cannot import name 'MSRA' from 'paddle.fluid.initializer' (/opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages/paddle/fluid/initializer.py) #55765

Liangwei-0521 · 2023-07-28T00:50:01Z

bug描述 Describe the Bug

Bug:
ImportError: cannot import name 'MSRA' from 'paddle.fluid.initializer' (/opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages/paddle/fluid/initializer.py)

运行代码：
import os
import paddleslim.quant as quant

  def _quant_embeddings(args):
      # input_prefix: 静态模型路径
      # 最终量化的模型路径：
      self.args.output_filename_prefix = "quant_emb"
      paddle.enable_static()
      place = paddle.set_device(self.args.device)
      exe = paddle.static.Executor(place)
      main_program, feed_target_names, fetch_targets = paddle.static.load_inference_model(input_prefix, exe)
  
      config = {"quantize_op_types": ["lookup_table_v2"], "lookup_table_v2": {"quantize_type": "moving_average_abs_max"}}
  
      quant_emb_program = quant.quant_embedding(main_program, place, config)
  
      input_dir = os.path.dirname(input_prefix)
  
      paddle.fluid.io.save_inference_model(
          input_dir,
          feed_target_names,
          fetch_targets,
          exe,
          quant_emb_program,
          model_filename=self.args.output_filename_prefix + ".pdmodel",
          params_filename=self.args.output_filename_prefix + ".pdiparams",
          export_for_deployment=True,
          program_only=False,
      )

版本：
paddle-bfloat 0.1.7
paddle2onnx 1.0.0
paddlefsl 1.1.0
paddlehub 2.3.0
paddlenlp 2.5.2
paddlepaddle 2.5.1
paddleslim 2.4.0

其他补充信息 Additional Supplementary Information

No response

The text was updated successfully, but these errors were encountered:

qili93 · 2023-07-28T02:20:14Z

@1998-Chen 您好，请提供一下 ImportError: cannot import name 'MSRA' from 'paddle.fluid.initializer' 前后完整的python报错的stack信息，以便分析这个错误是在Paddle框架中还是在PaddleSlim中出现的，谢谢！

另外可以尝试下把 paddle.fluid.io.save_inference_model 接口换成 paddle.static.save_inference_model 接口。

Paddle从2.5.0开始不再支持所有 paddle.fluid 开头的接口了，可以参考 Paddle 2.5 release信息 https://github.com/PaddlePaddle/Paddle/releases/tag/v2.5.0 中 2. 不兼容升级的 paddle.fluid API全面退场的内容。

Liangwei-0521 · 2023-07-28T03:15:25Z

任务描述：我们想基于paddleslim实现QAT量化以及embedding量化，QAT量化后用paddle inference推理无报错。然而，想进行embedding量化，发现高版本的paddle2.5.0报错，具体报错内容如下：

---------------------------------------------------------------------------
ImportError                               Traceback (most recent call last)
/tmp/ipykernel_383/2533658229.py in <module>
      1 import os
----> 2 import paddleslim.quant as quant
      3 
      4 def _quant_embeddings(args):
      5     # input_prefix: 静态模型路径

/opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages/paddleslim/__init__.py in <module>
     14 
     15 from __future__ import absolute_import
---> 16 from paddleslim import models
     17 from paddleslim import prune
     18 from paddleslim import nas

/opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages/paddleslim/models/__init__.py in <module>
     14 
     15 from __future__ import absolute_import
---> 16 from .util import image_classification
     17 from .slimfacenet import SlimFaceNet_A_x0_60, SlimFaceNet_B_x0_75, SlimFaceNet_C_x0_75
     18 from .slim_mobilenet import SlimMobileNet_v1, SlimMobileNet_v2, SlimMobileNet_v3, SlimMobileNet_v4, SlimMobileNet_v5

/opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages/paddleslim/models/util.py in <module>
      1 from __future__ import absolute_import
      2 import paddle.fluid as fluid
----> 3 from ..models import classification_models
      4 
      5 __all__ = ["image_classification"]

/opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages/paddleslim/models/classification_models.py in <module>
      1 from __future__ import absolute_import
----> 2 from .mobilenet import MobileNet
      3 from .resnet import ResNet34, ResNet50
      4 from .mobilenet_v2 import MobileNetV2
      5 __all__ = ["model_list", "MobileNet", "ResNet34", "ResNet50", "MobileNetV2"]

/opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages/paddleslim/models/mobilenet.py in <module>
      3 from __future__ import print_function
      4 import paddle.fluid as fluid
----> 5 from paddle.fluid.initializer import MSRA
      6 from paddle.fluid.param_attr import ParamAttr
      7 

ImportError: cannot import name 'MSRA' from 'paddle.fluid.initializer' (/opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages/paddle/fluid/initializer.py)

结合你所说的paddle2.5.0不再支持paddle.fluid全部接口，我们将paddlepaddle降低版本至version==2.4.0，上述错误可以解决，即使用低版本的paddle实现了paddleslim的embedding量化。但是，在低版本的paddle执行静态模型推理时，报以下错误：

---------------------------------------------------------------------------
NotImplementedError                       Traceback (most recent call last)
/tmp/ipykernel_2165/16845876.py in <module>
     28 # 执行测试
     29 for batch_data in batches:
---> 30     results.extend(predictor.predict(batch_data, tokenizer))
     31 end = time.time()
     32 run_time = end - start

/tmp/ipykernel_2165/1978965815.py in predict(self, data, tokenizer)
     79         self.input_handles[1].copy_from_cpu(token_type_ids)
     80         self.input_handles[2].copy_from_cpu(attention_mask)
---> 81         self.predictor.run()
     82         sim_score = self.output_handle.copy_to_cpu()
     83 

NotImplementedError: (Unimplemented) There are no kernels which are registered in the lookup_table_v2 operator.
  [Hint: Expected kernels_iter != all_op_kernels.end(), but received kernels_iter == all_op_kernels.end().] (at /paddle/paddle/fluid/framework/operator.cc:1895)
  [operator < lookup_table_v2 > error]

为此，我使用了高版本的paddlepaddle（version==2.5.0）进行推理量化后的模型，上述的报错消失，然而高版本的paddlepaddle实现模型推理的时出现以下的Warnig，在CPU下面测试并无加速，甚至比未量化的模型更慢:

