| program(1.0) | |
| [buildInfo = dict<tensor<string, []>, tensor<string, []>>({{"coremlc-component-MIL", "3405.2.1"}, {"coremlc-version", "3404.23.1"}, {"coremltools-component-torch", "2.6.0"}, {"coremltools-source-dialect", "TorchScript"}, {"coremltools-version", "8.3.0"}})] | |
| { | |
| func main<ios16>(tensor<int32, [1]> cache_length, tensor<fp16, [1, 448]> decoder_key_padding_mask, tensor<fp16, [1, 512, 1, 1500]> encoder_output_embeds, tensor<int32, [1]> input_ids, tensor<fp16, [1, 3072, 1, 448]> key_cache, tensor<fp16, [1, 448]> kv_cache_update_mask, tensor<fp16, [1, 3072, 1, 448]> value_cache) { | |
| tensor<int32, []> var_28_axis_0 = const()[name = tensor<string, []>("op_28_axis_0"), val = tensor<int32, []>(0)]; | |
| tensor<int32, []> var_28_batch_dims_0 = const()[name = tensor<string, []>("op_28_batch_dims_0"), val = tensor<int32, []>(0)]; | |
| tensor<fp16, [51865, 512]> embed_tokens_weight_to_fp16 = const()[name = tensor<string, []>("embed_tokens_weight_to_fp16"), val = tensor<fp16, [51865, 512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(64)))]; | |
| tensor<fp16, [1, 512]> var_28_cast_fp16 = gather(axis = var_28_axis_0, batch_dims = var_28_batch_dims_0, indices = input_ids, x = embed_tokens_weight_to_fp16)[name = tensor<string, []>("op_28_cast_fp16")]; | |
| tensor<int32, []> var_32_axis_0 = const()[name = tensor<string, []>("op_32_axis_0"), val = tensor<int32, []>(0)]; | |
| tensor<int32, []> var_32_batch_dims_0 = const()[name = tensor<string, []>("op_32_batch_dims_0"), val = tensor<int32, []>(0)]; | |
| tensor<fp16, [448, 512]> embed_positions_weight_to_fp16 = const()[name = tensor<string, []>("embed_positions_weight_to_fp16"), val = tensor<fp16, [448, 512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(53109888)))]; | |
| tensor<fp16, [1, 512]> var_32_cast_fp16 = gather(axis = var_32_axis_0, batch_dims = var_32_batch_dims_0, indices = cache_length, x = embed_positions_weight_to_fp16)[name = tensor<string, []>("op_32_cast_fp16")]; | |
| tensor<fp16, [1, 512]> hidden_states_1_cast_fp16 = add(x = var_28_cast_fp16, y = var_32_cast_fp16)[name = tensor<string, []>("hidden_states_1_cast_fp16")]; | |
| tensor<int32, [1]> var_46_axes_0 = const()[name = tensor<string, []>("op_46_axes_0"), val = tensor<int32, [1]>([2])]; | |
| tensor<fp16, [1, 512, 1]> var_46_cast_fp16 = expand_dims(axes = var_46_axes_0, x = hidden_states_1_cast_fp16)[name = tensor<string, []>("op_46_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, 512, 1, 1]> inputs_1_cast_fp16 = expand_dims(axes = inputs_1_axes_0, x = var_46_cast_fp16)[name = tensor<string, []>("inputs_1_cast_fp16")]; | |
| tensor<int32, [6]> tile_0 = const()[name = tensor<string, []>("tile_0"), val = tensor<int32, [6]>([512, 512, 512, 512, 512, 512])]; | |
| tensor<int32, []> var_51_axis_0 = const()[name = tensor<string, []>("op_51_axis_0"), val = tensor<int32, []>(1)]; | |
| tensor<fp16, [1, 512, 1, 448]> var_51_cast_fp16_0, tensor<fp16, [1, 512, 1, 448]> var_51_cast_fp16_1, tensor<fp16, [1, 512, 1, 448]> var_51_cast_fp16_2, tensor<fp16, [1, 512, 1, 448]> var_51_cast_fp16_3, tensor<fp16, [1, 512, 1, 448]> var_51_cast_fp16_4, tensor<fp16, [1, 512, 1, 448]> var_51_cast_fp16_5 = split(axis = var_51_axis_0, split_sizes = tile_0, x = key_cache)[name = tensor<string, []>("op_51_cast_fp16")]; | |
| tensor<int32, [6]> tile_1 = const()[name = tensor<string, []>("tile_1"), val = tensor<int32, [6]>([512, 512, 512, 512, 512, 512])]; | |
| tensor<int32, []> var_60_axis_0 = const()[name = tensor<string, []>("op_60_axis_0"), val = tensor<int32, []>(1)]; | |
| tensor<fp16, [1, 512, 1, 448]> var_60_cast_fp16_0, tensor<fp16, [1, 512, 1, 448]> var_60_cast_fp16_1, tensor<fp16, [1, 512, 1, 448]> var_60_cast_fp16_2, tensor<fp16, [1, 512, 1, 448]> var_60_cast_fp16_3, tensor<fp16, [1, 512, 1, 448]> var_60_cast_fp16_4, tensor<fp16, [1, 512, 1, 448]> var_60_cast_fp16_5 = split(axis = var_60_axis_0, split_sizes = tile_1, x = value_cache)[name = tensor<string, []>("op_60_cast_fp16")]; | |
| tensor<int32, []> var_72 = const()[name = tensor<string, []>("op_72"), 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_97_to_fp16 = const()[name = tensor<string, []>("op_97_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; | |
| tensor<fp16, [1, 512, 1, 1]> out_1_cast_fp16 = layer_norm(axes = out_1_axes_0, epsilon = var_97_to_fp16, x = inputs_1_cast_fp16)[name = tensor<string, []>("out_1_cast_fp16")]; | |
| tensor<fp16, [512]> obj_1_mean_0_to_fp16 = const()[name = tensor<string, []>("obj_1_mean_0_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(53568704)))]; | |
| tensor<fp16, [512]> obj_1_variance_0_to_fp16 = const()[name = tensor<string, []>("obj_1_variance_0_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(53569792)))]; | |
| tensor<fp16, [512]> obj_1_gamma_0_to_fp16 = const()[name = tensor<string, []>("obj_1_gamma_0_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(53570880)))]; | |
| tensor<fp16, [512]> obj_1_beta_0_to_fp16 = const()[name = tensor<string, []>("obj_1_beta_0_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(53571968)))]; | |
| 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, 512, 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, [512, 512, 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, [512, 512, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(53573056)))]; | |
| tensor<fp16, [512]> 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, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(54097408)))]; | |
| tensor<fp16, [1, 512, 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, [512, 512, 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, [512, 512, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(54098496)))]; | |
| tensor<fp16, [1, 512, 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, [512, 512, 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, [512, 512, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(54622848)))]; | |
| tensor<fp16, [512]> 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, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(55147200)))]; | |
| tensor<fp16, [1, 512, 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_132_axes_0 = const()[name = tensor<string, []>("op_132_axes_0"), val = tensor<int32, [1]>([1])]; | |
| tensor<fp16, [1, 1, 448]> var_132_cast_fp16 = expand_dims(axes = var_132_axes_0, x = kv_cache_update_mask)[name = tensor<string, []>("op_132_cast_fp16")]; | |
| tensor<int32, [1]> var_133_axes_0 = const()[name = tensor<string, []>("op_133_axes_0"), val = tensor<int32, [1]>([2])]; | |
| tensor<fp16, [1, 1, 1, 448]> var_133_cast_fp16 = expand_dims(axes = var_133_axes_0, x = var_132_cast_fp16)[name = tensor<string, []>("op_133_cast_fp16")]; | |
| tensor<fp16, []> var_73_to_fp16 = const()[name = tensor<string, []>("op_73_to_fp16"), val = tensor<fp16, []>(0x1p+0)]; | |
| tensor<fp16, [1, 1, 1, 448]> var_135_cast_fp16 = sub(x = var_73_to_fp16, y = var_133_cast_fp16)[name = tensor<string, []>("op_135_cast_fp16")]; | |
| tensor<fp16, [1, 512, 1, 448]> var_136_cast_fp16 = mul(x = var_51_cast_fp16_0, y = var_135_cast_fp16)[name = tensor<string, []>("op_136_cast_fp16")]; | |
| tensor<fp16, [1, 512, 1, 448]> var_137_cast_fp16 = mul(x = current_key_1_cast_fp16, y = var_133_cast_fp16)[name = tensor<string, []>("op_137_cast_fp16")]; | |
| tensor<fp16, [1, 512, 1, 448]> key_1_cast_fp16 = add(x = var_136_cast_fp16, y = var_137_cast_fp16)[name = tensor<string, []>("key_1_cast_fp16")]; | |
| tensor<fp16, [1, 512, 1, 448]> var_140_cast_fp16 = mul(x = var_60_cast_fp16_0, y = var_135_cast_fp16)[name = tensor<string, []>("op_140_cast_fp16")]; | |
| tensor<fp16, [1, 512, 1, 448]> var_141_cast_fp16 = mul(x = current_value_1_cast_fp16, y = var_133_cast_fp16)[name = tensor<string, []>("op_141_cast_fp16")]; | |
| tensor<fp16, [1, 512, 1, 448]> value_1_cast_fp16 = add(x = var_140_cast_fp16, y = var_141_cast_fp16)[name = tensor<string, []>("value_1_cast_fp16")]; | |
| tensor<int32, [4]> var_145 = const()[name = tensor<string, []>("op_145"), val = tensor<int32, [4]>([1, 8, 64, 1])]; | |
| tensor<fp16, [1, 8, 64, 1]> mh_q_1_cast_fp16 = reshape(shape = var_145, x = query_1_cast_fp16)[name = tensor<string, []>("mh_q_1_cast_fp16")]; | |
| tensor<fp16, []> var_147_to_fp16 = const()[name = tensor<string, []>("op_147_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; | |
| tensor<fp16, [1, 8, 64, 1]> var_148_cast_fp16 = mul(x = mh_q_1_cast_fp16, y = var_147_to_fp16)[name = tensor<string, []>("op_148_cast_fp16")]; | |
| tensor<int32, [4]> var_151 = const()[name = tensor<string, []>("op_151"), val = tensor<int32, [4]>([1, 8, 64, 448])]; | |
| tensor<fp16, [1, 8, 64, 448]> var_152_cast_fp16 = reshape(shape = var_151, x = key_1_cast_fp16)[name = tensor<string, []>("op_152_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, 8, 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_148_cast_fp16, y = var_152_cast_fp16)[name = tensor<string, []>("mh_w_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 = decoder_key_padding_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, [1, 8, 1, 448]> mh_w_3_cast_fp16 = add(x = mh_w_1_cast_fp16, y = var_157_cast_fp16)[name = tensor<string, []>("mh_w_3_cast_fp16")]; | |
| tensor<fp16, [1, 8, 1, 448]> var_160_cast_fp16 = softmax(axis = var_72, x = mh_w_3_cast_fp16)[name = tensor<string, []>("op_160_cast_fp16")]; | |
| tensor<int32, [4]> var_161 = const()[name = tensor<string, []>("op_161"), val = tensor<int32, [4]>([1, 8, 64, 448])]; | |
| tensor<fp16, [1, 8, 64, 448]> var_162_cast_fp16 = reshape(shape = var_161, x = value_1_cast_fp16)[name = tensor<string, []>("op_162_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, 8, 64, 1]> attn_1_cast_fp16 = matmul(transpose_x = attn_1_transpose_x_0, transpose_y = attn_1_transpose_y_0, x = var_162_cast_fp16, y = var_160_cast_fp16)[name = tensor<string, []>("attn_1_cast_fp16")]; | |
| tensor<int32, [4]> var_165 = const()[name = tensor<string, []>("op_165"), val = tensor<int32, [4]>([1, 512, 1, 1])]; | |
| tensor<fp16, [1, 512, 1, 1]> input_1_cast_fp16 = reshape(shape = var_165, 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, [512, 512, 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, [512, 512, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(55148288)))]; | |
| tensor<fp16, [512]> 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, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(55672640)))]; | |
| tensor<fp16, [1, 512, 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, 512, 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_187_to_fp16 = const()[name = tensor<string, []>("op_187_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; | |
| tensor<fp16, [1, 512, 1, 1]> out_3_cast_fp16 = layer_norm(axes = out_3_axes_0, epsilon = var_187_to_fp16, x = inputs_3_cast_fp16)[name = tensor<string, []>("out_3_cast_fp16")]; | |
| tensor<fp16, [512]> obj_9_gamma_0_to_fp16 = const()[name = tensor<string, []>("obj_9_gamma_0_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(55673728)))]; | |
| tensor<fp16, [512]> obj_9_beta_0_to_fp16 = const()[name = tensor<string, []>("obj_9_beta_0_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(55674816)))]; | |
