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_5_encoder_attn_o_proj_weight_to_fp16"), val = tensor<fp16, [768, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(183138624)))]; |
| tensor<fp16, [768]> layers_5_encoder_attn_o_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_5_encoder_attn_o_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(184318336)))]; |
| tensor<fp16, [1, 768, 1, 1]> obj_81_cast_fp16 = conv(bias = layers_5_encoder_attn_o_proj_bias_to_fp16, dilations = obj_81_dilations_0, groups = obj_81_groups_0, pad = obj_81_pad_0, pad_type = obj_81_pad_type_0, strides = obj_81_strides_0, weight = layers_5_encoder_attn_o_proj_weight_to_fp16, x = input_53_cast_fp16)[name = tensor<string, []>("obj_81_cast_fp16")]; |
| tensor<fp16, [1, 768, 1, 1]> inputs_35_cast_fp16 = add(x = inputs_33_cast_fp16, y = obj_81_cast_fp16)[name = tensor<string, []>("inputs_35_cast_fp16")]; |
| tensor<int32, [1]> out_35_axes_0 = const()[name = tensor<string, []>("out_35_axes_0"), val = tensor<int32, [1]>([1])]; |
| tensor<fp16, []> var_1378_to_fp16 = const()[name = tensor<string, []>("op_1378_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, 768, 1, 1]> out_35_cast_fp16 = layer_norm(axes = out_35_axes_0, epsilon = var_1378_to_fp16, x = inputs_35_cast_fp16)[name = tensor<string, []>("out_35_cast_fp16")]; |
| tensor<fp16, [768]> input_55_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_55_gamma_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(184319936)))]; |
| tensor<fp16, [768]> input_55_beta_0_to_fp16 = const()[name = tensor<string, []>("input_55_beta_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(184321536)))]; |
| tensor<fp16, []> input_55_epsilon_0_to_fp16 = const()[name = tensor<string, []>("input_55_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, 768, 1, 1]> input_55_cast_fp16 = batch_norm(beta = input_55_beta_0_to_fp16, epsilon = input_55_epsilon_0_to_fp16, gamma = input_55_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_35_cast_fp16)[name = tensor<string, []>("input_55_cast_fp16")]; |
| tensor<string, []> input_57_pad_type_0 = const()[name = tensor<string, []>("input_57_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> input_57_strides_0 = const()[name = tensor<string, []>("input_57_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> input_57_pad_0 = const()[name = tensor<string, []>("input_57_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> input_57_dilations_0 = const()[name = tensor<string, []>("input_57_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> input_57_groups_0 = const()[name = tensor<string, []>("input_57_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [3072, 768, 1, 1]> layers_5_fc1_weight_to_fp16 = const()[name = tensor<string, []>("layers_5_fc1_weight_to_fp16"), val = tensor<fp16, [3072, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(184323136)))]; |
| tensor<fp16, [3072]> layers_5_fc1_bias_to_fp16 = const()[name = tensor<string, []>("layers_5_fc1_bias_to_fp16"), val = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(189041792)))]; |
| tensor<fp16, [1, 3072, 1, 1]> input_57_cast_fp16 = conv(bias = layers_5_fc1_bias_to_fp16, dilations = input_57_dilations_0, groups = input_57_groups_0, pad = input_57_pad_0, pad_type = input_57_pad_type_0, strides = input_57_strides_0, weight = layers_5_fc1_weight_to_fp16, x = input_55_cast_fp16)[name = tensor<string, []>("input_57_cast_fp16")]; |
| tensor<string, []> input_59_mode_0 = const()[name = tensor<string, []>("input_59_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 3072, 1, 1]> input_59_cast_fp16 = gelu(mode = input_59_mode_0, x = input_57_cast_fp16)[name = tensor<string, []>("input_59_cast_fp16")]; |
| tensor<string, []> hidden_states_13_pad_type_0 = const()[name = tensor<string, []>("hidden_states_13_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> hidden_states_13_strides_0 = const()[name = tensor<string, []>("hidden_states_13_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> hidden_states_13_pad_0 = const()[name = tensor<string, []>("hidden_states_13_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> hidden_states_13_dilations_0 = const()[name = tensor<string, []>("hidden_states_13_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> hidden_states_13_groups_0 = const()[name = tensor<string, []>("hidden_states_13_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 3072, 1, 1]> layers_5_fc2_weight_to_fp16 = const()[name = tensor<string, []>("layers_5_fc2_weight_to_fp16"), val = tensor<fp16, [768, 3072, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(189048000)))]; |
| tensor<fp16, [768]> layers_5_fc2_bias_to_fp16 = const()[name = tensor<string, []>("layers_5_fc2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(193766656)))]; |
| tensor<fp16, [1, 768, 1, 1]> hidden_states_13_cast_fp16 = conv(bias = layers_5_fc2_bias_to_fp16, dilations = hidden_states_13_dilations_0, groups = hidden_states_13_groups_0, pad = hidden_states_13_pad_0, pad_type = hidden_states_13_pad_type_0, strides = hidden_states_13_strides_0, weight = layers_5_fc2_weight_to_fp16, x = input_59_cast_fp16)[name = tensor<string, []>("hidden_states_13_cast_fp16")]; |
| tensor<fp16, [1, 768, 1, 1]> inputs_37_cast_fp16 = add(x = inputs_35_cast_fp16, y = hidden_states_13_cast_fp16)[name = tensor<string, []>("inputs_37_cast_fp16")]; |
| tensor<int32, []> var_1414 = const()[name = tensor<string, []>("op_1414"), val = tensor<int32, []>(3)]; |
| tensor<int32, [1]> out_37_axes_0 = const()[name = tensor<string, []>("out_37_axes_0"), val = tensor<int32, [1]>([1])]; |
| tensor<fp16, []> var_1439_to_fp16 = const()[name = tensor<string, []>("op_1439_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, 768, 1, 1]> out_37_cast_fp16 = layer_norm(axes = out_37_axes_0, epsilon = var_1439_to_fp16, x = inputs_37_cast_fp16)[name = tensor<string, []>("out_37_cast_fp16")]; |
| tensor<fp16, [768]> obj_85_gamma_0_to_fp16 = const()[name = tensor<string, []>("obj_85_gamma_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(193768256)))]; |
| tensor<fp16, [768]> obj_85_beta_0_to_fp16 = const()[name = tensor<string, []>("obj_85_beta_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(193769856)))]; |
| tensor<fp16, []> obj_85_epsilon_0_to_fp16 = const()[name = tensor<string, []>("obj_85_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, 768, 1, 1]> obj_85_cast_fp16 = batch_norm(beta = obj_85_beta_0_to_fp16, epsilon = obj_85_epsilon_0_to_fp16, gamma = obj_85_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_37_cast_fp16)[name = tensor<string, []>("obj_85_cast_fp16")]; |
| tensor<string, []> query_25_pad_type_0 = const()[name = tensor<string, []>("query_25_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> query_25_strides_0 = const()[name = tensor<string, []>("query_25_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> query_25_pad_0 = const()[name = tensor<string, []>("query_25_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> query_25_dilations_0 = const()[name = tensor<string, []>("query_25_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> query_25_groups_0 = const()[name = tensor<string, []>("query_25_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 768, 1, 1]> layers_6_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_6_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [768, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(193771456)))]; |
| tensor<fp16, [768]> layers_6_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("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, []>(194951168)))]; |
| tensor<fp16, [1, 768, 1, 1]> query_25_cast_fp16 = conv(bias = layers_6_self_attn_q_proj_bias_to_fp16, dilations = query_25_dilations_0, groups = query_25_groups_0, pad = query_25_pad_0, pad_type = query_25_pad_type_0, strides = query_25_strides_0, weight = layers_6_self_attn_q_proj_weight_to_fp16, x = obj_85_cast_fp16)[name = tensor<string, []>("query_25_cast_fp16")]; |
| tensor<string, []> current_key_13_pad_type_0 = const()[name = tensor<string, []>("current_key_13_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> current_key_13_strides_0 = const()[name = tensor<string, []>("current_key_13_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> current_key_13_pad_0 = const()[name = tensor<string, []>("current_key_13_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> current_key_13_dilations_0 = const()[name = tensor<string, []>("current_key_13_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> current_key_13_groups_0 = const()[name = tensor<string, []>("current_key_13_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 768, 1, 1]> layers_6_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_6_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [768, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(194952768)))]; |
| tensor<fp16, [1, 768, 1, 1]> current_key_13_cast_fp16 = conv(dilations = current_key_13_dilations_0, groups = current_key_13_groups_0, pad = current_key_13_pad_0, pad_type = current_key_13_pad_type_0, strides = current_key_13_strides_0, weight = layers_6_self_attn_k_proj_weight_to_fp16, x = obj_85_cast_fp16)[name = tensor<string, []>("current_key_13_cast_fp16")]; |
| tensor<string, []> current_value_13_pad_type_0 = const()[name = tensor<string, []>("current_value_13_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> current_value_13_strides_0 = const()[name = tensor<string, []>("current_value_13_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> current_value_13_pad_0 = const()[name = tensor<string, []>("current_value_13_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> current_value_13_dilations_0 = const()[name = tensor<string, []>("current_value_13_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> current_value_13_groups_0 = const()[name = tensor<string, []>("current_value_13_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 768, 1, 1]> layers_6_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_6_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [768, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(196132480)))]; |
| tensor<fp16, [768]> layers_6_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("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, []>(197312192)))]; |
| tensor<fp16, [1, 768, 1, 1]> current_value_13_cast_fp16 = conv(bias = layers_6_self_attn_v_proj_bias_to_fp16, dilations = current_value_13_dilations_0, groups = current_value_13_groups_0, pad = current_value_13_pad_0, pad_type = current_value_13_pad_type_0, strides = current_value_13_strides_0, weight = layers_6_self_attn_v_proj_weight_to_fp16, x = obj_85_cast_fp16)[name = tensor<string, []>("current_value_13_cast_fp16")]; |
| tensor<fp16, [1, 768, 1, 448]> var_1478_cast_fp16 = mul(x = var_63_cast_fp16_6, y = var_159_cast_fp16)[name = tensor<string, []>("op_1478_cast_fp16")]; |
| tensor<fp16, [1, 768, 1, 448]> var_1479_cast_fp16 = mul(x = current_key_13_cast_fp16, y = var_157_cast_fp16)[name = tensor<string, []>("op_1479_cast_fp16")]; |
| tensor<fp16, [1, 768, 1, 448]> key_25_cast_fp16 = add(x = var_1478_cast_fp16, y = var_1479_cast_fp16)[name = tensor<string, []>("key_25_cast_fp16")]; |
| tensor<fp16, [1, 768, 1, 448]> var_1482_cast_fp16 = mul(x = var_78_cast_fp16_6, y = var_159_cast_fp16)[name = tensor<string, []>("op_1482_cast_fp16")]; |
| tensor<fp16, [1, 768, 1, 448]> var_1483_cast_fp16 = mul(x = current_value_13_cast_fp16, y = var_157_cast_fp16)[name = tensor<string, []>("op_1483_cast_fp16")]; |
| tensor<fp16, [1, 768, 1, 448]> value_25_cast_fp16 = add(x = var_1482_cast_fp16, y = var_1483_cast_fp16)[name = tensor<string, []>("value_25_cast_fp16")]; |
| tensor<int32, [4]> var_1487 = const()[name = tensor<string, []>("op_1487"), val = tensor<int32, [4]>([1, 12, 64, 1])]; |
| tensor<fp16, [1, 12, 64, 1]> mh_q_25_cast_fp16 = reshape(shape = var_1487, x = query_25_cast_fp16)[name = tensor<string, []>("mh_q_25_cast_fp16")]; |
| tensor<fp16, []> var_1489_to_fp16 = const()[name = tensor<string, []>("op_1489_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; |
| tensor<fp16, [1, 12, 64, 1]> var_1490_cast_fp16 = mul(x = mh_q_25_cast_fp16, y = var_1489_to_fp16)[name = tensor<string, []>("op_1490_cast_fp16")]; |
| tensor<int32, [4]> var_1493 = const()[name = tensor<string, []>("op_1493"), val = tensor<int32, [4]>([1, 12, 64, 448])]; |
| tensor<fp16, [1, 12, 64, 448]> var_1494_cast_fp16 = reshape(shape = var_1493, x = key_25_cast_fp16)[name = tensor<string, []>("op_1494_cast_fp16")]; |
| tensor<bool, []> mh_w_37_transpose_x_0 = const()[name = tensor<string, []>("mh_w_37_transpose_x_0"), val = tensor<bool, []>(true)]; |
| tensor<bool, []> mh_w_37_transpose_y_0 = const()[name = tensor<string, []>("mh_w_37_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [1, 12, 1, 448]> mh_w_37_cast_fp16 = matmul(transpose_x = mh_w_37_transpose_x_0, transpose_y = mh_w_37_transpose_y_0, x = var_1490_cast_fp16, y = var_1494_cast_fp16)[name = tensor<string, []>("mh_w_37_cast_fp16")]; |
| tensor<fp16, [1, 12, 1, 448]> mh_w_39_cast_fp16 = add(x = mh_w_37_cast_fp16, y = var_181_cast_fp16)[name = tensor<string, []>("mh_w_39_cast_fp16")]; |
| tensor<fp16, [1, 12, 1, 448]> var_1502_cast_fp16 = softmax(axis = var_1414, x = mh_w_39_cast_fp16)[name = tensor<string, []>("op_1502_cast_fp16")]; |
| tensor<int32, [4]> var_1503 = const()[name = tensor<string, []>("op_1503"), val = tensor<int32, [4]>([1, 12, 64, 448])]; |
| tensor<fp16, [1, 12, 64, 448]> var_1504_cast_fp16 = reshape(shape = var_1503, x = value_25_cast_fp16)[name = tensor<string, []>("op_1504_cast_fp16")]; |
| tensor<bool, []> attn_25_transpose_x_0 = const()[name = tensor<string, []>("attn_25_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_25_transpose_y_0 = const()[name = tensor<string, []>("attn_25_transpose_y_0"), val = tensor<bool, []>(true)]; |
| tensor<fp16, [1, 12, 64, 1]> attn_25_cast_fp16 = matmul(transpose_x = attn_25_transpose_x_0, transpose_y = attn_25_transpose_y_0, x = var_1504_cast_fp16, y = var_1502_cast_fp16)[name = tensor<string, []>("attn_25_cast_fp16")]; |
| tensor<int32, [4]> var_1507 = const()[name = tensor<string, []>("op_1507"), val = tensor<int32, [4]>([1, 768, 1, 1])]; |
| tensor<fp16, [1, 768, 1, 1]> input_61_cast_fp16 = reshape(shape = var_1507, x = attn_25_cast_fp16)[name = tensor<string, []>("input_61_cast_fp16")]; |
| tensor<string, []> obj_91_pad_type_0 = const()[name = tensor<string, []>("obj_91_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> obj_91_strides_0 = const()[name = tensor<string, []>("obj_91_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> obj_91_pad_0 = const()[name = tensor<string, []>("obj_91_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> obj_91_dilations_0 = const()[name = tensor<string, []>("obj_91_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> obj_91_groups_0 = const()[name = tensor<string, []>("obj_91_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 768, 1, 1]> layers_6_self_attn_o_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_6_self_attn_o_proj_weight_to_fp16"), val = tensor<fp16, [768, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(197313792)))]; |
| tensor<fp16, [768]> layers_6_self_attn_o_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_6_self_attn_o_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(198493504)))]; |
| tensor<fp16, [1, 768, 1, 1]> obj_91_cast_fp16 = conv(bias = layers_6_self_attn_o_proj_bias_to_fp16, dilations = obj_91_dilations_0, groups = obj_91_groups_0, pad = obj_91_pad_0, pad_type = obj_91_pad_type_0, strides = obj_91_strides_0, weight = layers_6_self_attn_o_proj_weight_to_fp16, x = input_61_cast_fp16)[name = tensor<string, []>("obj_91_cast_fp16")]; |
| tensor<fp16, [1, 768, 1, 1]> inputs_39_cast_fp16 = add(x = inputs_37_cast_fp16, y = obj_91_cast_fp16)[name = tensor<string, []>("inputs_39_cast_fp16")]; |
| tensor<int32, [1]> out_39_axes_0 = const()[name = tensor<string, []>("out_39_axes_0"), val = tensor<int32, [1]>([1])]; |
| tensor<fp16, []> var_1529_to_fp16 = const()[name = tensor<string, []>("op_1529_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, 768, 1, 1]> out_39_cast_fp16 = layer_norm(axes = out_39_axes_0, epsilon = var_1529_to_fp16, x = inputs_39_cast_fp16)[name = tensor<string, []>("out_39_cast_fp16")]; |
| tensor<fp16, [768]> obj_93_gamma_0_to_fp16 = const()[name = tensor<string, []>("obj_93_gamma_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(198495104)))]; |
| tensor<fp16, [768]> obj_93_beta_0_to_fp16 = const()[name = tensor<string, []>("obj_93_beta_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(198496704)))]; |
| tensor<fp16, []> obj_93_epsilon_0_to_fp16 = const()[name = tensor<string, []>("obj_93_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, 768, 1, 1]> obj_93_cast_fp16 = batch_norm(beta = obj_93_beta_0_to_fp16, epsilon = obj_93_epsilon_0_to_fp16, gamma = obj_93_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_39_cast_fp16)[name = tensor<string, []>("obj_93_cast_fp16")]; |
| tensor<string, []> query_27_pad_type_0 = const()[name = tensor<string, []>("query_27_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> query_27_strides_0 = const()[name = tensor<string, []>("query_27_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> query_27_pad_0 = const()[name = tensor<string, []>("query_27_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> query_27_dilations_0 = const()[name = tensor<string, []>("query_27_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> query_27_groups_0 = const()[name = tensor<string, []>("query_27_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 768, 1, 1]> layers_6_encoder_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_6_encoder_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [768, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(198498304)))]; |
| tensor<fp16, [768]> layers_6_encoder_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_6_encoder_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(199678016)))]; |
| tensor<fp16, [1, 768, 1, 1]> query_27_cast_fp16 = conv(bias = layers_6_encoder_attn_q_proj_bias_to_fp16, dilations = query_27_dilations_0, groups = query_27_groups_0, pad = query_27_pad_0, pad_type = query_27_pad_type_0, strides = query_27_strides_0, weight = layers_6_encoder_attn_q_proj_weight_to_fp16, x = obj_93_cast_fp16)[name = tensor<string, []>("query_27_cast_fp16")]; |
| tensor<string, []> key_27_pad_type_0 = const()[name = tensor<string, []>("key_27_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> key_27_strides_0 = const()[name = tensor<string, []>("key_27_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> key_27_pad_0 = const()[name = tensor<string, []>("key_27_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> key_27_dilations_0 = const()[name = tensor<string, []>("key_27_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> key_27_groups_0 = const()[name = tensor<string, []>("key_27_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 768, 1, 1]> layers_6_encoder_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_6_encoder_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [768, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(199679616)))]; |
| tensor<fp16, [1, 768, 1, 1500]> key_27_cast_fp16 = conv(dilations = key_27_dilations_0, groups = key_27_groups_0, pad = key_27_pad_0, pad_type = key_27_pad_type_0, strides = key_27_strides_0, weight = layers_6_encoder_attn_k_proj_weight_to_fp16, x = encoder_output_embeds)[name = tensor<string, []>("key_27_cast_fp16")]; |
| tensor<string, []> value_27_pad_type_0 = const()[name = tensor<string, []>("value_27_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> value_27_strides_0 = const()[name = tensor<string, []>("value_27_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> value_27_pad_0 = const()[name = tensor<string, []>("value_27_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> value_27_dilations_0 = const()[name = tensor<string, []>("value_27_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> value_27_groups_0 = const()[name = tensor<string, []>("value_27_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 768, 1, 1]> layers_6_encoder_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_6_encoder_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [768, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(200859328)))]; |
| tensor<fp16, [768]> layers_6_encoder_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_6_encoder_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(202039040)))]; |
| tensor<fp16, [1, 768, 1, 1500]> value_27_cast_fp16 = conv(bias = layers_6_encoder_attn_v_proj_bias_to_fp16, dilations = value_27_dilations_0, groups = value_27_groups_0, pad = value_27_pad_0, pad_type = value_27_pad_type_0, strides = value_27_strides_0, weight = layers_6_encoder_attn_v_proj_weight_to_fp16, x = encoder_output_embeds)[name = tensor<string, []>("value_27_cast_fp16")]; |
| tensor<int32, [4]> var_1565 = const()[name = tensor<string, []>("op_1565"), val = tensor<int32, [4]>([1, 12, 64, 1])]; |
| tensor<fp16, [1, 12, 64, 1]> mh_q_27_cast_fp16 = reshape(shape = var_1565, x = query_27_cast_fp16)[name = tensor<string, []>("mh_q_27_cast_fp16")]; |
