Remove cached torch_extensions on CI runners

[ydshieh]

What does this PR do?

• The test

  tests/deepspeed/test_deepspeed.py::TrainerIntegrationDeepSpeed::test_hf_scheduler_ds_optimizer
  

  failed since 2 weeks due to some cache issue. The error message is

  E   ImportError: /github/home/.cache/torch_extensions/py38_cu113/fused_adam/fused_adam.so: undefined symbol: _ZN3c104impl8GPUTrace13gpuTraceStateE`

• After I remove the cache (on the host runners, not inside the running docker) by

  sudo rm -rf /home/github_actions/actions-runner/_work_temp/_github_home/.cache/torch_extensions/py38_cu113/

  the test passes.

• This PR add the following in the workflow file

  rm -rf /github/home/.cache/torch_extensions/

  to avoid the same problem occurring in the future.

Remark: Notice the host directory

/home/github_actions/actions-runner/_work_temp/_github_home/

is mapped to

 /github/home/

inside the running docker (we can see this in the job run page).

[HuggingFaceDocBuilderDev]

The documentation is not available anymore as the PR was closed or merged.

[sgugger]

Thanks for the fix!

[stas00]

Thank you for figuring out how to fix this, @ydshieh