[2023-07-28 11:09:52,467] [    INFO] - We are using <class 'paddlenlp.transformers.ernie.fast_tokenizer.ErnieFastTokenizer'> to load './checkpoint/ernie-3.0-nano-zh'.
46665it [00:04, 11158.08it/s]
I0728 11:09:57.157755  2568 analysis_predictor.cc:1471] MKLDNN is enabled
--- Running analysis [ir_graph_build_pass]
I0728 11:09:57.163435  2568 executor.cc:187] Old Executor is Running.
--- Running analysis [ir_analysis_pass]
--- Running IR pass [mkldnn_placement_pass]
--- Running IR pass [simplify_with_basic_ops_pass]
--- Running IR pass [layer_norm_fuse_pass]
--- Running IR pass [attention_lstm_fuse_pass]
--- Running IR pass [seqconv_eltadd_relu_fuse_pass]
--- Running IR pass [seqpool_cvm_concat_fuse_pass]
--- Running IR pass [mul_lstm_fuse_pass]
--- Running IR pass [fc_gru_fuse_pass]
--- Running IR pass [mul_gru_fuse_pass]
--- Running IR pass [seq_concat_fc_fuse_pass]
--- Running IR pass [gpu_cpu_squeeze2_matmul_fuse_pass]
--- Running IR pass [gpu_cpu_reshape2_matmul_fuse_pass]
--- Running IR pass [gpu_cpu_flatten2_matmul_fuse_pass]
--- Running IR pass [matmul_v2_scale_fuse_pass]
--- Running IR pass [gpu_cpu_map_matmul_v2_to_mul_pass]
--- Running IR pass [gpu_cpu_map_matmul_v2_to_matmul_pass]
W0728 11:09:57.251009  2568 gpu_cpu_map_matmul_to_mul_pass.cc:425] matmul op not support broadcast, please check inputs'shape. 
W0728 11:09:57.251089  2568 gpu_cpu_map_matmul_to_mul_pass.cc:425] matmul op not support broadcast, please check inputs'shape. 
W0728 11:09:57.251137  2568 gpu_cpu_map_matmul_to_mul_pass.cc:425] matmul op not support broadcast, please check inputs'shape. 
W0728 11:09:57.251505  2568 gpu_cpu_map_matmul_to_mul_pass.cc:425] matmul op not support broadcast, please check inputs'shape. 
W0728 11:09:57.251556  2568 gpu_cpu_map_matmul_to_mul_pass.cc:425] matmul op not support broadcast, please check inputs'shape. 
W0728 11:09:57.251602  2568 gpu_cpu_map_matmul_to_mul_pass.cc:425] matmul op not support broadcast, please check inputs'shape. 
W0728 11:09:57.251648  2568 gpu_cpu_map_matmul_to_mul_pass.cc:425] matmul op not support broadcast, please check inputs'shape. 
W0728 11:09:57.251693  2568 gpu_cpu_map_matmul_to_mul_pass.cc:425] matmul op not support broadcast, please check inputs'shape. 
W0728 11:09:57.251739  2568 gpu_cpu_map_matmul_to_mul_pass.cc:425] matmul op not support broadcast, please check inputs'shape. 
W0728 11:09:57.251957  2568 gpu_cpu_map_matmul_to_mul_pass.cc:425] matmul op not support broadcast, please check inputs'shape. 
W0728 11:09:57.252005  2568 gpu_cpu_map_matmul_to_mul_pass.cc:425] matmul op not support broadcast, please check inputs'shape. 
W0728 11:09:57.252050  2568 gpu_cpu_map_matmul_to_mul_pass.cc:425] matmul op not support broadcast, please check inputs'shape. 
W0728 11:09:57.252095  2568 gpu_cpu_map_matmul_to_mul_pass.cc:425] matmul op not support broadcast, please check inputs'shape. 
W0728 11:09:57.252140  2568 gpu_cpu_map_matmul_to_mul_pass.cc:425] matmul op not support broadcast, please check inputs'shape. 
W0728 11:09:57.252185  2568 gpu_cpu_map_matmul_to_mul_pass.cc:425] matmul op not support broadcast, please check inputs'shape. 
W0728 11:09:57.252383  2568 gpu_cpu_map_matmul_to_mul_pass.cc:425] matmul op not support broadcast, please check inputs'shape. 
W0728 11:09:57.252430  2568 gpu_cpu_map_matmul_to_mul_pass.cc:425] matmul op not support broadcast, please check inputs'shape. 
W0728 11:09:57.252475  2568 gpu_cpu_map_matmul_to_mul_pass.cc:425] matmul op not support broadcast, please check inputs'shape. 
W0728 11:09:57.252521  2568 gpu_cpu_map_matmul_to_mul_pass.cc:425] matmul op not support broadcast, please check inputs'shape. 
W0728 11:09:57.252565  2568 gpu_cpu_map_matmul_to_mul_pass.cc:425] matmul op not support broadcast, please check inputs'shape. 
W0728 11:09:57.252610  2568 gpu_cpu_map_matmul_to_mul_pass.cc:425] matmul op not support broadcast, please check inputs'shape. 
W0728 11:09:57.252841  2568 gpu_cpu_map_matmul_to_mul_pass.cc:425] matmul op not support broadcast, please check inputs'shape. 
W0728 11:09:57.252889  2568 gpu_cpu_map_matmul_to_mul_pass.cc:425] matmul op not support broadcast, please check inputs'shape. 
W0728 11:09:57.252934  2568 gpu_cpu_map_matmul_to_mul_pass.cc:425] matmul op not support broadcast, please check inputs'shape. 
I0728 11:09:57.253104  2568 fuse_pass_base.cc:59] ---  detected 10 subgraphs
--- Running IR pass [matmul_scale_fuse_pass]
--- Running IR pass [gpu_cpu_map_matmul_to_mul_pass]
--- Running IR pass [fc_fuse_pass]
--- Running IR pass [repeated_fc_relu_fuse_pass]
--- Running IR pass [squared_mat_sub_fuse_pass]
--- Running IR pass [conv_bn_fuse_pass]
--- Running IR pass [conv_eltwiseadd_bn_fuse_pass]
--- Running IR pass [conv_transpose_bn_fuse_pass]
--- Running IR pass [conv_transpose_eltwiseadd_bn_fuse_pass]
--- Running IR pass [is_test_pass]
--- Running IR pass [constant_folding_pass]
--- Running IR pass [squeeze2_transpose2_onednn_fuse_pass]
--- fused 0 squeeze2 with transpose2
--- Running IR pass [depthwise_conv_mkldnn_pass]
--- Running IR pass [conv_bn_fuse_pass]
--- Running IR pass [conv_eltwiseadd_bn_fuse_pass]
--- Running IR pass [conv_affine_channel_mkldnn_fuse_pass]
--- Running IR pass [conv_transpose_bn_fuse_pass]
