OpenRLHFScalable framework for reinforcement learning with language models.Apache-2.0
Scalable framework for reinforcement learning with language models.
OpenRLHF is a Python framework for reinforcement learning from feedback and language-model training on distributed GPUs. It coordinates training and generation through Ray, DeepSpeed and vLLM, with tools for supervised fine-tuning, reward modeling and policy optimization. Researchers can supply custom rewards and connect multi-turn agent interactions to the training loop.
OpenRLHF official documentation · Reviewed 2026-10-28
- 01 / Feature noteDistributed training roles
Ray coordinates actor, reward, reference and critic components across GPU resources.
Feature source ↗ - 02 / Feature noteCustom reward functions
Training can use user-defined reward functions and multi-turn environment feedback.
Feature source ↗
GitHub stars
10,140
Detected license
Apache-2.0
Repository created
2023-07-30
Repository push
2026-11-03
2023-07-30GitHub release
2026-09-14 · v0.11.2Repository push
2026-11-03
GitHub stars
Unknown
i
Cached history uses a different month window; chart withheld until aligned.
Latest attempt: ok · Checked 2026-10-15T19:40:39.600225+00:00 · Last successful fetch 2026-10-15T19:40:39.600225+00:00 · Provider source
Hacker News stories
Unknown
i
OpenSSF Scorecard
Unknown
i
No result returned.
Checked 2026-10-27T00:21:53Z · Latest attempt: not found · Provider source
GitHub release
Published 2026-09-14
i
Checked 2026-11-04T04:04:05Z · Last successful fetch 2026-11-04T04:04:05Z · Value fetched 2026-11-04T04:04:05Z · Provider source
Unknown