| tensor<fp16, []> var_1567_to_fp16 = const()[name = tensor<string, []>("op_1567_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; |
| tensor<fp16, [1, 12, 64, 1]> var_1568_cast_fp16 = mul(x = mh_q_27_cast_fp16, y = var_1567_to_fp16)[name = tensor<string, []>("op_1568_cast_fp16")]; |
| tensor<int32, [4]> var_1571 = const()[name = tensor<string, []>("op_1571"), val = tensor<int32, [4]>([1, 12, 64, 1500])]; |
| tensor<fp16, [1, 12, 64, 1500]> var_1572_cast_fp16 = reshape(shape = var_1571, x = key_27_cast_fp16)[name = tensor<string, []>("op_1572_cast_fp16")]; |
| tensor<bool, []> mh_w_41_transpose_x_0 = const()[name = tensor<string, []>("mh_w_41_transpose_x_0"), val = tensor<bool, []>(true)]; |
| tensor<bool, []> mh_w_41_transpose_y_0 = const()[name = tensor<string, []>("mh_w_41_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [1, 12, 1, 1500]> mh_w_41_cast_fp16 = matmul(transpose_x = mh_w_41_transpose_x_0, transpose_y = mh_w_41_transpose_y_0, x = var_1568_cast_fp16, y = var_1572_cast_fp16)[name = tensor<string, []>("mh_w_41_cast_fp16")]; |
| tensor<fp16, [1, 12, 1, 1500]> obj_97_cast_fp16 = softmax(axis = var_1414, x = mh_w_41_cast_fp16)[name = tensor<string, []>("obj_97_cast_fp16")]; |
| tensor<int32, [4]> var_1576 = const()[name = tensor<string, []>("op_1576"), val = tensor<int32, [4]>([1, 12, 64, 1500])]; |
| tensor<fp16, [1, 12, 64, 1500]> var_1577_cast_fp16 = reshape(shape = var_1576, x = value_27_cast_fp16)[name = tensor<string, []>("op_1577_cast_fp16")]; |
| tensor<bool, []> attn_27_transpose_x_0 = const()[name = tensor<string, []>("attn_27_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_27_transpose_y_0 = const()[name = tensor<string, []>("attn_27_transpose_y_0"), val = tensor<bool, []>(true)]; |
| tensor<fp16, [1, 12, 64, 1]> attn_27_cast_fp16 = matmul(transpose_x = attn_27_transpose_x_0, transpose_y = attn_27_transpose_y_0, x = var_1577_cast_fp16, y = obj_97_cast_fp16)[name = tensor<string, []>("attn_27_cast_fp16")]; |
| tensor<int32, [4]> var_1580 = const()[name = tensor<string, []>("op_1580"), val = tensor<int32, [4]>([1, 768, 1, 1])]; |
| tensor<fp16, [1, 768, 1, 1]> input_63_cast_fp16 = reshape(shape = var_1580, x = attn_27_cast_fp16)[name = tensor<string, []>("input_63_cast_fp16")]; |
| tensor<string, []> obj_95_pad_type_0 = const()[name = tensor<string, []>("obj_95_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> obj_95_strides_0 = const()[name = tensor<string, []>("obj_95_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> obj_95_pad_0 = const()[name = tensor<string, []>("obj_95_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> obj_95_dilations_0 = const()[name = tensor<string, []>("obj_95_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> obj_95_groups_0 = const()[name = tensor<string, []>("obj_95_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 768, 1, 1]> layers_6_encoder_attn_o_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_6_encoder_attn_o_proj_weight_to_fp16"), val = tensor<fp16, [768, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(202040640)))]; |
| tensor<fp16, [768]> layers_6_encoder_attn_o_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_6_encoder_attn_o_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(203220352)))]; |
| tensor<fp16, [1, 768, 1, 1]> obj_95_cast_fp16 = conv(bias = layers_6_encoder_attn_o_proj_bias_to_fp16, dilations = obj_95_dilations_0, groups = obj_95_groups_0, pad = obj_95_pad_0, pad_type = obj_95_pad_type_0, strides = obj_95_strides_0, weight = layers_6_encoder_attn_o_proj_weight_to_fp16, x = input_63_cast_fp16)[name = tensor<string, []>("obj_95_cast_fp16")]; |
| tensor<fp16, [1, 768, 1, 1]> inputs_41_cast_fp16 = add(x = inputs_39_cast_fp16, y = obj_95_cast_fp16)[name = tensor<string, []>("inputs_41_cast_fp16")]; |
| tensor<int32, [1]> out_41_axes_0 = const()[name = tensor<string, []>("out_41_axes_0"), val = tensor<int32, [1]>([1])]; |
| tensor<fp16, []> var_1598_to_fp16 = const()[name = tensor<string, []>("op_1598_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, 768, 1, 1]> out_41_cast_fp16 = layer_norm(axes = out_41_axes_0, epsilon = var_1598_to_fp16, x = inputs_41_cast_fp16)[name = tensor<string, []>("out_41_cast_fp16")]; |
| tensor<fp16, [768]> input_65_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_65_gamma_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(203221952)))]; |
| tensor<fp16, [768]> input_65_beta_0_to_fp16 = const()[name = tensor<string, []>("input_65_beta_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(203223552)))]; |
| tensor<fp16, []> input_65_epsilon_0_to_fp16 = const()[name = tensor<string, []>("input_65_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, 768, 1, 1]> input_65_cast_fp16 = batch_norm(beta = input_65_beta_0_to_fp16, epsilon = input_65_epsilon_0_to_fp16, gamma = input_65_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_41_cast_fp16)[name = tensor<string, []>("input_65_cast_fp16")]; |
| tensor<string, []> input_67_pad_type_0 = const()[name = tensor<string, []>("input_67_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> input_67_strides_0 = const()[name = tensor<string, []>("input_67_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> input_67_pad_0 = const()[name = tensor<string, []>("input_67_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> input_67_dilations_0 = const()[name = tensor<string, []>("input_67_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> input_67_groups_0 = const()[name = tensor<string, []>("input_67_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [3072, 768, 1, 1]> layers_6_fc1_weight_to_fp16 = const()[name = tensor<string, []>("layers_6_fc1_weight_to_fp16"), val = tensor<fp16, [3072, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(203225152)))]; |
| tensor<fp16, [3072]> layers_6_fc1_bias_to_fp16 = const()[name = tensor<string, []>("layers_6_fc1_bias_to_fp16"), val = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(207943808)))]; |
| tensor<fp16, [1, 3072, 1, 1]> input_67_cast_fp16 = conv(bias = layers_6_fc1_bias_to_fp16, dilations = input_67_dilations_0, groups = input_67_groups_0, pad = input_67_pad_0, pad_type = input_67_pad_type_0, strides = input_67_strides_0, weight = layers_6_fc1_weight_to_fp16, x = input_65_cast_fp16)[name = tensor<string, []>("input_67_cast_fp16")]; |
| tensor<string, []> input_69_mode_0 = const()[name = tensor<string, []>("input_69_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 3072, 1, 1]> input_69_cast_fp16 = gelu(mode = input_69_mode_0, x = input_67_cast_fp16)[name = tensor<string, []>("input_69_cast_fp16")]; |
| tensor<string, []> hidden_states_15_pad_type_0 = const()[name = tensor<string, []>("hidden_states_15_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> hidden_states_15_strides_0 = const()[name = tensor<string, []>("hidden_states_15_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> hidden_states_15_pad_0 = const()[name = tensor<string, []>("hidden_states_15_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> hidden_states_15_dilations_0 = const()[name = tensor<string, []>("hidden_states_15_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> hidden_states_15_groups_0 = const()[name = tensor<string, []>("hidden_states_15_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 3072, 1, 1]> layers_6_fc2_weight_to_fp16 = const()[name = tensor<string, []>("layers_6_fc2_weight_to_fp16"), val = tensor<fp16, [768, 3072, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(207950016)))]; |
| tensor<fp16, [768]> layers_6_fc2_bias_to_fp16 = const()[name = tensor<string, []>("layers_6_fc2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(212668672)))]; |
| tensor<fp16, [1, 768, 1, 1]> hidden_states_15_cast_fp16 = conv(bias = layers_6_fc2_bias_to_fp16, dilations = hidden_states_15_dilations_0, groups = hidden_states_15_groups_0, pad = hidden_states_15_pad_0, pad_type = hidden_states_15_pad_type_0, strides = hidden_states_15_strides_0, weight = layers_6_fc2_weight_to_fp16, x = input_69_cast_fp16)[name = tensor<string, []>("hidden_states_15_cast_fp16")]; |
| tensor<fp16, [1, 768, 1, 1]> inputs_43_cast_fp16 = add(x = inputs_41_cast_fp16, y = hidden_states_15_cast_fp16)[name = tensor<string, []>("inputs_43_cast_fp16")]; |
| tensor<int32, []> var_1633 = const()[name = tensor<string, []>("op_1633"), val = tensor<int32, []>(3)]; |
| tensor<int32, [1]> out_43_axes_0 = const()[name = tensor<string, []>("out_43_axes_0"), val = tensor<int32, [1]>([1])]; |
| tensor<fp16, []> var_1658_to_fp16 = const()[name = tensor<string, []>("op_1658_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, 768, 1, 1]> out_43_cast_fp16 = layer_norm(axes = out_43_axes_0, epsilon = var_1658_to_fp16, x = inputs_43_cast_fp16)[name = tensor<string, []>("out_43_cast_fp16")]; |
| tensor<fp16, [768]> obj_99_gamma_0_to_fp16 = const()[name = tensor<string, []>("obj_99_gamma_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(212670272)))]; |
| tensor<fp16, [768]> obj_99_beta_0_to_fp16 = const()[name = tensor<string, []>("obj_99_beta_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(212671872)))]; |
| tensor<fp16, []> obj_99_epsilon_0_to_fp16 = const()[name = tensor<string, []>("obj_99_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, 768, 1, 1]> obj_99_cast_fp16 = batch_norm(beta = obj_99_beta_0_to_fp16, epsilon = obj_99_epsilon_0_to_fp16, gamma = obj_99_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_43_cast_fp16)[name = tensor<string, []>("obj_99_cast_fp16")]; |
| tensor<string, []> query_29_pad_type_0 = const()[name = tensor<string, []>("query_29_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> query_29_strides_0 = const()[name = tensor<string, []>("query_29_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> query_29_pad_0 = const()[name = tensor<string, []>("query_29_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> query_29_dilations_0 = const()[name = tensor<string, []>("query_29_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> query_29_groups_0 = const()[name = tensor<string, []>("query_29_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 768, 1, 1]> layers_7_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_7_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [768, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(212673472)))]; |
| tensor<fp16, [768]> layers_7_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("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, []>(213853184)))]; |
| tensor<fp16, [1, 768, 1, 1]> query_29_cast_fp16 = conv(bias = layers_7_self_attn_q_proj_bias_to_fp16, dilations = query_29_dilations_0, groups = query_29_groups_0, pad = query_29_pad_0, pad_type = query_29_pad_type_0, strides = query_29_strides_0, weight = layers_7_self_attn_q_proj_weight_to_fp16, x = obj_99_cast_fp16)[name = tensor<string, []>("query_29_cast_fp16")]; |
| tensor<string, []> current_key_15_pad_type_0 = const()[name = tensor<string, []>("current_key_15_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> current_key_15_strides_0 = const()[name = tensor<string, []>("current_key_15_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> current_key_15_pad_0 = const()[name = tensor<string, []>("current_key_15_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> current_key_15_dilations_0 = const()[name = tensor<string, []>("current_key_15_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> current_key_15_groups_0 = const()[name = tensor<string, []>("current_key_15_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 768, 1, 1]> layers_7_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_7_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [768, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(213854784)))]; |
| tensor<fp16, [1, 768, 1, 1]> current_key_15_cast_fp16 = conv(dilations = current_key_15_dilations_0, groups = current_key_15_groups_0, pad = current_key_15_pad_0, pad_type = current_key_15_pad_type_0, strides = current_key_15_strides_0, weight = layers_7_self_attn_k_proj_weight_to_fp16, x = obj_99_cast_fp16)[name = tensor<string, []>("current_key_15_cast_fp16")]; |
| tensor<string, []> current_value_15_pad_type_0 = const()[name = tensor<string, []>("current_value_15_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> current_value_15_strides_0 = const()[name = tensor<string, []>("current_value_15_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> current_value_15_pad_0 = const()[name = tensor<string, []>("current_value_15_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> current_value_15_dilations_0 = const()[name = tensor<string, []>("current_value_15_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> current_value_15_groups_0 = const()[name = tensor<string, []>("current_value_15_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 768, 1, 1]> layers_7_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_7_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [768, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(215034496)))]; |
| tensor<fp16, [768]> layers_7_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("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, []>(216214208)))]; |
| tensor<fp16, [1, 768, 1, 1]> current_value_15_cast_fp16 = conv(bias = layers_7_self_attn_v_proj_bias_to_fp16, dilations = current_value_15_dilations_0, groups = current_value_15_groups_0, pad = current_value_15_pad_0, pad_type = current_value_15_pad_type_0, strides = current_value_15_strides_0, weight = layers_7_self_attn_v_proj_weight_to_fp16, x = obj_99_cast_fp16)[name = tensor<string, []>("current_value_15_cast_fp16")]; |
| tensor<fp16, [1, 768, 1, 448]> var_1697_cast_fp16 = mul(x = var_63_cast_fp16_7, y = var_159_cast_fp16)[name = tensor<string, []>("op_1697_cast_fp16")]; |
| tensor<fp16, [1, 768, 1, 448]> var_1698_cast_fp16 = mul(x = current_key_15_cast_fp16, y = var_157_cast_fp16)[name = tensor<string, []>("op_1698_cast_fp16")]; |
| tensor<fp16, [1, 768, 1, 448]> key_29_cast_fp16 = add(x = var_1697_cast_fp16, y = var_1698_cast_fp16)[name = tensor<string, []>("key_29_cast_fp16")]; |
| tensor<fp16, [1, 768, 1, 448]> var_1701_cast_fp16 = mul(x = var_78_cast_fp16_7, y = var_159_cast_fp16)[name = tensor<string, []>("op_1701_cast_fp16")]; |
| tensor<fp16, [1, 768, 1, 448]> var_1702_cast_fp16 = mul(x = current_value_15_cast_fp16, y = var_157_cast_fp16)[name = tensor<string, []>("op_1702_cast_fp16")]; |
| tensor<fp16, [1, 768, 1, 448]> value_29_cast_fp16 = add(x = var_1701_cast_fp16, y = var_1702_cast_fp16)[name = tensor<string, []>("value_29_cast_fp16")]; |
| tensor<int32, [4]> var_1706 = const()[name = tensor<string, []>("op_1706"), val = tensor<int32, [4]>([1, 12, 64, 1])]; |
| tensor<fp16, [1, 12, 64, 1]> mh_q_29_cast_fp16 = reshape(shape = var_1706, x = query_29_cast_fp16)[name = tensor<string, []>("mh_q_29_cast_fp16")]; |
| tensor<fp16, []> var_1708_to_fp16 = const()[name = tensor<string, []>("op_1708_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; |
| tensor<fp16, [1, 12, 64, 1]> var_1709_cast_fp16 = mul(x = mh_q_29_cast_fp16, y = var_1708_to_fp16)[name = tensor<string, []>("op_1709_cast_fp16")]; |
| tensor<int32, [4]> var_1712 = const()[name = tensor<string, []>("op_1712"), val = tensor<int32, [4]>([1, 12, 64, 448])]; |
| tensor<fp16, [1, 12, 64, 448]> var_1713_cast_fp16 = reshape(shape = var_1712, x = key_29_cast_fp16)[name = tensor<string, []>("op_1713_cast_fp16")]; |
| tensor<bool, []> mh_w_43_transpose_x_0 = const()[name = tensor<string, []>("mh_w_43_transpose_x_0"), val = tensor<bool, []>(true)]; |
| tensor<bool, []> mh_w_43_transpose_y_0 = const()[name = tensor<string, []>("mh_w_43_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [1, 12, 1, 448]> mh_w_43_cast_fp16 = matmul(transpose_x = mh_w_43_transpose_x_0, transpose_y = mh_w_43_transpose_y_0, x = var_1709_cast_fp16, y = var_1713_cast_fp16)[name = tensor<string, []>("mh_w_43_cast_fp16")]; |
| tensor<fp16, [1, 12, 1, 448]> mh_w_45_cast_fp16 = add(x = mh_w_43_cast_fp16, y = var_181_cast_fp16)[name = tensor<string, []>("mh_w_45_cast_fp16")]; |
| tensor<fp16, [1, 12, 1, 448]> var_1721_cast_fp16 = softmax(axis = var_1633, x = mh_w_45_cast_fp16)[name = tensor<string, []>("op_1721_cast_fp16")]; |
| tensor<int32, [4]> var_1722 = const()[name = tensor<string, []>("op_1722"), val = tensor<int32, [4]>([1, 12, 64, 448])]; |
| tensor<fp16, [1, 12, 64, 448]> var_1723_cast_fp16 = reshape(shape = var_1722, x = value_29_cast_fp16)[name = tensor<string, []>("op_1723_cast_fp16")]; |
| tensor<bool, []> attn_29_transpose_x_0 = const()[name = tensor<string, []>("attn_29_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_29_transpose_y_0 = const()[name = tensor<string, []>("attn_29_transpose_y_0"), val = tensor<bool, []>(true)]; |
| tensor<fp16, [1, 12, 64, 1]> attn_29_cast_fp16 = matmul(transpose_x = attn_29_transpose_x_0, transpose_y = attn_29_transpose_y_0, x = var_1723_cast_fp16, y = var_1721_cast_fp16)[name = tensor<string, []>("attn_29_cast_fp16")]; |
| tensor<int32, [4]> var_1726 = const()[name = tensor<string, []>("op_1726"), val = tensor<int32, [4]>([1, 768, 1, 1])]; |
| tensor<fp16, [1, 768, 1, 1]> input_71_cast_fp16 = reshape(shape = var_1726, x = attn_29_cast_fp16)[name = tensor<string, []>("input_71_cast_fp16")]; |
| tensor<string, []> obj_105_pad_type_0 = const()[name = tensor<string, []>("obj_105_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> obj_105_strides_0 = const()[name = tensor<string, []>("obj_105_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> obj_105_pad_0 = const()[name = tensor<string, []>("obj_105_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> obj_105_dilations_0 = const()[name = tensor<string, []>("obj_105_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> obj_105_groups_0 = const()[name = tensor<string, []>("obj_105_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 768, 1, 1]> layers_7_self_attn_o_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_7_self_attn_o_proj_weight_to_fp16"), val = tensor<fp16, [768, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(216215808)))]; |
| tensor<fp16, [768]> layers_7_self_attn_o_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_7_self_attn_o_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(217395520)))]; |
| tensor<fp16, [1, 768, 1, 1]> obj_105_cast_fp16 = conv(bias = layers_7_self_attn_o_proj_bias_to_fp16, dilations = obj_105_dilations_0, groups = obj_105_groups_0, pad = obj_105_pad_0, pad_type = obj_105_pad_type_0, strides = obj_105_strides_0, weight = layers_7_self_attn_o_proj_weight_to_fp16, x = input_71_cast_fp16)[name = tensor<string, []>("obj_105_cast_fp16")]; |
| tensor<fp16, [1, 768, 1, 1]> inputs_45_cast_fp16 = add(x = inputs_43_cast_fp16, y = obj_105_cast_fp16)[name = tensor<string, []>("inputs_45_cast_fp16")]; |
| tensor<int32, [1]> out_45_axes_0 = const()[name = tensor<string, []>("out_45_axes_0"), val = tensor<int32, [1]>([1])]; |
| tensor<fp16, []> var_1748_to_fp16 = const()[name = tensor<string, []>("op_1748_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, 768, 1, 1]> out_45_cast_fp16 = layer_norm(axes = out_45_axes_0, epsilon = var_1748_to_fp16, x = inputs_45_cast_fp16)[name = tensor<string, []>("out_45_cast_fp16")]; |
| tensor<fp16, [768]> obj_107_gamma_0_to_fp16 = const()[name = tensor<string, []>("obj_107_gamma_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(217397120)))]; |
| tensor<fp16, [768]> obj_107_beta_0_to_fp16 = const()[name = tensor<string, []>("obj_107_beta_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(217398720)))]; |
| tensor<fp16, []> obj_107_epsilon_0_to_fp16 = const()[name = tensor<string, []>("obj_107_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, 768, 1, 1]> obj_107_cast_fp16 = batch_norm(beta = obj_107_beta_0_to_fp16, epsilon = obj_107_epsilon_0_to_fp16, gamma = obj_107_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_45_cast_fp16)[name = tensor<string, []>("obj_107_cast_fp16")]; |
| tensor<string, []> query_31_pad_type_0 = const()[name = tensor<string, []>("query_31_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> query_31_strides_0 = const()[name = tensor<string, []>("query_31_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> query_31_pad_0 = const()[name = tensor<string, []>("query_31_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> query_31_dilations_0 = const()[name = tensor<string, []>("query_31_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> query_31_groups_0 = const()[name = tensor<string, []>("query_31_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 768, 1, 1]> layers_7_encoder_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_7_encoder_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [768, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(217400320)))]; |
| tensor<fp16, [768]> layers_7_encoder_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_7_encoder_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(218580032)))]; |
| tensor<fp16, [1, 768, 1, 1]> query_31_cast_fp16 = conv(bias = layers_7_encoder_attn_q_proj_bias_to_fp16, dilations = query_31_dilations_0, groups = query_31_groups_0, pad = query_31_pad_0, pad_type = query_31_pad_type_0, strides = query_31_strides_0, weight = layers_7_encoder_attn_q_proj_weight_to_fp16, x = obj_107_cast_fp16)[name = tensor<string, []>("query_31_cast_fp16")]; |
