Search results

Load/offload optimizer states only when calling optimizer.step()?

Status: Open.

#5282 In verl-project/verl;

· jp1924 opened on Feb 11, 2026

目前的verl不支持qwen3-omni模型的GRPO训练

bugSomething isn't workingSomething isn't working

Status: Open.

#5281 In verl-project/verl;

· LiXinYuECNU opened on Feb 11, 2026

How to save gradients during training?

Status: Open.

#5280 In verl-project/verl;

· louiswng opened on Feb 11, 2026

Prompt token IDs not decoded for single turn agent in weave traces

bugSomething isn't workingSomething isn't working

Status: Open.

#5276 In verl-project/verl;

· jnash10 opened on Feb 11, 2026

Occasional KL loss spike with FSDP workers in grpo flow

bugSomething isn't workingSomething isn't working

Status: Open.

#5274 In verl-project/verl;

· hchings opened on Feb 10, 2026

labels are missing from verl/verl/models/mcore/model_forward_fused.py

bugSomething isn't workingSomething isn't working

Status: Open.

#5273 In verl-project/verl;

· sneng1997 opened on Feb 10, 2026

<code>extract_system_prompt_and_generation</code> may fail for sft in transformer 5.x

bugSomething isn't workingSomething isn't working

Status: Open.

#5270 In verl-project/verl;

· rakkit opened on Feb 10, 2026

[RFC][Feat Proposal][New Contributor] Integrating F-GRPO

Status: Open.

#5256 In verl-project/verl;

· Felix-Zhenghao opened on Feb 10, 2026

[RFC] use megatron-bridge's implementation for megatron checkpoint manager

Status: Open.

#5247 In verl-project/verl;

· ISEEKYAN opened on Feb 9, 2026

[RFC] support Nemo-AutoModel as an alternative training backend

Status: Open.

#5245 In verl-project/verl;

· ISEEKYAN opened on Feb 9, 2026

[RFC] support megatron-fsdp as a new alternative training backend

Status: Open.

#5244 In verl-project/verl;

· ISEEKYAN opened on Feb 9, 2026

[Fix] encode vllm prompt_ids via tokenizer (not processor) for multi-modal inputs

bugSomething isn't workingSomething isn't working

Status: Open.

#5241 In verl-project/verl;

· null-test-7 opened on Feb 9, 2026