--- Running IR pass [conv_transpose_eltwiseadd_bn_fuse_pass]
--- Running IR pass [conv_bias_mkldnn_fuse_pass]
--- Running IR pass [conv_transpose_bias_mkldnn_fuse_pass]
--- Running IR pass [conv_elementwise_add_mkldnn_fuse_pass]
--- Running IR pass [conv_activation_mkldnn_fuse_pass]
--- Running IR pass [scale_matmul_fuse_pass]
I0728 11:09:57.395366  2568 fuse_pass_base.cc:59] ---  detected 4 subgraphs
---    fused 4 scale with matmul
--- Running IR pass [reshape_transpose_matmul_mkldnn_fuse_pass]
I0728 11:09:57.409730  2568 fuse_pass_base.cc:59] ---  detected 12 subgraphs
---    fused 12 reshape + transpose + matmul with reshape's xshape with transpose's xshape
--- Running IR pass [matmul_transpose_reshape_mkldnn_fuse_pass]
I0728 11:09:57.421487  2568 fuse_pass_base.cc:59] ---  detected 4 subgraphs
---    fused 4 fused_matmul + transpose + reshape patterns
--- Running IR pass [matmul_elementwise_add_mkldnn_fuse_pass]
I0728 11:09:57.425412  2568 fuse_pass_base.cc:59] ---  detected 4 subgraphs
---    fused 4 fused_matmul (as x) with elementwise_add
I0728 11:09:57.428108  2568 fuse_pass_base.cc:59] ---  detected 2 subgraphs
---    fused 2 matmul (as x) with elementwise_add
W0728 11:09:57.432423  2568 op_compat_sensible_pass.cc:232]  Check the Attr(axis) of Op(elementwise_add) in pass(matmul_elementwise_add_mkldnn_fuse_pass) failed!
W0728 11:09:57.432436  2568 matmul_elementwise_add_mkldnn_fuse_pass.cc:64] op compat for matmul_elementwise_add_mkldnn_fuse_pass failed.
W0728 11:09:57.432500  2568 op_compat_sensible_pass.cc:232]  Check the Attr(axis) of Op(elementwise_add) in pass(matmul_elementwise_add_mkldnn_fuse_pass) failed!
W0728 11:09:57.432503  2568 matmul_elementwise_add_mkldnn_fuse_pass.cc:64] op compat for matmul_elementwise_add_mkldnn_fuse_pass failed.
W0728 11:09:57.432565  2568 op_compat_sensible_pass.cc:232]  Check the Attr(axis) of Op(elementwise_add) in pass(matmul_elementwise_add_mkldnn_fuse_pass) failed!
W0728 11:09:57.432569  2568 matmul_elementwise_add_mkldnn_fuse_pass.cc:64] op compat for matmul_elementwise_add_mkldnn_fuse_pass failed.
W0728 11:09:57.432653  2568 op_compat_sensible_pass.cc:232]  Check the Attr(axis) of Op(elementwise_add) in pass(matmul_elementwise_add_mkldnn_fuse_pass) failed!
W0728 11:09:57.432657  2568 matmul_elementwise_add_mkldnn_fuse_pass.cc:64] op compat for matmul_elementwise_add_mkldnn_fuse_pass failed.
W0728 11:09:57.432719  2568 op_compat_sensible_pass.cc:232]  Check the Attr(axis) of Op(elementwise_add) in pass(matmul_elementwise_add_mkldnn_fuse_pass) failed!
W0728 11:09:57.432723  2568 matmul_elementwise_add_mkldnn_fuse_pass.cc:64] op compat for matmul_elementwise_add_mkldnn_fuse_pass failed.
W0728 11:09:57.432791  2568 op_compat_sensible_pass.cc:232]  Check the Attr(axis) of Op(elementwise_add) in pass(matmul_elementwise_add_mkldnn_fuse_pass) failed!
W0728 11:09:57.432803  2568 matmul_elementwise_add_mkldnn_fuse_pass.cc:64] op compat for matmul_elementwise_add_mkldnn_fuse_pass failed.
W0728 11:09:57.432865  2568 op_compat_sensible_pass.cc:232]  Check the Attr(axis) of Op(elementwise_add) in pass(matmul_elementwise_add_mkldnn_fuse_pass) failed!
W0728 11:09:57.432868  2568 matmul_elementwise_add_mkldnn_fuse_pass.cc:64] op compat for matmul_elementwise_add_mkldnn_fuse_pass failed.
W0728 11:09:57.432930  2568 op_compat_sensible_pass.cc:232]  Check the Attr(axis) of Op(elementwise_add) in pass(matmul_elementwise_add_mkldnn_fuse_pass) failed!
W0728 11:09:57.432933  2568 matmul_elementwise_add_mkldnn_fuse_pass.cc:64] op compat for matmul_elementwise_add_mkldnn_fuse_pass failed.
W0728 11:09:57.432994  2568 op_compat_sensible_pass.cc:232]  Check the Attr(axis) of Op(elementwise_add) in pass(matmul_elementwise_add_mkldnn_fuse_pass) failed!
W0728 11:09:57.432998  2568 matmul_elementwise_add_mkldnn_fuse_pass.cc:64] op compat for matmul_elementwise_add_mkldnn_fuse_pass failed.
W0728 11:09:57.433059  2568 op_compat_sensible_pass.cc:232]  Check the Attr(axis) of Op(elementwise_add) in pass(matmul_elementwise_add_mkldnn_fuse_pass) failed!
W0728 11:09:57.433063  2568 matmul_elementwise_add_mkldnn_fuse_pass.cc:64] op compat for matmul_elementwise_add_mkldnn_fuse_pass failed.
W0728 11:09:57.433125  2568 op_compat_sensible_pass.cc:232]  Check the Attr(axis) of Op(elementwise_add) in pass(matmul_elementwise_add_mkldnn_fuse_pass) failed!
W0728 11:09:57.433127  2568 matmul_elementwise_add_mkldnn_fuse_pass.cc:64] op compat for matmul_elementwise_add_mkldnn_fuse_pass failed.
W0728 11:09:57.433189  2568 op_compat_sensible_pass.cc:232]  Check the Attr(axis) of Op(elementwise_add) in pass(matmul_elementwise_add_mkldnn_fuse_pass) failed!
W0728 11:09:57.433192  2568 matmul_elementwise_add_mkldnn_fuse_pass.cc:64] op compat for matmul_elementwise_add_mkldnn_fuse_pass failed.
W0728 11:09:57.433254  2568 op_compat_sensible_pass.cc:232]  Check the Attr(axis) of Op(elementwise_add) in pass(matmul_elementwise_add_mkldnn_fuse_pass) failed!
W0728 11:09:57.433257  2568 matmul_elementwise_add_mkldnn_fuse_pass.cc:64] op compat for matmul_elementwise_add_mkldnn_fuse_pass failed.
W0728 11:09:57.433318  2568 op_compat_sensible_pass.cc:232]  Check the Attr(axis) of Op(elementwise_add) in pass(matmul_elementwise_add_mkldnn_fuse_pass) failed!
