Error when doing fpAnalyser

Liang-Chou

Hi there,

I was doing yolov3 project on docker and follow the steps provided on the document center

http://doc.kneron.com/docs/#toolchain/yolo_example/

Steps before doing quantization seems to be okay for our model. But when I tried to run fpAnalyserCompilerIpevaluator_520.py to do quantization, there's error message raised as follow:

/workspace/miniconda/lib/python3.7/site-packages/numpy/__init__.py:156: UserWarning: mkl-service package failed to import, therefore Intel(R) MKL initialization ensuring its correct out-of-the box operation under condition when Gnu OpenMP had already been loaded by Python process is not assured. Please install mkl-service package, see http://github.com/IntelPython/mkl-service
  from . import _distributor_init
[2021-06-12 08:45:43] [error] [Thread: 26] [/projects_src/kneron_piano/knerex/updater/include/base/process_node/BiasAdjustmentBase.hpp:397] 
Warning: not enough memory to calculate bias adjustment, stopped.
 Increase virtual memory or reduce the number of input images, or both,
 Or set less_DRAM_mode in dump_level in config file to re-run.
Traceback (most recent call last):
  File "/workspace/scripts/fpAnalyserCompilerIpevaluator_520.py", line 43, in <module>
    bie_file = run_knerex(model_config, threads, 520)
  File "/workspace/scripts/utils/run_knerex.py", line 63, in run_knerex
    shutil.copy2(TMP_FOLDER + '/output.quan.wqbi.bie', output_path)
  File "/workspace/miniconda/lib/python3.7/shutil.py", line 266, in copy2
    copyfile(src, dst, follow_symlinks=follow_symlinks)
  File "/workspace/miniconda/lib/python3.7/shutil.py", line 120, in copyfile
    with open(src, 'rb') as fsrc:
FileNotFoundError: [Errno 2] No such file or directory: '/workspace/.tmp/output.quan.wqbi.bie'

which seems to be problem of memory insufficient, here's the things I've tried to eliminate the problem:

1. run the docker with all my laptop's memory (which is 8G) and add swap memory
2. reduce number of input images, to only 1 image
3. tried to add "dump_level": "less_DRAM_mode" into the config file input_params.json, but this step was only the attempt of myself, since I couldn't find the dump_level argument anywhere in the manual of KL520

All these attempts were failed, or maybe I have wrong config on dump_level, I can't see any solution to this.

Please help!

Ethon Lin

Once run with out of memory, the system could keep on some memory issues still.

Please exit from the docker environment, re-login and use 1 image case retry again.