| tensor<string, []> key_31_pad_type_0 = const()[name = tensor<string, []>("key_31_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> key_31_strides_0 = const()[name = tensor<string, []>("key_31_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> key_31_pad_0 = const()[name = tensor<string, []>("key_31_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> key_31_dilations_0 = const()[name = tensor<string, []>("key_31_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> key_31_groups_0 = const()[name = tensor<string, []>("key_31_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 768, 1, 1]> layers_7_encoder_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_7_encoder_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [768, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(218581632)))]; |
| tensor<fp16, [1, 768, 1, 1500]> key_31_cast_fp16 = conv(dilations = key_31_dilations_0, groups = key_31_groups_0, pad = key_31_pad_0, pad_type = key_31_pad_type_0, strides = key_31_strides_0, weight = layers_7_encoder_attn_k_proj_weight_to_fp16, x = encoder_output_embeds)[name = tensor<string, []>("key_31_cast_fp16")]; |
| tensor<string, []> value_31_pad_type_0 = const()[name = tensor<string, []>("value_31_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> value_31_strides_0 = const()[name = tensor<string, []>("value_31_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> value_31_pad_0 = const()[name = tensor<string, []>("value_31_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> value_31_dilations_0 = const()[name = tensor<string, []>("value_31_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> value_31_groups_0 = const()[name = tensor<string, []>("value_31_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 768, 1, 1]> layers_7_encoder_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_7_encoder_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [768, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(219761344)))]; |
| tensor<fp16, [768]> layers_7_encoder_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_7_encoder_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(220941056)))]; |
| tensor<fp16, [1, 768, 1, 1500]> value_31_cast_fp16 = conv(bias = layers_7_encoder_attn_v_proj_bias_to_fp16, dilations = value_31_dilations_0, groups = value_31_groups_0, pad = value_31_pad_0, pad_type = value_31_pad_type_0, strides = value_31_strides_0, weight = layers_7_encoder_attn_v_proj_weight_to_fp16, x = encoder_output_embeds)[name = tensor<string, []>("value_31_cast_fp16")]; |
| tensor<int32, [4]> var_1784 = const()[name = tensor<string, []>("op_1784"), val = tensor<int32, [4]>([1, 12, 64, 1])]; |
| tensor<fp16, [1, 12, 64, 1]> mh_q_31_cast_fp16 = reshape(shape = var_1784, x = query_31_cast_fp16)[name = tensor<string, []>("mh_q_31_cast_fp16")]; |
| tensor<fp16, []> var_1786_to_fp16 = const()[name = tensor<string, []>("op_1786_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; |
| tensor<fp16, [1, 12, 64, 1]> var_1787_cast_fp16 = mul(x = mh_q_31_cast_fp16, y = var_1786_to_fp16)[name = tensor<string, []>("op_1787_cast_fp16")]; |
| tensor<int32, [4]> var_1790 = const()[name = tensor<string, []>("op_1790"), val = tensor<int32, [4]>([1, 12, 64, 1500])]; |
| tensor<fp16, [1, 12, 64, 1500]> var_1791_cast_fp16 = reshape(shape = var_1790, x = key_31_cast_fp16)[name = tensor<string, []>("op_1791_cast_fp16")]; |
| tensor<bool, []> mh_w_47_transpose_x_0 = const()[name = tensor<string, []>("mh_w_47_transpose_x_0"), val = tensor<bool, []>(true)]; |
| tensor<bool, []> mh_w_47_transpose_y_0 = const()[name = tensor<string, []>("mh_w_47_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [1, 12, 1, 1500]> mh_w_47_cast_fp16 = matmul(transpose_x = mh_w_47_transpose_x_0, transpose_y = mh_w_47_transpose_y_0, x = var_1787_cast_fp16, y = var_1791_cast_fp16)[name = tensor<string, []>("mh_w_47_cast_fp16")]; |
| tensor<fp16, [1, 12, 1, 1500]> obj_111_cast_fp16 = softmax(axis = var_1633, x = mh_w_47_cast_fp16)[name = tensor<string, []>("obj_111_cast_fp16")]; |
| tensor<int32, [4]> var_1795 = const()[name = tensor<string, []>("op_1795"), val = tensor<int32, [4]>([1, 12, 64, 1500])]; |
| tensor<fp16, [1, 12, 64, 1500]> var_1796_cast_fp16 = reshape(shape = var_1795, x = value_31_cast_fp16)[name = tensor<string, []>("op_1796_cast_fp16")]; |
| tensor<bool, []> attn_31_transpose_x_0 = const()[name = tensor<string, []>("attn_31_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_31_transpose_y_0 = const()[name = tensor<string, []>("attn_31_transpose_y_0"), val = tensor<bool, []>(true)]; |
| tensor<fp16, [1, 12, 64, 1]> attn_31_cast_fp16 = matmul(transpose_x = attn_31_transpose_x_0, transpose_y = attn_31_transpose_y_0, x = var_1796_cast_fp16, y = obj_111_cast_fp16)[name = tensor<string, []>("attn_31_cast_fp16")]; |
| tensor<int32, [4]> var_1799 = const()[name = tensor<string, []>("op_1799"), val = tensor<int32, [4]>([1, 768, 1, 1])]; |
| tensor<fp16, [1, 768, 1, 1]> input_73_cast_fp16 = reshape(shape = var_1799, x = attn_31_cast_fp16)[name = tensor<string, []>("input_73_cast_fp16")]; |
| tensor<string, []> obj_109_pad_type_0 = const()[name = tensor<string, []>("obj_109_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> obj_109_strides_0 = const()[name = tensor<string, []>("obj_109_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> obj_109_pad_0 = const()[name = tensor<string, []>("obj_109_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> obj_109_dilations_0 = const()[name = tensor<string, []>("obj_109_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> obj_109_groups_0 = const()[name = tensor<string, []>("obj_109_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 768, 1, 1]> layers_7_encoder_attn_o_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_7_encoder_attn_o_proj_weight_to_fp16"), val = tensor<fp16, [768, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(220942656)))]; |
| tensor<fp16, [768]> layers_7_encoder_attn_o_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_7_encoder_attn_o_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(222122368)))]; |
| tensor<fp16, [1, 768, 1, 1]> obj_109_cast_fp16 = conv(bias = layers_7_encoder_attn_o_proj_bias_to_fp16, dilations = obj_109_dilations_0, groups = obj_109_groups_0, pad = obj_109_pad_0, pad_type = obj_109_pad_type_0, strides = obj_109_strides_0, weight = layers_7_encoder_attn_o_proj_weight_to_fp16, x = input_73_cast_fp16)[name = tensor<string, []>("obj_109_cast_fp16")]; |
| tensor<fp16, [1, 768, 1, 1]> inputs_47_cast_fp16 = add(x = inputs_45_cast_fp16, y = obj_109_cast_fp16)[name = tensor<string, []>("inputs_47_cast_fp16")]; |
| tensor<int32, [1]> out_47_axes_0 = const()[name = tensor<string, []>("out_47_axes_0"), val = tensor<int32, [1]>([1])]; |
| tensor<fp16, []> var_1817_to_fp16 = const()[name = tensor<string, []>("op_1817_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, 768, 1, 1]> out_47_cast_fp16 = layer_norm(axes = out_47_axes_0, epsilon = var_1817_to_fp16, x = inputs_47_cast_fp16)[name = tensor<string, []>("out_47_cast_fp16")]; |
| tensor<fp16, [768]> input_75_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_75_gamma_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(222123968)))]; |
| tensor<fp16, [768]> input_75_beta_0_to_fp16 = const()[name = tensor<string, []>("input_75_beta_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(222125568)))]; |
| tensor<fp16, []> input_75_epsilon_0_to_fp16 = const()[name = tensor<string, []>("input_75_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, 768, 1, 1]> input_75_cast_fp16 = batch_norm(beta = input_75_beta_0_to_fp16, epsilon = input_75_epsilon_0_to_fp16, gamma = input_75_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_47_cast_fp16)[name = tensor<string, []>("input_75_cast_fp16")]; |
| tensor<string, []> input_77_pad_type_0 = const()[name = tensor<string, []>("input_77_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> input_77_strides_0 = const()[name = tensor<string, []>("input_77_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> input_77_pad_0 = const()[name = tensor<string, []>("input_77_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> input_77_dilations_0 = const()[name = tensor<string, []>("input_77_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> input_77_groups_0 = const()[name = tensor<string, []>("input_77_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [3072, 768, 1, 1]> layers_7_fc1_weight_to_fp16 = const()[name = tensor<string, []>("layers_7_fc1_weight_to_fp16"), val = tensor<fp16, [3072, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(222127168)))]; |
| tensor<fp16, [3072]> layers_7_fc1_bias_to_fp16 = const()[name = tensor<string, []>("layers_7_fc1_bias_to_fp16"), val = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(226845824)))]; |
| tensor<fp16, [1, 3072, 1, 1]> input_77_cast_fp16 = conv(bias = layers_7_fc1_bias_to_fp16, dilations = input_77_dilations_0, groups = input_77_groups_0, pad = input_77_pad_0, pad_type = input_77_pad_type_0, strides = input_77_strides_0, weight = layers_7_fc1_weight_to_fp16, x = input_75_cast_fp16)[name = tensor<string, []>("input_77_cast_fp16")]; |
| tensor<string, []> input_79_mode_0 = const()[name = tensor<string, []>("input_79_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 3072, 1, 1]> input_79_cast_fp16 = gelu(mode = input_79_mode_0, x = input_77_cast_fp16)[name = tensor<string, []>("input_79_cast_fp16")]; |
| tensor<string, []> hidden_states_17_pad_type_0 = const()[name = tensor<string, []>("hidden_states_17_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> hidden_states_17_strides_0 = const()[name = tensor<string, []>("hidden_states_17_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> hidden_states_17_pad_0 = const()[name = tensor<string, []>("hidden_states_17_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> hidden_states_17_dilations_0 = const()[name = tensor<string, []>("hidden_states_17_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> hidden_states_17_groups_0 = const()[name = tensor<string, []>("hidden_states_17_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 3072, 1, 1]> layers_7_fc2_weight_to_fp16 = const()[name = tensor<string, []>("layers_7_fc2_weight_to_fp16"), val = tensor<fp16, [768, 3072, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(226852032)))]; |
| tensor<fp16, [768]> layers_7_fc2_bias_to_fp16 = const()[name = tensor<string, []>("layers_7_fc2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(231570688)))]; |
| tensor<fp16, [1, 768, 1, 1]> hidden_states_17_cast_fp16 = conv(bias = layers_7_fc2_bias_to_fp16, dilations = hidden_states_17_dilations_0, groups = hidden_states_17_groups_0, pad = hidden_states_17_pad_0, pad_type = hidden_states_17_pad_type_0, strides = hidden_states_17_strides_0, weight = layers_7_fc2_weight_to_fp16, x = input_79_cast_fp16)[name = tensor<string, []>("hidden_states_17_cast_fp16")]; |
| tensor<fp16, [1, 768, 1, 1]> inputs_49_cast_fp16 = add(x = inputs_47_cast_fp16, y = hidden_states_17_cast_fp16)[name = tensor<string, []>("inputs_49_cast_fp16")]; |
| tensor<int32, []> var_1852 = const()[name = tensor<string, []>("op_1852"), val = tensor<int32, []>(3)]; |
| tensor<int32, [1]> out_49_axes_0 = const()[name = tensor<string, []>("out_49_axes_0"), val = tensor<int32, [1]>([1])]; |
| tensor<fp16, []> var_1877_to_fp16 = const()[name = tensor<string, []>("op_1877_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, 768, 1, 1]> out_49_cast_fp16 = layer_norm(axes = out_49_axes_0, epsilon = var_1877_to_fp16, x = inputs_49_cast_fp16)[name = tensor<string, []>("out_49_cast_fp16")]; |
| tensor<fp16, [768]> obj_113_gamma_0_to_fp16 = const()[name = tensor<string, []>("obj_113_gamma_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(231572288)))]; |
| tensor<fp16, [768]> obj_113_beta_0_to_fp16 = const()[name = tensor<string, []>("obj_113_beta_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(231573888)))]; |
| tensor<fp16, []> obj_113_epsilon_0_to_fp16 = const()[name = tensor<string, []>("obj_113_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, 768, 1, 1]> obj_113_cast_fp16 = batch_norm(beta = obj_113_beta_0_to_fp16, epsilon = obj_113_epsilon_0_to_fp16, gamma = obj_113_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_49_cast_fp16)[name = tensor<string, []>("obj_113_cast_fp16")]; |
| tensor<string, []> query_33_pad_type_0 = const()[name = tensor<string, []>("query_33_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> query_33_strides_0 = const()[name = tensor<string, []>("query_33_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> query_33_pad_0 = const()[name = tensor<string, []>("query_33_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> query_33_dilations_0 = const()[name = tensor<string, []>("query_33_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> query_33_groups_0 = const()[name = tensor<string, []>("query_33_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 768, 1, 1]> layers_8_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_8_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [768, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(231575488)))]; |
| tensor<fp16, [768]> layers_8_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("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, []>(232755200)))]; |
| tensor<fp16, [1, 768, 1, 1]> query_33_cast_fp16 = conv(bias = layers_8_self_attn_q_proj_bias_to_fp16, dilations = query_33_dilations_0, groups = query_33_groups_0, pad = query_33_pad_0, pad_type = query_33_pad_type_0, strides = query_33_strides_0, weight = layers_8_self_attn_q_proj_weight_to_fp16, x = obj_113_cast_fp16)[name = tensor<string, []>("query_33_cast_fp16")]; |
| tensor<string, []> current_key_17_pad_type_0 = const()[name = tensor<string, []>("current_key_17_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> current_key_17_strides_0 = const()[name = tensor<string, []>("current_key_17_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> current_key_17_pad_0 = const()[name = tensor<string, []>("current_key_17_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> current_key_17_dilations_0 = const()[name = tensor<string, []>("current_key_17_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> current_key_17_groups_0 = const()[name = tensor<string, []>("current_key_17_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 768, 1, 1]> layers_8_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_8_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [768, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(232756800)))]; |
| tensor<fp16, [1, 768, 1, 1]> current_key_17_cast_fp16 = conv(dilations = current_key_17_dilations_0, groups = current_key_17_groups_0, pad = current_key_17_pad_0, pad_type = current_key_17_pad_type_0, strides = current_key_17_strides_0, weight = layers_8_self_attn_k_proj_weight_to_fp16, x = obj_113_cast_fp16)[name = tensor<string, []>("current_key_17_cast_fp16")]; |
| tensor<string, []> current_value_17_pad_type_0 = const()[name = tensor<string, []>("current_value_17_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> current_value_17_strides_0 = const()[name = tensor<string, []>("current_value_17_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> current_value_17_pad_0 = const()[name = tensor<string, []>("current_value_17_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> current_value_17_dilations_0 = const()[name = tensor<string, []>("current_value_17_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> current_value_17_groups_0 = const()[name = tensor<string, []>("current_value_17_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 768, 1, 1]> layers_8_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_8_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [768, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(233936512)))]; |
| tensor<fp16, [768]> layers_8_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("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, []>(235116224)))]; |
| tensor<fp16, [1, 768, 1, 1]> current_value_17_cast_fp16 = conv(bias = layers_8_self_attn_v_proj_bias_to_fp16, dilations = current_value_17_dilations_0, groups = current_value_17_groups_0, pad = current_value_17_pad_0, pad_type = current_value_17_pad_type_0, strides = current_value_17_strides_0, weight = layers_8_self_attn_v_proj_weight_to_fp16, x = obj_113_cast_fp16)[name = tensor<string, []>("current_value_17_cast_fp16")]; |
| tensor<fp16, [1, 768, 1, 448]> var_1916_cast_fp16 = mul(x = var_63_cast_fp16_8, y = var_159_cast_fp16)[name = tensor<string, []>("op_1916_cast_fp16")]; |
| tensor<fp16, [1, 768, 1, 448]> var_1917_cast_fp16 = mul(x = current_key_17_cast_fp16, y = var_157_cast_fp16)[name = tensor<string, []>("op_1917_cast_fp16")]; |
| tensor<fp16, [1, 768, 1, 448]> key_33_cast_fp16 = add(x = var_1916_cast_fp16, y = var_1917_cast_fp16)[name = tensor<string, []>("key_33_cast_fp16")]; |
| tensor<fp16, [1, 768, 1, 448]> var_1920_cast_fp16 = mul(x = var_78_cast_fp16_8, y = var_159_cast_fp16)[name = tensor<string, []>("op_1920_cast_fp16")]; |
| tensor<fp16, [1, 768, 1, 448]> var_1921_cast_fp16 = mul(x = current_value_17_cast_fp16, y = var_157_cast_fp16)[name = tensor<string, []>("op_1921_cast_fp16")]; |
| tensor<fp16, [1, 768, 1, 448]> value_33_cast_fp16 = add(x = var_1920_cast_fp16, y = var_1921_cast_fp16)[name = tensor<string, []>("value_33_cast_fp16")]; |
| tensor<int32, [4]> var_1925 = const()[name = tensor<string, []>("op_1925"), val = tensor<int32, [4]>([1, 12, 64, 1])]; |
| tensor<fp16, [1, 12, 64, 1]> mh_q_33_cast_fp16 = reshape(shape = var_1925, x = query_33_cast_fp16)[name = tensor<string, []>("mh_q_33_cast_fp16")]; |
| tensor<fp16, []> var_1927_to_fp16 = const()[name = tensor<string, []>("op_1927_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; |
| tensor<fp16, [1, 12, 64, 1]> var_1928_cast_fp16 = mul(x = mh_q_33_cast_fp16, y = var_1927_to_fp16)[name = tensor<string, []>("op_1928_cast_fp16")]; |
| tensor<int32, [4]> var_1931 = const()[name = tensor<string, []>("op_1931"), val = tensor<int32, [4]>([1, 12, 64, 448])]; |
| tensor<fp16, [1, 12, 64, 448]> var_1932_cast_fp16 = reshape(shape = var_1931, x = key_33_cast_fp16)[name = tensor<string, []>("op_1932_cast_fp16")]; |
| tensor<bool, []> mh_w_49_transpose_x_0 = const()[name = tensor<string, []>("mh_w_49_transpose_x_0"), val = tensor<bool, []>(true)]; |
| tensor<bool, []> mh_w_49_transpose_y_0 = const()[name = tensor<string, []>("mh_w_49_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [1, 12, 1, 448]> mh_w_49_cast_fp16 = matmul(transpose_x = mh_w_49_transpose_x_0, transpose_y = mh_w_49_transpose_y_0, x = var_1928_cast_fp16, y = var_1932_cast_fp16)[name = tensor<string, []>("mh_w_49_cast_fp16")]; |
| tensor<fp16, [1, 12, 1, 448]> mh_w_51_cast_fp16 = add(x = mh_w_49_cast_fp16, y = var_181_cast_fp16)[name = tensor<string, []>("mh_w_51_cast_fp16")]; |
| tensor<fp16, [1, 12, 1, 448]> var_1940_cast_fp16 = softmax(axis = var_1852, x = mh_w_51_cast_fp16)[name = tensor<string, []>("op_1940_cast_fp16")]; |
| tensor<int32, [4]> var_1941 = const()[name = tensor<string, []>("op_1941"), val = tensor<int32, [4]>([1, 12, 64, 448])]; |
| tensor<fp16, [1, 12, 64, 448]> var_1942_cast_fp16 = reshape(shape = var_1941, x = value_33_cast_fp16)[name = tensor<string, []>("op_1942_cast_fp16")]; |
| tensor<bool, []> attn_33_transpose_x_0 = const()[name = tensor<string, []>("attn_33_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_33_transpose_y_0 = const()[name = tensor<string, []>("attn_33_transpose_y_0"), val = tensor<bool, []>(true)]; |
| tensor<fp16, [1, 12, 64, 1]> attn_33_cast_fp16 = matmul(transpose_x = attn_33_transpose_x_0, transpose_y = attn_33_transpose_y_0, x = var_1942_cast_fp16, y = var_1940_cast_fp16)[name = tensor<string, []>("attn_33_cast_fp16")]; |
| tensor<int32, [4]> var_1945 = const()[name = tensor<string, []>("op_1945"), val = tensor<int32, [4]>([1, 768, 1, 1])]; |
| tensor<fp16, [1, 768, 1, 1]> input_81_cast_fp16 = reshape(shape = var_1945, x = attn_33_cast_fp16)[name = tensor<string, []>("input_81_cast_fp16")]; |
| tensor<string, []> obj_119_pad_type_0 = const()[name = tensor<string, []>("obj_119_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> obj_119_strides_0 = const()[name = tensor<string, []>("obj_119_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> obj_119_pad_0 = const()[name = tensor<string, []>("obj_119_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> obj_119_dilations_0 = const()[name = tensor<string, []>("obj_119_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> obj_119_groups_0 = const()[name = tensor<string, []>("obj_119_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 768, 1, 1]> layers_8_self_attn_o_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_8_self_attn_o_proj_weight_to_fp16"), val = tensor<fp16, [768, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(235117824)))]; |
| tensor<fp16, [768]> layers_8_self_attn_o_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_8_self_attn_o_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(236297536)))]; |
