Name | Modified | Size | Downloads / Week |
---|---|---|---|
Parent folder | |||
README.md | 2025-08-06 | 960 Bytes | |
Release v0.8.9 source code.tar.gz | 2025-08-06 | 369.5 kB | |
Release v0.8.9 source code.zip | 2025-08-06 | 453.2 kB | |
Totals: 3 Items | 823.6 kB | 0 |
What's Changed
- Add TensorBoard logging to PRM training by @xjli360 in https://github.com/OpenRLHF/OpenRLHF/pull/1096
-
Support vLLM off-policy importance sampling correction by @xiaoxigua999 and @MooMoo-Yang in https://github.com/OpenRLHF/OpenRLHF/pull/1098
- Requires vLLM version > 0.10 :
pip install -U vllm --pre --extra-index-url https://wheels.vllm.ai/nightly
- Requires vLLM version > 0.10 :
-
Fix weight broadcasting issue in Async RL with PyTorch 2.7.1 and vLLM 0.10 by @xiaoxigua999 in https://github.com/OpenRLHF/OpenRLHF/pull/1100
- Fix sequence-level loss calculation for GSPO by @xiaoxigua999
Full Changelog: https://github.com/OpenRLHF/OpenRLHF/compare/v0.8.8...v0.8.9