| 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, 512, 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, [512, 512, 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, [512, 512, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(55675904)))]; | |
| tensor<fp16, [512]> 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, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(56200256)))]; | |
| tensor<fp16, [1, 512, 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, [512, 512, 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, [512, 512, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(56201344)))]; | |
| tensor<fp16, [1, 512, 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, [512, 512, 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, [512, 512, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(56725696)))]; | |
| tensor<fp16, [512]> 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, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(57250048)))]; | |
| tensor<fp16, [1, 512, 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_223 = const()[name = tensor<string, []>("op_223"), val = tensor<int32, [4]>([1, 8, 64, 1])]; | |
| tensor<fp16, [1, 8, 64, 1]> mh_q_3_cast_fp16 = reshape(shape = var_223, x = query_3_cast_fp16)[name = tensor<string, []>("mh_q_3_cast_fp16")]; | |
| tensor<fp16, []> var_225_to_fp16 = const()[name = tensor<string, []>("op_225_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; | |
| tensor<fp16, [1, 8, 64, 1]> var_226_cast_fp16 = mul(x = mh_q_3_cast_fp16, y = var_225_to_fp16)[name = tensor<string, []>("op_226_cast_fp16")]; | |
| tensor<int32, [4]> var_229 = const()[name = tensor<string, []>("op_229"), val = tensor<int32, [4]>([1, 8, 64, 1500])]; | |
| tensor<fp16, [1, 8, 64, 1500]> var_230_cast_fp16 = reshape(shape = var_229, x = key_3_cast_fp16)[name = tensor<string, []>("op_230_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, 8, 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_226_cast_fp16, y = var_230_cast_fp16)[name = tensor<string, []>("mh_w_5_cast_fp16")]; | |
| tensor<fp16, [1, 8, 1, 1500]> obj_13_cast_fp16 = softmax(axis = var_72, x = mh_w_5_cast_fp16)[name = tensor<string, []>("obj_13_cast_fp16")]; | |
| tensor<int32, [4]> var_234 = const()[name = tensor<string, []>("op_234"), val = tensor<int32, [4]>([1, 8, 64, 1500])]; | |
| tensor<fp16, [1, 8, 64, 1500]> var_235_cast_fp16 = reshape(shape = var_234, x = value_3_cast_fp16)[name = tensor<string, []>("op_235_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, 8, 64, 1]> attn_3_cast_fp16 = matmul(transpose_x = attn_3_transpose_x_0, transpose_y = attn_3_transpose_y_0, x = var_235_cast_fp16, y = obj_13_cast_fp16)[name = tensor<string, []>("attn_3_cast_fp16")]; | |
| tensor<int32, [4]> var_238 = const()[name = tensor<string, []>("op_238"), val = tensor<int32, [4]>([1, 512, 1, 1])]; | |
| tensor<fp16, [1, 512, 1, 1]> input_3_cast_fp16 = reshape(shape = var_238, 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, [512, 512, 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, [512, 512, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(57251136)))]; | |
| tensor<fp16, [512]> 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, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(57775488)))]; | |
| tensor<fp16, [1, 512, 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, 512, 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_256_to_fp16 = const()[name = tensor<string, []>("op_256_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; | |
| tensor<fp16, [1, 512, 1, 1]> out_5_cast_fp16 = layer_norm(axes = out_5_axes_0, epsilon = var_256_to_fp16, x = inputs_5_cast_fp16)[name = tensor<string, []>("out_5_cast_fp16")]; | |
| tensor<fp16, [512]> input_5_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_5_gamma_0_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(57776576)))]; | |
| tensor<fp16, [512]> input_5_beta_0_to_fp16 = const()[name = tensor<string, []>("input_5_beta_0_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(57777664)))]; | |
| 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, 512, 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, [2048, 512, 1, 1]> layers_0_fc1_weight_to_fp16 = const()[name = tensor<string, []>("layers_0_fc1_weight_to_fp16"), val = tensor<fp16, [2048, 512, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(57778752)))]; | |
| tensor<fp16, [2048]> layers_0_fc1_bias_to_fp16 = const()[name = tensor<string, []>("layers_0_fc1_bias_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(59875968)))]; | |
| tensor<fp16, [1, 2048, 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, 2048, 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, [512, 2048, 1, 1]> layers_0_fc2_weight_to_fp16 = const()[name = tensor<string, []>("layers_0_fc2_weight_to_fp16"), val = tensor<fp16, [512, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(59880128)))]; | |
| tensor<fp16, [512]> layers_0_fc2_bias_to_fp16 = const()[name = tensor<string, []>("layers_0_fc2_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(61977344)))]; | |
| tensor<fp16, [1, 512, 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, 512, 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_291 = const()[name = tensor<string, []>("op_291"), 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_316_to_fp16 = const()[name = tensor<string, []>("op_316_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; | |
| tensor<fp16, [1, 512, 1, 1]> out_7_cast_fp16 = layer_norm(axes = out_7_axes_0, epsilon = var_316_to_fp16, x = inputs_7_cast_fp16)[name = tensor<string, []>("out_7_cast_fp16")]; | |
| tensor<fp16, [512]> obj_15_gamma_0_to_fp16 = const()[name = tensor<string, []>("obj_15_gamma_0_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(61978432)))]; | |
| tensor<fp16, [512]> obj_15_beta_0_to_fp16 = const()[name = tensor<string, []>("obj_15_beta_0_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(61979520)))]; | |
| 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, 512, 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, [512, 512, 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, [512, 512, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(61980608)))]; | |
| tensor<fp16, [512]> 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, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(62504960)))]; | |
| tensor<fp16, [1, 512, 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, [512, 512, 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, [512, 512, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(62506048)))]; | |
| tensor<fp16, [1, 512, 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, [512, 512, 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, [512, 512, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(63030400)))]; | |
| tensor<fp16, [512]> 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, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(63554752)))]; | |
| tensor<fp16, [1, 512, 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, 512, 1, 448]> var_355_cast_fp16 = mul(x = var_51_cast_fp16_1, y = var_135_cast_fp16)[name = tensor<string, []>("op_355_cast_fp16")]; | |
| tensor<fp16, [1, 512, 1, 448]> var_356_cast_fp16 = mul(x = current_key_3_cast_fp16, y = var_133_cast_fp16)[name = tensor<string, []>("op_356_cast_fp16")]; | |
| tensor<fp16, [1, 512, 1, 448]> key_5_cast_fp16 = add(x = var_355_cast_fp16, y = var_356_cast_fp16)[name = tensor<string, []>("key_5_cast_fp16")]; | |
| tensor<fp16, [1, 512, 1, 448]> var_359_cast_fp16 = mul(x = var_60_cast_fp16_1, y = var_135_cast_fp16)[name = tensor<string, []>("op_359_cast_fp16")]; | |
| tensor<fp16, [1, 512, 1, 448]> var_360_cast_fp16 = mul(x = current_value_3_cast_fp16, y = var_133_cast_fp16)[name = tensor<string, []>("op_360_cast_fp16")]; | |
| tensor<fp16, [1, 512, 1, 448]> value_5_cast_fp16 = add(x = var_359_cast_fp16, y = var_360_cast_fp16)[name = tensor<string, []>("value_5_cast_fp16")]; | |
| tensor<int32, [4]> var_364 = const()[name = tensor<string, []>("op_364"), val = tensor<int32, [4]>([1, 8, 64, 1])]; | |
| tensor<fp16, [1, 8, 64, 1]> mh_q_5_cast_fp16 = reshape(shape = var_364, x = query_5_cast_fp16)[name = tensor<string, []>("mh_q_5_cast_fp16")]; | |
| tensor<fp16, []> var_366_to_fp16 = const()[name = tensor<string, []>("op_366_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; | |
| tensor<fp16, [1, 8, 64, 1]> var_367_cast_fp16 = mul(x = mh_q_5_cast_fp16, y = var_366_to_fp16)[name = tensor<string, []>("op_367_cast_fp16")]; | |
| tensor<int32, [4]> var_370 = const()[name = tensor<string, []>("op_370"), val = tensor<int32, [4]>([1, 8, 64, 448])]; | |
| tensor<fp16, [1, 8, 64, 448]> var_371_cast_fp16 = reshape(shape = var_370, x = key_5_cast_fp16)[name = tensor<string, []>("op_371_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, 8, 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_367_cast_fp16, y = var_371_cast_fp16)[name = tensor<string, []>("mh_w_7_cast_fp16")]; | |
| tensor<fp16, [1, 8, 1, 448]> mh_w_9_cast_fp16 = add(x = mh_w_7_cast_fp16, y = var_157_cast_fp16)[name = tensor<string, []>("mh_w_9_cast_fp16")]; | |
| tensor<fp16, [1, 8, 1, 448]> var_379_cast_fp16 = softmax(axis = var_291, x = mh_w_9_cast_fp16)[name = tensor<string, []>("op_379_cast_fp16")]; | |
| tensor<int32, [4]> var_380 = const()[name = tensor<string, []>("op_380"), val = tensor<int32, [4]>([1, 8, 64, 448])]; | |
| tensor<fp16, [1, 8, 64, 448]> var_381_cast_fp16 = reshape(shape = var_380, x = value_5_cast_fp16)[name = tensor<string, []>("op_381_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, 8, 64, 1]> attn_5_cast_fp16 = matmul(transpose_x = attn_5_transpose_x_0, transpose_y = attn_5_transpose_y_0, x = var_381_cast_fp16, y = var_379_cast_fp16)[name = tensor<string, []>("attn_5_cast_fp16")]; | |
| tensor<int32, [4]> var_384 = const()[name = tensor<string, []>("op_384"), val = tensor<int32, [4]>([1, 512, 1, 1])]; | |
| tensor<fp16, [1, 512, 1, 1]> input_11_cast_fp16 = reshape(shape = var_384, 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, [512, 512, 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, [512, 512, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(63555840)))]; | |
| tensor<fp16, [512]> 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, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(64080192)))]; | |
| tensor<fp16, [1, 512, 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, 512, 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_406_to_fp16 = const()[name = tensor<string, []>("op_406_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; | |
| tensor<fp16, [1, 512, 1, 1]> out_9_cast_fp16 = layer_norm(axes = out_9_axes_0, epsilon = var_406_to_fp16, x = inputs_9_cast_fp16)[name = tensor<string, []>("out_9_cast_fp16")]; | |
| tensor<fp16, [512]> obj_23_gamma_0_to_fp16 = const()[name = tensor<string, []>("obj_23_gamma_0_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(64081280)))]; | |