| tensor<fp16, [1, 768, 1, 1]> obj_119_cast_fp16 = conv(bias = layers_8_self_attn_o_proj_bias_to_fp16, dilations = obj_119_dilations_0, groups = obj_119_groups_0, pad = obj_119_pad_0, pad_type = obj_119_pad_type_0, strides = obj_119_strides_0, weight = layers_8_self_attn_o_proj_weight_to_fp16, x = input_81_cast_fp16)[name = tensor<string, []>("obj_119_cast_fp16")]; |
| tensor<fp16, [1, 768, 1, 1]> inputs_51_cast_fp16 = add(x = inputs_49_cast_fp16, y = obj_119_cast_fp16)[name = tensor<string, []>("inputs_51_cast_fp16")]; |
| tensor<int32, [1]> out_51_axes_0 = const()[name = tensor<string, []>("out_51_axes_0"), val = tensor<int32, [1]>([1])]; |
| tensor<fp16, []> var_1967_to_fp16 = const()[name = tensor<string, []>("op_1967_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, 768, 1, 1]> out_51_cast_fp16 = layer_norm(axes = out_51_axes_0, epsilon = var_1967_to_fp16, x = inputs_51_cast_fp16)[name = tensor<string, []>("out_51_cast_fp16")]; |
| tensor<fp16, [768]> obj_121_gamma_0_to_fp16 = const()[name = tensor<string, []>("obj_121_gamma_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(236299136)))]; |
| tensor<fp16, [768]> obj_121_beta_0_to_fp16 = const()[name = tensor<string, []>("obj_121_beta_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(236300736)))]; |
| tensor<fp16, []> obj_121_epsilon_0_to_fp16 = const()[name = tensor<string, []>("obj_121_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, 768, 1, 1]> obj_121_cast_fp16 = batch_norm(beta = obj_121_beta_0_to_fp16, epsilon = obj_121_epsilon_0_to_fp16, gamma = obj_121_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_51_cast_fp16)[name = tensor<string, []>("obj_121_cast_fp16")]; |
| tensor<string, []> query_35_pad_type_0 = const()[name = tensor<string, []>("query_35_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> query_35_strides_0 = const()[name = tensor<string, []>("query_35_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> query_35_pad_0 = const()[name = tensor<string, []>("query_35_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> query_35_dilations_0 = const()[name = tensor<string, []>("query_35_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> query_35_groups_0 = const()[name = tensor<string, []>("query_35_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 768, 1, 1]> layers_8_encoder_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_8_encoder_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [768, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(236302336)))]; |
| tensor<fp16, [768]> layers_8_encoder_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_8_encoder_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(237482048)))]; |
| tensor<fp16, [1, 768, 1, 1]> query_35_cast_fp16 = conv(bias = layers_8_encoder_attn_q_proj_bias_to_fp16, dilations = query_35_dilations_0, groups = query_35_groups_0, pad = query_35_pad_0, pad_type = query_35_pad_type_0, strides = query_35_strides_0, weight = layers_8_encoder_attn_q_proj_weight_to_fp16, x = obj_121_cast_fp16)[name = tensor<string, []>("query_35_cast_fp16")]; |
| tensor<string, []> key_35_pad_type_0 = const()[name = tensor<string, []>("key_35_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> key_35_strides_0 = const()[name = tensor<string, []>("key_35_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> key_35_pad_0 = const()[name = tensor<string, []>("key_35_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> key_35_dilations_0 = const()[name = tensor<string, []>("key_35_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> key_35_groups_0 = const()[name = tensor<string, []>("key_35_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 768, 1, 1]> layers_8_encoder_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_8_encoder_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [768, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(237483648)))]; |
| tensor<fp16, [1, 768, 1, 1500]> key_35_cast_fp16 = conv(dilations = key_35_dilations_0, groups = key_35_groups_0, pad = key_35_pad_0, pad_type = key_35_pad_type_0, strides = key_35_strides_0, weight = layers_8_encoder_attn_k_proj_weight_to_fp16, x = encoder_output_embeds)[name = tensor<string, []>("key_35_cast_fp16")]; |
| tensor<string, []> value_35_pad_type_0 = const()[name = tensor<string, []>("value_35_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> value_35_strides_0 = const()[name = tensor<string, []>("value_35_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> value_35_pad_0 = const()[name = tensor<string, []>("value_35_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> value_35_dilations_0 = const()[name = tensor<string, []>("value_35_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> value_35_groups_0 = const()[name = tensor<string, []>("value_35_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 768, 1, 1]> layers_8_encoder_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_8_encoder_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [768, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(238663360)))]; |
| tensor<fp16, [768]> layers_8_encoder_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_8_encoder_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(239843072)))]; |
| tensor<fp16, [1, 768, 1, 1500]> value_35_cast_fp16 = conv(bias = layers_8_encoder_attn_v_proj_bias_to_fp16, dilations = value_35_dilations_0, groups = value_35_groups_0, pad = value_35_pad_0, pad_type = value_35_pad_type_0, strides = value_35_strides_0, weight = layers_8_encoder_attn_v_proj_weight_to_fp16, x = encoder_output_embeds)[name = tensor<string, []>("value_35_cast_fp16")]; |
| tensor<int32, [4]> var_2003 = const()[name = tensor<string, []>("op_2003"), val = tensor<int32, [4]>([1, 12, 64, 1])]; |
| tensor<fp16, [1, 12, 64, 1]> mh_q_35_cast_fp16 = reshape(shape = var_2003, x = query_35_cast_fp16)[name = tensor<string, []>("mh_q_35_cast_fp16")]; |
| tensor<fp16, []> var_2005_to_fp16 = const()[name = tensor<string, []>("op_2005_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; |
| tensor<fp16, [1, 12, 64, 1]> var_2006_cast_fp16 = mul(x = mh_q_35_cast_fp16, y = var_2005_to_fp16)[name = tensor<string, []>("op_2006_cast_fp16")]; |
| tensor<int32, [4]> var_2009 = const()[name = tensor<string, []>("op_2009"), val = tensor<int32, [4]>([1, 12, 64, 1500])]; |
| tensor<fp16, [1, 12, 64, 1500]> var_2010_cast_fp16 = reshape(shape = var_2009, x = key_35_cast_fp16)[name = tensor<string, []>("op_2010_cast_fp16")]; |
| tensor<bool, []> mh_w_53_transpose_x_0 = const()[name = tensor<string, []>("mh_w_53_transpose_x_0"), val = tensor<bool, []>(true)]; |
| tensor<bool, []> mh_w_53_transpose_y_0 = const()[name = tensor<string, []>("mh_w_53_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [1, 12, 1, 1500]> mh_w_53_cast_fp16 = matmul(transpose_x = mh_w_53_transpose_x_0, transpose_y = mh_w_53_transpose_y_0, x = var_2006_cast_fp16, y = var_2010_cast_fp16)[name = tensor<string, []>("mh_w_53_cast_fp16")]; |
| tensor<fp16, [1, 12, 1, 1500]> obj_125_cast_fp16 = softmax(axis = var_1852, x = mh_w_53_cast_fp16)[name = tensor<string, []>("obj_125_cast_fp16")]; |
| tensor<int32, [4]> var_2014 = const()[name = tensor<string, []>("op_2014"), val = tensor<int32, [4]>([1, 12, 64, 1500])]; |
| tensor<fp16, [1, 12, 64, 1500]> var_2015_cast_fp16 = reshape(shape = var_2014, x = value_35_cast_fp16)[name = tensor<string, []>("op_2015_cast_fp16")]; |
| tensor<bool, []> attn_35_transpose_x_0 = const()[name = tensor<string, []>("attn_35_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_35_transpose_y_0 = const()[name = tensor<string, []>("attn_35_transpose_y_0"), val = tensor<bool, []>(true)]; |
| tensor<fp16, [1, 12, 64, 1]> attn_35_cast_fp16 = matmul(transpose_x = attn_35_transpose_x_0, transpose_y = attn_35_transpose_y_0, x = var_2015_cast_fp16, y = obj_125_cast_fp16)[name = tensor<string, []>("attn_35_cast_fp16")]; |
| tensor<int32, [4]> var_2018 = const()[name = tensor<string, []>("op_2018"), val = tensor<int32, [4]>([1, 768, 1, 1])]; |
| tensor<fp16, [1, 768, 1, 1]> input_83_cast_fp16 = reshape(shape = var_2018, x = attn_35_cast_fp16)[name = tensor<string, []>("input_83_cast_fp16")]; |
| tensor<string, []> obj_123_pad_type_0 = const()[name = tensor<string, []>("obj_123_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> obj_123_strides_0 = const()[name = tensor<string, []>("obj_123_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> obj_123_pad_0 = const()[name = tensor<string, []>("obj_123_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> obj_123_dilations_0 = const()[name = tensor<string, []>("obj_123_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> obj_123_groups_0 = const()[name = tensor<string, []>("obj_123_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 768, 1, 1]> layers_8_encoder_attn_o_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_8_encoder_attn_o_proj_weight_to_fp16"), val = tensor<fp16, [768, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(239844672)))]; |
| tensor<fp16, [768]> layers_8_encoder_attn_o_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_8_encoder_attn_o_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(241024384)))]; |
| tensor<fp16, [1, 768, 1, 1]> obj_123_cast_fp16 = conv(bias = layers_8_encoder_attn_o_proj_bias_to_fp16, dilations = obj_123_dilations_0, groups = obj_123_groups_0, pad = obj_123_pad_0, pad_type = obj_123_pad_type_0, strides = obj_123_strides_0, weight = layers_8_encoder_attn_o_proj_weight_to_fp16, x = input_83_cast_fp16)[name = tensor<string, []>("obj_123_cast_fp16")]; |
| tensor<fp16, [1, 768, 1, 1]> inputs_53_cast_fp16 = add(x = inputs_51_cast_fp16, y = obj_123_cast_fp16)[name = tensor<string, []>("inputs_53_cast_fp16")]; |
| tensor<int32, [1]> out_53_axes_0 = const()[name = tensor<string, []>("out_53_axes_0"), val = tensor<int32, [1]>([1])]; |
| tensor<fp16, []> var_2039_to_fp16 = const()[name = tensor<string, []>("op_2039_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, 768, 1, 1]> out_53_cast_fp16 = layer_norm(axes = out_53_axes_0, epsilon = var_2039_to_fp16, x = inputs_53_cast_fp16)[name = tensor<string, []>("out_53_cast_fp16")]; |
| tensor<fp16, [768]> input_85_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_85_gamma_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(241025984)))]; |
| tensor<fp16, [768]> input_85_beta_0_to_fp16 = const()[name = tensor<string, []>("input_85_beta_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(241027584)))]; |
| tensor<fp16, []> input_85_epsilon_0_to_fp16 = const()[name = tensor<string, []>("input_85_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, 768, 1, 1]> input_85_cast_fp16 = batch_norm(beta = input_85_beta_0_to_fp16, epsilon = input_85_epsilon_0_to_fp16, gamma = input_85_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_53_cast_fp16)[name = tensor<string, []>("input_85_cast_fp16")]; |
| tensor<string, []> input_87_pad_type_0 = const()[name = tensor<string, []>("input_87_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> input_87_strides_0 = const()[name = tensor<string, []>("input_87_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> input_87_pad_0 = const()[name = tensor<string, []>("input_87_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> input_87_dilations_0 = const()[name = tensor<string, []>("input_87_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> input_87_groups_0 = const()[name = tensor<string, []>("input_87_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [3072, 768, 1, 1]> layers_8_fc1_weight_to_fp16 = const()[name = tensor<string, []>("layers_8_fc1_weight_to_fp16"), val = tensor<fp16, [3072, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(241029184)))]; |
| tensor<fp16, [3072]> layers_8_fc1_bias_to_fp16 = const()[name = tensor<string, []>("layers_8_fc1_bias_to_fp16"), val = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(245747840)))]; |
| tensor<fp16, [1, 3072, 1, 1]> input_87_cast_fp16 = conv(bias = layers_8_fc1_bias_to_fp16, dilations = input_87_dilations_0, groups = input_87_groups_0, pad = input_87_pad_0, pad_type = input_87_pad_type_0, strides = input_87_strides_0, weight = layers_8_fc1_weight_to_fp16, x = input_85_cast_fp16)[name = tensor<string, []>("input_87_cast_fp16")]; |
| tensor<string, []> input_89_mode_0 = const()[name = tensor<string, []>("input_89_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 3072, 1, 1]> input_89_cast_fp16 = gelu(mode = input_89_mode_0, x = input_87_cast_fp16)[name = tensor<string, []>("input_89_cast_fp16")]; |
| tensor<string, []> hidden_states_19_pad_type_0 = const()[name = tensor<string, []>("hidden_states_19_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> hidden_states_19_strides_0 = const()[name = tensor<string, []>("hidden_states_19_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> hidden_states_19_pad_0 = const()[name = tensor<string, []>("hidden_states_19_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> hidden_states_19_dilations_0 = const()[name = tensor<string, []>("hidden_states_19_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> hidden_states_19_groups_0 = const()[name = tensor<string, []>("hidden_states_19_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 3072, 1, 1]> layers_8_fc2_weight_to_fp16 = const()[name = tensor<string, []>("layers_8_fc2_weight_to_fp16"), val = tensor<fp16, [768, 3072, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(245754048)))]; |
| tensor<fp16, [768]> layers_8_fc2_bias_to_fp16 = const()[name = tensor<string, []>("layers_8_fc2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(250472704)))]; |
| tensor<fp16, [1, 768, 1, 1]> hidden_states_19_cast_fp16 = conv(bias = layers_8_fc2_bias_to_fp16, dilations = hidden_states_19_dilations_0, groups = hidden_states_19_groups_0, pad = hidden_states_19_pad_0, pad_type = hidden_states_19_pad_type_0, strides = hidden_states_19_strides_0, weight = layers_8_fc2_weight_to_fp16, x = input_89_cast_fp16)[name = tensor<string, []>("hidden_states_19_cast_fp16")]; |
| tensor<fp16, [1, 768, 1, 1]> inputs_55_cast_fp16 = add(x = inputs_53_cast_fp16, y = hidden_states_19_cast_fp16)[name = tensor<string, []>("inputs_55_cast_fp16")]; |
| tensor<int32, []> var_2075 = const()[name = tensor<string, []>("op_2075"), val = tensor<int32, []>(3)]; |
| tensor<int32, [1]> out_55_axes_0 = const()[name = tensor<string, []>("out_55_axes_0"), val = tensor<int32, [1]>([1])]; |
| tensor<fp16, []> var_2100_to_fp16 = const()[name = tensor<string, []>("op_2100_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, 768, 1, 1]> out_55_cast_fp16 = layer_norm(axes = out_55_axes_0, epsilon = var_2100_to_fp16, x = inputs_55_cast_fp16)[name = tensor<string, []>("out_55_cast_fp16")]; |
| tensor<fp16, [768]> obj_127_gamma_0_to_fp16 = const()[name = tensor<string, []>("obj_127_gamma_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(250474304)))]; |
| tensor<fp16, [768]> obj_127_beta_0_to_fp16 = const()[name = tensor<string, []>("obj_127_beta_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(250475904)))]; |
| tensor<fp16, []> obj_127_epsilon_0_to_fp16 = const()[name = tensor<string, []>("obj_127_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, 768, 1, 1]> obj_127_cast_fp16 = batch_norm(beta = obj_127_beta_0_to_fp16, epsilon = obj_127_epsilon_0_to_fp16, gamma = obj_127_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_55_cast_fp16)[name = tensor<string, []>("obj_127_cast_fp16")]; |
| tensor<string, []> query_37_pad_type_0 = const()[name = tensor<string, []>("query_37_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> query_37_strides_0 = const()[name = tensor<string, []>("query_37_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> query_37_pad_0 = const()[name = tensor<string, []>("query_37_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> query_37_dilations_0 = const()[name = tensor<string, []>("query_37_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> query_37_groups_0 = const()[name = tensor<string, []>("query_37_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 768, 1, 1]> layers_9_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_9_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [768, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(250477504)))]; |
| tensor<fp16, [768]> layers_9_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("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, []>(251657216)))]; |
| tensor<fp16, [1, 768, 1, 1]> query_37_cast_fp16 = conv(bias = layers_9_self_attn_q_proj_bias_to_fp16, dilations = query_37_dilations_0, groups = query_37_groups_0, pad = query_37_pad_0, pad_type = query_37_pad_type_0, strides = query_37_strides_0, weight = layers_9_self_attn_q_proj_weight_to_fp16, x = obj_127_cast_fp16)[name = tensor<string, []>("query_37_cast_fp16")]; |
| tensor<string, []> current_key_19_pad_type_0 = const()[name = tensor<string, []>("current_key_19_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> current_key_19_strides_0 = const()[name = tensor<string, []>("current_key_19_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> current_key_19_pad_0 = const()[name = tensor<string, []>("current_key_19_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> current_key_19_dilations_0 = const()[name = tensor<string, []>("current_key_19_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> current_key_19_groups_0 = const()[name = tensor<string, []>("current_key_19_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 768, 1, 1]> layers_9_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_9_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [768, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(251658816)))]; |
| tensor<fp16, [1, 768, 1, 1]> current_key_19_cast_fp16 = conv(dilations = current_key_19_dilations_0, groups = current_key_19_groups_0, pad = current_key_19_pad_0, pad_type = current_key_19_pad_type_0, strides = current_key_19_strides_0, weight = layers_9_self_attn_k_proj_weight_to_fp16, x = obj_127_cast_fp16)[name = tensor<string, []>("current_key_19_cast_fp16")]; |
| tensor<string, []> current_value_19_pad_type_0 = const()[name = tensor<string, []>("current_value_19_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> current_value_19_strides_0 = const()[name = tensor<string, []>("current_value_19_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> current_value_19_pad_0 = const()[name = tensor<string, []>("current_value_19_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> current_value_19_dilations_0 = const()[name = tensor<string, []>("current_value_19_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> current_value_19_groups_0 = const()[name = tensor<string, []>("current_value_19_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 768, 1, 1]> layers_9_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_9_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [768, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(252838528)))]; |
| tensor<fp16, [768]> layers_9_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("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, []>(254018240)))]; |
| tensor<fp16, [1, 768, 1, 1]> current_value_19_cast_fp16 = conv(bias = layers_9_self_attn_v_proj_bias_to_fp16, dilations = current_value_19_dilations_0, groups = current_value_19_groups_0, pad = current_value_19_pad_0, pad_type = current_value_19_pad_type_0, strides = current_value_19_strides_0, weight = layers_9_self_attn_v_proj_weight_to_fp16, x = obj_127_cast_fp16)[name = tensor<string, []>("current_value_19_cast_fp16")]; |
| tensor<fp16, [1, 768, 1, 448]> var_2139_cast_fp16 = mul(x = var_63_cast_fp16_9, y = var_159_cast_fp16)[name = tensor<string, []>("op_2139_cast_fp16")]; |
| tensor<fp16, [1, 768, 1, 448]> var_2140_cast_fp16 = mul(x = current_key_19_cast_fp16, y = var_157_cast_fp16)[name = tensor<string, []>("op_2140_cast_fp16")]; |
| tensor<fp16, [1, 768, 1, 448]> key_37_cast_fp16 = add(x = var_2139_cast_fp16, y = var_2140_cast_fp16)[name = tensor<string, []>("key_37_cast_fp16")]; |
| tensor<fp16, [1, 768, 1, 448]> var_2143_cast_fp16 = mul(x = var_78_cast_fp16_9, y = var_159_cast_fp16)[name = tensor<string, []>("op_2143_cast_fp16")]; |
| tensor<fp16, [1, 768, 1, 448]> var_2144_cast_fp16 = mul(x = current_value_19_cast_fp16, y = var_157_cast_fp16)[name = tensor<string, []>("op_2144_cast_fp16")]; |
| tensor<fp16, [1, 768, 1, 448]> value_37_cast_fp16 = add(x = var_2143_cast_fp16, y = var_2144_cast_fp16)[name = tensor<string, []>("value_37_cast_fp16")]; |
| tensor<int32, [4]> var_2148 = const()[name = tensor<string, []>("op_2148"), val = tensor<int32, [4]>([1, 12, 64, 1])]; |
| tensor<fp16, [1, 12, 64, 1]> mh_q_37_cast_fp16 = reshape(shape = var_2148, x = query_37_cast_fp16)[name = tensor<string, []>("mh_q_37_cast_fp16")]; |
| tensor<fp16, []> var_2150_to_fp16 = const()[name = tensor<string, []>("op_2150_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; |
| tensor<fp16, [1, 12, 64, 1]> var_2151_cast_fp16 = mul(x = mh_q_37_cast_fp16, y = var_2150_to_fp16)[name = tensor<string, []>("op_2151_cast_fp16")]; |
| tensor<int32, [4]> var_2154 = const()[name = tensor<string, []>("op_2154"), val = tensor<int32, [4]>([1, 12, 64, 448])]; |
