Skip to content

Pull requests: huggingface/trl

New pull request New
172 Open 3,422 Closed
172 Open 3,422 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

feat(dpo): implement Adaptive Beta-DPO (arXiv:2407.08639)
#6123 opened Jun 19, 2026 by mukund1985 Loading…
3 of 8 tasks
1
4
fix(online_dpo): add prediction_step to fix eval_strategy=steps crash
#6122 opened Jun 19, 2026 by mukund1985 Loading…
3 of 8 tasks
1
2
fix(ppo): exclude padding tokens from entropy calculation
#6121 opened Jun 19, 2026 by mukund1985 Loading…
3 of 8 tasks
1
feat(ppo): add save_value_model flag to PPOConfig
#6120 opened Jun 19, 2026 by mukund1985 Loading…
3 of 8 tasks
1
Add support for vLLM 0.22.1
#6119 opened Jun 19, 2026 by qgallouedec Member Loading…
5
Handle missing vllm_ascend imports gracefully 🐛 bug Something isn't working
#6118 opened Jun 19, 2026 by burtenshaw Collaborator Loading…
2
Make evaluate() accept the same dataset types as the trainer
#6116 opened Jun 19, 2026 by qgallouedec Member Loading…
1
5
Support LFM2-VL multimodal inputs in GRPO and RLOO
#6114 opened Jun 19, 2026 by zwischenraum Loading…
4 of 8 tasks
2
Migrate to native vLLM weight sync
#6110 opened Jun 19, 2026 by qgallouedec Member • Draft
Add support for vLLM 0.20.0
#6108 opened Jun 18, 2026 by qgallouedec Member Loading…
1
fix: address 4 remaining GMPO bugs from code review
#6095 opened Jun 18, 2026 by abderahmane-ai Contributor Loading…
4 of 6 tasks
1
4
Add packing-aware dynamic batching to AsyncGRPO
#6092 opened Jun 17, 2026 by AmineDiro Member Loading…
9
Auto-load processing_class in CPO/ORPO trainers when omitted
#6087 opened Jun 17, 2026 by DaoyuanLi2816 Contributor Loading…
2 of 4 tasks
fix(data): correct extract_prompt index for divergence at start and prefix completions
#6079 opened Jun 16, 2026 by he-yufeng Loading…
1
Refactor GRPO loss component computation
#6075 opened Jun 16, 2026 by alexbowe • Draft
4 of 8 tasks
[async GRPO] Per-generation reset() observation
#6072 opened Jun 16, 2026 by qgallouedec Member Loading…
6
async_grpo: inject the environment_factory reset() observation into the prompt
#6068 opened Jun 15, 2026 by adithya-s-k Collaborator Loading…
1
12
Add get_cosine_scaled_reward
#6066 opened Jun 15, 2026 by qgallouedec Member Loading…
5
Warn when sequence-level importance sampling is combined with a token-summed loss type
#6042 opened Jun 13, 2026 by discobot Contributor Loading…
4 of 8 tasks
1
2
fix: preserve OnlineDPO vLLM completion ids
#6038 opened Jun 13, 2026 by he-yufeng Loading…
4 of 8 tasks
1
refactor(sft): build labels during dataset preparation instead of collation
#6037 opened Jun 12, 2026 by 0xadvait Loading…
5 of 8 tasks
11
test: don't hard-code bf16=True on devices that lack bf16 support
#6036 opened Jun 12, 2026 by behroozazarkhalili Collaborator Loading…
1
5
fix: load image-text policy for async grpo
#6032 opened Jun 12, 2026 by he-yufeng Loading…
5 of 8 tasks
1
@AmineDiro
2
fix: pass AsyncGRPO environment rewards
#6031 opened Jun 12, 2026 by he-yufeng Loading…
5 of 8 tasks
1
5
fix(sft): drop examples with no trainable tokens after max_length truncation
#6025 opened Jun 12, 2026 by 0xadvait • Draft
5 of 8 tasks
1
4
ProTip! Add no:assignee to see everything that’s not assigned.