[sergiopaniego]

What does this PR do?

The GRPOConfig docstring labels token_truncate as the default vllm_importance_sampling_mode, but the default is sequence_mask, as the field and the docstring header already say. This removes the stale label.

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

• 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.

---

Note

Low Risk
Docstring-only change with no effect on code paths or configuration defaults.

Overview
Corrects a documentation-only mismatch in GRPOConfig: the vllm_importance_sampling_mode enum list no longer marks "token_truncate" as the default.

The actual default remains "sequence_mask", as already stated on the parameter line and in the dataclass field. The "token_truncate" bullet is also lightly reworded (drops the stale “(default)” label and tightens the clipping range text).

No runtime or training behavior changes.

^{Reviewed by Cursor Bugbot for commit e340667. Bugbot is set up for automated code reviews on this repo. Configure here.}

[bot-ci-comment]

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.

[albertvillanova]

LGTM, thanks!

Same stale default in the TIP at docs/source/grpo_trainer.md:221: "By default, Truncated Importance Sampling is activated". Since #4530 the default is sequence-level Masked IS. Could you fix it here too?

[qgallouedec]

👌