TensorBoard callback no longer adds hparams

[bri25yu]

What does this PR do?

The TensorBoardCallback.on_train_begin function calls add_hparams with an empty metric_dict parameter, meaning that the only information that is logged is from args.sanitized_dict(). This is duplicated information from the previous self.tb_writer.add_text("args", args.to_json_string()). As a result, the add_hparams call is unnecessary and this PR removes it.

Adjacently fixes #21821.

I'm aware of the following TensorBoard related documentation:

• https://huggingface.co/docs/hub/tensorboard
• https://huggingface.co/docs/transformers/main/en/main_classes/callback#transformers.integrations.TensorBoardCallback

None of these docs need to be updated in this PR.

A sanity check test:

"""
Minimal replication of https://github.com/huggingface/transformers/issues/21821
"""
from os import listdir
from shutil import rmtree

from transformers import TrainingArguments
from transformers.integrations import TensorBoardCallback


output_dir = "output_dir"

args = TrainingArguments(output_dir=output_dir, logging_dir=output_dir)

def has_extra_file():
    return len(listdir(output_dir)) > 1


class DummyControl:
    should_training_stop = None


class DummyState:
    is_world_process_zero = True
    is_hyper_param_search = False


class NoHParamsTensorBoardCallback(TensorBoardCallback):
    # This is a copy of `TensorBoardCallback.on_train_begin` unless specified otherwise
    def on_train_begin(self, args, state, control, **kwargs):
        if not state.is_world_process_zero:
            return

        log_dir = None

        if state.is_hyper_param_search:
            trial_name = state.trial_name
            if trial_name is not None:
                log_dir = os.path.join(args.logging_dir, trial_name)

        if self.tb_writer is None:
            self._init_summary_writer(args, log_dir)

        if self.tb_writer is not None:
            self.tb_writer.add_text("args", args.to_json_string())
            if "model" in kwargs:
                model = kwargs["model"]
                if hasattr(model, "config") and model.config is not None:
                    model_config_json = model.config.to_json_string()
                    self.tb_writer.add_text("model_config", model_config_json)

            ###########################
            # START no hparams call
            ###########################

            # Original code:
            # # Version of TensorBoard coming from tensorboardX does not have this method.
            # if hasattr(self.tb_writer, "add_hparams"):
            #     self.tb_writer.add_hparams(args.to_sanitized_dict(), metric_dict={})

            ###########################
            # END no hparams call
            ###########################


rmtree(output_dir, ignore_errors=True)
TensorBoardCallback().on_train_begin(args, DummyState(), DummyControl())
print(f"With the call to `add_hparams`, has extra file is {has_extra_file()}")

rmtree(output_dir, ignore_errors=True)
NoHParamsTensorBoardCallback().on_train_begin(args, DummyState(), DummyControl())
print(f"WithOUT the call to `add_hparams`, has extra file is {has_extra_file()}")

rmtree(output_dir, ignore_errors=True)  # Cleanup

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?

Anyone in the community is free to review the PR once the tests have passed.

[sgugger]

Thanks for cleaning this up!

[HuggingFaceDocBuilderDev]

The docs for this PR live here. All of your documentation changes will be reflected on that endpoint.