-
Notifications
You must be signed in to change notification settings - Fork 4.2k
Pull requests: verl-project/verl
Author
Label
Projects
Milestones
Reviews
Assignee
Sort
Pull requests list
[megatron] fix: align teacher tensor seqlen with student for forward_kl_topk CP split
#7012
opened Jul 10, 2026 by
gaohongkui
Loading…
4 of 7 tasks
[wandb] fix: fixes async training finish sequence
#7011
opened Jul 10, 2026 by
theely
Contributor
Loading…
[fully_async] fix: release state lock while waiting for rollout capacity
#7010
opened Jul 10, 2026 by
kdubovikov
Loading…
7 of 8 tasks
ci: update Ascend A2 Docker image tag
#7009
opened Jul 10, 2026 by
20000912gyf-dotcom
•
Draft
8 tasks
[single_controller] fix: remove transfer queue and restore RayPPOTrainer
#7000
opened Jul 9, 2026 by
zw0610
Collaborator
Loading…
2 tasks done
[vllm] fix: release only kv cache during sleep
#6991
opened Jul 9, 2026 by
zhangxin81
Loading…
5 tasks done
[vllm, rollout, perf]feat: add extra-prefix-cache for agentic system prefix rollout
#6990
opened Jul 9, 2026 by
HanlinDu
Contributor
Loading…
8 tasks
[fully_async] fix: pre-compute total_training_steps before trainer init
#6987
opened Jul 9, 2026 by
gaohongkui
Loading…
4 of 7 tasks
[algo] feat: add TOPR (Tapered Off-Policy REINFORCE) policy loss
#6986
opened Jul 9, 2026 by
abtonmoy
Contributor
Loading…
4 tasks done
[checkpoint_engine][rollout] Delta weight sync over NCCL for disaggregated rollout (+ sharded-snapshot variant)
#6974
opened Jul 8, 2026 by
ChangyiYang
Contributor
Loading…
[ci] feat: add GRPO Qwen3.5-35B Megatron vLLM nightly test for Ascend
Ascend
#6973
opened Jul 8, 2026 by
chengminhua
Contributor
•
Draft
8 tasks
[perf] fix: correct Qwen-VL head_dim consistency and merger token count (#6903)
#6944
opened Jul 6, 2026 by
chethanuk
Loading…
5 of 6 tasks
[rollout, vllm] feat: KV-cache-aware request load balancer
#6940
opened Jul 6, 2026 by
zhaizhiqiangA
•
Draft
5 of 8 tasks
[megatron] feat: migrate fused logprob/entropy from GPTModel.forward monkey-patch to Megatron output_processor hook
#6933
opened Jul 6, 2026 by
chengcuiping
Contributor
Loading…
8 of 11 tasks
[misc] chore: add optional vLLM cache config for fully-async NPU script
Ascend
#6932
opened Jul 6, 2026 by
lxb007981
Contributor
Loading…
4 of 8 tasks
[trainer, fsdp, worker, algo, cfg, recipe] feat: add reverse-KL on-policy distillation with FSDP teacher
#6920
opened Jul 2, 2026 by
YangxSong
Loading…
5 of 6 tasks
[trainer] fix: gracefully finalize tracking and dataloader workers at fit() exits
#6917
opened Jul 2, 2026 by
seokhyunan
Loading…
8 tasks done
[opd] feat: add fusenode on teacher forward and student train in fully_async_opd
#6914
opened Jul 2, 2026 by
msz12345
Loading…
8 tasks
[data] fix: make overlong prompt filter picklable
#6888
opened Jun 29, 2026 by
zhangdw156
Loading…
7 of 8 tasks
[trainer, rollout] feat: opt-in rollout-level dispatch for the V1 agent-loop trainer
#6874
opened Jun 28, 2026 by
huaiyizhao
Contributor
Loading…
8 tasks
[fully_async, reward, docs] fix: skip reward function call when use_task_rewards=False
#6870
opened Jun 28, 2026 by
kenkenpa2126
Contributor
Loading…
7 of 8 tasks
Previous Next
ProTip!
Adding no:label will show everything without a label.