W0728 11:09:57.433321  2568 matmul_elementwise_add_mkldnn_fuse_pass.cc:64] op compat for matmul_elementwise_add_mkldnn_fuse_pass failed.
W0728 11:09:57.433383  2568 op_compat_sensible_pass.cc:232]  Check the Attr(axis) of Op(elementwise_add) in pass(matmul_elementwise_add_mkldnn_fuse_pass) failed!
W0728 11:09:57.433387  2568 matmul_elementwise_add_mkldnn_fuse_pass.cc:64] op compat for matmul_elementwise_add_mkldnn_fuse_pass failed.
W0728 11:09:57.433447  2568 op_compat_sensible_pass.cc:232]  Check the Attr(axis) of Op(elementwise_add) in pass(matmul_elementwise_add_mkldnn_fuse_pass) failed!
W0728 11:09:57.433450  2568 matmul_elementwise_add_mkldnn_fuse_pass.cc:64] op compat for matmul_elementwise_add_mkldnn_fuse_pass failed.
W0728 11:09:57.433512  2568 op_compat_sensible_pass.cc:232]  Check the Attr(axis) of Op(elementwise_add) in pass(matmul_elementwise_add_mkldnn_fuse_pass) failed!
W0728 11:09:57.433516  2568 matmul_elementwise_add_mkldnn_fuse_pass.cc:64] op compat for matmul_elementwise_add_mkldnn_fuse_pass failed.
W0728 11:09:57.433576  2568 op_compat_sensible_pass.cc:232]  Check the Attr(axis) of Op(elementwise_add) in pass(matmul_elementwise_add_mkldnn_fuse_pass) failed!
W0728 11:09:57.433579  2568 matmul_elementwise_add_mkldnn_fuse_pass.cc:64] op compat for matmul_elementwise_add_mkldnn_fuse_pass failed.
W0728 11:09:57.433640  2568 op_compat_sensible_pass.cc:232]  Check the Attr(axis) of Op(elementwise_add) in pass(matmul_elementwise_add_mkldnn_fuse_pass) failed!
W0728 11:09:57.433643  2568 matmul_elementwise_add_mkldnn_fuse_pass.cc:64] op compat for matmul_elementwise_add_mkldnn_fuse_pass failed.
W0728 11:09:57.433707  2568 op_compat_sensible_pass.cc:232]  Check the Attr(axis) of Op(elementwise_add) in pass(matmul_elementwise_add_mkldnn_fuse_pass) failed!
W0728 11:09:57.433710  2568 matmul_elementwise_add_mkldnn_fuse_pass.cc:64] op compat for matmul_elementwise_add_mkldnn_fuse_pass failed.
W0728 11:09:57.433779  2568 op_compat_sensible_pass.cc:232]  Check the Attr(axis) of Op(elementwise_add) in pass(matmul_elementwise_add_mkldnn_fuse_pass) failed!
W0728 11:09:57.433781  2568 matmul_elementwise_add_mkldnn_fuse_pass.cc:64] op compat for matmul_elementwise_add_mkldnn_fuse_pass failed.
W0728 11:09:57.433842  2568 op_compat_sensible_pass.cc:232]  Check the Attr(axis) of Op(elementwise_add) in pass(matmul_elementwise_add_mkldnn_fuse_pass) failed!
W0728 11:09:57.433845  2568 matmul_elementwise_add_mkldnn_fuse_pass.cc:64] op compat for matmul_elementwise_add_mkldnn_fuse_pass failed.
W0728 11:09:57.433907  2568 op_compat_sensible_pass.cc:232]  Check the Attr(axis) of Op(elementwise_add) in pass(matmul_elementwise_add_mkldnn_fuse_pass) failed!
W0728 11:09:57.433909  2568 matmul_elementwise_add_mkldnn_fuse_pass.cc:64] op compat for matmul_elementwise_add_mkldnn_fuse_pass failed.
W0728 11:09:57.433971  2568 op_compat_sensible_pass.cc:232]  Check the Attr(axis) of Op(elementwise_add) in pass(matmul_elementwise_add_mkldnn_fuse_pass) failed!
W0728 11:09:57.433974  2568 matmul_elementwise_add_mkldnn_fuse_pass.cc:64] op compat for matmul_elementwise_add_mkldnn_fuse_pass failed.
--- Running IR pass [matmul_activation_mkldnn_fuse_pass]
I0728 11:09:57.439098  2568 fuse_pass_base.cc:59] ---  detected 1 subgraphs
---    fused 1 fused_matmul with sigmoid activation
I0728 11:09:57.440340  2568 fuse_pass_base.cc:59] ---  detected 1 subgraphs
---    fused 1 fused_matmul with tanh activation
--- Running IR pass [fc_mkldnn_pass]
--- Running IR pass [fc_act_mkldnn_fuse_pass]
--- Running IR pass [fc_elementwise_add_mkldnn_fuse_pass]
--- Running IR pass [batch_norm_act_fuse_pass]
--- Running IR pass [softplus_activation_onednn_fuse_pass]
--- Running IR pass [shuffle_channel_mkldnn_detect_pass]
--- Running IR pass [elementwise_act_onednn_fuse_pass]
I0728 11:09:57.462807  2568 fuse_pass_base.cc:59] ---  detected 4 subgraphs
---    fused 4 elementwise_add with gelu activation
--- Running IR pass [layer_norm_onednn_optimization_pass]
I0728 11:09:57.475025  2568 fuse_pass_base.cc:59] ---  detected 9 subgraphs
---    optimized 9 layer_norms by merging Scale and Bias
--- Running IR pass [operator_scale_onednn_fuse_pass]
--- Running IR pass [operator_unsqueeze2_onednn_fuse_pass]
--- Running IR pass [operator_reshape2_onednn_fuse_pass]
--- Running analysis [save_optimized_model_pass]
W0728 11:09:57.483620  2568 save_optimized_model_pass.cc:28] save_optim_cache_model is turned off, skip save_optimized_model_pass
--- Running analysis [ir_params_sync_among_devices_pass]
--- Running analysis [adjust_cudnn_workspace_size_pass]
--- Running analysis [inference_op_replace_pass]
--- Running analysis [ir_graph_to_program_pass]
I0728 11:09:57.519243  2568 analysis_predictor.cc:1660] ======= optimize end =======
I0728 11:09:57.521556  2568 naive_executor.cc:164] ---  skip [feed], feed -> attention_mask_pos
I0728 11:09:57.521575  2568 naive_executor.cc:164] ---  skip [feed], feed -> token_type_ids_pos
I0728 11:09:57.521579  2568 naive_executor.cc:164] ---  skip [feed], feed -> input_ids_pos
I0728 11:09:57.522771  2568 naive_executor.cc:164] ---  skip [sigmoid_0.tmp_0], fetch -> fetch
I0728 11:09:57.525840  2568 onednn_context.cc:81] oneDNN v2.7.3

