Skip to content

Pull requests: huggingface/trl

New pull request New
178 Open 3,925 Closed
178 Open 3,925 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

[GRPO + RLOO] Un-expand tool-returned images (vLLM VLM fix 2/2)
#6719 opened Aug 12, 2026 by Strongich Contributor • Draft
1
[GRPO + RLOO] Feed vLLM unexpanded image placeholders (vLLM VLM fix 1/2)
#6718 opened Aug 12, 2026 by Strongich Contributor • Draft
Honor model_init_kwargs dtype in AsyncGRPOTrainer
#6717 opened Aug 12, 2026 by baremetaldevx86 Loading…
5 of 8 tasks
1
Rework AsyncGRPO metrics metrics and docs
#6715 opened Aug 12, 2026 by AmineDiro Member Loading…
7
Raise a clear error when generation returns no completions
#6714 opened Aug 12, 2026 by albertvillanova Member Loading…
1
Add loss_type support to AsyncGRPOTrainer
#6713 opened Aug 12, 2026 by kashif Collaborator Loading…
6
Hotfix CI: Skip transformers continuous batching test on pre-Ampere GPUs
#6712 opened Aug 12, 2026 by albertvillanova Member Loading…
2
⚡️ Guard the RLOO completions table against prompts with no image
#6710 opened Aug 11, 2026 by vineethsaivs Loading…
3 of 8 tasks
2
Fix Embed and lm_head FLOPS accounting
#6709 opened Aug 11, 2026 by Harindhar10 Loading…
3 of 8 tasks
1
fix(examples): use OpenEnv's proxy path helpers in the opencode example
#6706 opened Aug 11, 2026 by behroozazarkhalili Collaborator Loading…
1
1
Add AsyncDistillationTrainer
#6705 opened Aug 11, 2026 by kashif Collaborator Loading…
12
Fix extract_prompt boundary cases
#6703 opened Aug 11, 2026 by RerankerGuo Contributor Loading…
4 of 8 tasks
fix(grpo): fail fast when continuing from a non-trainable PEFT adapter (silent no-op)
#6701 opened Aug 10, 2026 by Meredith2328 Loading…
5 of 8 tasks
1
7
Fix Qwen3.5 multimodal inputs in Liger GRPO
#6699 opened Aug 10, 2026 by DimensionSTP Loading…
5 of 8 tasks
1
Fix ZeRO-3 chunk synchronization in chunked CE
#6698 opened Aug 10, 2026 by DimensionSTP Loading…
5 of 8 tasks
1
3
Handle None token logprobs from vLLM in GRPO importance sampling
#6693 opened Aug 9, 2026 by behroozazarkhalili Collaborator Loading…
1
6
feat: add Falcon3 training chat template with generation markers
#6691 opened Aug 9, 2026 by RudrenduPaul Contributor Loading…
5 of 6 tasks
feat: add OLMo 3 training chat template with generation markers
#6690 opened Aug 9, 2026 by RudrenduPaul Contributor Loading…
4 of 5 tasks
feat: add Mistral training chat template with generation markers
#6689 opened Aug 9, 2026 by RudrenduPaul Contributor Loading…
4 of 8 tasks
Use the vLLM base image for TRL Docker images
#6687 opened Aug 8, 2026 by YingqiDuan Loading…
5 of 8 tasks
1
[SFT] Treat past_key_values and attentions as optional on the model output
#6686 opened Aug 8, 2026 by Hakureirm Loading…
5 of 8 tasks
1
4
Use spawn context for vLLM server workers
#6682 opened Aug 8, 2026 by Sevendogs077 Loading…
5 of 8 tasks
1
2
Add an experimental GPG trainer
#6681 opened Aug 8, 2026 by behroozazarkhalili Collaborator Loading…
1
4
Document the GRPO reward-function batch layout and group-relative rewards
#6680 opened Aug 8, 2026 by behroozazarkhalili Collaborator Loading…
1
1
Optimize metrics computation
#6678 opened Aug 7, 2026 by michaelbenayoun Member Loading…
4
ProTip! no:milestone will show everything without a milestone.