Search results

GRPO Training Fails with OOM Errors Despite 8x80GB GPUs

Status: Open.

#6449 In verl-project/verl;

· waleedahmed0001 opened on May 22, 2026

Bump recipe submodule commit

Status: Open.

#6440 In verl-project/verl;

· cmunley1 opened on May 21, 2026

【Ascend】社区任务发布

Ascend

Status: Open.

#6439 In verl-project/verl;

· wucong25 opened on May 21, 2026

[trainer] main_ppo_sync can silently train on partial rollout batches after agent-loop failures

bugSomething isn't workingSomething isn't working

Status: Open.

#6437 In verl-project/verl;

· Baiyu-Su opened on May 21, 2026

Could you update the document? Specifically, for versions 0.7 and 0.8?

Status: Open.

#6436 In verl-project/verl;

· li624120638 opened on May 21, 2026

Grad norm is nan when training qwen3.6-35b and enable sequence parallel

bugSomething isn't workingSomething isn't working

Status: Open.

#6431 In verl-project/verl;

· bingnoi77 opened on May 21, 2026

HCCL checkpoint engine 中存在两个实现疑问：注册名使用 nccl，且 BroadcastOperation 未采用线程池异步执行

Ascend

Status: Open.

#6425 In verl-project/verl;

· tsinghuazl22 opened on May 21, 2026

[rollout] agent loop should delegate think stripping to inference engine's native reasoning parser

Status: Open.

#6424 In verl-project/verl;

· wzhgba opened on May 21, 2026

[rename] engin_workers.py ActorRolloutRefWorker method: update_weights

Status: Open.

#6418 In verl-project/verl;

· wengeezhang opened on May 20, 2026

希望可以增加单独检查权重转换逻辑

help wantedExtra attention is neededExtra attention is needed

Status: Open.

#6414 In verl-project/verl;

· LebudiPrince opened on May 20, 2026

[Hardware] Intel XPU plugin support — is verl-hardware-plugin ready for contributions?

Status: Open.

#6403 In verl-project/verl;

· kahlun opened on May 19, 2026

[RFC] Prefix-Tree Shared Attention for Multi-Turn RL Training

Status: Open.

#6401 In verl-project/verl;

· arvyanh opened on May 19, 2026