Change logging level from warning to info for max_steps overriding num_train_epochs

[qgallouedec]

What does this PR do?

Passing max_step is a supported feature, and therefore should not raise a warning. I suggest lowering the level to info.

Alternatively, we could have an additional parameter like ignore_max_step_warning but this seems overengineered.

Considering that the use of max_steps is already documented:

transformers/src/transformers/training_args.py

Line 307 in ce1d328

If set to a positive number, the total number of training steps to perform. Overrides `num_train_epochs`.

lowering the level to info seems to be the best solution.

Technically, we can ignore this warnings with

TrainingArguments(
    ...,
    max_steps=100,
    num_train_epochs=-1
)

But it's not intuitive at all IMO.

---

As for warnings, I'd say that a good practice is that they should always be removable by the user:

• either by modifying their code, in the event of a feature being misused, deprecated, etc. E.g example:
  

  transformers/src/transformers/models/speech_to_text/feature_extraction_speech_to_text.py

  Lines 235 to 238 in ce1d328

  	logger.warning(
  	"It is strongly recommended to pass the `sampling_rate` argument to this function. "
  	"Failing to do so can result in silent errors that might be hard to debug."
  	)

• or by acknowledging, in the case of very important information upgraded to warning; we add a parameter for acknowledging. E.g:
  example: allow_missing_key in load_pytorch_checkpoint_in_tf2_model

Related: huggingface/trl#2350

Before submitting

• This PR fixes a typo or improves the docs (you can dismiss the other checks if that's the case).
• Did you read the contributor guideline,
  Pull Request section?
• Was this discussed/approved via a Github issue or the forum? Please add a link
  to it if that's the case.
• Did you make sure to update the documentation with your changes? Here are the
  documentation guidelines, and
  here are tips on formatting docstrings.
• Did you write any new necessary tests?

Who can review?

@muellerzr and @SunMarc

[SunMarc]

LGTM ! cc @muellerzr for a second look

[SunMarc]

I don't mind switching to info since we already have documented that setting max_steps will overwrite num_train_epochs. Another compromise would be to indicate to the user to pass num_train_epochs = -1 in order to remove the warning but then the code becomes bloated for nothing.

[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.

[muellerzr]

Agree with Marc, this change seems fine. cc @LysandreJik for final :)

[LysandreJik]

Sounds good to me, thanks all!

[LysandreJik]

Unrelated errors, merging