Skip to content

Pull requests: huggingface/trl

New pull request New
163 Open 3,223 Closed
163 Open 3,223 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

AsyncGRPOTrainer: add PEFT/LoRA support
#5896 opened May 31, 2026 by rycerzes Contributor Loading…
5 of 8 tasks
AsyncGRPOTrainer: add ProcessorMixin handling
#5895 opened May 31, 2026 by rycerzes Contributor Loading…
5 of 8 tasks
2
AsyncGRPOTrainer: add sampling parameters (top_p, top_k, min_p, repetition_penalty)
#5894 opened May 31, 2026 by rycerzes Contributor Loading…
5 of 8 tasks
AsyncGRPOTrainer: add model_init_kwargs support
#5893 opened May 31, 2026 by rycerzes Contributor Loading…
5 of 8 tasks
async grpo native weight sync with vllm>=0.22.0
#5892 opened May 30, 2026 by AmineDiro Member Loading…
2
Fix GRPO use_liger_kernel under DeepSpeed ZeRO-3
#5891 opened May 30, 2026 by kashif Collaborator Loading…
1
13
[GRPO] Match the Liger dapo/cispo/vespo normalizer to the non-Liger path
#5890 opened May 30, 2026 by kashif Collaborator • Draft
3
fix(grpo,rloo): apply generation_config override in use_transformers_paged path
#5888 opened May 30, 2026 by Sumu004 Loading…
4 of 8 tasks
1
Cross-tokenizer alignment via byte offsets in GOLD trainer
#5885 opened May 29, 2026 by kashif Collaborator Loading…
4 of 8 tasks
1
35
[2/2] refactor: decoupled self distillation trainers; cleanup
#5883 opened May 29, 2026 by LeonEricsson Collaborator Loading…
8 tasks
4
DPOTrainer: eagerly delete intermediate logits tensors to reduce peak memory
#5882 opened May 29, 2026 by flutist Contributor Loading…
2
[DPOTrainer] Drop images when max_length truncation causes token/feature mismatch
#5881 opened May 29, 2026 by flutist Contributor Loading…
4
Simplify reference model handling in GRPO/RLOO
#5877 opened May 29, 2026 by albertvillanova Member Loading…
3
Simplify reference model handling in DPO
#5876 opened May 29, 2026 by albertvillanova Member Loading…
6
Fix loss_type="chunked_nll" under DeepSpeed ZeRO-3
#5873 opened May 27, 2026 by qgallouedec Member Loading…
7
Add Idefics3 original and training chat template with generation markers
#5871 opened May 27, 2026 by aazizyan Contributor Loading…
5
Removed generate_rollout_completions
#5870 opened May 27, 2026 by sergiopaniego Member Loading…
8 tasks
10
Add SmolVLM original and training chat template with generation markers
#5868 opened May 27, 2026 by aazizyan Contributor Loading…
1
[1/2] refactor: decoupled self distillation trainers (sdpo, sdft, ...)
#5862 opened May 27, 2026 by LeonEricsson Collaborator Loading…
4 of 12 tasks
18
[AsyncGRPO] Rollout worker: set aiohttp limit to max(100, max_inflight_tasks)
#5861 opened May 27, 2026 by ggcr Loading…
3 of 8 tasks
1
5
Handle empty conversational fields in dataset format checks
#5860 opened May 27, 2026 by emery-Xu Loading…
3 of 8 tasks
2
Fix Qwen3.5 vLLM weight name remapping
#5858 opened May 27, 2026 by haimianxing Loading…
5 of 6 tasks
1
10
Support non-lm_head output projections in chunked SFT loss (GPTNeoX)
#5857 opened May 26, 2026 by qgallouedec Member Loading…
4
Padding-free training in AsyncGRPO
#5854 opened May 26, 2026 by qgallouedec Member Loading…
10
[WIP] Chunked DPO loss (MVP)
#5853 opened May 26, 2026 by qgallouedec Member Loading…
5
ProTip! Updated in the last three days: updated:>2026-05-28.