以上便是，项目中出现的错误链，因此，想询问如何基于高版本的paddle实现paddleslim量化（embedding 量化）。 @qili93

qili93 · 2023-07-28T03:43:05Z

你好，PaddleSlim的版本需要和 PaddlePaddle版本保持一致，目前 PaddleSlim 尚未发布对应v2.5.0或者v2.5.1的版本，可以尝试使用 PaddleSlim 的 release/2.5 分支得到支持 PaddlePaddle 2.5 版本的 PaddleSlim 安装包，具体方式为

# 下载 release/2.5 分支的代码
git clone https://github.com/PaddlePaddle/PaddleSlim.git -b release/2.5

# 源码编译安装
cd PaddleSlim
python setup.py install

PaddlePaddle仍旧使用 v2.5.0 或者 v2.5.1 的版本即可，请确认下更新 PaddleSlim 版本之后问题是否仍旧存在，谢谢！

Liangwei-0521 · 2023-07-28T08:37:47Z

回复：
根据您提供的建议，我使用 PaddleSlim 的 release/2.5 分支解决了上述版本冲突的问题。所用版本信息如下：

paddlenlp                      2.5.2
paddlepaddle                   2.5.0
paddlepaddle-gpu               2.5.0
paddleslim                     2.5.0

根据推理时输出的信息，

[2023-07-28 16:18:20,142] [    INFO] - We are using <class 'paddlenlp.transformers.ernie.fast_tokenizer.ErnieFastTokenizer'> to load '/home/aistudio/checkpoint/ernie-3.0-nano-zh'.
46665it [00:04, 10982.13it/s]
I0728 16:18:25.369899  1837 analysis_predictor.cc:1471] MKLDNN is enabled
--- Running analysis [ir_graph_build_pass]
I0728 16:18:25.375305  1837 executor.cc:187] Old Executor is Running.
--- Running analysis [ir_analysis_pass]
--- Running IR pass [mkldnn_placement_pass]
--- Running IR pass [simplify_with_basic_ops_pass]
--- Running IR pass [layer_norm_fuse_pass]
--- Running IR pass [attention_lstm_fuse_pass]
--- Running IR pass [seqconv_eltadd_relu_fuse_pass]
--- Running IR pass [seqpool_cvm_concat_fuse_pass]
--- Running IR pass [mul_lstm_fuse_pass]
--- Running IR pass [fc_gru_fuse_pass]
--- Running IR pass [mul_gru_fuse_pass]
--- Running IR pass [seq_concat_fc_fuse_pass]
--- Running IR pass [gpu_cpu_squeeze2_matmul_fuse_pass]
--- Running IR pass [gpu_cpu_reshape2_matmul_fuse_pass]
--- Running IR pass [gpu_cpu_flatten2_matmul_fuse_pass]
--- Running IR pass [matmul_v2_scale_fuse_pass]
--- Running IR pass [gpu_cpu_map_matmul_v2_to_mul_pass]
--- Running IR pass [gpu_cpu_map_matmul_v2_to_matmul_pass]
W0728 16:18:25.427968  1837 gpu_cpu_map_matmul_to_mul_pass.cc:425] matmul op not support broadcast, please check inputs'shape. 
W0728 16:18:25.428051  1837 gpu_cpu_map_matmul_to_mul_pass.cc:425] matmul op not support broadcast, please check inputs'shape. 
W0728 16:18:25.428100  1837 gpu_cpu_map_matmul_to_mul_pass.cc:425] matmul op not support broadcast, please check inputs'shape. 
W0728 16:18:25.428479  1837 gpu_cpu_map_matmul_to_mul_pass.cc:425] matmul op not support broadcast, please check inputs'shape. 
W0728 16:18:25.428534  1837 gpu_cpu_map_matmul_to_mul_pass.cc:425] matmul op not support broadcast, please check inputs'shape. 
W0728 16:18:25.428581  1837 gpu_cpu_map_matmul_to_mul_pass.cc:425] matmul op not support broadcast, please check inputs'shape. 
W0728 16:18:25.428628  1837 gpu_cpu_map_matmul_to_mul_pass.cc:425] matmul op not support broadcast, please check inputs'shape. 
W0728 16:18:25.428675  1837 gpu_cpu_map_matmul_to_mul_pass.cc:425] matmul op not support broadcast, please check inputs'shape. 
W0728 16:18:25.428722  1837 gpu_cpu_map_matmul_to_mul_pass.cc:425] matmul op not support broadcast, please check inputs'shape. 
W0728 16:18:25.428958  1837 gpu_cpu_map_matmul_to_mul_pass.cc:425] matmul op not support broadcast, please check inputs'shape. 
W0728 16:18:25.429009  1837 gpu_cpu_map_matmul_to_mul_pass.cc:425] matmul op not support broadcast, please check inputs'shape. 
W0728 16:18:25.429056  1837 gpu_cpu_map_matmul_to_mul_pass.cc:425] matmul op not support broadcast, please check inputs'shape. 
W0728 16:18:25.429103  1837 gpu_cpu_map_matmul_to_mul_pass.cc:425] matmul op not support broadcast, please check inputs'shape. 
W0728 16:18:25.429150  1837 gpu_cpu_map_matmul_to_mul_pass.cc:425] matmul op not support broadcast, please check inputs'shape. 
W0728 16:18:25.429196  1837 gpu_cpu_map_matmul_to_mul_pass.cc:425] matmul op not support broadcast, please check inputs'shape. 
W0728 16:18:25.429419  1837 gpu_cpu_map_matmul_to_mul_pass.cc:425] matmul op not support broadcast, please check inputs'shape. 
W0728 16:18:25.429469  1837 gpu_cpu_map_matmul_to_mul_pass.cc:425] matmul op not support broadcast, please check inputs'shape. 
W0728 16:18:25.429517  1837 gpu_cpu_map_matmul_to_mul_pass.cc:425] matmul op not support broadcast, please check inputs'shape. 
W0728 16:18:25.429563  1837 gpu_cpu_map_matmul_to_mul_pass.cc:425] matmul op not support broadcast, please check inputs'shape. 
W0728 16:18:25.429611  1837 gpu_cpu_map_matmul_to_mul_pass.cc:425] matmul op not support broadcast, please check inputs'shape. 
W0728 16:18:25.429658  1837 gpu_cpu_map_matmul_to_mul_pass.cc:425] matmul op not support broadcast, please check inputs'shape. 