| tensor<fp16, [512]> obj_23_beta_0_to_fp16 = const()[name = tensor<string, []>("obj_23_beta_0_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(64082368)))]; | |
| 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, 512, 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, [512, 512, 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, [512, 512, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(64083456)))]; | |
| tensor<fp16, [512]> 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, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(64607808)))]; | |
| tensor<fp16, [1, 512, 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, [512, 512, 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, [512, 512, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(64608896)))]; | |
| tensor<fp16, [1, 512, 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, [512, 512, 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, [512, 512, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(65133248)))]; | |
| tensor<fp16, [512]> 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, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(65657600)))]; | |
| tensor<fp16, [1, 512, 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_442 = const()[name = tensor<string, []>("op_442"), val = tensor<int32, [4]>([1, 8, 64, 1])]; | |
| tensor<fp16, [1, 8, 64, 1]> mh_q_7_cast_fp16 = reshape(shape = var_442, x = query_7_cast_fp16)[name = tensor<string, []>("mh_q_7_cast_fp16")]; | |
| tensor<fp16, []> var_444_to_fp16 = const()[name = tensor<string, []>("op_444_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; | |
| tensor<fp16, [1, 8, 64, 1]> var_445_cast_fp16 = mul(x = mh_q_7_cast_fp16, y = var_444_to_fp16)[name = tensor<string, []>("op_445_cast_fp16")]; | |
| tensor<int32, [4]> var_448 = const()[name = tensor<string, []>("op_448"), val = tensor<int32, [4]>([1, 8, 64, 1500])]; | |
| tensor<fp16, [1, 8, 64, 1500]> var_449_cast_fp16 = reshape(shape = var_448, x = key_7_cast_fp16)[name = tensor<string, []>("op_449_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, 8, 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_445_cast_fp16, y = var_449_cast_fp16)[name = tensor<string, []>("mh_w_11_cast_fp16")]; | |
| tensor<fp16, [1, 8, 1, 1500]> obj_27_cast_fp16 = softmax(axis = var_291, x = mh_w_11_cast_fp16)[name = tensor<string, []>("obj_27_cast_fp16")]; | |
| tensor<int32, [4]> var_453 = const()[name = tensor<string, []>("op_453"), val = tensor<int32, [4]>([1, 8, 64, 1500])]; | |
| tensor<fp16, [1, 8, 64, 1500]> var_454_cast_fp16 = reshape(shape = var_453, x = value_7_cast_fp16)[name = tensor<string, []>("op_454_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, 8, 64, 1]> attn_7_cast_fp16 = matmul(transpose_x = attn_7_transpose_x_0, transpose_y = attn_7_transpose_y_0, x = var_454_cast_fp16, y = obj_27_cast_fp16)[name = tensor<string, []>("attn_7_cast_fp16")]; | |
| tensor<int32, [4]> var_457 = const()[name = tensor<string, []>("op_457"), val = tensor<int32, [4]>([1, 512, 1, 1])]; | |
| tensor<fp16, [1, 512, 1, 1]> input_13_cast_fp16 = reshape(shape = var_457, 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, [512, 512, 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, [512, 512, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(65658688)))]; | |
| tensor<fp16, [512]> 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, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(66183040)))]; | |
| tensor<fp16, [1, 512, 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, 512, 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_475_to_fp16 = const()[name = tensor<string, []>("op_475_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; | |
| tensor<fp16, [1, 512, 1, 1]> out_11_cast_fp16 = layer_norm(axes = out_11_axes_0, epsilon = var_475_to_fp16, x = inputs_11_cast_fp16)[name = tensor<string, []>("out_11_cast_fp16")]; | |
| tensor<fp16, [512]> input_15_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_15_gamma_0_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(66184128)))]; | |
| tensor<fp16, [512]> input_15_beta_0_to_fp16 = const()[name = tensor<string, []>("input_15_beta_0_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(66185216)))]; | |
| 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, 512, 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, [2048, 512, 1, 1]> layers_1_fc1_weight_to_fp16 = const()[name = tensor<string, []>("layers_1_fc1_weight_to_fp16"), val = tensor<fp16, [2048, 512, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(66186304)))]; | |
| tensor<fp16, [2048]> layers_1_fc1_bias_to_fp16 = const()[name = tensor<string, []>("layers_1_fc1_bias_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(68283520)))]; | |
| tensor<fp16, [1, 2048, 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, 2048, 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, [512, 2048, 1, 1]> layers_1_fc2_weight_to_fp16 = const()[name = tensor<string, []>("layers_1_fc2_weight_to_fp16"), val = tensor<fp16, [512, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(68287680)))]; | |
| tensor<fp16, [512]> layers_1_fc2_bias_to_fp16 = const()[name = tensor<string, []>("layers_1_fc2_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(70384896)))]; | |
| tensor<fp16, [1, 512, 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, 512, 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_510 = const()[name = tensor<string, []>("op_510"), 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_535_to_fp16 = const()[name = tensor<string, []>("op_535_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; | |
| tensor<fp16, [1, 512, 1, 1]> out_13_cast_fp16 = layer_norm(axes = out_13_axes_0, epsilon = var_535_to_fp16, x = inputs_13_cast_fp16)[name = tensor<string, []>("out_13_cast_fp16")]; | |
| tensor<fp16, [512]> obj_29_gamma_0_to_fp16 = const()[name = tensor<string, []>("obj_29_gamma_0_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(70385984)))]; | |
| tensor<fp16, [512]> obj_29_beta_0_to_fp16 = const()[name = tensor<string, []>("obj_29_beta_0_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(70387072)))]; | |
| 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, 512, 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, [512, 512, 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, [512, 512, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(70388160)))]; | |
| tensor<fp16, [512]> 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, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(70912512)))]; | |
| tensor<fp16, [1, 512, 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, [512, 512, 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, [512, 512, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(70913600)))]; | |
| tensor<fp16, [1, 512, 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, [512, 512, 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, [512, 512, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(71437952)))]; | |
| tensor<fp16, [512]> 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, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(71962304)))]; | |
| tensor<fp16, [1, 512, 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, 512, 1, 448]> var_574_cast_fp16 = mul(x = var_51_cast_fp16_2, y = var_135_cast_fp16)[name = tensor<string, []>("op_574_cast_fp16")]; | |
| tensor<fp16, [1, 512, 1, 448]> var_575_cast_fp16 = mul(x = current_key_5_cast_fp16, y = var_133_cast_fp16)[name = tensor<string, []>("op_575_cast_fp16")]; | |
| tensor<fp16, [1, 512, 1, 448]> key_9_cast_fp16 = add(x = var_574_cast_fp16, y = var_575_cast_fp16)[name = tensor<string, []>("key_9_cast_fp16")]; | |
| tensor<fp16, [1, 512, 1, 448]> var_578_cast_fp16 = mul(x = var_60_cast_fp16_2, y = var_135_cast_fp16)[name = tensor<string, []>("op_578_cast_fp16")]; | |
| tensor<fp16, [1, 512, 1, 448]> var_579_cast_fp16 = mul(x = current_value_5_cast_fp16, y = var_133_cast_fp16)[name = tensor<string, []>("op_579_cast_fp16")]; | |
| tensor<fp16, [1, 512, 1, 448]> value_9_cast_fp16 = add(x = var_578_cast_fp16, y = var_579_cast_fp16)[name = tensor<string, []>("value_9_cast_fp16")]; | |
| tensor<int32, [4]> var_583 = const()[name = tensor<string, []>("op_583"), val = tensor<int32, [4]>([1, 8, 64, 1])]; | |
| tensor<fp16, [1, 8, 64, 1]> mh_q_9_cast_fp16 = reshape(shape = var_583, x = query_9_cast_fp16)[name = tensor<string, []>("mh_q_9_cast_fp16")]; | |
| tensor<fp16, []> var_585_to_fp16 = const()[name = tensor<string, []>("op_585_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; | |
| tensor<fp16, [1, 8, 64, 1]> var_586_cast_fp16 = mul(x = mh_q_9_cast_fp16, y = var_585_to_fp16)[name = tensor<string, []>("op_586_cast_fp16")]; | |
| tensor<int32, [4]> var_589 = const()[name = tensor<string, []>("op_589"), val = tensor<int32, [4]>([1, 8, 64, 448])]; | |
| tensor<fp16, [1, 8, 64, 448]> var_590_cast_fp16 = reshape(shape = var_589, x = key_9_cast_fp16)[name = tensor<string, []>("op_590_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, 8, 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_586_cast_fp16, y = var_590_cast_fp16)[name = tensor<string, []>("mh_w_13_cast_fp16")]; | |
| tensor<fp16, [1, 8, 1, 448]> mh_w_15_cast_fp16 = add(x = mh_w_13_cast_fp16, y = var_157_cast_fp16)[name = tensor<string, []>("mh_w_15_cast_fp16")]; | |
| tensor<fp16, [1, 8, 1, 448]> var_598_cast_fp16 = softmax(axis = var_510, x = mh_w_15_cast_fp16)[name = tensor<string, []>("op_598_cast_fp16")]; | |
| tensor<int32, [4]> var_599 = const()[name = tensor<string, []>("op_599"), val = tensor<int32, [4]>([1, 8, 64, 448])]; | |
| tensor<fp16, [1, 8, 64, 448]> var_600_cast_fp16 = reshape(shape = var_599, x = value_9_cast_fp16)[name = tensor<string, []>("op_600_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, 8, 64, 1]> attn_9_cast_fp16 = matmul(transpose_x = attn_9_transpose_x_0, transpose_y = attn_9_transpose_y_0, x = var_600_cast_fp16, y = var_598_cast_fp16)[name = tensor<string, []>("attn_9_cast_fp16")]; | |
| tensor<int32, [4]> var_603 = const()[name = tensor<string, []>("op_603"), val = tensor<int32, [4]>([1, 512, 1, 1])]; | |
| tensor<fp16, [1, 512, 1, 1]> input_21_cast_fp16 = reshape(shape = var_603, 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, [512, 512, 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, [512, 512, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(71963392)))]; | |
| tensor<fp16, [512]> 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, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(72487744)))]; | |
| tensor<fp16, [1, 512, 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, 512, 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_625_to_fp16 = const()[name = tensor<string, []>("op_625_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; | |
| tensor<fp16, [1, 512, 1, 1]> out_15_cast_fp16 = layer_norm(axes = out_15_axes_0, epsilon = var_625_to_fp16, x = inputs_15_cast_fp16)[name = tensor<string, []>("out_15_cast_fp16")]; | |
| tensor<fp16, [512]> obj_37_gamma_0_to_fp16 = const()[name = tensor<string, []>("obj_37_gamma_0_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(72488832)))]; | |
