Search results

Unify cache clearing between Dtensor and Megatron backend

enhancementNew feature or requestNew feature or request

Status: Open.

#1888 In NVIDIA-NeMo/RL;

· hemildesai opened on Feb 5, 2026

Online Eagle model training during RL

enhancementNew feature or requestNew feature or request

Status: Open.

#1878 In NVIDIA-NeMo/RL;

· isomap opened on Feb 4, 2026

Ray data transfer bottleneck with long sequences/large batch size in distillation

bugSomething isn't workingSomething isn't working

Status: Open.

#1877 In NVIDIA-NeMo/RL;

· zpqiu opened on Feb 4, 2026

SFT Long context recipe

enhancementNew feature or requestNew feature or request

Status: Open.

#1875 In NVIDIA-NeMo/RL;

· snowmanwwg opened on Feb 4, 2026

Bump cuDNN to 9.14+ to support sequence packing fused attention in various models

enhancementNew feature or requestNew feature or request

Status: Open.

#1870 In NVIDIA-NeMo/RL;

· guyueh1 opened on Feb 3, 2026

DTensorPolicyWorker AdamW device mismatch (cpu vs cuda) after optimizer offload in non-colocated GRPO

community-request

Status: Open.

#1869 In NVIDIA-NeMo/RL;

· banyan-god opened on Feb 3, 2026

Cleanup default_task_data_spec in AllTaskProcessedDataset

data module

enhancementNew feature or requestNew feature or request

Status: Open.

#1867 In NVIDIA-NeMo/RL;

· yuki-97 opened on Feb 3, 2026

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