Code stuck infinitely when performing Fine-Tuning on CodeSearch even with suggested fixes about codebert

[fengzhangyin]

1. After you did run the process data script, the 17 batch files are test set. And the number of files does not matter.
2. Although there is no any feedback, I think it's training. You can sample some examples (such as 1000, now the number of training examples is 824342. ) from training file and validation file to speed up the detection of whether the code is correct. And you can try to print something during traing: https://github.com/microsoft/CodeBERT/blob/master/CodeBERT/codesearch/run_classifier.py#L98-L105
3. You can change the transformers version to 2.5.0 (the version I used) and try it again.

[atonkamanda]

Switching to transformers==2.5.0 didn't fix the problem.

I will try my luck with the CodeXGlue pipeline as it take less time to train anyway https://github.com/microsoft/CodeXGLUE/tree/main/Text-Code/NL-code-search-WebQuery

Thank you very much for your help.