W0728 16:18:25.429883  1837 gpu_cpu_map_matmul_to_mul_pass.cc:425] matmul op not support broadcast, please check inputs'shape. 
W0728 16:18:25.429934  1837 gpu_cpu_map_matmul_to_mul_pass.cc:425] matmul op not support broadcast, please check inputs'shape. 
W0728 16:18:25.429989  1837 gpu_cpu_map_matmul_to_mul_pass.cc:425] matmul op not support broadcast, please check inputs'shape. 
I0728 16:18:25.430182  1837 fuse_pass_base.cc:59] ---  detected 10 subgraphs
--- Running IR pass [matmul_scale_fuse_pass]
--- Running IR pass [gpu_cpu_map_matmul_to_mul_pass]
--- Running IR pass [fc_fuse_pass]
--- Running IR pass [repeated_fc_relu_fuse_pass]
--- Running IR pass [squared_mat_sub_fuse_pass]
--- Running IR pass [conv_bn_fuse_pass]
--- Running IR pass [conv_eltwiseadd_bn_fuse_pass]
--- Running IR pass [conv_transpose_bn_fuse_pass]
--- Running IR pass [conv_transpose_eltwiseadd_bn_fuse_pass]
--- Running IR pass [is_test_pass]
--- Running IR pass [constant_folding_pass]
--- Running IR pass [squeeze2_transpose2_onednn_fuse_pass]
--- fused 0 squeeze2 with transpose2
--- Running IR pass [depthwise_conv_mkldnn_pass]
--- Running IR pass [conv_bn_fuse_pass]
--- Running IR pass [conv_eltwiseadd_bn_fuse_pass]
--- Running IR pass [conv_affine_channel_mkldnn_fuse_pass]
--- Running IR pass [conv_transpose_bn_fuse_pass]
--- Running IR pass [conv_transpose_eltwiseadd_bn_fuse_pass]
--- Running IR pass [conv_bias_mkldnn_fuse_pass]
--- Running IR pass [conv_transpose_bias_mkldnn_fuse_pass]
--- Running IR pass [conv_elementwise_add_mkldnn_fuse_pass]
--- Running IR pass [conv_activation_mkldnn_fuse_pass]
--- Running IR pass [scale_matmul_fuse_pass]
I0728 16:18:25.540614  1837 fuse_pass_base.cc:59] ---  detected 4 subgraphs
---    fused 4 scale with matmul
--- Running IR pass [reshape_transpose_matmul_mkldnn_fuse_pass]
I0728 16:18:25.551424  1837 fuse_pass_base.cc:59] ---  detected 12 subgraphs
---    fused 12 reshape + transpose + matmul with reshape's xshape
--- Running IR pass [matmul_transpose_reshape_mkldnn_fuse_pass]
--- Running IR pass [matmul_elementwise_add_mkldnn_fuse_pass]
I0728 16:18:25.565969  1837 fuse_pass_base.cc:59] ---  detected 4 subgraphs
---    fused 4 fused_matmul (as x) with elementwise_add
I0728 16:18:25.568967  1837 fuse_pass_base.cc:59] ---  detected 2 subgraphs
---    fused 2 matmul (as x) with elementwise_add
I0728 16:18:25.576004  1837 fuse_pass_base.cc:59] ---  detected 24 subgraphs
---    fused 24 matmul_v2 (as x) with elementwise_add
--- Running IR pass [matmul_activation_mkldnn_fuse_pass]
--- Running IR pass [fc_mkldnn_pass]
--- Running IR pass [fc_act_mkldnn_fuse_pass]
--- Running IR pass [fc_elementwise_add_mkldnn_fuse_pass]
--- Running IR pass [batch_norm_act_fuse_pass]
--- Running IR pass [softplus_activation_onednn_fuse_pass]
--- Running IR pass [shuffle_channel_mkldnn_detect_pass]
--- Running IR pass [elementwise_act_onednn_fuse_pass]
--- Running IR pass [layer_norm_onednn_optimization_pass]
I0728 16:18:25.623422  1837 fuse_pass_base.cc:59] ---  detected 9 subgraphs
---    optimized 9 layer_norms by merging Scale and Bias
--- Running IR pass [operator_scale_onednn_fuse_pass]
--- Running IR pass [operator_unsqueeze2_onednn_fuse_pass]
--- Running IR pass [operator_reshape2_onednn_fuse_pass]
--- Running analysis [save_optimized_model_pass]
W0728 16:18:25.632709  1837 save_optimized_model_pass.cc:28] save_optim_cache_model is turned off, skip save_optimized_model_pass
--- Running analysis [ir_params_sync_among_devices_pass]
--- Running analysis [adjust_cudnn_workspace_size_pass]
--- Running analysis [inference_op_replace_pass]
--- Running analysis [ir_graph_to_program_pass]
I0728 16:18:25.670608  1837 analysis_predictor.cc:1660] ======= optimize end =======
I0728 16:18:25.6708  1837 analysis_predictor.cc:1660] ======= optimize end =======
I0728 16:18:25.673341  1837 naive_executor.cc:164] ---  skip [feed], feed -> attention_mask_pos
I0728 16:18:25.673357  1833341  1837 naive_executor.cc:164] ---  skip [feed], feed -> attention_mask_pos
I0728 16:18:25.673357  1837 naive_executor.cc:164] ---  skip [feed], feed -> token_type_ids_pos
I0728 16:18:25.673369  1837 naive_executor.cc:164] ---  skip [feed], feed -> input_ids_pos
I0728 16:18:25.675235  1837 naive_executor.cc:164] ---  skip [save_infer_model/scale_0.tmp_0], fetch -> fetch
I0728 16:18:25.677532  1837 onednn_context.cc:81] oneDNN v2.7.3

关于“failed!”的Warning解决，但是关于gpu_cpu_map_matmul_to_mul_pass.cc:425] matmul op not support broadcast, please check inputs'shape.的问题想进一步咨询：出现关于shape的warning，初步认为会不会是实现QAT量化时在保存模型时指定input_spec均为[None, None]导致，还是另有原因（QAT的实现方法是否正确？）

