Drop decoder_input_ids in DPOTrainer

[qgallouedec]

What does this PR do?

Since we only use the logits (and not the returned loss), it seems like decoder_input_ids is useless as an input for encoder-decoder models.

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? 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.
• 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. Feel free to tag
members/contributors who may be interested in your PR.

[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.

[qgallouedec]

The new curves are significantly smoother compared to before.

I believe this is due to padding completions without applying a loss mask, causing the loss to be calculated over the entire sequence, including the padding tokens:

trl/trl/trainer/dpo_trainer.py

Lines 278 to 283 in 7e5924d

	batch["chosen_labels"] = [
	seq + [tokenizer.pad_token_id] * (max_length - len(seq)) for seq in batch["chosen_labels"]
	]
	batch["rejected_labels"] = [
	seq + [tokenizer.pad_token_id] * (max_length - len(seq)) for seq in batch["rejected_labels"]
	]