GRPO + vLLM hang fixed on non-NVLink; dataset fingerprinting corrected
v1.7.1
1 feature1 enhancement5 fixesThis release1 featureNew capabilities1 enhancementImprovements to existing features5 fixesBug fixesAI-tallied from the release notes
From the original release noteView original ↗
What's Changed
- Fix GRPO + vLLM colocate + PEFT hang on non-NVLink hardware by @albertvillanova in https://github.com/huggingface/trl/pull/6139
- Fix dataset fingerprinting in DPO/SFT tokenization by @qgallouedec in https://github.com/huggingface/trl/pull/6206
- Pass GPU device_ids to barrier fix in GRPO + vLLM colocate + PEFT by @albertvillanova in https://github.com/huggingface/trl/pull/6187
- Integrate the new response parsing API by @qgallouedec in https://github.com/huggingface/trl/pull/5791
- Fix
add_response_schematests for the newparse_responseprefix requirement by @qgallouedec in https://github.com/huggingface/trl/pull/6236 - Add prompt-learning guard for PEFT with Liger in GRPO by @albertvillanova in https://github.com/huggingface/trl/pull/6186
- Fix activation offload storage dedupe reuse by @winglian in https://github.com/huggingface/trl/pull/6241
Full Changelog: https://github.com/huggingface/trl/compare/v1.7.0...v1.7.1
Fetched July 4, 2026

