Skip to content

Pull requests: OpenRLHF/OpenRLHF

New pull request New
40 Open 448 Closed
40 Open 448 Closed
Author
Filter by author
Loading
Label
Filter by label
Loading
Use alt + click/return to exclude labels
or ⇧ + click/return for logical OR
Projects
Filter by project
Loading
Milestones
Filter by milestone
Loading
Reviews
Filter by reviews
No reviews Review required Approved review Changes requested
Assignee
Filter by who’s assigned
Assigned to nobody Loading
Sort
Sort by
Newest Oldest Most commented Least commented Recently updated Least recently updated Best match
Most reactions
👍 👎 😄 🎉 😕 ❤️ 🚀 👀

Pull requests list

Added REBEL-inspired offline reward-gap regression loss to DPO trainer
#1247 opened Jun 5, 2026 by LeoPhilly Loading…
Fix overlong penalty action token length
#1246 opened May 30, 2026 by Jiang020609 Loading…
1
3
Security: Arbitrary Code Execution via trust_remote_code=True in Dataset Loading
#1241 opened May 20, 2026 by tuanaiseo Loading…
3 tasks
5
fix: ignore masked invalid values in policy loss reductions
#1240 opened May 17, 2026 by haoyang9804 Loading…
3
fix: ignore zero-action samples in group reward baselines
#1239 opened May 17, 2026 by haoyang9804 Loading…
4
Add TokenSpeed-backed PPO rollout engine
#1237 opened May 7, 2026 by 4teven Loading…
2
Fix vLLM GPU allocation for colocated tensor-parallel engines
#1231 opened Apr 30, 2026 by fuyuan-li Loading…
1
3
Add Trackio logger backend
#1230 opened Apr 27, 2026 by abidlabs Loading…
6
Clarify reward normalization behavior for custom reward sources
#1229 opened Apr 26, 2026 by taivu1998 Loading…
2
Add SFT tools field support for chat templates
#1228 opened Apr 26, 2026 by taivu1998 Loading…
1
3
Fix multi-turn SFT masks for Qwen3 thinking templates
#1227 opened Apr 26, 2026 by taivu1998 Loading…
1
3
Replace Deepspeed backend with Automodel
#1226 opened Apr 26, 2026 by hijkzzz Collaborator Loading…
6
feat: full async PPO training with partial rollout agent support
#1218 opened Apr 11, 2026 by LYMDLUT Collaborator Loading…
5
fix: true loss aggregation across dp ranks
#1216 opened Apr 10, 2026 by alek6kun Loading…
10
Fast Evolutionary Algorithm Support
#1214 opened Apr 5, 2026 by DavidKoplow Loading…
69
feat: add --from_scratch option to initialize model with random weights
#1209 opened Apr 1, 2026 by konghw-git Contributor Loading…
2 tasks done
2
adding CFPO to OpenRLHF
#1184 opened Feb 9, 2026 by asparius Loading…
1
feat: add AgentTokenHandler with defensive token concatenation for agentic training (#1128)
#1181 opened Jan 30, 2026 by ichbinlucaskim Loading…
13
feat: Switch vLLM rollout sampling to oversampling.
#1179 opened Jan 20, 2026 by Freder-chen Contributor Loading…
6
feat: add the support of fsdp2 and remove deepspeed (new version of PR 1115)
#1176 opened Jan 16, 2026 by LYMDLUT Collaborator Loading…
27
Default overlap_comm on for ZeRO-2+ RLHF runs
#1154 opened Nov 26, 2025 by MagellaX Loading…
4
Enhance PPO logging with entropy, reward stats, and grad norm insights
#1148 opened Nov 8, 2025 by MagellaX Loading…
7
Star Attention topology support with model integration and --attn_topology flag
#1122 opened Aug 26, 2025 by MagellaX Loading…
3
Add FSDP backend and --dist_backend flag across CLIs; introduce FSDPStrategy
#1115 opened Aug 23, 2025 by MagellaX Loading…
72
CLI support for top_k
#1104 opened Aug 13, 2025 by JoNeedsSleep Loading…
3
ProTip! Updated in the last three days: updated:>2026-06-08.