我们实现QAT的代码如下：

   import paddleslim as slim
   from paddleslim import QAT
    quant_config = {
        'weight_preprocess_type': None,
        'activation_preprocess_type': None,
        'weight_quantize_type': 'channel_wise_abs_max',
        'activation_quantize_type': 'moving_average_abs_max',
        'weight_bits': 8,
        'activation_bits': 8,
        'dtype': 'int8',
        'window_size': 10000,
        'moving_rate': 0.9,
        'quantizable_layer_type': ['Linear'],
    }
    quanter = QAT(config=quant_config)
    # 量化模型
    quanter.quantize(model)  ## model = quanter.quantize(model)

    # 训练模型
    train(args, model, train_data_loader, dev_data_loader, dev_data)

    # 保存量化模型
    print('------quant------')
    quanter.save_quantized_model(
        model,
        path=args.quant_save_path,
        input_spec=[paddle.static.InputSpec(shape=[None, None], dtype="int64"),    # inputs_ids_pos
                    paddle.static.InputSpec(shape=[None, None], dtype="int64"),             # token_type_ids_pos
                    paddle.static.InputSpec(shape=[None, None], dtype="int64"),             # attention mask
    ])
    print('------embedding quant------')

以上便是进一步需要咨询的问题。 @qili93

Liangwei-0521 · 2023-07-28T10:50:25Z

经过测试，模型的大小减至原先模型大小的四分之一，但是推理速度严重地比传统模型慢，分两种情况：
1、QAT量化：比较传统模型，延迟10%
2、QAT量化+embedding量化（低版本paddleslim）：比较传统模型，延迟一倍。
3、QAT量化+embedding量化（高版本paddleslim，即上述git链接）：比较传统模型，延迟多倍。
请问是否可以提供基于paddlenlp的ernie-3.0-nano-zh模型，利用paddleslim进行QAT量化+embedding量化的最新参考代码/文档。 @qili93

qili93 · 2023-07-29T11:04:19Z

请 @wanghaoshuang 帮忙看下Slim QAT量化的问题，谢谢！

wanghaoshuang · 2023-07-31T06:38:30Z

经过测试，模型的大小减至原先模型大小的四分之一，但是推理速度严重地比传统模型慢，分两种情况：
1、QAT量化：比较传统模型，延迟10%
2、QAT量化+embedding量化（低版本paddleslim）：比较传统模型，延迟一倍。
3、QAT量化+embedding量化（高版本paddleslim，即上述git链接）：比较传统模型，延迟多倍。
请问是否可以提供基于paddlenlp的ernie-3.0-nano-zh模型，利用paddleslim进行QAT量化+embedding量化的最新参考代码/文档。

您 QAT 实现的没问题。
PaddleSlim 产出的量化模型后，是否有加速还依赖推理库的支持。目前 BERT模型只支持在 PaddleInference + GPU 上的量化推理加速，没有在 CPU推理上优化，所以不能保证在 CPU 上有加速。

Liangwei-0521 · 2023-07-31T07:28:22Z

根据您的建议，我们将模型放置GPU尝试进行推理（paddlepaddle-gpu=2.5.1)，存在以下的报错：

@wanghaoshuang

--- Running analysis [ir_graph_build_pass]
--- Running analysis [ir_analysis_pass]
--- Running IR pass [map_op_to_another_pass]
--- Running IR pass [identity_scale_op_clean_pass]
I0731 15:22:02.315886   870 fuse_pass_base.cc:59] ---  detected 1 subgraphs
--- Running IR pass [is_test_pass]
--- Running IR pass [simplify_with_basic_ops_pass]
--- Running IR pass [delete_quant_dequant_linear_op_pass]
I0731 15:22:02.343159   870 fuse_pass_base.cc:59] ---  detected 52 subgraphs
--- Running IR pass [delete_weight_dequant_linear_op_pass]
---------------------------------------------------------------------------
NotImplementedError                       Traceback (most recent call last)
/tmp/ipykernel_870/2712444669.py in <module>
     23 predictor = Predictor(model_dir, device, max_seq_length,
     24                         batch_size, use_tensorrt, precision,
---> 25                         cpu_threads, enable_mkldnn)
     26 results = []
     27 rst_dict = {}

/tmp/ipykernel_870/1978965815.py in __init__(self, model_dir, device, max_seq_length, batch_size, use_tensorrt, precision, cpu_threads, enable_mkldnn)
     50 
     51         config.switch_use_feed_fetch_ops(False)
---> 52         self.predictor = paddle.inference.create_predictor(config)
     53         self.input_handles = [
     54             self.predictor.get_input_handle(name)

NotImplementedError: (Unimplemented) Delete Weight Dequant Linear Op Pass is not supported for per-channel quantization (at ../paddle/fluid/framework/ir/delete_weight_dequant_linear_op_pass.cc:130)

具体的推理代码如下：

class Predictor(object):
    def __init__(self,
                 model_dir,
                 device="gpu",
                 max_seq_length=128,
                 batch_size=32,
                 use_tensorrt=False,
                 precision="fp32",
                 cpu_threads=10,
                 enable_mkldnn=False):
        self.max_seq_length = max_seq_length
        self.batch_size = batch_size

        model_file = model_dir + "rank_quant_emb.pdmodel"
        params_file = model_dir + "rank_quant_emb.pdiparams"
        if not os.path.exists(model_file):
            raise ValueError("not find model file path {}".format(model_file))
        if not os.path.exists(params_file):
            raise ValueError("not find params file path {}".format(params_file))
        config = paddle.inference.Config(model_file, params_file)

        if device == "gpu":
            # set GPU configs accordingly
            # such as intialize the gpu memory, enable tensorrt
            config.enable_use_gpu(100, 0)
            precision_map = {
                "fp16": inference.PrecisionType.Half,
                "fp32": inference.PrecisionType.Float32,
                "int8": inference.PrecisionType.Int8
            }
            precision_mode = precision_map[precision]

            if use_tensorrt:
                config.enable_tensorrt_engine(
                    max_batch_size=batch_size,
                    min_subgraph_size=30,
                    precision_mode=precision_mode)
        elif device == "cpu":
            # set CPU configs accordingly,
            # such as enable_mkldnn, set_cpu_math_library_num_threads
            config.disable_gpu()
            if enable_mkldnn:
                # cache 10 different shapes for mkldnn to avoid memory leak
                config.set_mkldnn_cache_capacity(10)
                config.enable_mkldnn()
            config.set_cpu_math_library_num_threads(cpu_threads)
        elif device == "xpu":
            # set XPU configs accordingly
            config.enable_xpu(100)

