保存模型时提示“RuntimeError: record_file is empty, no layers to be quantized”

问题描述

训练后量化场景，量化因子记录文件中无法找到某层的shift_bit字段，无法进行后续量化模型保存动作，量化流程终止。提示信息如下：

      
       
         
         
           Traceback (most recent call last):
  File "./GenAIModel/amt/tools/tmp_amct_tools.py", line 457, in wrapper
    func(*args, **kwargs)
  File "/raid/AMCT/demo/new_project/test/test_tf_iter37_d16w8.py", line 1806, in test_xxx
    run_case(config, acc=True)
  File "/raid/AMCT/demo/new_project/test/test_tf_iter37_d16w8.py", line 130, in run_case
    res = tf_api_compress(model, input_data, quant_cfg=quant_cfg,data_config=data_config, nuq=nuq, per_channel=per_channel)
  File "./GenAIModel/amt/model_process/infer_quant.py", line 754, in tf_api_compress
    save_model(model, out_puts, record_file, f'{save_path}/{case_name}' )
  File "/raid/AMCT/daily/daily_env/lib/python3.9/site-packages/amct_tensorflow/common/utils/check_params.py", line 43, in wrapper
    return func(*args, **kwargs)
  File "/raid/AMCT/daily/daily_env/lib/python3.9/site-packages/amct_tensorflow/interface/save_model.py", line 68, in save_model
    save_model_inner(pb_model, outputs, record_file, save_path)
  File "/raid/AMCT/daily/daily_env/lib/python3.9/site-packages/amct_tensorflow/interface/save_model.py", line 102, in save_model_inner
    raise RuntimeError(
RuntimeError: record_file is empty, no layers to be quantized. please ensure calibration is finished by checking information!

          

        

      
     

可能原因

网络的尾层节点为量化节点时，如果量化推理时不修改推理的输出节点，则会导致该问题。此场景下，调用quantize_model时，会有warning信息提示输出节点发生变化。

处理建议

该场景下调用quantize_model时，会有warning信息提示输出节点发生变化，在执行校准的过程中，根据提示信息，修改输出节点即可。

父主题： FAQ