Error when trying to run unit tests on AMD

[401qingkong]

Hi， I install the rocm deepspeed via the source code. And now I want to test the op in my platform. And I test with the following commands:
cd tests; pytest
But I find many import mistakes shows like below:
/path/of/my/python/installed/python3.7/importlib/__init__.py:127: in import_module return _bootstrap._gcd_import(name[level:], package, level) tests/unit/inference/test_checkpoint_sharding.py:10: in <module> from transformers import AutoConfig, AutoModelForCausalLM E ModuleNotFoundError: No module named 'transformers'
I want to know how to run the test in the project. Hope for your replay

[tjruwase]

Have you installed transformers?

[401qingkong]

Have you installed transformers?

what's your transformers mean? I compile the project from source with the args DS_BUILD_TRANSFORMER=1. Do you mean I need to install the transformers package from pip?
pip install transformers

[tjruwase]

Yes, you need to do pip install transformers. Sorry, DS_BUILD_TRANSFORMER= is unrelated to transformers package.

[loadams]

Hi @401qingkong - not sure if you still have this issue, but if you do, the steps are here for getting things running on AMD: https://github.com/microsoft/DeepSpeed/blob/master/.github/workflows/amd-mi200.yml#L40

Please note the sections of installing transformers and apex, both of which are needed here.

Please let us know if this helps or you have other issues.

[loadams]

Closing this issue for now, if you are still having issues, please re-open or open a new issue and tag me.