| tensor<fp16, [512]> obj_37_beta_0_to_fp16 = const()[name = tensor<string, []>("obj_37_beta_0_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(72489920)))]; | |
| 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, 512, 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, [512, 512, 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, [512, 512, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(72491008)))]; | |
| tensor<fp16, [512]> 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, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(73015360)))]; | |
| tensor<fp16, [1, 512, 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, [512, 512, 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, [512, 512, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(73016448)))]; | |
| tensor<fp16, [1, 512, 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, [512, 512, 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, [512, 512, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(73540800)))]; | |
| tensor<fp16, [512]> 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, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(74065152)))]; | |
| tensor<fp16, [1, 512, 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_661 = const()[name = tensor<string, []>("op_661"), val = tensor<int32, [4]>([1, 8, 64, 1])]; | |
| tensor<fp16, [1, 8, 64, 1]> mh_q_11_cast_fp16 = reshape(shape = var_661, x = query_11_cast_fp16)[name = tensor<string, []>("mh_q_11_cast_fp16")]; | |
| tensor<fp16, []> var_663_to_fp16 = const()[name = tensor<string, []>("op_663_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; | |
| tensor<fp16, [1, 8, 64, 1]> var_664_cast_fp16 = mul(x = mh_q_11_cast_fp16, y = var_663_to_fp16)[name = tensor<string, []>("op_664_cast_fp16")]; | |
| tensor<int32, [4]> var_667 = const()[name = tensor<string, []>("op_667"), val = tensor<int32, [4]>([1, 8, 64, 1500])]; | |
| tensor<fp16, [1, 8, 64, 1500]> var_668_cast_fp16 = reshape(shape = var_667, x = key_11_cast_fp16)[name = tensor<string, []>("op_668_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, 8, 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_664_cast_fp16, y = var_668_cast_fp16)[name = tensor<string, []>("mh_w_17_cast_fp16")]; | |
| tensor<fp16, [1, 8, 1, 1500]> obj_41_cast_fp16 = softmax(axis = var_510, x = mh_w_17_cast_fp16)[name = tensor<string, []>("obj_41_cast_fp16")]; | |
| tensor<int32, [4]> var_672 = const()[name = tensor<string, []>("op_672"), val = tensor<int32, [4]>([1, 8, 64, 1500])]; | |
| tensor<fp16, [1, 8, 64, 1500]> var_673_cast_fp16 = reshape(shape = var_672, x = value_11_cast_fp16)[name = tensor<string, []>("op_673_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, 8, 64, 1]> attn_11_cast_fp16 = matmul(transpose_x = attn_11_transpose_x_0, transpose_y = attn_11_transpose_y_0, x = var_673_cast_fp16, y = obj_41_cast_fp16)[name = tensor<string, []>("attn_11_cast_fp16")]; | |
| tensor<int32, [4]> var_676 = const()[name = tensor<string, []>("op_676"), val = tensor<int32, [4]>([1, 512, 1, 1])]; | |
| tensor<fp16, [1, 512, 1, 1]> input_23_cast_fp16 = reshape(shape = var_676, 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, [512, 512, 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, [512, 512, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(74066240)))]; | |
| tensor<fp16, [512]> 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, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(74590592)))]; | |
| tensor<fp16, [1, 512, 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, 512, 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_694_to_fp16 = const()[name = tensor<string, []>("op_694_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; | |
| tensor<fp16, [1, 512, 1, 1]> out_17_cast_fp16 = layer_norm(axes = out_17_axes_0, epsilon = var_694_to_fp16, x = inputs_17_cast_fp16)[name = tensor<string, []>("out_17_cast_fp16")]; | |
| tensor<fp16, [512]> input_25_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_25_gamma_0_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(74591680)))]; | |
| tensor<fp16, [512]> input_25_beta_0_to_fp16 = const()[name = tensor<string, []>("input_25_beta_0_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(74592768)))]; | |
| 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, 512, 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, [2048, 512, 1, 1]> layers_2_fc1_weight_to_fp16 = const()[name = tensor<string, []>("layers_2_fc1_weight_to_fp16"), val = tensor<fp16, [2048, 512, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(74593856)))]; | |
| tensor<fp16, [2048]> layers_2_fc1_bias_to_fp16 = const()[name = tensor<string, []>("layers_2_fc1_bias_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(76691072)))]; | |
| tensor<fp16, [1, 2048, 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, 2048, 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, [512, 2048, 1, 1]> layers_2_fc2_weight_to_fp16 = const()[name = tensor<string, []>("layers_2_fc2_weight_to_fp16"), val = tensor<fp16, [512, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(76695232)))]; | |
| tensor<fp16, [512]> layers_2_fc2_bias_to_fp16 = const()[name = tensor<string, []>("layers_2_fc2_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(78792448)))]; | |
| tensor<fp16, [1, 512, 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, 512, 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_729 = const()[name = tensor<string, []>("op_729"), 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_754_to_fp16 = const()[name = tensor<string, []>("op_754_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; | |
| tensor<fp16, [1, 512, 1, 1]> out_19_cast_fp16 = layer_norm(axes = out_19_axes_0, epsilon = var_754_to_fp16, x = inputs_19_cast_fp16)[name = tensor<string, []>("out_19_cast_fp16")]; | |
| tensor<fp16, [512]> obj_43_gamma_0_to_fp16 = const()[name = tensor<string, []>("obj_43_gamma_0_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(78793536)))]; | |
| tensor<fp16, [512]> obj_43_beta_0_to_fp16 = const()[name = tensor<string, []>("obj_43_beta_0_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(78794624)))]; | |
| 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, 512, 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, [512, 512, 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, [512, 512, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(78795712)))]; | |
| tensor<fp16, [512]> 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, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(79320064)))]; | |
| tensor<fp16, [1, 512, 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, [512, 512, 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, [512, 512, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(79321152)))]; | |
| tensor<fp16, [1, 512, 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, [512, 512, 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, [512, 512, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(79845504)))]; | |
| tensor<fp16, [512]> 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, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(80369856)))]; | |
| tensor<fp16, [1, 512, 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, 512, 1, 448]> var_793_cast_fp16 = mul(x = var_51_cast_fp16_3, y = var_135_cast_fp16)[name = tensor<string, []>("op_793_cast_fp16")]; | |
| tensor<fp16, [1, 512, 1, 448]> var_794_cast_fp16 = mul(x = current_key_7_cast_fp16, y = var_133_cast_fp16)[name = tensor<string, []>("op_794_cast_fp16")]; | |
| tensor<fp16, [1, 512, 1, 448]> key_13_cast_fp16 = add(x = var_793_cast_fp16, y = var_794_cast_fp16)[name = tensor<string, []>("key_13_cast_fp16")]; | |
| tensor<fp16, [1, 512, 1, 448]> var_797_cast_fp16 = mul(x = var_60_cast_fp16_3, y = var_135_cast_fp16)[name = tensor<string, []>("op_797_cast_fp16")]; | |
| tensor<fp16, [1, 512, 1, 448]> var_798_cast_fp16 = mul(x = current_value_7_cast_fp16, y = var_133_cast_fp16)[name = tensor<string, []>("op_798_cast_fp16")]; | |
| tensor<fp16, [1, 512, 1, 448]> value_13_cast_fp16 = add(x = var_797_cast_fp16, y = var_798_cast_fp16)[name = tensor<string, []>("value_13_cast_fp16")]; | |
| tensor<int32, [4]> var_802 = const()[name = tensor<string, []>("op_802"), val = tensor<int32, [4]>([1, 8, 64, 1])]; | |
| tensor<fp16, [1, 8, 64, 1]> mh_q_13_cast_fp16 = reshape(shape = var_802, x = query_13_cast_fp16)[name = tensor<string, []>("mh_q_13_cast_fp16")]; | |
| tensor<fp16, []> var_804_to_fp16 = const()[name = tensor<string, []>("op_804_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; | |
| tensor<fp16, [1, 8, 64, 1]> var_805_cast_fp16 = mul(x = mh_q_13_cast_fp16, y = var_804_to_fp16)[name = tensor<string, []>("op_805_cast_fp16")]; | |
| tensor<int32, [4]> var_808 = const()[name = tensor<string, []>("op_808"), val = tensor<int32, [4]>([1, 8, 64, 448])]; | |
| tensor<fp16, [1, 8, 64, 448]> var_809_cast_fp16 = reshape(shape = var_808, x = key_13_cast_fp16)[name = tensor<string, []>("op_809_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, 8, 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_805_cast_fp16, y = var_809_cast_fp16)[name = tensor<string, []>("mh_w_19_cast_fp16")]; | |
| tensor<fp16, [1, 8, 1, 448]> mh_w_21_cast_fp16 = add(x = mh_w_19_cast_fp16, y = var_157_cast_fp16)[name = tensor<string, []>("mh_w_21_cast_fp16")]; | |
| tensor<fp16, [1, 8, 1, 448]> var_817_cast_fp16 = softmax(axis = var_729, x = mh_w_21_cast_fp16)[name = tensor<string, []>("op_817_cast_fp16")]; | |
| tensor<int32, [4]> var_818 = const()[name = tensor<string, []>("op_818"), val = tensor<int32, [4]>([1, 8, 64, 448])]; | |
| tensor<fp16, [1, 8, 64, 448]> var_819_cast_fp16 = reshape(shape = var_818, x = value_13_cast_fp16)[name = tensor<string, []>("op_819_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, 8, 64, 1]> attn_13_cast_fp16 = matmul(transpose_x = attn_13_transpose_x_0, transpose_y = attn_13_transpose_y_0, x = var_819_cast_fp16, y = var_817_cast_fp16)[name = tensor<string, []>("attn_13_cast_fp16")]; | |
| tensor<int32, [4]> var_822 = const()[name = tensor<string, []>("op_822"), val = tensor<int32, [4]>([1, 512, 1, 1])]; | |
| tensor<fp16, [1, 512, 1, 1]> input_31_cast_fp16 = reshape(shape = var_822, 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, [512, 512, 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, [512, 512, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(80370944)))]; | |
| tensor<fp16, [512]> 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, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(80895296)))]; | |
| tensor<fp16, [1, 512, 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, 512, 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_844_to_fp16 = const()[name = tensor<string, []>("op_844_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; | |
| tensor<fp16, [1, 512, 1, 1]> out_21_cast_fp16 = layer_norm(axes = out_21_axes_0, epsilon = var_844_to_fp16, x = inputs_21_cast_fp16)[name = tensor<string, []>("out_21_cast_fp16")]; | |
