Skip to content

Pull requests: volcengine/verl

New pull request New
288 Open 2,438 Closed
288 Open 2,438 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

[ci] test: switch the vlm rl test case in the npu environment to use the model engine
#4844 opened Jan 8, 2026 by ji-huazhong Loading…
8 tasks
2
[ckpt] fix: Megatron save ckpt after validation
#4841 opened Jan 8, 2026 by wdl339 Loading…
6
[algo, rollout, sglang] feat: Support router replay with sglang
#4840 opened Jan 8, 2026 by moehanabi Loading…
4 of 8 tasks
9
[worker] fix: FSDP Engine LoRA layered summon
#4836 opened Jan 8, 2026 by JacobHelwig Loading…
4
[vllm] fix: pad data_hp to be multiples of block_size
#4835 opened Jan 8, 2026 by Agoniii Loading…
2 of 8 tasks
[tool] Update JSON dump to ensure ASCII characters are preserved
#4819 opened Jan 7, 2026 by Mencuis Loading…
2
[rollout][DONT-MERGE] debug: Add TRTLLM as new rollout engine.
#4812 opened Jan 6, 2026 by Superjomn • Draft
8 tasks
3
Fix: use prime reward manager for GSPO script
#4809 opened Jan 6, 2026 by aub123 Loading…
4
[trainer, worker] fix: Pass a dict config to ray to avoid pickling issues with Deepseek V3.1's tokenizer
#4808 opened Jan 6, 2026 by jthomy Loading…
4 of 6 tasks
4
[ray, worker] feat: create one placement group across all nodes in resource pool and implement IP-based worker sorting
#4792 opened Jan 5, 2026 by jianjunzhong Loading…
4 of 7 tasks
3
[sglang] feat: Support SGLang remote_instance
#4784 opened Jan 4, 2026 by EduardDurech Loading…
2
[rollout] fix: vllm-rollout update weights
#4780 opened Jan 4, 2026 by Annarine Loading…
7 tasks
2
[checkpoint] fix: normalize LoRA state_dict keys when saving HF model
#4770 opened Jan 2, 2026 by yurekami Loading…
4 of 5 tasks
1
6
[trainer,data,algo] feat: Unify SFT and PPO Pipelines
#4766 opened Jan 2, 2026 by JacobHelwig Loading…
6
[WIP][tool] feature: scheduling analysis based on profiling data
#4746 opened Dec 31, 2025 by tardis-key • Draft
7 tasks
4
fix(fully_async): use rollout.total_epochs for total_rollout_steps calculation
#4736 opened Dec 30, 2025 by FarrenZhang Loading…
2 tasks done
[hardware] feat: support qwen3vl ON ASCEND NPU
#4734 opened Dec 30, 2025 by Seren-hao Loading…
7 tasks
8
fix(fsdp): handle param offloading in get_actor_weights_info
#4726 opened Dec 29, 2025 by yurekami Loading…
2 of 3 tasks
1
4
tests: Add gradient accumulation tests for all loss aggregation modes
#4721 opened Dec 29, 2025 by yurekami Loading…
1 of 2 tasks
1
3
docs: Add comprehensive OOM resolution guide
#4720 opened Dec 29, 2025 by yurekami Loading…
3 tasks
1
7
[recipe, fsdp] feat: support GPT-OSS-20B DAPO training script on ASCEND NPU
#4716 opened Dec 29, 2025 by mikequan0425 • Draft
2 of 7 tasks
3
[sglang] fix: sglang TP+DP support / port bug
#4715 opened Dec 29, 2025 by hustmf Loading…
7 tasks
7
[perf] fix: correct torch profiler logic for distributed trace collection
#4707 opened Dec 28, 2025 by AkiRusProd Loading…
4
[merger] fix: support vision_model keys in Megatron merger for VL models
#4701 opened Dec 28, 2025 by yurekami Loading…
1 task
1
3
feat(trainer): add logging of extra reward metrics
#4698 opened Dec 27, 2025 by yurekami Loading…
2 tasks
1
3
ProTip! no:milestone will show everything without a milestone.