        config.switch_use_feed_fetch_ops(False)
        self.predictor = paddle.inference.create_predictor(config)
        self.input_handles = [
            self.predictor.get_input_handle(name)
            for name in self.predictor.get_input_names()
        ]
        self.output_handle = self.predictor.get_output_handle(
            self.predictor.get_output_names()[0])

    def predict(self, data, tokenizer):

        examples = []
        for text in data:
            input_ids, token_type_ids, attention_mask = convert_example_ranking(
                text,
                tokenizer,
                max_seq_length=self.max_seq_length,
                is_test=True)
            examples.append((input_ids, token_type_ids, attention_mask))

        batchify_fn = lambda samples, fn=Tuple(
            Pad(axis=0, pad_val=tokenizer.pad_token_id, dtype='int64'),  # input_ids
            Pad(axis=0, pad_val=tokenizer.pad_token_type_id, dtype='int64'),  # token_type_ids
            Pad(axis=0, pad_val=tokenizer.pad_token_id, dtype='int64'),
        ): fn(samples)

        input_ids, token_type_ids, attention_mask = batchify_fn(examples)
        self.input_handles[0].copy_from_cpu(input_ids)
        self.input_handles[1].copy_from_cpu(token_type_ids)
        self.input_handles[2].copy_from_cpu(attention_mask)
        self.predictor.run()
        sim_score = self.output_handle.copy_to_cpu()
        return sim_score

执行推理代码

import time 

model_dir = '/home/aistudio/checkpoint/'
device = 'gpu'
max_seq_length = 128
batch_size = 32  # 32
# 可以安装对应的Tensorrt之后进行加速
use_tensorrt = False
# 精度，也可以选择fp16，精度几乎无损
precision = 'fp16'
# cpu的线程数目
cpu_threads = 1
# 可以在CPU的情况下进行加速
enable_mkldnn = False
tokenizer = ppnlp.transformers.ErnieTokenizer.from_pretrained('./checkpoint/ernie-3.0-nano-zh')
dev_data = load_dataset(read_text, data_path='/home/aistudio/rank_data/data_v4/all_test.csv', mode='dev', lazy=False) 
batches = [
    dev_data[idx:idx + batch_size]
    for idx in range(0, len(dev_data), batch_size)
]
predictor = Predictor(model_dir, device, max_seq_length,
                        batch_size, use_tensorrt, precision,
                        cpu_threads, enable_mkldnn)
results = []
rst_dict = {}
start = time.time()
# 执行测试
for batch_data in batches:
    results.extend(predictor.predict(batch_data, tokenizer))
end = time.time()
run_time = end - start
print('运行时间：', run_time)

lyuwenyu · 2023-08-02T02:11:59Z

paddlepaddle==2.5.1 在安装时候指定一下版本号试一下

lrp123456 · 2023-09-05T04:42:58Z

#55765 (comment)
请问该问题同学是否解决了，我出现了相同的问题，无法解决

paddle-bot · 2024-09-10T06:33:51Z

Since you haven't replied for more than a year, we have closed this issue/pr.
If the problem is not solved or there is a follow-up one, please reopen it at any time and we will continue to follow up.
由于您超过一年未回复，我们将关闭这个issue/pr。
若问题未解决或有后续问题，请随时重新打开，我们会继续跟进。

Liangwei-0521 added status/new-issue 新建 type/bug-report 报bug labels Jul 28, 2023

paddle-bot bot assigned qili93 Jul 28, 2023

paddle-bot bot added type/debug 帮用户debug status/following-up 跟进中 and removed type/bug-report 报bug status/new-issue 新建 labels Aug 2, 2023

quzhixue-Kimi mentioned this issue Sep 13, 2023

用 CSMSC 数据集训练 FastSpeech2 模型, Cannot import name 'MSRA' from 'paddle.fluid.initializer (..../site-package/fluid/initializer.py)' PaddlePaddle/PaddleSpeech#3508

Closed

yrqs mentioned this issue Sep 27, 2023

paddslim在python3.11下无法使用 PaddlePaddle/PaddleSlim#1797

Open

zah-tane mentioned this issue Nov 17, 2023

[General Issue] Can't get PaddleRS working PaddlePaddle/PaddleRS#177

Closed

paddle-bot bot closed this as completed Sep 10, 2024

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

cannot import name 'MSRA' from 'paddle.fluid.initializer' (/opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages/paddle/fluid/initializer.py) #55765

cannot import name 'MSRA' from 'paddle.fluid.initializer' (/opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages/paddle/fluid/initializer.py) #55765

Liangwei-0521 commented Jul 28, 2023

qili93 commented Jul 28, 2023 •

edited

Loading

Liangwei-0521 commented Jul 28, 2023 •

edited

Loading

qili93 commented Jul 28, 2023

Liangwei-0521 commented Jul 28, 2023

Liangwei-0521 commented Jul 28, 2023 •

edited

Loading

qili93 commented Jul 29, 2023

wanghaoshuang commented Jul 31, 2023

Liangwei-0521 commented Jul 31, 2023 •

edited

Loading

lyuwenyu commented Aug 2, 2023

lrp123456 commented Sep 5, 2023

paddle-bot bot commented Sep 10, 2024

cannot import name 'MSRA' from 'paddle.fluid.initializer' (/opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages/paddle/fluid/initializer.py) #55765

cannot import name 'MSRA' from 'paddle.fluid.initializer' (/opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages/paddle/fluid/initializer.py) #55765

Comments

Liangwei-0521 commented Jul 28, 2023

bug描述 Describe the Bug

其他补充信息 Additional Supplementary Information

qili93 commented Jul 28, 2023 • edited Loading

Liangwei-0521 commented Jul 28, 2023 • edited Loading

qili93 commented Jul 28, 2023

Liangwei-0521 commented Jul 28, 2023

Liangwei-0521 commented Jul 28, 2023 • edited Loading

qili93 commented Jul 29, 2023

wanghaoshuang commented Jul 31, 2023

Liangwei-0521 commented Jul 31, 2023 • edited Loading

根据您的建议，我们将模型放置GPU尝试进行推理（paddlepaddle-gpu=2.5.1)，存在以下的报错：

具体的推理代码如下：

执行推理代码

lyuwenyu commented Aug 2, 2023

lrp123456 commented Sep 5, 2023

paddle-bot bot commented Sep 10, 2024

qili93 commented Jul 28, 2023 •

edited

Loading

Liangwei-0521 commented Jul 28, 2023 •

edited

Loading

Liangwei-0521 commented Jul 28, 2023 •

edited

Loading

Liangwei-0521 commented Jul 31, 2023 •

edited

Loading