fix: fail when required args not passed when prompt_tuning_init==TEXT

[kmehant]

When prompt_tuning_init==TEXT is set, there are some args needed to be provided. For instance, tokenizer_name_or_path should be set to some tokenizer name or path, failing which would result in errors such as

HTTPError: 401 Client Error: Unauthorized for url: https://huggingface.co/None/resolve/main/tokenizer_config.json

tokenizer_name_or_path is taken as None. Such errors can be handled and reported meaningfully post config initialization and this PR makes that contribution.

[BenjaminBossan]

Thanks for adding this quality of life feature. Two points:

• Could you please break the long lines? We want to keep a 120 char limit.
• Would you mind adding tests to check that the error is raised? We can help with that if needed.

[kmehant]

@BenjaminBossan

Would you mind adding tests to check that the error is raised? We can help with that if needed.

Sure @BenjaminBossan, I tried looking for tests for prompt tuning implementation, but I could not find them, can you point me to that?

[BenjaminBossan]

I tried looking for tests for prompt tuning implementation, but I could not find them, can you point me to that?

The tests are a bit scattered around, as we group by model architecture and not by method. But a good fit would be to add them to tests/test_initialization.py.

[kmehant]

@BenjaminBossan Have added the tests and made required changes.

[BenjaminBossan]

Thanks a lot for adding the test, could you please run make style and push the changes?

[kmehant]

#1519 (comment)

@BenjaminBossan I have run make style and pushed the changes thanks.

[HuggingFaceDocBuilderDev]

The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update.

[BenjaminBossan]

Thanks for the update. I found some very small things still, after that we should be good to go.

[BenjaminBossan]

Thanks a lot for adding these early checks.