| tensor<fp16, [1, 12, 64, 448]> var_2155_cast_fp16 = reshape(shape = var_2154, x = key_37_cast_fp16)[name = tensor<string, []>("op_2155_cast_fp16")]; |
| tensor<bool, []> mh_w_55_transpose_x_0 = const()[name = tensor<string, []>("mh_w_55_transpose_x_0"), val = tensor<bool, []>(true)]; |
| tensor<bool, []> mh_w_55_transpose_y_0 = const()[name = tensor<string, []>("mh_w_55_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [1, 12, 1, 448]> mh_w_55_cast_fp16 = matmul(transpose_x = mh_w_55_transpose_x_0, transpose_y = mh_w_55_transpose_y_0, x = var_2151_cast_fp16, y = var_2155_cast_fp16)[name = tensor<string, []>("mh_w_55_cast_fp16")]; |
| tensor<fp16, [1, 12, 1, 448]> mh_w_57_cast_fp16 = add(x = mh_w_55_cast_fp16, y = var_181_cast_fp16)[name = tensor<string, []>("mh_w_57_cast_fp16")]; |
| tensor<fp16, [1, 12, 1, 448]> var_2163_cast_fp16 = softmax(axis = var_2075, x = mh_w_57_cast_fp16)[name = tensor<string, []>("op_2163_cast_fp16")]; |
| tensor<int32, [4]> var_2164 = const()[name = tensor<string, []>("op_2164"), val = tensor<int32, [4]>([1, 12, 64, 448])]; |
| tensor<fp16, [1, 12, 64, 448]> var_2165_cast_fp16 = reshape(shape = var_2164, x = value_37_cast_fp16)[name = tensor<string, []>("op_2165_cast_fp16")]; |
| tensor<bool, []> attn_37_transpose_x_0 = const()[name = tensor<string, []>("attn_37_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_37_transpose_y_0 = const()[name = tensor<string, []>("attn_37_transpose_y_0"), val = tensor<bool, []>(true)]; |
| tensor<fp16, [1, 12, 64, 1]> attn_37_cast_fp16 = matmul(transpose_x = attn_37_transpose_x_0, transpose_y = attn_37_transpose_y_0, x = var_2165_cast_fp16, y = var_2163_cast_fp16)[name = tensor<string, []>("attn_37_cast_fp16")]; |
| tensor<int32, [4]> var_2168 = const()[name = tensor<string, []>("op_2168"), val = tensor<int32, [4]>([1, 768, 1, 1])]; |
| tensor<fp16, [1, 768, 1, 1]> input_91_cast_fp16 = reshape(shape = var_2168, x = attn_37_cast_fp16)[name = tensor<string, []>("input_91_cast_fp16")]; |
| tensor<string, []> obj_133_pad_type_0 = const()[name = tensor<string, []>("obj_133_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> obj_133_strides_0 = const()[name = tensor<string, []>("obj_133_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> obj_133_pad_0 = const()[name = tensor<string, []>("obj_133_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> obj_133_dilations_0 = const()[name = tensor<string, []>("obj_133_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> obj_133_groups_0 = const()[name = tensor<string, []>("obj_133_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 768, 1, 1]> layers_9_self_attn_o_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_9_self_attn_o_proj_weight_to_fp16"), val = tensor<fp16, [768, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(254019840)))]; |
| tensor<fp16, [768]> layers_9_self_attn_o_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_9_self_attn_o_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(255199552)))]; |
| tensor<fp16, [1, 768, 1, 1]> obj_133_cast_fp16 = conv(bias = layers_9_self_attn_o_proj_bias_to_fp16, dilations = obj_133_dilations_0, groups = obj_133_groups_0, pad = obj_133_pad_0, pad_type = obj_133_pad_type_0, strides = obj_133_strides_0, weight = layers_9_self_attn_o_proj_weight_to_fp16, x = input_91_cast_fp16)[name = tensor<string, []>("obj_133_cast_fp16")]; |
| tensor<fp16, [1, 768, 1, 1]> inputs_57_cast_fp16 = add(x = inputs_55_cast_fp16, y = obj_133_cast_fp16)[name = tensor<string, []>("inputs_57_cast_fp16")]; |
| tensor<int32, [1]> out_57_axes_0 = const()[name = tensor<string, []>("out_57_axes_0"), val = tensor<int32, [1]>([1])]; |
| tensor<fp16, []> var_2190_to_fp16 = const()[name = tensor<string, []>("op_2190_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, 768, 1, 1]> out_57_cast_fp16 = layer_norm(axes = out_57_axes_0, epsilon = var_2190_to_fp16, x = inputs_57_cast_fp16)[name = tensor<string, []>("out_57_cast_fp16")]; |
| tensor<fp16, [768]> obj_135_gamma_0_to_fp16 = const()[name = tensor<string, []>("obj_135_gamma_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(255201152)))]; |
| tensor<fp16, [768]> obj_135_beta_0_to_fp16 = const()[name = tensor<string, []>("obj_135_beta_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(255202752)))]; |
| tensor<fp16, []> obj_135_epsilon_0_to_fp16 = const()[name = tensor<string, []>("obj_135_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, 768, 1, 1]> obj_135_cast_fp16 = batch_norm(beta = obj_135_beta_0_to_fp16, epsilon = obj_135_epsilon_0_to_fp16, gamma = obj_135_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_57_cast_fp16)[name = tensor<string, []>("obj_135_cast_fp16")]; |
| tensor<string, []> query_39_pad_type_0 = const()[name = tensor<string, []>("query_39_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> query_39_strides_0 = const()[name = tensor<string, []>("query_39_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> query_39_pad_0 = const()[name = tensor<string, []>("query_39_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> query_39_dilations_0 = const()[name = tensor<string, []>("query_39_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> query_39_groups_0 = const()[name = tensor<string, []>("query_39_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 768, 1, 1]> layers_9_encoder_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_9_encoder_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [768, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(255204352)))]; |
| tensor<fp16, [768]> layers_9_encoder_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_9_encoder_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(256384064)))]; |
| tensor<fp16, [1, 768, 1, 1]> query_39_cast_fp16 = conv(bias = layers_9_encoder_attn_q_proj_bias_to_fp16, dilations = query_39_dilations_0, groups = query_39_groups_0, pad = query_39_pad_0, pad_type = query_39_pad_type_0, strides = query_39_strides_0, weight = layers_9_encoder_attn_q_proj_weight_to_fp16, x = obj_135_cast_fp16)[name = tensor<string, []>("query_39_cast_fp16")]; |
| tensor<string, []> key_39_pad_type_0 = const()[name = tensor<string, []>("key_39_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> key_39_strides_0 = const()[name = tensor<string, []>("key_39_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> key_39_pad_0 = const()[name = tensor<string, []>("key_39_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> key_39_dilations_0 = const()[name = tensor<string, []>("key_39_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> key_39_groups_0 = const()[name = tensor<string, []>("key_39_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 768, 1, 1]> layers_9_encoder_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_9_encoder_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [768, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(256385664)))]; |
| tensor<fp16, [1, 768, 1, 1500]> key_39_cast_fp16 = conv(dilations = key_39_dilations_0, groups = key_39_groups_0, pad = key_39_pad_0, pad_type = key_39_pad_type_0, strides = key_39_strides_0, weight = layers_9_encoder_attn_k_proj_weight_to_fp16, x = encoder_output_embeds)[name = tensor<string, []>("key_39_cast_fp16")]; |
| tensor<string, []> value_39_pad_type_0 = const()[name = tensor<string, []>("value_39_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> value_39_strides_0 = const()[name = tensor<string, []>("value_39_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> value_39_pad_0 = const()[name = tensor<string, []>("value_39_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> value_39_dilations_0 = const()[name = tensor<string, []>("value_39_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> value_39_groups_0 = const()[name = tensor<string, []>("value_39_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 768, 1, 1]> layers_9_encoder_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_9_encoder_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [768, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(257565376)))]; |
| tensor<fp16, [768]> layers_9_encoder_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_9_encoder_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(258745088)))]; |
| tensor<fp16, [1, 768, 1, 1500]> value_39_cast_fp16 = conv(bias = layers_9_encoder_attn_v_proj_bias_to_fp16, dilations = value_39_dilations_0, groups = value_39_groups_0, pad = value_39_pad_0, pad_type = value_39_pad_type_0, strides = value_39_strides_0, weight = layers_9_encoder_attn_v_proj_weight_to_fp16, x = encoder_output_embeds)[name = tensor<string, []>("value_39_cast_fp16")]; |
| tensor<int32, [4]> var_2226 = const()[name = tensor<string, []>("op_2226"), val = tensor<int32, [4]>([1, 12, 64, 1])]; |
| tensor<fp16, [1, 12, 64, 1]> mh_q_39_cast_fp16 = reshape(shape = var_2226, x = query_39_cast_fp16)[name = tensor<string, []>("mh_q_39_cast_fp16")]; |
| tensor<fp16, []> var_2228_to_fp16 = const()[name = tensor<string, []>("op_2228_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; |
| tensor<fp16, [1, 12, 64, 1]> var_2229_cast_fp16 = mul(x = mh_q_39_cast_fp16, y = var_2228_to_fp16)[name = tensor<string, []>("op_2229_cast_fp16")]; |
| tensor<int32, [4]> var_2232 = const()[name = tensor<string, []>("op_2232"), val = tensor<int32, [4]>([1, 12, 64, 1500])]; |
| tensor<fp16, [1, 12, 64, 1500]> var_2233_cast_fp16 = reshape(shape = var_2232, x = key_39_cast_fp16)[name = tensor<string, []>("op_2233_cast_fp16")]; |
| tensor<bool, []> mh_w_59_transpose_x_0 = const()[name = tensor<string, []>("mh_w_59_transpose_x_0"), val = tensor<bool, []>(true)]; |
| tensor<bool, []> mh_w_59_transpose_y_0 = const()[name = tensor<string, []>("mh_w_59_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [1, 12, 1, 1500]> mh_w_59_cast_fp16 = matmul(transpose_x = mh_w_59_transpose_x_0, transpose_y = mh_w_59_transpose_y_0, x = var_2229_cast_fp16, y = var_2233_cast_fp16)[name = tensor<string, []>("mh_w_59_cast_fp16")]; |
| tensor<fp16, [1, 12, 1, 1500]> obj_139_cast_fp16 = softmax(axis = var_2075, x = mh_w_59_cast_fp16)[name = tensor<string, []>("obj_139_cast_fp16")]; |
| tensor<int32, [4]> var_2237 = const()[name = tensor<string, []>("op_2237"), val = tensor<int32, [4]>([1, 12, 64, 1500])]; |
| tensor<fp16, [1, 12, 64, 1500]> var_2238_cast_fp16 = reshape(shape = var_2237, x = value_39_cast_fp16)[name = tensor<string, []>("op_2238_cast_fp16")]; |
| tensor<bool, []> attn_39_transpose_x_0 = const()[name = tensor<string, []>("attn_39_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_39_transpose_y_0 = const()[name = tensor<string, []>("attn_39_transpose_y_0"), val = tensor<bool, []>(true)]; |
| tensor<fp16, [1, 12, 64, 1]> attn_39_cast_fp16 = matmul(transpose_x = attn_39_transpose_x_0, transpose_y = attn_39_transpose_y_0, x = var_2238_cast_fp16, y = obj_139_cast_fp16)[name = tensor<string, []>("attn_39_cast_fp16")]; |
| tensor<int32, [4]> var_2241 = const()[name = tensor<string, []>("op_2241"), val = tensor<int32, [4]>([1, 768, 1, 1])]; |
| tensor<fp16, [1, 768, 1, 1]> input_93_cast_fp16 = reshape(shape = var_2241, x = attn_39_cast_fp16)[name = tensor<string, []>("input_93_cast_fp16")]; |
| tensor<string, []> obj_137_pad_type_0 = const()[name = tensor<string, []>("obj_137_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> obj_137_strides_0 = const()[name = tensor<string, []>("obj_137_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> obj_137_pad_0 = const()[name = tensor<string, []>("obj_137_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> obj_137_dilations_0 = const()[name = tensor<string, []>("obj_137_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> obj_137_groups_0 = const()[name = tensor<string, []>("obj_137_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 768, 1, 1]> layers_9_encoder_attn_o_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_9_encoder_attn_o_proj_weight_to_fp16"), val = tensor<fp16, [768, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(258746688)))]; |
| tensor<fp16, [768]> layers_9_encoder_attn_o_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_9_encoder_attn_o_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(259926400)))]; |
| tensor<fp16, [1, 768, 1, 1]> obj_137_cast_fp16 = conv(bias = layers_9_encoder_attn_o_proj_bias_to_fp16, dilations = obj_137_dilations_0, groups = obj_137_groups_0, pad = obj_137_pad_0, pad_type = obj_137_pad_type_0, strides = obj_137_strides_0, weight = layers_9_encoder_attn_o_proj_weight_to_fp16, x = input_93_cast_fp16)[name = tensor<string, []>("obj_137_cast_fp16")]; |
| tensor<fp16, [1, 768, 1, 1]> inputs_59_cast_fp16 = add(x = inputs_57_cast_fp16, y = obj_137_cast_fp16)[name = tensor<string, []>("inputs_59_cast_fp16")]; |
| tensor<int32, [1]> out_59_axes_0 = const()[name = tensor<string, []>("out_59_axes_0"), val = tensor<int32, [1]>([1])]; |
| tensor<fp16, []> var_2262_to_fp16 = const()[name = tensor<string, []>("op_2262_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, 768, 1, 1]> out_59_cast_fp16 = layer_norm(axes = out_59_axes_0, epsilon = var_2262_to_fp16, x = inputs_59_cast_fp16)[name = tensor<string, []>("out_59_cast_fp16")]; |
| tensor<fp16, [768]> input_95_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_95_gamma_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(259928000)))]; |
| tensor<fp16, [768]> input_95_beta_0_to_fp16 = const()[name = tensor<string, []>("input_95_beta_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(259929600)))]; |
| tensor<fp16, []> input_95_epsilon_0_to_fp16 = const()[name = tensor<string, []>("input_95_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, 768, 1, 1]> input_95_cast_fp16 = batch_norm(beta = input_95_beta_0_to_fp16, epsilon = input_95_epsilon_0_to_fp16, gamma = input_95_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_59_cast_fp16)[name = tensor<string, []>("input_95_cast_fp16")]; |
| tensor<string, []> input_97_pad_type_0 = const()[name = tensor<string, []>("input_97_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> input_97_strides_0 = const()[name = tensor<string, []>("input_97_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> input_97_pad_0 = const()[name = tensor<string, []>("input_97_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> input_97_dilations_0 = const()[name = tensor<string, []>("input_97_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> input_97_groups_0 = const()[name = tensor<string, []>("input_97_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [3072, 768, 1, 1]> layers_9_fc1_weight_to_fp16 = const()[name = tensor<string, []>("layers_9_fc1_weight_to_fp16"), val = tensor<fp16, [3072, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(259931200)))]; |
| tensor<fp16, [3072]> layers_9_fc1_bias_to_fp16 = const()[name = tensor<string, []>("layers_9_fc1_bias_to_fp16"), val = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(264649856)))]; |
| tensor<fp16, [1, 3072, 1, 1]> input_97_cast_fp16 = conv(bias = layers_9_fc1_bias_to_fp16, dilations = input_97_dilations_0, groups = input_97_groups_0, pad = input_97_pad_0, pad_type = input_97_pad_type_0, strides = input_97_strides_0, weight = layers_9_fc1_weight_to_fp16, x = input_95_cast_fp16)[name = tensor<string, []>("input_97_cast_fp16")]; |
| tensor<string, []> input_99_mode_0 = const()[name = tensor<string, []>("input_99_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 3072, 1, 1]> input_99_cast_fp16 = gelu(mode = input_99_mode_0, x = input_97_cast_fp16)[name = tensor<string, []>("input_99_cast_fp16")]; |
| tensor<string, []> hidden_states_21_pad_type_0 = const()[name = tensor<string, []>("hidden_states_21_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> hidden_states_21_strides_0 = const()[name = tensor<string, []>("hidden_states_21_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> hidden_states_21_pad_0 = const()[name = tensor<string, []>("hidden_states_21_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> hidden_states_21_dilations_0 = const()[name = tensor<string, []>("hidden_states_21_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> hidden_states_21_groups_0 = const()[name = tensor<string, []>("hidden_states_21_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 3072, 1, 1]> layers_9_fc2_weight_to_fp16 = const()[name = tensor<string, []>("layers_9_fc2_weight_to_fp16"), val = tensor<fp16, [768, 3072, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(264656064)))]; |
| tensor<fp16, [768]> layers_9_fc2_bias_to_fp16 = const()[name = tensor<string, []>("layers_9_fc2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(269374720)))]; |
| tensor<fp16, [1, 768, 1, 1]> hidden_states_21_cast_fp16 = conv(bias = layers_9_fc2_bias_to_fp16, dilations = hidden_states_21_dilations_0, groups = hidden_states_21_groups_0, pad = hidden_states_21_pad_0, pad_type = hidden_states_21_pad_type_0, strides = hidden_states_21_strides_0, weight = layers_9_fc2_weight_to_fp16, x = input_99_cast_fp16)[name = tensor<string, []>("hidden_states_21_cast_fp16")]; |
| tensor<fp16, [1, 768, 1, 1]> inputs_61_cast_fp16 = add(x = inputs_59_cast_fp16, y = hidden_states_21_cast_fp16)[name = tensor<string, []>("inputs_61_cast_fp16")]; |
| tensor<int32, []> var_2298 = const()[name = tensor<string, []>("op_2298"), val = tensor<int32, []>(3)]; |
| tensor<int32, [1]> out_61_axes_0 = const()[name = tensor<string, []>("out_61_axes_0"), val = tensor<int32, [1]>([1])]; |
| tensor<fp16, []> var_2323_to_fp16 = const()[name = tensor<string, []>("op_2323_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, 768, 1, 1]> out_61_cast_fp16 = layer_norm(axes = out_61_axes_0, epsilon = var_2323_to_fp16, x = inputs_61_cast_fp16)[name = tensor<string, []>("out_61_cast_fp16")]; |
| tensor<fp16, [768]> obj_141_gamma_0_to_fp16 = const()[name = tensor<string, []>("obj_141_gamma_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(269376320)))]; |
| tensor<fp16, [768]> obj_141_beta_0_to_fp16 = const()[name = tensor<string, []>("obj_141_beta_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(269377920)))]; |
| tensor<fp16, []> obj_141_epsilon_0_to_fp16 = const()[name = tensor<string, []>("obj_141_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, 768, 1, 1]> obj_141_cast_fp16 = batch_norm(beta = obj_141_beta_0_to_fp16, epsilon = obj_141_epsilon_0_to_fp16, gamma = obj_141_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_61_cast_fp16)[name = tensor<string, []>("obj_141_cast_fp16")]; |
| tensor<string, []> query_41_pad_type_0 = const()[name = tensor<string, []>("query_41_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> query_41_strides_0 = const()[name = tensor<string, []>("query_41_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> query_41_pad_0 = const()[name = tensor<string, []>("query_41_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> query_41_dilations_0 = const()[name = tensor<string, []>("query_41_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> query_41_groups_0 = const()[name = tensor<string, []>("query_41_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 768, 1, 1]> layers_10_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_10_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [768, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(269379520)))]; |
| tensor<fp16, [768]> layers_10_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("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, []>(270559232)))]; |
| tensor<fp16, [1, 768, 1, 1]> query_41_cast_fp16 = conv(bias = layers_10_self_attn_q_proj_bias_to_fp16, dilations = query_41_dilations_0, groups = query_41_groups_0, pad = query_41_pad_0, pad_type = query_41_pad_type_0, strides = query_41_strides_0, weight = layers_10_self_attn_q_proj_weight_to_fp16, x = obj_141_cast_fp16)[name = tensor<string, []>("query_41_cast_fp16")]; |
| tensor<string, []> current_key_21_pad_type_0 = const()[name = tensor<string, []>("current_key_21_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> current_key_21_strides_0 = const()[name = tensor<string, []>("current_key_21_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> current_key_21_pad_0 = const()[name = tensor<string, []>("current_key_21_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> current_key_21_dilations_0 = const()[name = tensor<string, []>("current_key_21_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> current_key_21_groups_0 = const()[name = tensor<string, []>("current_key_21_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 768, 1, 1]> layers_10_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_10_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [768, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(270560832)))]; |
| tensor<fp16, [1, 768, 1, 1]> current_key_21_cast_fp16 = conv(dilations = current_key_21_dilations_0, groups = current_key_21_groups_0, pad = current_key_21_pad_0, pad_type = current_key_21_pad_type_0, strides = current_key_21_strides_0, weight = layers_10_self_attn_k_proj_weight_to_fp16, x = obj_141_cast_fp16)[name = tensor<string, []>("current_key_21_cast_fp16")]; |