| tensor<fp16, [512]> obj_51_gamma_0_to_fp16 = const()[name = tensor<string, []>("obj_51_gamma_0_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(80896384)))]; | |
| tensor<fp16, [512]> obj_51_beta_0_to_fp16 = const()[name = tensor<string, []>("obj_51_beta_0_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(80897472)))]; | |
| 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, 512, 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, [512, 512, 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, [512, 512, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(80898560)))]; | |
| tensor<fp16, [512]> 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, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(81422912)))]; | |
| tensor<fp16, [1, 512, 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, [512, 512, 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, [512, 512, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(81424000)))]; | |
| tensor<fp16, [1, 512, 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, [512, 512, 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, [512, 512, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(81948352)))]; | |
| tensor<fp16, [512]> 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, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(82472704)))]; | |
| tensor<fp16, [1, 512, 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_880 = const()[name = tensor<string, []>("op_880"), val = tensor<int32, [4]>([1, 8, 64, 1])]; | |
| tensor<fp16, [1, 8, 64, 1]> mh_q_15_cast_fp16 = reshape(shape = var_880, x = query_15_cast_fp16)[name = tensor<string, []>("mh_q_15_cast_fp16")]; | |
| tensor<fp16, []> var_882_to_fp16 = const()[name = tensor<string, []>("op_882_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; | |
| tensor<fp16, [1, 8, 64, 1]> var_883_cast_fp16 = mul(x = mh_q_15_cast_fp16, y = var_882_to_fp16)[name = tensor<string, []>("op_883_cast_fp16")]; | |
| tensor<int32, [4]> var_886 = const()[name = tensor<string, []>("op_886"), val = tensor<int32, [4]>([1, 8, 64, 1500])]; | |
| tensor<fp16, [1, 8, 64, 1500]> var_887_cast_fp16 = reshape(shape = var_886, x = key_15_cast_fp16)[name = tensor<string, []>("op_887_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, 8, 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_883_cast_fp16, y = var_887_cast_fp16)[name = tensor<string, []>("mh_w_23_cast_fp16")]; | |
| tensor<fp16, [1, 8, 1, 1500]> obj_55_cast_fp16 = softmax(axis = var_729, x = mh_w_23_cast_fp16)[name = tensor<string, []>("obj_55_cast_fp16")]; | |
| tensor<int32, [4]> var_891 = const()[name = tensor<string, []>("op_891"), val = tensor<int32, [4]>([1, 8, 64, 1500])]; | |
| tensor<fp16, [1, 8, 64, 1500]> var_892_cast_fp16 = reshape(shape = var_891, x = value_15_cast_fp16)[name = tensor<string, []>("op_892_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, 8, 64, 1]> attn_15_cast_fp16 = matmul(transpose_x = attn_15_transpose_x_0, transpose_y = attn_15_transpose_y_0, x = var_892_cast_fp16, y = obj_55_cast_fp16)[name = tensor<string, []>("attn_15_cast_fp16")]; | |
| tensor<int32, [4]> var_895 = const()[name = tensor<string, []>("op_895"), val = tensor<int32, [4]>([1, 512, 1, 1])]; | |
| tensor<fp16, [1, 512, 1, 1]> input_33_cast_fp16 = reshape(shape = var_895, 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, [512, 512, 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, [512, 512, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(82473792)))]; | |
| tensor<fp16, [512]> 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, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(82998144)))]; | |
| tensor<fp16, [1, 512, 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, 512, 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_916_to_fp16 = const()[name = tensor<string, []>("op_916_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; | |
| tensor<fp16, [1, 512, 1, 1]> out_23_cast_fp16 = layer_norm(axes = out_23_axes_0, epsilon = var_916_to_fp16, x = inputs_23_cast_fp16)[name = tensor<string, []>("out_23_cast_fp16")]; | |
| tensor<fp16, [512]> input_35_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_35_gamma_0_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(82999232)))]; | |
| tensor<fp16, [512]> input_35_beta_0_to_fp16 = const()[name = tensor<string, []>("input_35_beta_0_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(83000320)))]; | |
| 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, 512, 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, [2048, 512, 1, 1]> layers_3_fc1_weight_to_fp16 = const()[name = tensor<string, []>("layers_3_fc1_weight_to_fp16"), val = tensor<fp16, [2048, 512, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(83001408)))]; | |
| tensor<fp16, [2048]> layers_3_fc1_bias_to_fp16 = const()[name = tensor<string, []>("layers_3_fc1_bias_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(85098624)))]; | |
| tensor<fp16, [1, 2048, 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, 2048, 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, [512, 2048, 1, 1]> layers_3_fc2_weight_to_fp16 = const()[name = tensor<string, []>("layers_3_fc2_weight_to_fp16"), val = tensor<fp16, [512, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(85102784)))]; | |
| tensor<fp16, [512]> layers_3_fc2_bias_to_fp16 = const()[name = tensor<string, []>("layers_3_fc2_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(87200000)))]; | |
| tensor<fp16, [1, 512, 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, 512, 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_952 = const()[name = tensor<string, []>("op_952"), 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_977_to_fp16 = const()[name = tensor<string, []>("op_977_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; | |
| tensor<fp16, [1, 512, 1, 1]> out_25_cast_fp16 = layer_norm(axes = out_25_axes_0, epsilon = var_977_to_fp16, x = inputs_25_cast_fp16)[name = tensor<string, []>("out_25_cast_fp16")]; | |
| tensor<fp16, [512]> obj_57_gamma_0_to_fp16 = const()[name = tensor<string, []>("obj_57_gamma_0_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(87201088)))]; | |
| tensor<fp16, [512]> obj_57_beta_0_to_fp16 = const()[name = tensor<string, []>("obj_57_beta_0_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(87202176)))]; | |
| 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, 512, 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, [512, 512, 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, [512, 512, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(87203264)))]; | |
| tensor<fp16, [512]> 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, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(87727616)))]; | |
| tensor<fp16, [1, 512, 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, [512, 512, 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, [512, 512, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(87728704)))]; | |
| tensor<fp16, [1, 512, 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, [512, 512, 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, [512, 512, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(88253056)))]; | |
| tensor<fp16, [512]> 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, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(88777408)))]; | |
| tensor<fp16, [1, 512, 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, 512, 1, 448]> var_1016_cast_fp16 = mul(x = var_51_cast_fp16_4, y = var_135_cast_fp16)[name = tensor<string, []>("op_1016_cast_fp16")]; | |
| tensor<fp16, [1, 512, 1, 448]> var_1017_cast_fp16 = mul(x = current_key_9_cast_fp16, y = var_133_cast_fp16)[name = tensor<string, []>("op_1017_cast_fp16")]; | |
| tensor<fp16, [1, 512, 1, 448]> key_17_cast_fp16 = add(x = var_1016_cast_fp16, y = var_1017_cast_fp16)[name = tensor<string, []>("key_17_cast_fp16")]; | |
| tensor<fp16, [1, 512, 1, 448]> var_1020_cast_fp16 = mul(x = var_60_cast_fp16_4, y = var_135_cast_fp16)[name = tensor<string, []>("op_1020_cast_fp16")]; | |
| tensor<fp16, [1, 512, 1, 448]> var_1021_cast_fp16 = mul(x = current_value_9_cast_fp16, y = var_133_cast_fp16)[name = tensor<string, []>("op_1021_cast_fp16")]; | |
| tensor<fp16, [1, 512, 1, 448]> value_17_cast_fp16 = add(x = var_1020_cast_fp16, y = var_1021_cast_fp16)[name = tensor<string, []>("value_17_cast_fp16")]; | |
| tensor<int32, [4]> var_1025 = const()[name = tensor<string, []>("op_1025"), val = tensor<int32, [4]>([1, 8, 64, 1])]; | |
| tensor<fp16, [1, 8, 64, 1]> mh_q_17_cast_fp16 = reshape(shape = var_1025, x = query_17_cast_fp16)[name = tensor<string, []>("mh_q_17_cast_fp16")]; | |
| tensor<fp16, []> var_1027_to_fp16 = const()[name = tensor<string, []>("op_1027_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; | |
| tensor<fp16, [1, 8, 64, 1]> var_1028_cast_fp16 = mul(x = mh_q_17_cast_fp16, y = var_1027_to_fp16)[name = tensor<string, []>("op_1028_cast_fp16")]; | |
| tensor<int32, [4]> var_1031 = const()[name = tensor<string, []>("op_1031"), val = tensor<int32, [4]>([1, 8, 64, 448])]; | |
| tensor<fp16, [1, 8, 64, 448]> var_1032_cast_fp16 = reshape(shape = var_1031, x = key_17_cast_fp16)[name = tensor<string, []>("op_1032_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, 8, 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_1028_cast_fp16, y = var_1032_cast_fp16)[name = tensor<string, []>("mh_w_25_cast_fp16")]; | |
| tensor<fp16, [1, 8, 1, 448]> mh_w_27_cast_fp16 = add(x = mh_w_25_cast_fp16, y = var_157_cast_fp16)[name = tensor<string, []>("mh_w_27_cast_fp16")]; | |
| tensor<fp16, [1, 8, 1, 448]> var_1040_cast_fp16 = softmax(axis = var_952, x = mh_w_27_cast_fp16)[name = tensor<string, []>("op_1040_cast_fp16")]; | |
| tensor<int32, [4]> var_1041 = const()[name = tensor<string, []>("op_1041"), val = tensor<int32, [4]>([1, 8, 64, 448])]; | |
| tensor<fp16, [1, 8, 64, 448]> var_1042_cast_fp16 = reshape(shape = var_1041, x = value_17_cast_fp16)[name = tensor<string, []>("op_1042_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, 8, 64, 1]> attn_17_cast_fp16 = matmul(transpose_x = attn_17_transpose_x_0, transpose_y = attn_17_transpose_y_0, x = var_1042_cast_fp16, y = var_1040_cast_fp16)[name = tensor<string, []>("attn_17_cast_fp16")]; | |
| tensor<int32, [4]> var_1045 = const()[name = tensor<string, []>("op_1045"), val = tensor<int32, [4]>([1, 512, 1, 1])]; | |
| tensor<fp16, [1, 512, 1, 1]> input_41_cast_fp16 = reshape(shape = var_1045, 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, [512, 512, 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, [512, 512, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(88778496)))]; | |
| tensor<fp16, [512]> 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, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(89302848)))]; | |
| tensor<fp16, [1, 512, 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, 512, 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_1067_to_fp16 = const()[name = tensor<string, []>("op_1067_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; | |
| tensor<fp16, [1, 512, 1, 1]> out_27_cast_fp16 = layer_norm(axes = out_27_axes_0, epsilon = var_1067_to_fp16, x = inputs_27_cast_fp16)[name = tensor<string, []>("out_27_cast_fp16")]; | |
