Search results

Handle summary metric semantics more consistently

enhancementNew feature or requestNew feature or request

Status: Open.

#1865 In NVIDIA-NeMo/RL;

· terrykong opened on Feb 3, 2026

[mcore] HybridEP support for GB200 DeepSeek GRPO and SFT

deepseekRelated to deepseek 671bRelated to deepseek 671b

enhancementNew feature or requestNew feature or request

PerformanceRelated to improving performanceRelated to improving performance

Status: Open.

#1859 In NVIDIA-NeMo/RL;

· guyueh1 opened on Feb 1, 2026

vLLM async engine: <code>deque mutated during iteration</code> during rollout

bugSomething isn't workingSomething isn't working

community-request

Status: Open.

#1857 In NVIDIA-NeMo/RL;

· YijiawWang opened on Feb 1, 2026

Pipeclean GB200 nemotron 3

enhancementNew feature or requestNew feature or request

Status: Open.

#1853 In NVIDIA-NeMo/RL;

· terrykong opened on Jan 30, 2026

Training stops after long checkpointing

bugSomething isn't workingSomething isn't working

community-request

needs-follow-upIssue needs follow-upIssue needs follow-up

Status: Open.

#1847 In NVIDIA-NeMo/RL;

· xavier-owkin opened on Jan 29, 2026

Training stalls after several hours with no noticeable error

bugSomething isn't workingSomething isn't working

Status: Open.

#1846 In NVIDIA-NeMo/RL;

· roclark opened on Jan 29, 2026

Low performance of a multi-node SFT example from README.md

bugSomething isn't workingSomething isn't working

Status: Open.

#1844 In NVIDIA-NeMo/RL;

· OlegSudakov opened on Jan 29, 2026

Sequence packing + sequence parallel causes size mismatch in DTensorPolicyWorkerV2 logprobs

bugSomething isn't workingSomething isn't working

Status: Open.

#1835 In NVIDIA-NeMo/RL;

· RayenTian opened on Jan 28, 2026

MLFLow: Metrics are not reported to <code>experiment_name</code> consistently.

bugSomething isn't workingSomething isn't working

community-request

needs-follow-upIssue needs follow-upIssue needs follow-up

Status: Open.

#1832 In NVIDIA-NeMo/RL;

· dhineshkumar-r opened on Jan 27, 2026

Save checkpoint without optimizer states

community-request

enhancementNew feature or requestNew feature or request

needs-follow-upIssue needs follow-upIssue needs follow-up

Status: Open.

#1828 In NVIDIA-NeMo/RL;

· odedovadia opened on Jan 26, 2026

On-policy distillation with different tokenizers

enhancementNew feature or requestNew feature or request

x-shop

Status: Open.

#1827 In NVIDIA-NeMo/RL;

· snowmanwwg opened on Jan 26, 2026

Bump cudnn to 9.14 to support packed sequence fused attention

enhancementNew feature or requestNew feature or request

Status: Open.

#1819 In NVIDIA-NeMo/RL;

· guyueh1 opened on Jan 23, 2026