| tensor<string, []> current_value_21_pad_type_0 = const()[name = tensor<string, []>("current_value_21_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> current_value_21_strides_0 = const()[name = tensor<string, []>("current_value_21_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> current_value_21_pad_0 = const()[name = tensor<string, []>("current_value_21_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> current_value_21_dilations_0 = const()[name = tensor<string, []>("current_value_21_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> current_value_21_groups_0 = const()[name = tensor<string, []>("current_value_21_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 768, 1, 1]> layers_10_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_10_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [768, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(271740544)))]; |
| tensor<fp16, [768]> layers_10_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("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, []>(272920256)))]; |
| tensor<fp16, [1, 768, 1, 1]> current_value_21_cast_fp16 = conv(bias = layers_10_self_attn_v_proj_bias_to_fp16, dilations = current_value_21_dilations_0, groups = current_value_21_groups_0, pad = current_value_21_pad_0, pad_type = current_value_21_pad_type_0, strides = current_value_21_strides_0, weight = layers_10_self_attn_v_proj_weight_to_fp16, x = obj_141_cast_fp16)[name = tensor<string, []>("current_value_21_cast_fp16")]; |
| tensor<fp16, [1, 768, 1, 448]> var_2362_cast_fp16 = mul(x = var_63_cast_fp16_10, y = var_159_cast_fp16)[name = tensor<string, []>("op_2362_cast_fp16")]; |
| tensor<fp16, [1, 768, 1, 448]> var_2363_cast_fp16 = mul(x = current_key_21_cast_fp16, y = var_157_cast_fp16)[name = tensor<string, []>("op_2363_cast_fp16")]; |
| tensor<fp16, [1, 768, 1, 448]> key_41_cast_fp16 = add(x = var_2362_cast_fp16, y = var_2363_cast_fp16)[name = tensor<string, []>("key_41_cast_fp16")]; |
| tensor<fp16, [1, 768, 1, 448]> var_2366_cast_fp16 = mul(x = var_78_cast_fp16_10, y = var_159_cast_fp16)[name = tensor<string, []>("op_2366_cast_fp16")]; |
| tensor<fp16, [1, 768, 1, 448]> var_2367_cast_fp16 = mul(x = current_value_21_cast_fp16, y = var_157_cast_fp16)[name = tensor<string, []>("op_2367_cast_fp16")]; |
| tensor<fp16, [1, 768, 1, 448]> value_41_cast_fp16 = add(x = var_2366_cast_fp16, y = var_2367_cast_fp16)[name = tensor<string, []>("value_41_cast_fp16")]; |
| tensor<int32, [4]> var_2371 = const()[name = tensor<string, []>("op_2371"), val = tensor<int32, [4]>([1, 12, 64, 1])]; |
| tensor<fp16, [1, 12, 64, 1]> mh_q_41_cast_fp16 = reshape(shape = var_2371, x = query_41_cast_fp16)[name = tensor<string, []>("mh_q_41_cast_fp16")]; |
| tensor<fp16, []> var_2373_to_fp16 = const()[name = tensor<string, []>("op_2373_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; |
| tensor<fp16, [1, 12, 64, 1]> var_2374_cast_fp16 = mul(x = mh_q_41_cast_fp16, y = var_2373_to_fp16)[name = tensor<string, []>("op_2374_cast_fp16")]; |
| tensor<int32, [4]> var_2377 = const()[name = tensor<string, []>("op_2377"), val = tensor<int32, [4]>([1, 12, 64, 448])]; |
| tensor<fp16, [1, 12, 64, 448]> var_2378_cast_fp16 = reshape(shape = var_2377, x = key_41_cast_fp16)[name = tensor<string, []>("op_2378_cast_fp16")]; |
| tensor<bool, []> mh_w_61_transpose_x_0 = const()[name = tensor<string, []>("mh_w_61_transpose_x_0"), val = tensor<bool, []>(true)]; |
| tensor<bool, []> mh_w_61_transpose_y_0 = const()[name = tensor<string, []>("mh_w_61_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [1, 12, 1, 448]> mh_w_61_cast_fp16 = matmul(transpose_x = mh_w_61_transpose_x_0, transpose_y = mh_w_61_transpose_y_0, x = var_2374_cast_fp16, y = var_2378_cast_fp16)[name = tensor<string, []>("mh_w_61_cast_fp16")]; |
| tensor<fp16, [1, 12, 1, 448]> mh_w_63_cast_fp16 = add(x = mh_w_61_cast_fp16, y = var_181_cast_fp16)[name = tensor<string, []>("mh_w_63_cast_fp16")]; |
| tensor<fp16, [1, 12, 1, 448]> var_2386_cast_fp16 = softmax(axis = var_2298, x = mh_w_63_cast_fp16)[name = tensor<string, []>("op_2386_cast_fp16")]; |
| tensor<int32, [4]> var_2387 = const()[name = tensor<string, []>("op_2387"), val = tensor<int32, [4]>([1, 12, 64, 448])]; |
| tensor<fp16, [1, 12, 64, 448]> var_2388_cast_fp16 = reshape(shape = var_2387, x = value_41_cast_fp16)[name = tensor<string, []>("op_2388_cast_fp16")]; |
| tensor<bool, []> attn_41_transpose_x_0 = const()[name = tensor<string, []>("attn_41_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_41_transpose_y_0 = const()[name = tensor<string, []>("attn_41_transpose_y_0"), val = tensor<bool, []>(true)]; |
| tensor<fp16, [1, 12, 64, 1]> attn_41_cast_fp16 = matmul(transpose_x = attn_41_transpose_x_0, transpose_y = attn_41_transpose_y_0, x = var_2388_cast_fp16, y = var_2386_cast_fp16)[name = tensor<string, []>("attn_41_cast_fp16")]; |
| tensor<int32, [4]> var_2391 = const()[name = tensor<string, []>("op_2391"), val = tensor<int32, [4]>([1, 768, 1, 1])]; |
| tensor<fp16, [1, 768, 1, 1]> input_101_cast_fp16 = reshape(shape = var_2391, x = attn_41_cast_fp16)[name = tensor<string, []>("input_101_cast_fp16")]; |
| tensor<string, []> obj_147_pad_type_0 = const()[name = tensor<string, []>("obj_147_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> obj_147_strides_0 = const()[name = tensor<string, []>("obj_147_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> obj_147_pad_0 = const()[name = tensor<string, []>("obj_147_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> obj_147_dilations_0 = const()[name = tensor<string, []>("obj_147_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> obj_147_groups_0 = const()[name = tensor<string, []>("obj_147_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 768, 1, 1]> layers_10_self_attn_o_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_10_self_attn_o_proj_weight_to_fp16"), val = tensor<fp16, [768, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(272921856)))]; |
| tensor<fp16, [768]> layers_10_self_attn_o_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_10_self_attn_o_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(274101568)))]; |
| tensor<fp16, [1, 768, 1, 1]> obj_147_cast_fp16 = conv(bias = layers_10_self_attn_o_proj_bias_to_fp16, dilations = obj_147_dilations_0, groups = obj_147_groups_0, pad = obj_147_pad_0, pad_type = obj_147_pad_type_0, strides = obj_147_strides_0, weight = layers_10_self_attn_o_proj_weight_to_fp16, x = input_101_cast_fp16)[name = tensor<string, []>("obj_147_cast_fp16")]; |
| tensor<fp16, [1, 768, 1, 1]> inputs_63_cast_fp16 = add(x = inputs_61_cast_fp16, y = obj_147_cast_fp16)[name = tensor<string, []>("inputs_63_cast_fp16")]; |
| tensor<int32, [1]> out_63_axes_0 = const()[name = tensor<string, []>("out_63_axes_0"), val = tensor<int32, [1]>([1])]; |
| tensor<fp16, []> var_2413_to_fp16 = const()[name = tensor<string, []>("op_2413_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, 768, 1, 1]> out_63_cast_fp16 = layer_norm(axes = out_63_axes_0, epsilon = var_2413_to_fp16, x = inputs_63_cast_fp16)[name = tensor<string, []>("out_63_cast_fp16")]; |
| tensor<fp16, [768]> obj_149_gamma_0_to_fp16 = const()[name = tensor<string, []>("obj_149_gamma_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(274103168)))]; |
| tensor<fp16, [768]> obj_149_beta_0_to_fp16 = const()[name = tensor<string, []>("obj_149_beta_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(274104768)))]; |
| tensor<fp16, []> obj_149_epsilon_0_to_fp16 = const()[name = tensor<string, []>("obj_149_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, 768, 1, 1]> obj_149_cast_fp16 = batch_norm(beta = obj_149_beta_0_to_fp16, epsilon = obj_149_epsilon_0_to_fp16, gamma = obj_149_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_63_cast_fp16)[name = tensor<string, []>("obj_149_cast_fp16")]; |
| tensor<string, []> query_43_pad_type_0 = const()[name = tensor<string, []>("query_43_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> query_43_strides_0 = const()[name = tensor<string, []>("query_43_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> query_43_pad_0 = const()[name = tensor<string, []>("query_43_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> query_43_dilations_0 = const()[name = tensor<string, []>("query_43_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> query_43_groups_0 = const()[name = tensor<string, []>("query_43_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 768, 1, 1]> layers_10_encoder_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_10_encoder_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [768, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(274106368)))]; |
| tensor<fp16, [768]> layers_10_encoder_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_10_encoder_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(275286080)))]; |
| tensor<fp16, [1, 768, 1, 1]> query_43_cast_fp16 = conv(bias = layers_10_encoder_attn_q_proj_bias_to_fp16, dilations = query_43_dilations_0, groups = query_43_groups_0, pad = query_43_pad_0, pad_type = query_43_pad_type_0, strides = query_43_strides_0, weight = layers_10_encoder_attn_q_proj_weight_to_fp16, x = obj_149_cast_fp16)[name = tensor<string, []>("query_43_cast_fp16")]; |
| tensor<string, []> key_43_pad_type_0 = const()[name = tensor<string, []>("key_43_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> key_43_strides_0 = const()[name = tensor<string, []>("key_43_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> key_43_pad_0 = const()[name = tensor<string, []>("key_43_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> key_43_dilations_0 = const()[name = tensor<string, []>("key_43_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> key_43_groups_0 = const()[name = tensor<string, []>("key_43_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 768, 1, 1]> layers_10_encoder_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_10_encoder_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [768, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(275287680)))]; |
| tensor<fp16, [1, 768, 1, 1500]> key_43_cast_fp16 = conv(dilations = key_43_dilations_0, groups = key_43_groups_0, pad = key_43_pad_0, pad_type = key_43_pad_type_0, strides = key_43_strides_0, weight = layers_10_encoder_attn_k_proj_weight_to_fp16, x = encoder_output_embeds)[name = tensor<string, []>("key_43_cast_fp16")]; |
| tensor<string, []> value_43_pad_type_0 = const()[name = tensor<string, []>("value_43_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> value_43_strides_0 = const()[name = tensor<string, []>("value_43_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> value_43_pad_0 = const()[name = tensor<string, []>("value_43_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> value_43_dilations_0 = const()[name = tensor<string, []>("value_43_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> value_43_groups_0 = const()[name = tensor<string, []>("value_43_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 768, 1, 1]> layers_10_encoder_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_10_encoder_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [768, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(276467392)))]; |
| tensor<fp16, [768]> layers_10_encoder_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_10_encoder_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(277647104)))]; |
| tensor<fp16, [1, 768, 1, 1500]> value_43_cast_fp16 = conv(bias = layers_10_encoder_attn_v_proj_bias_to_fp16, dilations = value_43_dilations_0, groups = value_43_groups_0, pad = value_43_pad_0, pad_type = value_43_pad_type_0, strides = value_43_strides_0, weight = layers_10_encoder_attn_v_proj_weight_to_fp16, x = encoder_output_embeds)[name = tensor<string, []>("value_43_cast_fp16")]; |
| tensor<int32, [4]> var_2449 = const()[name = tensor<string, []>("op_2449"), val = tensor<int32, [4]>([1, 12, 64, 1])]; |
| tensor<fp16, [1, 12, 64, 1]> mh_q_43_cast_fp16 = reshape(shape = var_2449, x = query_43_cast_fp16)[name = tensor<string, []>("mh_q_43_cast_fp16")]; |
| tensor<fp16, []> var_2451_to_fp16 = const()[name = tensor<string, []>("op_2451_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; |
| tensor<fp16, [1, 12, 64, 1]> var_2452_cast_fp16 = mul(x = mh_q_43_cast_fp16, y = var_2451_to_fp16)[name = tensor<string, []>("op_2452_cast_fp16")]; |
| tensor<int32, [4]> var_2455 = const()[name = tensor<string, []>("op_2455"), val = tensor<int32, [4]>([1, 12, 64, 1500])]; |
| tensor<fp16, [1, 12, 64, 1500]> var_2456_cast_fp16 = reshape(shape = var_2455, x = key_43_cast_fp16)[name = tensor<string, []>("op_2456_cast_fp16")]; |
| tensor<bool, []> mh_w_65_transpose_x_0 = const()[name = tensor<string, []>("mh_w_65_transpose_x_0"), val = tensor<bool, []>(true)]; |
| tensor<bool, []> mh_w_65_transpose_y_0 = const()[name = tensor<string, []>("mh_w_65_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [1, 12, 1, 1500]> mh_w_65_cast_fp16 = matmul(transpose_x = mh_w_65_transpose_x_0, transpose_y = mh_w_65_transpose_y_0, x = var_2452_cast_fp16, y = var_2456_cast_fp16)[name = tensor<string, []>("mh_w_65_cast_fp16")]; |
| tensor<fp16, [1, 12, 1, 1500]> obj_153_cast_fp16 = softmax(axis = var_2298, x = mh_w_65_cast_fp16)[name = tensor<string, []>("obj_153_cast_fp16")]; |
| tensor<int32, [4]> var_2460 = const()[name = tensor<string, []>("op_2460"), val = tensor<int32, [4]>([1, 12, 64, 1500])]; |
| tensor<fp16, [1, 12, 64, 1500]> var_2461_cast_fp16 = reshape(shape = var_2460, x = value_43_cast_fp16)[name = tensor<string, []>("op_2461_cast_fp16")]; |
| tensor<bool, []> attn_43_transpose_x_0 = const()[name = tensor<string, []>("attn_43_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_43_transpose_y_0 = const()[name = tensor<string, []>("attn_43_transpose_y_0"), val = tensor<bool, []>(true)]; |
| tensor<fp16, [1, 12, 64, 1]> attn_43_cast_fp16 = matmul(transpose_x = attn_43_transpose_x_0, transpose_y = attn_43_transpose_y_0, x = var_2461_cast_fp16, y = obj_153_cast_fp16)[name = tensor<string, []>("attn_43_cast_fp16")]; |
| tensor<int32, [4]> var_2464 = const()[name = tensor<string, []>("op_2464"), val = tensor<int32, [4]>([1, 768, 1, 1])]; |
| tensor<fp16, [1, 768, 1, 1]> input_103_cast_fp16 = reshape(shape = var_2464, x = attn_43_cast_fp16)[name = tensor<string, []>("input_103_cast_fp16")]; |
| tensor<string, []> obj_151_pad_type_0 = const()[name = tensor<string, []>("obj_151_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> obj_151_strides_0 = const()[name = tensor<string, []>("obj_151_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> obj_151_pad_0 = const()[name = tensor<string, []>("obj_151_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> obj_151_dilations_0 = const()[name = tensor<string, []>("obj_151_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> obj_151_groups_0 = const()[name = tensor<string, []>("obj_151_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 768, 1, 1]> layers_10_encoder_attn_o_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_10_encoder_attn_o_proj_weight_to_fp16"), val = tensor<fp16, [768, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(277648704)))]; |
| tensor<fp16, [768]> layers_10_encoder_attn_o_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_10_encoder_attn_o_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(278828416)))]; |
| tensor<fp16, [1, 768, 1, 1]> obj_151_cast_fp16 = conv(bias = layers_10_encoder_attn_o_proj_bias_to_fp16, dilations = obj_151_dilations_0, groups = obj_151_groups_0, pad = obj_151_pad_0, pad_type = obj_151_pad_type_0, strides = obj_151_strides_0, weight = layers_10_encoder_attn_o_proj_weight_to_fp16, x = input_103_cast_fp16)[name = tensor<string, []>("obj_151_cast_fp16")]; |
| tensor<fp16, [1, 768, 1, 1]> inputs_65_cast_fp16 = add(x = inputs_63_cast_fp16, y = obj_151_cast_fp16)[name = tensor<string, []>("inputs_65_cast_fp16")]; |
| tensor<int32, [1]> out_65_axes_0 = const()[name = tensor<string, []>("out_65_axes_0"), val = tensor<int32, [1]>([1])]; |
| tensor<fp16, []> var_2485_to_fp16 = const()[name = tensor<string, []>("op_2485_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, 768, 1, 1]> out_65_cast_fp16 = layer_norm(axes = out_65_axes_0, epsilon = var_2485_to_fp16, x = inputs_65_cast_fp16)[name = tensor<string, []>("out_65_cast_fp16")]; |
| tensor<fp16, [768]> input_105_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_105_gamma_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(278830016)))]; |
| tensor<fp16, [768]> input_105_beta_0_to_fp16 = const()[name = tensor<string, []>("input_105_beta_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(278831616)))]; |
| tensor<fp16, []> input_105_epsilon_0_to_fp16 = const()[name = tensor<string, []>("input_105_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, 768, 1, 1]> input_105_cast_fp16 = batch_norm(beta = input_105_beta_0_to_fp16, epsilon = input_105_epsilon_0_to_fp16, gamma = input_105_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_65_cast_fp16)[name = tensor<string, []>("input_105_cast_fp16")]; |
| tensor<string, []> input_107_pad_type_0 = const()[name = tensor<string, []>("input_107_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> input_107_strides_0 = const()[name = tensor<string, []>("input_107_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> input_107_pad_0 = const()[name = tensor<string, []>("input_107_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> input_107_dilations_0 = const()[name = tensor<string, []>("input_107_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> input_107_groups_0 = const()[name = tensor<string, []>("input_107_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [3072, 768, 1, 1]> layers_10_fc1_weight_to_fp16 = const()[name = tensor<string, []>("layers_10_fc1_weight_to_fp16"), val = tensor<fp16, [3072, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(278833216)))]; |
| tensor<fp16, [3072]> layers_10_fc1_bias_to_fp16 = const()[name = tensor<string, []>("layers_10_fc1_bias_to_fp16"), val = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(283551872)))]; |
| tensor<fp16, [1, 3072, 1, 1]> input_107_cast_fp16 = conv(bias = layers_10_fc1_bias_to_fp16, dilations = input_107_dilations_0, groups = input_107_groups_0, pad = input_107_pad_0, pad_type = input_107_pad_type_0, strides = input_107_strides_0, weight = layers_10_fc1_weight_to_fp16, x = input_105_cast_fp16)[name = tensor<string, []>("input_107_cast_fp16")]; |
| tensor<string, []> input_109_mode_0 = const()[name = tensor<string, []>("input_109_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 3072, 1, 1]> input_109_cast_fp16 = gelu(mode = input_109_mode_0, x = input_107_cast_fp16)[name = tensor<string, []>("input_109_cast_fp16")]; |
| tensor<string, []> hidden_states_23_pad_type_0 = const()[name = tensor<string, []>("hidden_states_23_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> hidden_states_23_strides_0 = const()[name = tensor<string, []>("hidden_states_23_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> hidden_states_23_pad_0 = const()[name = tensor<string, []>("hidden_states_23_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> hidden_states_23_dilations_0 = const()[name = tensor<string, []>("hidden_states_23_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> hidden_states_23_groups_0 = const()[name = tensor<string, []>("hidden_states_23_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 3072, 1, 1]> layers_10_fc2_weight_to_fp16 = const()[name = tensor<string, []>("layers_10_fc2_weight_to_fp16"), val = tensor<fp16, [768, 3072, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(283558080)))]; |
| tensor<fp16, [768]> layers_10_fc2_bias_to_fp16 = const()[name = tensor<string, []>("layers_10_fc2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(288276736)))]; |
| tensor<fp16, [1, 768, 1, 1]> hidden_states_23_cast_fp16 = conv(bias = layers_10_fc2_bias_to_fp16, dilations = hidden_states_23_dilations_0, groups = hidden_states_23_groups_0, pad = hidden_states_23_pad_0, pad_type = hidden_states_23_pad_type_0, strides = hidden_states_23_strides_0, weight = layers_10_fc2_weight_to_fp16, x = input_109_cast_fp16)[name = tensor<string, []>("hidden_states_23_cast_fp16")]; |
| tensor<fp16, [1, 768, 1, 1]> inputs_67_cast_fp16 = add(x = inputs_65_cast_fp16, y = hidden_states_23_cast_fp16)[name = tensor<string, []>("inputs_67_cast_fp16")]; |
| tensor<int32, []> var_2521 = const()[name = tensor<string, []>("op_2521"), val = tensor<int32, []>(3)]; |
| tensor<int32, [1]> out_67_axes_0 = const()[name = tensor<string, []>("out_67_axes_0"), val = tensor<int32, [1]>([1])]; |
| tensor<fp16, []> var_2546_to_fp16 = const()[name = tensor<string, []>("op_2546_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, 768, 1, 1]> out_67_cast_fp16 = layer_norm(axes = out_67_axes_0, epsilon = var_2546_to_fp16, x = inputs_67_cast_fp16)[name = tensor<string, []>("out_67_cast_fp16")]; |
| tensor<fp16, [768]> obj_155_gamma_0_to_fp16 = const()[name = tensor<string, []>("obj_155_gamma_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(288278336)))]; |
| tensor<fp16, [768]> obj_155_beta_0_to_fp16 = const()[name = tensor<string, []>("obj_155_beta_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(288279936)))]; |