| tensor<fp16, [512]> obj_65_gamma_0_to_fp16 = const()[name = tensor<string, []>("obj_65_gamma_0_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(89303936)))]; | |
| tensor<fp16, [512]> obj_65_beta_0_to_fp16 = const()[name = tensor<string, []>("obj_65_beta_0_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(89305024)))]; | |
| 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, 512, 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, [512, 512, 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, [512, 512, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(89306112)))]; | |
| tensor<fp16, [512]> 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, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(89830464)))]; | |
| tensor<fp16, [1, 512, 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, [512, 512, 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, [512, 512, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(89831552)))]; | |
| tensor<fp16, [1, 512, 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, [512, 512, 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, [512, 512, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(90355904)))]; | |
| tensor<fp16, [512]> 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, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(90880256)))]; | |
| tensor<fp16, [1, 512, 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_1103 = const()[name = tensor<string, []>("op_1103"), val = tensor<int32, [4]>([1, 8, 64, 1])]; | |
| tensor<fp16, [1, 8, 64, 1]> mh_q_19_cast_fp16 = reshape(shape = var_1103, x = query_19_cast_fp16)[name = tensor<string, []>("mh_q_19_cast_fp16")]; | |
| tensor<fp16, []> var_1105_to_fp16 = const()[name = tensor<string, []>("op_1105_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; | |
| tensor<fp16, [1, 8, 64, 1]> var_1106_cast_fp16 = mul(x = mh_q_19_cast_fp16, y = var_1105_to_fp16)[name = tensor<string, []>("op_1106_cast_fp16")]; | |
| tensor<int32, [4]> var_1109 = const()[name = tensor<string, []>("op_1109"), val = tensor<int32, [4]>([1, 8, 64, 1500])]; | |
| tensor<fp16, [1, 8, 64, 1500]> var_1110_cast_fp16 = reshape(shape = var_1109, x = key_19_cast_fp16)[name = tensor<string, []>("op_1110_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, 8, 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_1106_cast_fp16, y = var_1110_cast_fp16)[name = tensor<string, []>("mh_w_29_cast_fp16")]; | |
| tensor<fp16, [1, 8, 1, 1500]> obj_69_cast_fp16 = softmax(axis = var_952, x = mh_w_29_cast_fp16)[name = tensor<string, []>("obj_69_cast_fp16")]; | |
| tensor<int32, [4]> var_1114 = const()[name = tensor<string, []>("op_1114"), val = tensor<int32, [4]>([1, 8, 64, 1500])]; | |
| tensor<fp16, [1, 8, 64, 1500]> var_1115_cast_fp16 = reshape(shape = var_1114, x = value_19_cast_fp16)[name = tensor<string, []>("op_1115_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, 8, 64, 1]> attn_19_cast_fp16 = matmul(transpose_x = attn_19_transpose_x_0, transpose_y = attn_19_transpose_y_0, x = var_1115_cast_fp16, y = obj_69_cast_fp16)[name = tensor<string, []>("attn_19_cast_fp16")]; | |
| tensor<int32, [4]> var_1118 = const()[name = tensor<string, []>("op_1118"), val = tensor<int32, [4]>([1, 512, 1, 1])]; | |
| tensor<fp16, [1, 512, 1, 1]> input_43_cast_fp16 = reshape(shape = var_1118, 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, [512, 512, 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, [512, 512, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(90881344)))]; | |
| tensor<fp16, [512]> 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, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(91405696)))]; | |
| tensor<fp16, [1, 512, 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, 512, 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_1139_to_fp16 = const()[name = tensor<string, []>("op_1139_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; | |
| tensor<fp16, [1, 512, 1, 1]> out_29_cast_fp16 = layer_norm(axes = out_29_axes_0, epsilon = var_1139_to_fp16, x = inputs_29_cast_fp16)[name = tensor<string, []>("out_29_cast_fp16")]; | |
| tensor<fp16, [512]> input_45_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_45_gamma_0_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(91406784)))]; | |
| tensor<fp16, [512]> input_45_beta_0_to_fp16 = const()[name = tensor<string, []>("input_45_beta_0_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(91407872)))]; | |
| 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, 512, 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, [2048, 512, 1, 1]> layers_4_fc1_weight_to_fp16 = const()[name = tensor<string, []>("layers_4_fc1_weight_to_fp16"), val = tensor<fp16, [2048, 512, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(91408960)))]; | |
| tensor<fp16, [2048]> layers_4_fc1_bias_to_fp16 = const()[name = tensor<string, []>("layers_4_fc1_bias_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(93506176)))]; | |
| tensor<fp16, [1, 2048, 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, 2048, 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, [512, 2048, 1, 1]> layers_4_fc2_weight_to_fp16 = const()[name = tensor<string, []>("layers_4_fc2_weight_to_fp16"), val = tensor<fp16, [512, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(93510336)))]; | |
| tensor<fp16, [512]> layers_4_fc2_bias_to_fp16 = const()[name = tensor<string, []>("layers_4_fc2_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(95607552)))]; | |
| tensor<fp16, [1, 512, 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, 512, 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_1175 = const()[name = tensor<string, []>("op_1175"), 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_1200_to_fp16 = const()[name = tensor<string, []>("op_1200_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; | |
| tensor<fp16, [1, 512, 1, 1]> out_31_cast_fp16 = layer_norm(axes = out_31_axes_0, epsilon = var_1200_to_fp16, x = inputs_31_cast_fp16)[name = tensor<string, []>("out_31_cast_fp16")]; | |
| tensor<fp16, [512]> obj_71_gamma_0_to_fp16 = const()[name = tensor<string, []>("obj_71_gamma_0_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(95608640)))]; | |
| tensor<fp16, [512]> obj_71_beta_0_to_fp16 = const()[name = tensor<string, []>("obj_71_beta_0_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(95609728)))]; | |
| 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, 512, 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, [512, 512, 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, [512, 512, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(95610816)))]; | |
| tensor<fp16, [512]> 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, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(96135168)))]; | |
| tensor<fp16, [1, 512, 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_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, [512, 512, 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, [512, 512, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(96136256)))]; | |
| tensor<fp16, [1, 512, 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_5_self_attn_k_proj_weight_to_fp16, x = obj_71_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, [512, 512, 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, [512, 512, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(96660608)))]; | |
| tensor<fp16, [512]> 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, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(97184960)))]; | |
| tensor<fp16, [1, 512, 1, 1]> current_value_cast_fp16 = conv(bias = layers_5_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_5_self_attn_v_proj_weight_to_fp16, x = obj_71_cast_fp16)[name = tensor<string, []>("current_value_cast_fp16")]; | |
| tensor<fp16, [1, 512, 1, 448]> var_1239_cast_fp16 = mul(x = var_51_cast_fp16_5, y = var_135_cast_fp16)[name = tensor<string, []>("op_1239_cast_fp16")]; | |
| tensor<fp16, [1, 512, 1, 448]> var_1240_cast_fp16 = mul(x = current_key_cast_fp16, y = var_133_cast_fp16)[name = tensor<string, []>("op_1240_cast_fp16")]; | |
| tensor<fp16, [1, 512, 1, 448]> key_21_cast_fp16 = add(x = var_1239_cast_fp16, y = var_1240_cast_fp16)[name = tensor<string, []>("key_21_cast_fp16")]; | |
| tensor<fp16, [1, 512, 1, 448]> var_1243_cast_fp16 = mul(x = var_60_cast_fp16_5, y = var_135_cast_fp16)[name = tensor<string, []>("op_1243_cast_fp16")]; | |
| tensor<fp16, [1, 512, 1, 448]> var_1244_cast_fp16 = mul(x = current_value_cast_fp16, y = var_133_cast_fp16)[name = tensor<string, []>("op_1244_cast_fp16")]; | |
| tensor<fp16, [1, 512, 1, 448]> value_21_cast_fp16 = add(x = var_1243_cast_fp16, y = var_1244_cast_fp16)[name = tensor<string, []>("value_21_cast_fp16")]; | |
| tensor<int32, [4]> var_1248 = const()[name = tensor<string, []>("op_1248"), val = tensor<int32, [4]>([1, 8, 64, 1])]; | |
| tensor<fp16, [1, 8, 64, 1]> mh_q_21_cast_fp16 = reshape(shape = var_1248, x = query_21_cast_fp16)[name = tensor<string, []>("mh_q_21_cast_fp16")]; | |
| tensor<fp16, []> var_1250_to_fp16 = const()[name = tensor<string, []>("op_1250_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; | |
| tensor<fp16, [1, 8, 64, 1]> var_1251_cast_fp16 = mul(x = mh_q_21_cast_fp16, y = var_1250_to_fp16)[name = tensor<string, []>("op_1251_cast_fp16")]; | |
| tensor<int32, [4]> var_1254 = const()[name = tensor<string, []>("op_1254"), val = tensor<int32, [4]>([1, 8, 64, 448])]; | |
| tensor<fp16, [1, 8, 64, 448]> var_1255_cast_fp16 = reshape(shape = var_1254, x = key_21_cast_fp16)[name = tensor<string, []>("op_1255_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, 8, 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_1251_cast_fp16, y = var_1255_cast_fp16)[name = tensor<string, []>("mh_w_31_cast_fp16")]; | |
| tensor<fp16, [1, 8, 1, 448]> mh_w_33_cast_fp16 = add(x = mh_w_31_cast_fp16, y = var_157_cast_fp16)[name = tensor<string, []>("mh_w_33_cast_fp16")]; | |
| tensor<fp16, [1, 8, 1, 448]> var_1263_cast_fp16 = softmax(axis = var_1175, x = mh_w_33_cast_fp16)[name = tensor<string, []>("op_1263_cast_fp16")]; | |
| tensor<int32, [4]> var_1264 = const()[name = tensor<string, []>("op_1264"), val = tensor<int32, [4]>([1, 8, 64, 448])]; | |
| tensor<fp16, [1, 8, 64, 448]> var_1265_cast_fp16 = reshape(shape = var_1264, x = value_21_cast_fp16)[name = tensor<string, []>("op_1265_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, 8, 64, 1]> attn_21_cast_fp16 = matmul(transpose_x = attn_21_transpose_x_0, transpose_y = attn_21_transpose_y_0, x = var_1265_cast_fp16, y = var_1263_cast_fp16)[name = tensor<string, []>("attn_21_cast_fp16")]; | |
| tensor<int32, [4]> var_1268 = const()[name = tensor<string, []>("op_1268"), val = tensor<int32, [4]>([1, 512, 1, 1])]; | |
| tensor<fp16, [1, 512, 1, 1]> input_51_cast_fp16 = reshape(shape = var_1268, 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, [512, 512, 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, [512, 512, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(97186048)))]; | |
