What does this PR do?

Fixes #5539.

DataCollatorForPreference currently decides whether to emit a margin tensor by checking only examples[0]. That makes the result depend on batch order:

• if the first example has no margin, later margins are silently dropped
• if the first example has margin but a later example does not, collation raises KeyError

This PR checks the whole batch instead. If any example provides margin, the collator emits a margin tensor and defaults missing margins to 0.0, matching the optional-field behavior documented for this collator.

I added regression tests for both mixed-order cases.

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?
• Did you write any new necessary tests?

AI writing disclosure

We welcome the use of AI tools to help with contributions. For transparency and to help us improve our review process, please indicate the level of AI involvement in this PR.

• No AI usage: the PR was written entirely by a human.
• AI-assisted: some parts were suggested or improved by AI, but the PR was written and reviewed by a human.
• AI-generated: the PR was mostly or fully generated by an AI tool.

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.

Tests

python -m pytest tests/test_reward_trainer.py::TestDataCollatorForPreference -q
python -m ruff check trl\trainer\reward_trainer.py tests\test_reward_trainer.py
python -m py_compile trl\trainer\reward_trainer.py tests\test_reward_trainer.py
git diff --check