| tensor<fp16, []> obj_155_epsilon_0_to_fp16 = const()[name = tensor<string, []>("obj_155_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, 768, 1, 1]> obj_155_cast_fp16 = batch_norm(beta = obj_155_beta_0_to_fp16, epsilon = obj_155_epsilon_0_to_fp16, gamma = obj_155_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_67_cast_fp16)[name = tensor<string, []>("obj_155_cast_fp16")]; |
| tensor<string, []> query_45_pad_type_0 = const()[name = tensor<string, []>("query_45_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> query_45_strides_0 = const()[name = tensor<string, []>("query_45_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> query_45_pad_0 = const()[name = tensor<string, []>("query_45_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> query_45_dilations_0 = const()[name = tensor<string, []>("query_45_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> query_45_groups_0 = const()[name = tensor<string, []>("query_45_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 768, 1, 1]> layers_11_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_11_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [768, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(288281536)))]; |
| tensor<fp16, [768]> layers_11_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("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, []>(289461248)))]; |
| tensor<fp16, [1, 768, 1, 1]> query_45_cast_fp16 = conv(bias = layers_11_self_attn_q_proj_bias_to_fp16, dilations = query_45_dilations_0, groups = query_45_groups_0, pad = query_45_pad_0, pad_type = query_45_pad_type_0, strides = query_45_strides_0, weight = layers_11_self_attn_q_proj_weight_to_fp16, x = obj_155_cast_fp16)[name = tensor<string, []>("query_45_cast_fp16")]; |
| tensor<string, []> current_key_pad_type_0 = const()[name = tensor<string, []>("current_key_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> current_key_strides_0 = const()[name = tensor<string, []>("current_key_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> current_key_pad_0 = const()[name = tensor<string, []>("current_key_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> current_key_dilations_0 = const()[name = tensor<string, []>("current_key_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> current_key_groups_0 = const()[name = tensor<string, []>("current_key_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 768, 1, 1]> layers_11_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_11_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [768, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(289462848)))]; |
| tensor<fp16, [1, 768, 1, 1]> current_key_cast_fp16 = conv(dilations = current_key_dilations_0, groups = current_key_groups_0, pad = current_key_pad_0, pad_type = current_key_pad_type_0, strides = current_key_strides_0, weight = layers_11_self_attn_k_proj_weight_to_fp16, x = obj_155_cast_fp16)[name = tensor<string, []>("current_key_cast_fp16")]; |
| tensor<string, []> current_value_pad_type_0 = const()[name = tensor<string, []>("current_value_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> current_value_strides_0 = const()[name = tensor<string, []>("current_value_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> current_value_pad_0 = const()[name = tensor<string, []>("current_value_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> current_value_dilations_0 = const()[name = tensor<string, []>("current_value_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> current_value_groups_0 = const()[name = tensor<string, []>("current_value_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 768, 1, 1]> layers_11_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_11_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [768, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(290642560)))]; |
| tensor<fp16, [768]> layers_11_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("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, []>(291822272)))]; |
| tensor<fp16, [1, 768, 1, 1]> current_value_cast_fp16 = conv(bias = layers_11_self_attn_v_proj_bias_to_fp16, dilations = current_value_dilations_0, groups = current_value_groups_0, pad = current_value_pad_0, pad_type = current_value_pad_type_0, strides = current_value_strides_0, weight = layers_11_self_attn_v_proj_weight_to_fp16, x = obj_155_cast_fp16)[name = tensor<string, []>("current_value_cast_fp16")]; |
| tensor<fp16, [1, 768, 1, 448]> var_2585_cast_fp16 = mul(x = var_63_cast_fp16_11, y = var_159_cast_fp16)[name = tensor<string, []>("op_2585_cast_fp16")]; |
| tensor<fp16, [1, 768, 1, 448]> var_2586_cast_fp16 = mul(x = current_key_cast_fp16, y = var_157_cast_fp16)[name = tensor<string, []>("op_2586_cast_fp16")]; |
| tensor<fp16, [1, 768, 1, 448]> key_45_cast_fp16 = add(x = var_2585_cast_fp16, y = var_2586_cast_fp16)[name = tensor<string, []>("key_45_cast_fp16")]; |
| tensor<fp16, [1, 768, 1, 448]> var_2589_cast_fp16 = mul(x = var_78_cast_fp16_11, y = var_159_cast_fp16)[name = tensor<string, []>("op_2589_cast_fp16")]; |
| tensor<fp16, [1, 768, 1, 448]> var_2590_cast_fp16 = mul(x = current_value_cast_fp16, y = var_157_cast_fp16)[name = tensor<string, []>("op_2590_cast_fp16")]; |
| tensor<fp16, [1, 768, 1, 448]> value_45_cast_fp16 = add(x = var_2589_cast_fp16, y = var_2590_cast_fp16)[name = tensor<string, []>("value_45_cast_fp16")]; |
| tensor<int32, [4]> var_2594 = const()[name = tensor<string, []>("op_2594"), val = tensor<int32, [4]>([1, 12, 64, 1])]; |
| tensor<fp16, [1, 12, 64, 1]> mh_q_45_cast_fp16 = reshape(shape = var_2594, x = query_45_cast_fp16)[name = tensor<string, []>("mh_q_45_cast_fp16")]; |
| tensor<fp16, []> var_2596_to_fp16 = const()[name = tensor<string, []>("op_2596_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; |
| tensor<fp16, [1, 12, 64, 1]> var_2597_cast_fp16 = mul(x = mh_q_45_cast_fp16, y = var_2596_to_fp16)[name = tensor<string, []>("op_2597_cast_fp16")]; |
| tensor<int32, [4]> var_2600 = const()[name = tensor<string, []>("op_2600"), val = tensor<int32, [4]>([1, 12, 64, 448])]; |
| tensor<fp16, [1, 12, 64, 448]> var_2601_cast_fp16 = reshape(shape = var_2600, x = key_45_cast_fp16)[name = tensor<string, []>("op_2601_cast_fp16")]; |
| tensor<bool, []> mh_w_67_transpose_x_0 = const()[name = tensor<string, []>("mh_w_67_transpose_x_0"), val = tensor<bool, []>(true)]; |
| tensor<bool, []> mh_w_67_transpose_y_0 = const()[name = tensor<string, []>("mh_w_67_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [1, 12, 1, 448]> mh_w_67_cast_fp16 = matmul(transpose_x = mh_w_67_transpose_x_0, transpose_y = mh_w_67_transpose_y_0, x = var_2597_cast_fp16, y = var_2601_cast_fp16)[name = tensor<string, []>("mh_w_67_cast_fp16")]; |
| tensor<fp16, [1, 12, 1, 448]> mh_w_69_cast_fp16 = add(x = mh_w_67_cast_fp16, y = var_181_cast_fp16)[name = tensor<string, []>("mh_w_69_cast_fp16")]; |
| tensor<fp16, [1, 12, 1, 448]> var_2609_cast_fp16 = softmax(axis = var_2521, x = mh_w_69_cast_fp16)[name = tensor<string, []>("op_2609_cast_fp16")]; |
| tensor<int32, [4]> var_2610 = const()[name = tensor<string, []>("op_2610"), val = tensor<int32, [4]>([1, 12, 64, 448])]; |
| tensor<fp16, [1, 12, 64, 448]> var_2611_cast_fp16 = reshape(shape = var_2610, x = value_45_cast_fp16)[name = tensor<string, []>("op_2611_cast_fp16")]; |
| tensor<bool, []> attn_45_transpose_x_0 = const()[name = tensor<string, []>("attn_45_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_45_transpose_y_0 = const()[name = tensor<string, []>("attn_45_transpose_y_0"), val = tensor<bool, []>(true)]; |
| tensor<fp16, [1, 12, 64, 1]> attn_45_cast_fp16 = matmul(transpose_x = attn_45_transpose_x_0, transpose_y = attn_45_transpose_y_0, x = var_2611_cast_fp16, y = var_2609_cast_fp16)[name = tensor<string, []>("attn_45_cast_fp16")]; |
| tensor<int32, [4]> var_2614 = const()[name = tensor<string, []>("op_2614"), val = tensor<int32, [4]>([1, 768, 1, 1])]; |
| tensor<fp16, [1, 768, 1, 1]> input_111_cast_fp16 = reshape(shape = var_2614, x = attn_45_cast_fp16)[name = tensor<string, []>("input_111_cast_fp16")]; |
| tensor<string, []> obj_161_pad_type_0 = const()[name = tensor<string, []>("obj_161_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> obj_161_strides_0 = const()[name = tensor<string, []>("obj_161_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> obj_161_pad_0 = const()[name = tensor<string, []>("obj_161_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> obj_161_dilations_0 = const()[name = tensor<string, []>("obj_161_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> obj_161_groups_0 = const()[name = tensor<string, []>("obj_161_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 768, 1, 1]> layers_11_self_attn_o_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_11_self_attn_o_proj_weight_to_fp16"), val = tensor<fp16, [768, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(291823872)))]; |
| tensor<fp16, [768]> layers_11_self_attn_o_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_11_self_attn_o_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(293003584)))]; |
| tensor<fp16, [1, 768, 1, 1]> obj_161_cast_fp16 = conv(bias = layers_11_self_attn_o_proj_bias_to_fp16, dilations = obj_161_dilations_0, groups = obj_161_groups_0, pad = obj_161_pad_0, pad_type = obj_161_pad_type_0, strides = obj_161_strides_0, weight = layers_11_self_attn_o_proj_weight_to_fp16, x = input_111_cast_fp16)[name = tensor<string, []>("obj_161_cast_fp16")]; |
| tensor<fp16, [1, 768, 1, 1]> inputs_69_cast_fp16 = add(x = inputs_67_cast_fp16, y = obj_161_cast_fp16)[name = tensor<string, []>("inputs_69_cast_fp16")]; |
| tensor<int32, [1]> out_69_axes_0 = const()[name = tensor<string, []>("out_69_axes_0"), val = tensor<int32, [1]>([1])]; |
| tensor<fp16, []> var_2636_to_fp16 = const()[name = tensor<string, []>("op_2636_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, 768, 1, 1]> out_69_cast_fp16 = layer_norm(axes = out_69_axes_0, epsilon = var_2636_to_fp16, x = inputs_69_cast_fp16)[name = tensor<string, []>("out_69_cast_fp16")]; |
| tensor<fp16, [768]> obj_163_gamma_0_to_fp16 = const()[name = tensor<string, []>("obj_163_gamma_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(293005184)))]; |
| tensor<fp16, [768]> obj_163_beta_0_to_fp16 = const()[name = tensor<string, []>("obj_163_beta_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(293006784)))]; |
| tensor<fp16, []> obj_163_epsilon_0_to_fp16 = const()[name = tensor<string, []>("obj_163_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, 768, 1, 1]> obj_163_cast_fp16 = batch_norm(beta = obj_163_beta_0_to_fp16, epsilon = obj_163_epsilon_0_to_fp16, gamma = obj_163_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_69_cast_fp16)[name = tensor<string, []>("obj_163_cast_fp16")]; |
| tensor<string, []> query_pad_type_0 = const()[name = tensor<string, []>("query_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> query_strides_0 = const()[name = tensor<string, []>("query_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> query_pad_0 = const()[name = tensor<string, []>("query_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> query_dilations_0 = const()[name = tensor<string, []>("query_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> query_groups_0 = const()[name = tensor<string, []>("query_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 768, 1, 1]> layers_11_encoder_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_11_encoder_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [768, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(293008384)))]; |
| tensor<fp16, [768]> layers_11_encoder_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_11_encoder_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(294188096)))]; |
| tensor<fp16, [1, 768, 1, 1]> query_cast_fp16 = conv(bias = layers_11_encoder_attn_q_proj_bias_to_fp16, dilations = query_dilations_0, groups = query_groups_0, pad = query_pad_0, pad_type = query_pad_type_0, strides = query_strides_0, weight = layers_11_encoder_attn_q_proj_weight_to_fp16, x = obj_163_cast_fp16)[name = tensor<string, []>("query_cast_fp16")]; |
| tensor<string, []> key_pad_type_0 = const()[name = tensor<string, []>("key_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> key_strides_0 = const()[name = tensor<string, []>("key_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> key_pad_0 = const()[name = tensor<string, []>("key_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> key_dilations_0 = const()[name = tensor<string, []>("key_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> key_groups_0 = const()[name = tensor<string, []>("key_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 768, 1, 1]> layers_11_encoder_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_11_encoder_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [768, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(294189696)))]; |
| tensor<fp16, [1, 768, 1, 1500]> key_cast_fp16 = conv(dilations = key_dilations_0, groups = key_groups_0, pad = key_pad_0, pad_type = key_pad_type_0, strides = key_strides_0, weight = layers_11_encoder_attn_k_proj_weight_to_fp16, x = encoder_output_embeds)[name = tensor<string, []>("key_cast_fp16")]; |
| tensor<string, []> value_pad_type_0 = const()[name = tensor<string, []>("value_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> value_strides_0 = const()[name = tensor<string, []>("value_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> value_pad_0 = const()[name = tensor<string, []>("value_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> value_dilations_0 = const()[name = tensor<string, []>("value_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> value_groups_0 = const()[name = tensor<string, []>("value_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 768, 1, 1]> layers_11_encoder_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_11_encoder_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [768, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(295369408)))]; |
| tensor<fp16, [768]> layers_11_encoder_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_11_encoder_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(296549120)))]; |
| tensor<fp16, [1, 768, 1, 1500]> value_cast_fp16 = conv(bias = layers_11_encoder_attn_v_proj_bias_to_fp16, dilations = value_dilations_0, groups = value_groups_0, pad = value_pad_0, pad_type = value_pad_type_0, strides = value_strides_0, weight = layers_11_encoder_attn_v_proj_weight_to_fp16, x = encoder_output_embeds)[name = tensor<string, []>("value_cast_fp16")]; |
| tensor<int32, [4]> var_2672 = const()[name = tensor<string, []>("op_2672"), val = tensor<int32, [4]>([1, 12, 64, 1])]; |
| tensor<fp16, [1, 12, 64, 1]> mh_q_cast_fp16 = reshape(shape = var_2672, x = query_cast_fp16)[name = tensor<string, []>("mh_q_cast_fp16")]; |
| tensor<fp16, []> var_2674_to_fp16 = const()[name = tensor<string, []>("op_2674_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; |
| tensor<fp16, [1, 12, 64, 1]> var_2675_cast_fp16 = mul(x = mh_q_cast_fp16, y = var_2674_to_fp16)[name = tensor<string, []>("op_2675_cast_fp16")]; |
| tensor<int32, [4]> var_2678 = const()[name = tensor<string, []>("op_2678"), val = tensor<int32, [4]>([1, 12, 64, 1500])]; |
| tensor<fp16, [1, 12, 64, 1500]> var_2679_cast_fp16 = reshape(shape = var_2678, x = key_cast_fp16)[name = tensor<string, []>("op_2679_cast_fp16")]; |
| tensor<bool, []> mh_w_transpose_x_0 = const()[name = tensor<string, []>("mh_w_transpose_x_0"), val = tensor<bool, []>(true)]; |
| tensor<bool, []> mh_w_transpose_y_0 = const()[name = tensor<string, []>("mh_w_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [1, 12, 1, 1500]> mh_w_cast_fp16 = matmul(transpose_x = mh_w_transpose_x_0, transpose_y = mh_w_transpose_y_0, x = var_2675_cast_fp16, y = var_2679_cast_fp16)[name = tensor<string, []>("mh_w_cast_fp16")]; |
| tensor<fp16, [1, 12, 1, 1500]> obj_167_cast_fp16 = softmax(axis = var_2521, x = mh_w_cast_fp16)[name = tensor<string, []>("obj_167_cast_fp16")]; |
| tensor<int32, [4]> var_2683 = const()[name = tensor<string, []>("op_2683"), val = tensor<int32, [4]>([1, 12, 64, 1500])]; |
| tensor<fp16, [1, 12, 64, 1500]> var_2684_cast_fp16 = reshape(shape = var_2683, x = value_cast_fp16)[name = tensor<string, []>("op_2684_cast_fp16")]; |
| tensor<bool, []> attn_transpose_x_0 = const()[name = tensor<string, []>("attn_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_transpose_y_0 = const()[name = tensor<string, []>("attn_transpose_y_0"), val = tensor<bool, []>(true)]; |
| tensor<fp16, [1, 12, 64, 1]> attn_cast_fp16 = matmul(transpose_x = attn_transpose_x_0, transpose_y = attn_transpose_y_0, x = var_2684_cast_fp16, y = obj_167_cast_fp16)[name = tensor<string, []>("attn_cast_fp16")]; |
| tensor<int32, [4]> var_2687 = const()[name = tensor<string, []>("op_2687"), val = tensor<int32, [4]>([1, 768, 1, 1])]; |
| tensor<fp16, [1, 768, 1, 1]> input_113_cast_fp16 = reshape(shape = var_2687, x = attn_cast_fp16)[name = tensor<string, []>("input_113_cast_fp16")]; |
| tensor<string, []> obj_165_pad_type_0 = const()[name = tensor<string, []>("obj_165_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> obj_165_strides_0 = const()[name = tensor<string, []>("obj_165_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> obj_165_pad_0 = const()[name = tensor<string, []>("obj_165_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> obj_165_dilations_0 = const()[name = tensor<string, []>("obj_165_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> obj_165_groups_0 = const()[name = tensor<string, []>("obj_165_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 768, 1, 1]> layers_11_encoder_attn_o_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_11_encoder_attn_o_proj_weight_to_fp16"), val = tensor<fp16, [768, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(296550720)))]; |
| tensor<fp16, [768]> layers_11_encoder_attn_o_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_11_encoder_attn_o_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(297730432)))]; |
| tensor<fp16, [1, 768, 1, 1]> obj_165_cast_fp16 = conv(bias = layers_11_encoder_attn_o_proj_bias_to_fp16, dilations = obj_165_dilations_0, groups = obj_165_groups_0, pad = obj_165_pad_0, pad_type = obj_165_pad_type_0, strides = obj_165_strides_0, weight = layers_11_encoder_attn_o_proj_weight_to_fp16, x = input_113_cast_fp16)[name = tensor<string, []>("obj_165_cast_fp16")]; |
| tensor<fp16, [1, 768, 1, 1]> inputs_71_cast_fp16 = add(x = inputs_69_cast_fp16, y = obj_165_cast_fp16)[name = tensor<string, []>("inputs_71_cast_fp16")]; |
| tensor<int32, [1]> out_71_axes_0 = const()[name = tensor<string, []>("out_71_axes_0"), val = tensor<int32, [1]>([1])]; |
| tensor<fp16, []> var_2705_to_fp16 = const()[name = tensor<string, []>("op_2705_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, 768, 1, 1]> out_71_cast_fp16 = layer_norm(axes = out_71_axes_0, epsilon = var_2705_to_fp16, x = inputs_71_cast_fp16)[name = tensor<string, []>("out_71_cast_fp16")]; |
| tensor<fp16, [768]> input_115_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_115_gamma_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(297732032)))]; |
| tensor<fp16, [768]> input_115_beta_0_to_fp16 = const()[name = tensor<string, []>("input_115_beta_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(297733632)))]; |
| tensor<fp16, []> input_115_epsilon_0_to_fp16 = const()[name = tensor<string, []>("input_115_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, 768, 1, 1]> input_115_cast_fp16 = batch_norm(beta = input_115_beta_0_to_fp16, epsilon = input_115_epsilon_0_to_fp16, gamma = input_115_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_71_cast_fp16)[name = tensor<string, []>("input_115_cast_fp16")]; |
| tensor<string, []> input_117_pad_type_0 = const()[name = tensor<string, []>("input_117_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> input_117_strides_0 = const()[name = tensor<string, []>("input_117_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> input_117_pad_0 = const()[name = tensor<string, []>("input_117_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> input_117_dilations_0 = const()[name = tensor<string, []>("input_117_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> input_117_groups_0 = const()[name = tensor<string, []>("input_117_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [3072, 768, 1, 1]> layers_11_fc1_weight_to_fp16 = const()[name = tensor<string, []>("layers_11_fc1_weight_to_fp16"), val = tensor<fp16, [3072, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(297735232)))]; |
| tensor<fp16, [3072]> layers_11_fc1_bias_to_fp16 = const()[name = tensor<string, []>("layers_11_fc1_bias_to_fp16"), val = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(302453888)))]; |
| tensor<fp16, [1, 3072, 1, 1]> input_117_cast_fp16 = conv(bias = layers_11_fc1_bias_to_fp16, dilations = input_117_dilations_0, groups = input_117_groups_0, pad = input_117_pad_0, pad_type = input_117_pad_type_0, strides = input_117_strides_0, weight = layers_11_fc1_weight_to_fp16, x = input_115_cast_fp16)[name = tensor<string, []>("input_117_cast_fp16")]; |