| tensor<fp16, [512]> 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, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(97710400)))]; | |
| tensor<fp16, [1, 512, 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, 512, 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_1290_to_fp16 = const()[name = tensor<string, []>("op_1290_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; | |
| tensor<fp16, [1, 512, 1, 1]> out_33_cast_fp16 = layer_norm(axes = out_33_axes_0, epsilon = var_1290_to_fp16, x = inputs_33_cast_fp16)[name = tensor<string, []>("out_33_cast_fp16")]; | |
| tensor<fp16, [512]> obj_79_gamma_0_to_fp16 = const()[name = tensor<string, []>("obj_79_gamma_0_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(97711488)))]; | |
| tensor<fp16, [512]> obj_79_beta_0_to_fp16 = const()[name = tensor<string, []>("obj_79_beta_0_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(97712576)))]; | |
| 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, 512, 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_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, [512, 512, 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, [512, 512, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(97713664)))]; | |
| tensor<fp16, [512]> 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, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(98238016)))]; | |
| tensor<fp16, [1, 512, 1, 1]> query_cast_fp16 = conv(bias = layers_5_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_5_encoder_attn_q_proj_weight_to_fp16, x = obj_79_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, [512, 512, 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, [512, 512, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(98239104)))]; | |
| tensor<fp16, [1, 512, 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_5_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, [512, 512, 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, [512, 512, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(98763456)))]; | |
| tensor<fp16, [512]> 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, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(99287808)))]; | |
| tensor<fp16, [1, 512, 1, 1500]> value_cast_fp16 = conv(bias = layers_5_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_5_encoder_attn_v_proj_weight_to_fp16, x = encoder_output_embeds)[name = tensor<string, []>("value_cast_fp16")]; | |
| tensor<int32, [4]> var_1326 = const()[name = tensor<string, []>("op_1326"), val = tensor<int32, [4]>([1, 8, 64, 1])]; | |
| tensor<fp16, [1, 8, 64, 1]> mh_q_cast_fp16 = reshape(shape = var_1326, x = query_cast_fp16)[name = tensor<string, []>("mh_q_cast_fp16")]; | |
| tensor<fp16, []> var_1328_to_fp16 = const()[name = tensor<string, []>("op_1328_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; | |
| tensor<fp16, [1, 8, 64, 1]> var_1329_cast_fp16 = mul(x = mh_q_cast_fp16, y = var_1328_to_fp16)[name = tensor<string, []>("op_1329_cast_fp16")]; | |
| tensor<int32, [4]> var_1332 = const()[name = tensor<string, []>("op_1332"), val = tensor<int32, [4]>([1, 8, 64, 1500])]; | |
| tensor<fp16, [1, 8, 64, 1500]> var_1333_cast_fp16 = reshape(shape = var_1332, x = key_cast_fp16)[name = tensor<string, []>("op_1333_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, 8, 1, 1500]> mh_w_cast_fp16 = matmul(transpose_x = mh_w_transpose_x_0, transpose_y = mh_w_transpose_y_0, x = var_1329_cast_fp16, y = var_1333_cast_fp16)[name = tensor<string, []>("mh_w_cast_fp16")]; | |
| tensor<fp16, [1, 8, 1, 1500]> obj_83_cast_fp16 = softmax(axis = var_1175, x = mh_w_cast_fp16)[name = tensor<string, []>("obj_83_cast_fp16")]; | |
| tensor<int32, [4]> var_1337 = const()[name = tensor<string, []>("op_1337"), val = tensor<int32, [4]>([1, 8, 64, 1500])]; | |
| tensor<fp16, [1, 8, 64, 1500]> var_1338_cast_fp16 = reshape(shape = var_1337, x = value_cast_fp16)[name = tensor<string, []>("op_1338_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, 8, 64, 1]> attn_cast_fp16 = matmul(transpose_x = attn_transpose_x_0, transpose_y = attn_transpose_y_0, x = var_1338_cast_fp16, y = obj_83_cast_fp16)[name = tensor<string, []>("attn_cast_fp16")]; | |
| tensor<int32, [4]> var_1341 = const()[name = tensor<string, []>("op_1341"), val = tensor<int32, [4]>([1, 512, 1, 1])]; | |
| tensor<fp16, [1, 512, 1, 1]> input_53_cast_fp16 = reshape(shape = var_1341, x = attn_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, [512, 512, 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, [512, 512, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(99288896)))]; | |
| tensor<fp16, [512]> 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, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(99813248)))]; | |
| tensor<fp16, [1, 512, 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, 512, 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_1362_to_fp16 = const()[name = tensor<string, []>("op_1362_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; | |
| tensor<fp16, [1, 512, 1, 1]> out_35_cast_fp16 = layer_norm(axes = out_35_axes_0, epsilon = var_1362_to_fp16, x = inputs_35_cast_fp16)[name = tensor<string, []>("out_35_cast_fp16")]; | |
| tensor<fp16, [512]> input_55_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_55_gamma_0_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(99814336)))]; | |
| tensor<fp16, [512]> input_55_beta_0_to_fp16 = const()[name = tensor<string, []>("input_55_beta_0_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(99815424)))]; | |
| 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, 512, 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, [2048, 512, 1, 1]> layers_5_fc1_weight_to_fp16 = const()[name = tensor<string, []>("layers_5_fc1_weight_to_fp16"), val = tensor<fp16, [2048, 512, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(99816512)))]; | |
| tensor<fp16, [2048]> layers_5_fc1_bias_to_fp16 = const()[name = tensor<string, []>("layers_5_fc1_bias_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(101913728)))]; | |
| tensor<fp16, [1, 2048, 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_mode_0 = const()[name = tensor<string, []>("input_mode_0"), val = tensor<string, []>("EXACT")]; | |
| tensor<fp16, [1, 2048, 1, 1]> input_cast_fp16 = gelu(mode = input_mode_0, x = input_57_cast_fp16)[name = tensor<string, []>("input_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, [512, 2048, 1, 1]> layers_5_fc2_weight_to_fp16 = const()[name = tensor<string, []>("layers_5_fc2_weight_to_fp16"), val = tensor<fp16, [512, 2048, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(101917888)))]; | |
| tensor<fp16, [512]> layers_5_fc2_bias_to_fp16 = const()[name = tensor<string, []>("layers_5_fc2_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(104015104)))]; | |
| tensor<fp16, [1, 512, 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_cast_fp16)[name = tensor<string, []>("hidden_states_13_cast_fp16")]; | |
| tensor<fp16, [1, 512, 1, 1]> inputs_cast_fp16 = add(x = inputs_35_cast_fp16, y = hidden_states_13_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_1405_to_fp16 = const()[name = tensor<string, []>("op_1405_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; | |
| tensor<fp16, [1, 512, 1, 1]> out_cast_fp16 = layer_norm(axes = out_axes_0, epsilon = var_1405_to_fp16, x = inputs_cast_fp16)[name = tensor<string, []>("out_cast_fp16")]; | |
| tensor<fp16, [512]> hidden_states_gamma_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_gamma_0_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(104016192)))]; | |
| tensor<fp16, [512]> hidden_states_beta_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_beta_0_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(104017280)))]; | |
| 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, 512, 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_1416_axes_0 = const()[name = tensor<string, []>("op_1416_axes_0"), val = tensor<int32, [1]>([2])]; | |
| tensor<fp16, [1, 512, 1]> var_1416_cast_fp16 = squeeze(axes = var_1416_axes_0, x = hidden_states_cast_fp16)[name = tensor<string, []>("op_1416_cast_fp16")]; | |
| tensor<int32, [3]> var_1419_perm_0 = const()[name = tensor<string, []>("op_1419_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, []>(104018368)))]; | |
| tensor<fp16, [1, 1, 512]> var_1419_cast_fp16 = transpose(perm = var_1419_perm_0, x = var_1416_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_1419_cast_fp16)[name = tensor<string, []>("linear_0_cast_fp16")]; | |
| tensor<int32, []> var_1423 = const()[name = tensor<string, []>("op_1423"), val = tensor<int32, []>(1)]; | |
| tensor<bool, []> obj_87_interleave_0 = const()[name = tensor<string, []>("obj_87_interleave_0"), val = tensor<bool, []>(false)]; | |
| tensor<fp16, [1, 3072, 1, 1]> key_cache_updates = concat(axis = var_1423, interleave = obj_87_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_cast_fp16))[name = tensor<string, []>("obj_87_cast_fp16")]; | |
| tensor<int32, []> var_1426 = const()[name = tensor<string, []>("op_1426"), val = tensor<int32, []>(1)]; | |
| tensor<bool, []> obj_89_interleave_0 = const()[name = tensor<string, []>("obj_89_interleave_0"), val = tensor<bool, []>(false)]; | |
| tensor<fp16, [1, 3072, 1, 1]> value_cache_updates = concat(axis = var_1426, interleave = obj_89_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_cast_fp16))[name = tensor<string, []>("obj_89_cast_fp16")]; | |
| tensor<int32, [4]> var_1437_begin_0 = const()[name = tensor<string, []>("op_1437_begin_0"), val = tensor<int32, [4]>([0, 1, 0, 0])]; | |
| tensor<int32, [4]> var_1437_end_0 = const()[name = tensor<string, []>("op_1437_end_0"), val = tensor<int32, [4]>([1, 2, 1, 1500])]; | |
| tensor<bool, [4]> var_1437_end_mask_0 = const()[name = tensor<string, []>("op_1437_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])]; | |
| tensor<fp16, [1, 1, 1, 1500]> var_1437_cast_fp16 = slice_by_index(begin = var_1437_begin_0, end = var_1437_end_0, end_mask = var_1437_end_mask_0, x = obj_55_cast_fp16)[name = tensor<string, []>("op_1437_cast_fp16")]; | |
| tensor<int32, [4]> var_1440_begin_0 = const()[name = tensor<string, []>("op_1440_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; | |
| tensor<int32, [4]> var_1440_end_0 = const()[name = tensor<string, []>("op_1440_end_0"), val = tensor<int32, [4]>([1, 1, 1, 1500])]; | |
| tensor<bool, [4]> var_1440_end_mask_0 = const()[name = tensor<string, []>("op_1440_end_mask_0"), val = tensor<bool, [4]>([true, true, false, true])]; | |
| tensor<bool, [4]> var_1440_squeeze_mask_0 = const()[name = tensor<string, []>("op_1440_squeeze_mask_0"), val = tensor<bool, [4]>([false, false, true, false])]; | |
| tensor<fp16, [1, 1, 1500]> var_1440_cast_fp16 = slice_by_index(begin = var_1440_begin_0, end = var_1440_end_0, end_mask = var_1440_end_mask_0, squeeze_mask = var_1440_squeeze_mask_0, x = var_1437_cast_fp16)[name = tensor<string, []>("op_1440_cast_fp16")]; | |
| tensor<int32, [4]> var_1455_begin_0 = const()[name = tensor<string, []>("op_1455_begin_0"), val = tensor<int32, [4]>([0, 2, 0, 0])]; | |
| tensor<int32, [4]> var_1455_end_0 = const()[name = tensor<string, []>("op_1455_end_0"), val = tensor<int32, [4]>([1, 3, 1, 1500])]; | |
