allow ref model use ds stage3 only

[gromzhu]

allow ref model and avtive model use different ds stage. i use 7b model to test, ref model use stage3 and active model use stage2 is 25% fast than ref model and active model both use stage3. if ref model and active model both use stage2 will cause oom

[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.

[Trangle]

Nice

[younesbelkada]

The code quality didn't pass https://github.com/huggingface/trl/actions/runs/9543664472/job/26304378146?pr=1730 can you double check ? @gromzhu 🙏

[gromzhu]

The code quality didn't pass https://github.com/huggingface/trl/actions/runs/9543664472/job/26304378146?pr=1730 can you double check ? @gromzhu 🙏

sorry for this mistake，i don‘t know how to run precommit local properly.

[github-actions[bot]]

This issue has been automatically marked as stale because it has not had recent activity. If you think this still needs to be addressed please comment on this thread.