| tensor<string, []> input_mode_0 = const()[name = tensor<string, []>("input_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 3072, 1, 1]> input_cast_fp16 = gelu(mode = input_mode_0, x = input_117_cast_fp16)[name = tensor<string, []>("input_cast_fp16")]; |
| tensor<string, []> hidden_states_25_pad_type_0 = const()[name = tensor<string, []>("hidden_states_25_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> hidden_states_25_strides_0 = const()[name = tensor<string, []>("hidden_states_25_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> hidden_states_25_pad_0 = const()[name = tensor<string, []>("hidden_states_25_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> hidden_states_25_dilations_0 = const()[name = tensor<string, []>("hidden_states_25_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> hidden_states_25_groups_0 = const()[name = tensor<string, []>("hidden_states_25_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 3072, 1, 1]> layers_11_fc2_weight_to_fp16 = const()[name = tensor<string, []>("layers_11_fc2_weight_to_fp16"), val = tensor<fp16, [768, 3072, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(302460096)))]; |
| tensor<fp16, [768]> layers_11_fc2_bias_to_fp16 = const()[name = tensor<string, []>("layers_11_fc2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(307178752)))]; |
| tensor<fp16, [1, 768, 1, 1]> hidden_states_25_cast_fp16 = conv(bias = layers_11_fc2_bias_to_fp16, dilations = hidden_states_25_dilations_0, groups = hidden_states_25_groups_0, pad = hidden_states_25_pad_0, pad_type = hidden_states_25_pad_type_0, strides = hidden_states_25_strides_0, weight = layers_11_fc2_weight_to_fp16, x = input_cast_fp16)[name = tensor<string, []>("hidden_states_25_cast_fp16")]; |
| tensor<fp16, [1, 768, 1, 1]> inputs_cast_fp16 = add(x = inputs_71_cast_fp16, y = hidden_states_25_cast_fp16)[name = tensor<string, []>("inputs_cast_fp16")]; |
| tensor<int32, [1]> out_axes_0 = const()[name = tensor<string, []>("out_axes_0"), val = tensor<int32, [1]>([1])]; |
| tensor<fp16, []> var_2747_to_fp16 = const()[name = tensor<string, []>("op_2747_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, 768, 1, 1]> out_cast_fp16 = layer_norm(axes = out_axes_0, epsilon = var_2747_to_fp16, x = inputs_cast_fp16)[name = tensor<string, []>("out_cast_fp16")]; |
| tensor<fp16, [768]> hidden_states_gamma_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_gamma_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(307180352)))]; |
| tensor<fp16, [768]> hidden_states_beta_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_beta_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(307181952)))]; |
| tensor<fp16, []> hidden_states_epsilon_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, 768, 1, 1]> hidden_states_cast_fp16 = batch_norm(beta = hidden_states_beta_0_to_fp16, epsilon = hidden_states_epsilon_0_to_fp16, gamma = hidden_states_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_cast_fp16)[name = tensor<string, []>("hidden_states_cast_fp16")]; |
| tensor<int32, [1]> var_2758_axes_0 = const()[name = tensor<string, []>("op_2758_axes_0"), val = tensor<int32, [1]>([2])]; |
| tensor<fp16, [1, 768, 1]> var_2758_cast_fp16 = squeeze(axes = var_2758_axes_0, x = hidden_states_cast_fp16)[name = tensor<string, []>("op_2758_cast_fp16")]; |
| tensor<int32, [3]> var_2761_perm_0 = const()[name = tensor<string, []>("op_2761_perm_0"), val = tensor<int32, [3]>([0, 2, 1])]; |
| tensor<fp16, [51865]> linear_0_bias_0_to_fp16 = const()[name = tensor<string, []>("linear_0_bias_0_to_fp16"), val = tensor<fp16, [51865]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(307183552)))]; |
| tensor<fp16, [1, 1, 768]> var_2761_cast_fp16 = transpose(perm = var_2761_perm_0, x = var_2758_cast_fp16)[name = tensor<string, []>("transpose_0")]; |
| tensor<fp16, [1, 1, 51865]> logits = linear(bias = linear_0_bias_0_to_fp16, weight = embed_tokens_weight_to_fp16, x = var_2761_cast_fp16)[name = tensor<string, []>("linear_0_cast_fp16")]; |
| tensor<int32, []> var_2765 = const()[name = tensor<string, []>("op_2765"), val = tensor<int32, []>(1)]; |
| tensor<bool, []> obj_171_interleave_0 = const()[name = tensor<string, []>("obj_171_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [1, 9216, 1, 1]> key_cache_updates = concat(axis = var_2765, interleave = obj_171_interleave_0, values = (current_key_1_cast_fp16, current_key_3_cast_fp16, current_key_5_cast_fp16, current_key_7_cast_fp16, current_key_9_cast_fp16, current_key_11_cast_fp16, current_key_13_cast_fp16, current_key_15_cast_fp16, current_key_17_cast_fp16, current_key_19_cast_fp16, current_key_21_cast_fp16, current_key_cast_fp16))[name = tensor<string, []>("obj_171_cast_fp16")]; |
| tensor<int32, []> var_2768 = const()[name = tensor<string, []>("op_2768"), val = tensor<int32, []>(1)]; |
| tensor<bool, []> obj_173_interleave_0 = const()[name = tensor<string, []>("obj_173_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [1, 9216, 1, 1]> value_cache_updates = concat(axis = var_2768, interleave = obj_173_interleave_0, values = (current_value_1_cast_fp16, current_value_3_cast_fp16, current_value_5_cast_fp16, current_value_7_cast_fp16, current_value_9_cast_fp16, current_value_11_cast_fp16, current_value_13_cast_fp16, current_value_15_cast_fp16, current_value_17_cast_fp16, current_value_19_cast_fp16, current_value_21_cast_fp16, current_value_cast_fp16))[name = tensor<string, []>("obj_173_cast_fp16")]; |
| tensor<int32, [4]> var_2779_begin_0 = const()[name = tensor<string, []>("op_2779_begin_0"), val = tensor<int32, [4]>([0, 3, 0, 0])]; |
| tensor<int32, [4]> var_2779_end_0 = const()[name = tensor<string, []>("op_2779_end_0"), val = tensor<int32, [4]>([1, 4, 1, 1500])]; |
| tensor<bool, [4]> var_2779_end_mask_0 = const()[name = tensor<string, []>("op_2779_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])]; |
| tensor<fp16, [1, 1, 1, 1500]> var_2779_cast_fp16 = slice_by_index(begin = var_2779_begin_0, end = var_2779_end_0, end_mask = var_2779_end_mask_0, x = obj_83_cast_fp16)[name = tensor<string, []>("op_2779_cast_fp16")]; |
| tensor<int32, [4]> var_2782_begin_0 = const()[name = tensor<string, []>("op_2782_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [4]> var_2782_end_0 = const()[name = tensor<string, []>("op_2782_end_0"), val = tensor<int32, [4]>([1, 1, 1, 1500])]; |
| tensor<bool, [4]> var_2782_end_mask_0 = const()[name = tensor<string, []>("op_2782_end_mask_0"), val = tensor<bool, [4]>([true, true, false, true])]; |
| tensor<bool, [4]> var_2782_squeeze_mask_0 = const()[name = tensor<string, []>("op_2782_squeeze_mask_0"), val = tensor<bool, [4]>([false, false, true, false])]; |
| tensor<fp16, [1, 1, 1500]> var_2782_cast_fp16 = slice_by_index(begin = var_2782_begin_0, end = var_2782_end_0, end_mask = var_2782_end_mask_0, squeeze_mask = var_2782_squeeze_mask_0, x = var_2779_cast_fp16)[name = tensor<string, []>("op_2782_cast_fp16")]; |
| tensor<int32, [4]> var_2797_begin_0 = const()[name = tensor<string, []>("op_2797_begin_0"), val = tensor<int32, [4]>([0, 9, 0, 0])]; |
| tensor<int32, [4]> var_2797_end_0 = const()[name = tensor<string, []>("op_2797_end_0"), val = tensor<int32, [4]>([1, 10, 1, 1500])]; |
| tensor<bool, [4]> var_2797_end_mask_0 = const()[name = tensor<string, []>("op_2797_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])]; |
| tensor<fp16, [1, 1, 1, 1500]> var_2797_cast_fp16 = slice_by_index(begin = var_2797_begin_0, end = var_2797_end_0, end_mask = var_2797_end_mask_0, x = obj_83_cast_fp16)[name = tensor<string, []>("op_2797_cast_fp16")]; |
| tensor<int32, [4]> var_2800_begin_0 = const()[name = tensor<string, []>("op_2800_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [4]> var_2800_end_0 = const()[name = tensor<string, []>("op_2800_end_0"), val = tensor<int32, [4]>([1, 1, 1, 1500])]; |
| tensor<bool, [4]> var_2800_end_mask_0 = const()[name = tensor<string, []>("op_2800_end_mask_0"), val = tensor<bool, [4]>([true, true, false, true])]; |
| tensor<bool, [4]> var_2800_squeeze_mask_0 = const()[name = tensor<string, []>("op_2800_squeeze_mask_0"), val = tensor<bool, [4]>([false, false, true, false])]; |
| tensor<fp16, [1, 1, 1500]> var_2800_cast_fp16 = slice_by_index(begin = var_2800_begin_0, end = var_2800_end_0, end_mask = var_2800_end_mask_0, squeeze_mask = var_2800_squeeze_mask_0, x = var_2797_cast_fp16)[name = tensor<string, []>("op_2800_cast_fp16")]; |
| tensor<int32, [4]> var_2815_begin_0 = const()[name = tensor<string, []>("op_2815_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [4]> var_2815_end_0 = const()[name = tensor<string, []>("op_2815_end_0"), val = tensor<int32, [4]>([1, 1, 1, 1500])]; |
| tensor<bool, [4]> var_2815_end_mask_0 = const()[name = tensor<string, []>("op_2815_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])]; |
| tensor<fp16, [1, 1, 1, 1500]> var_2815_cast_fp16 = slice_by_index(begin = var_2815_begin_0, end = var_2815_end_0, end_mask = var_2815_end_mask_0, x = obj_125_cast_fp16)[name = tensor<string, []>("op_2815_cast_fp16")]; |
| tensor<int32, [4]> var_2818_begin_0 = const()[name = tensor<string, []>("op_2818_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [4]> var_2818_end_0 = const()[name = tensor<string, []>("op_2818_end_0"), val = tensor<int32, [4]>([1, 1, 1, 1500])]; |
| tensor<bool, [4]> var_2818_end_mask_0 = const()[name = tensor<string, []>("op_2818_end_mask_0"), val = tensor<bool, [4]>([true, true, false, true])]; |
| tensor<bool, [4]> var_2818_squeeze_mask_0 = const()[name = tensor<string, []>("op_2818_squeeze_mask_0"), val = tensor<bool, [4]>([false, false, true, false])]; |
| tensor<fp16, [1, 1, 1500]> var_2818_cast_fp16 = slice_by_index(begin = var_2818_begin_0, end = var_2818_end_0, end_mask = var_2818_end_mask_0, squeeze_mask = var_2818_squeeze_mask_0, x = var_2815_cast_fp16)[name = tensor<string, []>("op_2818_cast_fp16")]; |
| tensor<int32, [4]> var_2833_begin_0 = const()[name = tensor<string, []>("op_2833_begin_0"), val = tensor<int32, [4]>([0, 4, 0, 0])]; |
| tensor<int32, [4]> var_2833_end_0 = const()[name = tensor<string, []>("op_2833_end_0"), val = tensor<int32, [4]>([1, 5, 1, 1500])]; |
| tensor<bool, [4]> var_2833_end_mask_0 = const()[name = tensor<string, []>("op_2833_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])]; |
| tensor<fp16, [1, 1, 1, 1500]> var_2833_cast_fp16 = slice_by_index(begin = var_2833_begin_0, end = var_2833_end_0, end_mask = var_2833_end_mask_0, x = obj_125_cast_fp16)[name = tensor<string, []>("op_2833_cast_fp16")]; |
| tensor<int32, [4]> var_2836_begin_0 = const()[name = tensor<string, []>("op_2836_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [4]> var_2836_end_0 = const()[name = tensor<string, []>("op_2836_end_0"), val = tensor<int32, [4]>([1, 1, 1, 1500])]; |
| tensor<bool, [4]> var_2836_end_mask_0 = const()[name = tensor<string, []>("op_2836_end_mask_0"), val = tensor<bool, [4]>([true, true, false, true])]; |
| tensor<bool, [4]> var_2836_squeeze_mask_0 = const()[name = tensor<string, []>("op_2836_squeeze_mask_0"), val = tensor<bool, [4]>([false, false, true, false])]; |
| tensor<fp16, [1, 1, 1500]> var_2836_cast_fp16 = slice_by_index(begin = var_2836_begin_0, end = var_2836_end_0, end_mask = var_2836_end_mask_0, squeeze_mask = var_2836_squeeze_mask_0, x = var_2833_cast_fp16)[name = tensor<string, []>("op_2836_cast_fp16")]; |
| tensor<int32, [4]> var_2851_begin_0 = const()[name = tensor<string, []>("op_2851_begin_0"), val = tensor<int32, [4]>([0, 7, 0, 0])]; |
| tensor<int32, [4]> var_2851_end_0 = const()[name = tensor<string, []>("op_2851_end_0"), val = tensor<int32, [4]>([1, 8, 1, 1500])]; |
| tensor<bool, [4]> var_2851_end_mask_0 = const()[name = tensor<string, []>("op_2851_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])]; |
| tensor<fp16, [1, 1, 1, 1500]> var_2851_cast_fp16 = slice_by_index(begin = var_2851_begin_0, end = var_2851_end_0, end_mask = var_2851_end_mask_0, x = obj_125_cast_fp16)[name = tensor<string, []>("op_2851_cast_fp16")]; |
| tensor<int32, [4]> var_2854_begin_0 = const()[name = tensor<string, []>("op_2854_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [4]> var_2854_end_0 = const()[name = tensor<string, []>("op_2854_end_0"), val = tensor<int32, [4]>([1, 1, 1, 1500])]; |
| tensor<bool, [4]> var_2854_end_mask_0 = const()[name = tensor<string, []>("op_2854_end_mask_0"), val = tensor<bool, [4]>([true, true, false, true])]; |
| tensor<bool, [4]> var_2854_squeeze_mask_0 = const()[name = tensor<string, []>("op_2854_squeeze_mask_0"), val = tensor<bool, [4]>([false, false, true, false])]; |
| tensor<fp16, [1, 1, 1500]> var_2854_cast_fp16 = slice_by_index(begin = var_2854_begin_0, end = var_2854_end_0, end_mask = var_2854_end_mask_0, squeeze_mask = var_2854_squeeze_mask_0, x = var_2851_cast_fp16)[name = tensor<string, []>("op_2854_cast_fp16")]; |
| tensor<int32, [4]> var_2869_begin_0 = const()[name = tensor<string, []>("op_2869_begin_0"), val = tensor<int32, [4]>([0, 8, 0, 0])]; |
| tensor<int32, [4]> var_2869_end_0 = const()[name = tensor<string, []>("op_2869_end_0"), val = tensor<int32, [4]>([1, 9, 1, 1500])]; |
| tensor<bool, [4]> var_2869_end_mask_0 = const()[name = tensor<string, []>("op_2869_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])]; |
| tensor<fp16, [1, 1, 1, 1500]> var_2869_cast_fp16 = slice_by_index(begin = var_2869_begin_0, end = var_2869_end_0, end_mask = var_2869_end_mask_0, x = obj_125_cast_fp16)[name = tensor<string, []>("op_2869_cast_fp16")]; |
| tensor<int32, [4]> var_2872_begin_0 = const()[name = tensor<string, []>("op_2872_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [4]> var_2872_end_0 = const()[name = tensor<string, []>("op_2872_end_0"), val = tensor<int32, [4]>([1, 1, 1, 1500])]; |
| tensor<bool, [4]> var_2872_end_mask_0 = const()[name = tensor<string, []>("op_2872_end_mask_0"), val = tensor<bool, [4]>([true, true, false, true])]; |
| tensor<bool, [4]> var_2872_squeeze_mask_0 = const()[name = tensor<string, []>("op_2872_squeeze_mask_0"), val = tensor<bool, [4]>([false, false, true, false])]; |
| tensor<fp16, [1, 1, 1500]> var_2872_cast_fp16 = slice_by_index(begin = var_2872_begin_0, end = var_2872_end_0, end_mask = var_2872_end_mask_0, squeeze_mask = var_2872_squeeze_mask_0, x = var_2869_cast_fp16)[name = tensor<string, []>("op_2872_cast_fp16")]; |
| tensor<int32, [4]> var_2887_begin_0 = const()[name = tensor<string, []>("op_2887_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [4]> var_2887_end_0 = const()[name = tensor<string, []>("op_2887_end_0"), val = tensor<int32, [4]>([1, 1, 1, 1500])]; |
| tensor<bool, [4]> var_2887_end_mask_0 = const()[name = tensor<string, []>("op_2887_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])]; |
| tensor<fp16, [1, 1, 1, 1500]> var_2887_cast_fp16 = slice_by_index(begin = var_2887_begin_0, end = var_2887_end_0, end_mask = var_2887_end_mask_0, x = obj_139_cast_fp16)[name = tensor<string, []>("op_2887_cast_fp16")]; |
| tensor<int32, [4]> var_2890_begin_0 = const()[name = tensor<string, []>("op_2890_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [4]> var_2890_end_0 = const()[name = tensor<string, []>("op_2890_end_0"), val = tensor<int32, [4]>([1, 1, 1, 1500])]; |
| tensor<bool, [4]> var_2890_end_mask_0 = const()[name = tensor<string, []>("op_2890_end_mask_0"), val = tensor<bool, [4]>([true, true, false, true])]; |
| tensor<bool, [4]> var_2890_squeeze_mask_0 = const()[name = tensor<string, []>("op_2890_squeeze_mask_0"), val = tensor<bool, [4]>([false, false, true, false])]; |
| tensor<fp16, [1, 1, 1500]> var_2890_cast_fp16 = slice_by_index(begin = var_2890_begin_0, end = var_2890_end_0, end_mask = var_2890_end_mask_0, squeeze_mask = var_2890_squeeze_mask_0, x = var_2887_cast_fp16)[name = tensor<string, []>("op_2890_cast_fp16")]; |
| tensor<int32, [4]> var_2905_begin_0 = const()[name = tensor<string, []>("op_2905_begin_0"), val = tensor<int32, [4]>([0, 7, 0, 0])]; |
| tensor<int32, [4]> var_2905_end_0 = const()[name = tensor<string, []>("op_2905_end_0"), val = tensor<int32, [4]>([1, 8, 1, 1500])]; |
| tensor<bool, [4]> var_2905_end_mask_0 = const()[name = tensor<string, []>("op_2905_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])]; |
| tensor<fp16, [1, 1, 1, 1500]> var_2905_cast_fp16 = slice_by_index(begin = var_2905_begin_0, end = var_2905_end_0, end_mask = var_2905_end_mask_0, x = obj_139_cast_fp16)[name = tensor<string, []>("op_2905_cast_fp16")]; |
| tensor<int32, [4]> var_2908_begin_0 = const()[name = tensor<string, []>("op_2908_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [4]> var_2908_end_0 = const()[name = tensor<string, []>("op_2908_end_0"), val = tensor<int32, [4]>([1, 1, 1, 1500])]; |
| tensor<bool, [4]> var_2908_end_mask_0 = const()[name = tensor<string, []>("op_2908_end_mask_0"), val = tensor<bool, [4]>([true, true, false, true])]; |
| tensor<bool, [4]> var_2908_squeeze_mask_0 = const()[name = tensor<string, []>("op_2908_squeeze_mask_0"), val = tensor<bool, [4]>([false, false, true, false])]; |
| tensor<fp16, [1, 1, 1500]> var_2908_cast_fp16 = slice_by_index(begin = var_2908_begin_0, end = var_2908_end_0, end_mask = var_2908_end_mask_0, squeeze_mask = var_2908_squeeze_mask_0, x = var_2905_cast_fp16)[name = tensor<string, []>("op_2908_cast_fp16")]; |
| tensor<int32, [4]> var_2923_begin_0 = const()[name = tensor<string, []>("op_2923_begin_0"), val = tensor<int32, [4]>([0, 9, 0, 0])]; |
| tensor<int32, [4]> var_2923_end_0 = const()[name = tensor<string, []>("op_2923_end_0"), val = tensor<int32, [4]>([1, 10, 1, 1500])]; |
| tensor<bool, [4]> var_2923_end_mask_0 = const()[name = tensor<string, []>("op_2923_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])]; |
| tensor<fp16, [1, 1, 1, 1500]> var_2923_cast_fp16 = slice_by_index(begin = var_2923_begin_0, end = var_2923_end_0, end_mask = var_2923_end_mask_0, x = obj_139_cast_fp16)[name = tensor<string, []>("op_2923_cast_fp16")]; |
| tensor<int32, [4]> var_2926_begin_0 = const()[name = tensor<string, []>("op_2926_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [4]> var_2926_end_0 = const()[name = tensor<string, []>("op_2926_end_0"), val = tensor<int32, [4]>([1, 1, 1, 1500])]; |
| tensor<bool, [4]> var_2926_end_mask_0 = const()[name = tensor<string, []>("op_2926_end_mask_0"), val = tensor<bool, [4]>([true, true, false, true])]; |
| tensor<bool, [4]> var_2926_squeeze_mask_0 = const()[name = tensor<string, []>("op_2926_squeeze_mask_0"), val = tensor<bool, [4]>([false, false, true, false])]; |
| tensor<fp16, [1, 1, 1500]> var_2926_cast_fp16 = slice_by_index(begin = var_2926_begin_0, end = var_2926_end_0, end_mask = var_2926_end_mask_0, squeeze_mask = var_2926_squeeze_mask_0, x = var_2923_cast_fp16)[name = tensor<string, []>("op_2926_cast_fp16")]; |
| tensor<int32, [4]> var_2941_begin_0 = const()[name = tensor<string, []>("op_2941_begin_0"), val = tensor<int32, [4]>([0, 5, 0, 0])]; |
| tensor<int32, [4]> var_2941_end_0 = const()[name = tensor<string, []>("op_2941_end_0"), val = tensor<int32, [4]>([1, 6, 1, 1500])]; |
| tensor<bool, [4]> var_2941_end_mask_0 = const()[name = tensor<string, []>("op_2941_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])]; |
| tensor<fp16, [1, 1, 1, 1500]> var_2941_cast_fp16 = slice_by_index(begin = var_2941_begin_0, end = var_2941_end_0, end_mask = var_2941_end_mask_0, x = obj_153_cast_fp16)[name = tensor<string, []>("op_2941_cast_fp16")]; |
| tensor<int32, [4]> var_2944_begin_0 = const()[name = tensor<string, []>("op_2944_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [4]> var_2944_end_0 = const()[name = tensor<string, []>("op_2944_end_0"), val = tensor<int32, [4]>([1, 1, 1, 1500])]; |
| tensor<bool, [4]> var_2944_end_mask_0 = const()[name = tensor<string, []>("op_2944_end_mask_0"), val = tensor<bool, [4]>([true, true, false, true])]; |
| tensor<bool, [4]> var_2944_squeeze_mask_0 = const()[name = tensor<string, []>("op_2944_squeeze_mask_0"), val = tensor<bool, [4]>([false, false, true, false])]; |
| tensor<fp16, [1, 1, 1500]> var_2944_cast_fp16 = slice_by_index(begin = var_2944_begin_0, end = var_2944_end_0, end_mask = var_2944_end_mask_0, squeeze_mask = var_2944_squeeze_mask_0, x = var_2941_cast_fp16)[name = tensor<string, []>("op_2944_cast_fp16")]; |
| tensor<int32, []> var_2951 = const()[name = tensor<string, []>("op_2951"), val = tensor<int32, []>(1)]; |
| tensor<bool, []> var_2952_interleave_0 = const()[name = tensor<string, []>("op_2952_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [1, 10, 1500]> var_2952_cast_fp16 = concat(axis = var_2951, interleave = var_2952_interleave_0, values = (var_2782_cast_fp16, var_2800_cast_fp16, var_2818_cast_fp16, var_2836_cast_fp16, var_2854_cast_fp16, var_2872_cast_fp16, var_2890_cast_fp16, var_2908_cast_fp16, var_2926_cast_fp16, var_2944_cast_fp16))[name = tensor<string, []>("op_2952_cast_fp16")]; |
| tensor<int32, [1]> obj_axes_0 = const()[name = tensor<string, []>("obj_axes_0"), val = tensor<int32, [1]>([1])]; |
| tensor<bool, []> obj_keep_dims_0 = const()[name = tensor<string, []>("obj_keep_dims_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [1, 1500]> alignment_heads_weights = reduce_mean(axes = obj_axes_0, keep_dims = obj_keep_dims_0, x = var_2952_cast_fp16)[name = tensor<string, []>("obj_cast_fp16")]; |
| } -> (logits, key_cache_updates, value_cache_updates, alignment_heads_weights); |
| } |