| tensor<bool, [4]> var_1455_end_mask_0 = const()[name = tensor<string, []>("op_1455_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])]; | |
| tensor<fp16, [1, 1, 1, 1500]> var_1455_cast_fp16 = slice_by_index(begin = var_1455_begin_0, end = var_1455_end_0, end_mask = var_1455_end_mask_0, x = obj_69_cast_fp16)[name = tensor<string, []>("op_1455_cast_fp16")]; | |
| tensor<int32, [4]> var_1458_begin_0 = const()[name = tensor<string, []>("op_1458_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; | |
| tensor<int32, [4]> var_1458_end_0 = const()[name = tensor<string, []>("op_1458_end_0"), val = tensor<int32, [4]>([1, 1, 1, 1500])]; | |
| tensor<bool, [4]> var_1458_end_mask_0 = const()[name = tensor<string, []>("op_1458_end_mask_0"), val = tensor<bool, [4]>([true, true, false, true])]; | |
| tensor<bool, [4]> var_1458_squeeze_mask_0 = const()[name = tensor<string, []>("op_1458_squeeze_mask_0"), val = tensor<bool, [4]>([false, false, true, false])]; | |
| tensor<fp16, [1, 1, 1500]> var_1458_cast_fp16 = slice_by_index(begin = var_1458_begin_0, end = var_1458_end_0, end_mask = var_1458_end_mask_0, squeeze_mask = var_1458_squeeze_mask_0, x = var_1455_cast_fp16)[name = tensor<string, []>("op_1458_cast_fp16")]; | |
| tensor<int32, [4]> var_1473_begin_0 = const()[name = tensor<string, []>("op_1473_begin_0"), val = tensor<int32, [4]>([0, 3, 0, 0])]; | |
| tensor<int32, [4]> var_1473_end_0 = const()[name = tensor<string, []>("op_1473_end_0"), val = tensor<int32, [4]>([1, 4, 1, 1500])]; | |
| tensor<bool, [4]> var_1473_end_mask_0 = const()[name = tensor<string, []>("op_1473_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])]; | |
| tensor<fp16, [1, 1, 1, 1500]> var_1473_cast_fp16 = slice_by_index(begin = var_1473_begin_0, end = var_1473_end_0, end_mask = var_1473_end_mask_0, x = obj_69_cast_fp16)[name = tensor<string, []>("op_1473_cast_fp16")]; | |
| tensor<int32, [4]> var_1476_begin_0 = const()[name = tensor<string, []>("op_1476_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; | |
| tensor<int32, [4]> var_1476_end_0 = const()[name = tensor<string, []>("op_1476_end_0"), val = tensor<int32, [4]>([1, 1, 1, 1500])]; | |
| tensor<bool, [4]> var_1476_end_mask_0 = const()[name = tensor<string, []>("op_1476_end_mask_0"), val = tensor<bool, [4]>([true, true, false, true])]; | |
| tensor<bool, [4]> var_1476_squeeze_mask_0 = const()[name = tensor<string, []>("op_1476_squeeze_mask_0"), val = tensor<bool, [4]>([false, false, true, false])]; | |
| tensor<fp16, [1, 1, 1500]> var_1476_cast_fp16 = slice_by_index(begin = var_1476_begin_0, end = var_1476_end_0, end_mask = var_1476_end_mask_0, squeeze_mask = var_1476_squeeze_mask_0, x = var_1473_cast_fp16)[name = tensor<string, []>("op_1476_cast_fp16")]; | |
| tensor<int32, [4]> var_1491_begin_0 = const()[name = tensor<string, []>("op_1491_begin_0"), val = tensor<int32, [4]>([0, 7, 0, 0])]; | |
| tensor<int32, [4]> var_1491_end_0 = const()[name = tensor<string, []>("op_1491_end_0"), val = tensor<int32, [4]>([1, 1, 1, 1500])]; | |
| tensor<bool, [4]> var_1491_end_mask_0 = const()[name = tensor<string, []>("op_1491_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])]; | |
| tensor<fp16, [1, 1, 1, 1500]> var_1491_cast_fp16 = slice_by_index(begin = var_1491_begin_0, end = var_1491_end_0, end_mask = var_1491_end_mask_0, x = obj_69_cast_fp16)[name = tensor<string, []>("op_1491_cast_fp16")]; | |
| tensor<int32, [4]> var_1494_begin_0 = const()[name = tensor<string, []>("op_1494_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; | |
| tensor<int32, [4]> var_1494_end_0 = const()[name = tensor<string, []>("op_1494_end_0"), val = tensor<int32, [4]>([1, 1, 1, 1500])]; | |
| tensor<bool, [4]> var_1494_end_mask_0 = const()[name = tensor<string, []>("op_1494_end_mask_0"), val = tensor<bool, [4]>([true, true, false, true])]; | |
| tensor<bool, [4]> var_1494_squeeze_mask_0 = const()[name = tensor<string, []>("op_1494_squeeze_mask_0"), val = tensor<bool, [4]>([false, false, true, false])]; | |
| tensor<fp16, [1, 1, 1500]> var_1494_cast_fp16 = slice_by_index(begin = var_1494_begin_0, end = var_1494_end_0, end_mask = var_1494_end_mask_0, squeeze_mask = var_1494_squeeze_mask_0, x = var_1491_cast_fp16)[name = tensor<string, []>("op_1494_cast_fp16")]; | |
| tensor<int32, [4]> var_1509_begin_0 = const()[name = tensor<string, []>("op_1509_begin_0"), val = tensor<int32, [4]>([0, 1, 0, 0])]; | |
| tensor<int32, [4]> var_1509_end_0 = const()[name = tensor<string, []>("op_1509_end_0"), val = tensor<int32, [4]>([1, 2, 1, 1500])]; | |
| tensor<bool, [4]> var_1509_end_mask_0 = const()[name = tensor<string, []>("op_1509_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])]; | |
| tensor<fp16, [1, 1, 1, 1500]> var_1509_cast_fp16 = slice_by_index(begin = var_1509_begin_0, end = var_1509_end_0, end_mask = var_1509_end_mask_0, x = obj_83_cast_fp16)[name = tensor<string, []>("op_1509_cast_fp16")]; | |
| tensor<int32, [4]> var_1512_begin_0 = const()[name = tensor<string, []>("op_1512_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; | |
| tensor<int32, [4]> var_1512_end_0 = const()[name = tensor<string, []>("op_1512_end_0"), val = tensor<int32, [4]>([1, 1, 1, 1500])]; | |
| tensor<bool, [4]> var_1512_end_mask_0 = const()[name = tensor<string, []>("op_1512_end_mask_0"), val = tensor<bool, [4]>([true, true, false, true])]; | |
| tensor<bool, [4]> var_1512_squeeze_mask_0 = const()[name = tensor<string, []>("op_1512_squeeze_mask_0"), val = tensor<bool, [4]>([false, false, true, false])]; | |
| tensor<fp16, [1, 1, 1500]> var_1512_cast_fp16 = slice_by_index(begin = var_1512_begin_0, end = var_1512_end_0, end_mask = var_1512_end_mask_0, squeeze_mask = var_1512_squeeze_mask_0, x = var_1509_cast_fp16)[name = tensor<string, []>("op_1512_cast_fp16")]; | |
| tensor<int32, [4]> var_1527_begin_0 = const()[name = tensor<string, []>("op_1527_begin_0"), val = tensor<int32, [4]>([0, 2, 0, 0])]; | |
| tensor<int32, [4]> var_1527_end_0 = const()[name = tensor<string, []>("op_1527_end_0"), val = tensor<int32, [4]>([1, 3, 1, 1500])]; | |
| tensor<bool, [4]> var_1527_end_mask_0 = const()[name = tensor<string, []>("op_1527_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])]; | |
| tensor<fp16, [1, 1, 1, 1500]> var_1527_cast_fp16 = slice_by_index(begin = var_1527_begin_0, end = var_1527_end_0, end_mask = var_1527_end_mask_0, x = obj_83_cast_fp16)[name = tensor<string, []>("op_1527_cast_fp16")]; | |
| tensor<int32, [4]> var_1530_begin_0 = const()[name = tensor<string, []>("op_1530_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; | |
| tensor<int32, [4]> var_1530_end_0 = const()[name = tensor<string, []>("op_1530_end_0"), val = tensor<int32, [4]>([1, 1, 1, 1500])]; | |
| tensor<bool, [4]> var_1530_end_mask_0 = const()[name = tensor<string, []>("op_1530_end_mask_0"), val = tensor<bool, [4]>([true, true, false, true])]; | |
| tensor<bool, [4]> var_1530_squeeze_mask_0 = const()[name = tensor<string, []>("op_1530_squeeze_mask_0"), val = tensor<bool, [4]>([false, false, true, false])]; | |
| tensor<fp16, [1, 1, 1500]> var_1530_cast_fp16 = slice_by_index(begin = var_1530_begin_0, end = var_1530_end_0, end_mask = var_1530_end_mask_0, squeeze_mask = var_1530_squeeze_mask_0, x = var_1527_cast_fp16)[name = tensor<string, []>("op_1530_cast_fp16")]; | |
| tensor<int32, [4]> var_1545_begin_0 = const()[name = tensor<string, []>("op_1545_begin_0"), val = tensor<int32, [4]>([0, 4, 0, 0])]; | |
| tensor<int32, [4]> var_1545_end_0 = const()[name = tensor<string, []>("op_1545_end_0"), val = tensor<int32, [4]>([1, 5, 1, 1500])]; | |
| tensor<bool, [4]> var_1545_end_mask_0 = const()[name = tensor<string, []>("op_1545_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])]; | |
| tensor<fp16, [1, 1, 1, 1500]> var_1545_cast_fp16 = slice_by_index(begin = var_1545_begin_0, end = var_1545_end_0, end_mask = var_1545_end_mask_0, x = obj_83_cast_fp16)[name = tensor<string, []>("op_1545_cast_fp16")]; | |
| tensor<int32, [4]> var_1548_begin_0 = const()[name = tensor<string, []>("op_1548_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; | |
| tensor<int32, [4]> var_1548_end_0 = const()[name = tensor<string, []>("op_1548_end_0"), val = tensor<int32, [4]>([1, 1, 1, 1500])]; | |
| tensor<bool, [4]> var_1548_end_mask_0 = const()[name = tensor<string, []>("op_1548_end_mask_0"), val = tensor<bool, [4]>([true, true, false, true])]; | |
| tensor<bool, [4]> var_1548_squeeze_mask_0 = const()[name = tensor<string, []>("op_1548_squeeze_mask_0"), val = tensor<bool, [4]>([false, false, true, false])]; | |
| tensor<fp16, [1, 1, 1500]> var_1548_cast_fp16 = slice_by_index(begin = var_1548_begin_0, end = var_1548_end_0, end_mask = var_1548_end_mask_0, squeeze_mask = var_1548_squeeze_mask_0, x = var_1545_cast_fp16)[name = tensor<string, []>("op_1548_cast_fp16")]; | |
| tensor<int32, [4]> var_1563_begin_0 = const()[name = tensor<string, []>("op_1563_begin_0"), val = tensor<int32, [4]>([0, 6, 0, 0])]; | |
| tensor<int32, [4]> var_1563_end_0 = const()[name = tensor<string, []>("op_1563_end_0"), val = tensor<int32, [4]>([1, 7, 1, 1500])]; | |
| tensor<bool, [4]> var_1563_end_mask_0 = const()[name = tensor<string, []>("op_1563_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])]; | |
| tensor<fp16, [1, 1, 1, 1500]> var_1563_cast_fp16 = slice_by_index(begin = var_1563_begin_0, end = var_1563_end_0, end_mask = var_1563_end_mask_0, x = obj_83_cast_fp16)[name = tensor<string, []>("op_1563_cast_fp16")]; | |
| tensor<int32, [4]> var_1566_begin_0 = const()[name = tensor<string, []>("op_1566_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; | |
| tensor<int32, [4]> var_1566_end_0 = const()[name = tensor<string, []>("op_1566_end_0"), val = tensor<int32, [4]>([1, 1, 1, 1500])]; | |
| tensor<bool, [4]> var_1566_end_mask_0 = const()[name = tensor<string, []>("op_1566_end_mask_0"), val = tensor<bool, [4]>([true, true, false, true])]; | |
| tensor<bool, [4]> var_1566_squeeze_mask_0 = const()[name = tensor<string, []>("op_1566_squeeze_mask_0"), val = tensor<bool, [4]>([false, false, true, false])]; | |
| tensor<fp16, [1, 1, 1500]> var_1566_cast_fp16 = slice_by_index(begin = var_1566_begin_0, end = var_1566_end_0, end_mask = var_1566_end_mask_0, squeeze_mask = var_1566_squeeze_mask_0, x = var_1563_cast_fp16)[name = tensor<string, []>("op_1566_cast_fp16")]; | |
| tensor<int32, []> var_1573 = const()[name = tensor<string, []>("op_1573"), val = tensor<int32, []>(1)]; | |
| tensor<bool, []> var_1574_interleave_0 = const()[name = tensor<string, []>("op_1574_interleave_0"), val = tensor<bool, []>(false)]; | |
| tensor<fp16, [1, 8, 1500]> var_1574_cast_fp16 = concat(axis = var_1573, interleave = var_1574_interleave_0, values = (var_1440_cast_fp16, var_1458_cast_fp16, var_1476_cast_fp16, var_1494_cast_fp16, var_1512_cast_fp16, var_1530_cast_fp16, var_1548_cast_fp16, var_1566_cast_fp16))[name = tensor<string, []>("op_1574_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_1574_cast_fp16)[name = tensor<string, []>("obj_cast_fp16")]; | |
| } -> (logits, key_cache_updates, value_cache_updates, alignment_heads_weights); | |
| } |