List view

[dtensor] PPO

enhancementNew feature or requestNew feature or request

Status: Open.

#2047 In NVIDIA-NeMo/RL;

· terrykong opened on Mar 2, 2026

[mcore] PPO

enhancementNew feature or requestNew feature or request

Status: Open.

#2048 In NVIDIA-NeMo/RL;

· terrykong opened on Mar 2, 2026

[super-pr] Polish lambda ray.sub and upstream

enhancementNew feature or requestNew feature or request

super-v3

Status: Open.

#1765 In NVIDIA-NeMo/RL;

· terrykong opened on Jan 13, 2026

Remove Gym dependency contract to smooth out Gym-side dependency changes

enhancementNew feature or requestNew feature or request

Status: Open.

#1769 In NVIDIA-NeMo/RL;

· bxyu-nvidia opened on Jan 13, 2026

perf: FP8 Refit Optimization

PerformanceRelated to improving performanceRelated to improving performance

Status: Open (in progress).

NVIDIA-NeMo/RLnumber 2037

#2037 In NVIDIA-NeMo/RL;

· Jianbing-D opened on Feb 28, 2026

feat: Support Ray Compiled Graph for SFT

PerformanceRelated to improving performanceRelated to improving performance

Status: Open (in progress).

NVIDIA-NeMo/RLnumber 1612

#1612 In NVIDIA-NeMo/RL;

· katec846 opened on Dec 9, 2025

feat: Add HybridEP support for MoE expert parallelism

CI:L2Run doctests, unit tests, functional tests, and convergence testsRun doctests, unit tests, functional tests, and convergence tests

deepseekRelated to deepseek 671bRelated to deepseek 671b

PerformanceRelated to improving performanceRelated to improving performance

Status: Open (in progress).

NVIDIA-NeMo/RLnumber 1942

#1942 In NVIDIA-NeMo/RL;

· seonjinn opened on Feb 13, 2026

perf: Support mcore virtual pipeline parallel

CI:L2Run doctests, unit tests, functional tests, and convergence testsRun doctests, unit tests, functional tests, and convergence tests

deepseekRelated to deepseek 671bRelated to deepseek 671b

PerformanceRelated to improving performanceRelated to improving performance

Status: Open (in progress).

NVIDIA-NeMo/RLnumber 1126

#1126 In NVIDIA-NeMo/RL;

· guyueh1 opened on Sep 15, 2025

feat: skip logprob and reference logprob computation under certain conditions

CI:L2Run doctests, unit tests, functional tests, and convergence testsRun doctests, unit tests, functional tests, and convergence tests

deepseekRelated to deepseek 671bRelated to deepseek 671b

PerformanceRelated to improving performanceRelated to improving performance

super-v3

Status: Open (in progress).

NVIDIA-NeMo/RLnumber 1891

#1891 In NVIDIA-NeMo/RL;

· guyueh1 opened on Feb 6, 2026

feat: use ray.put to reduce overhead of repeated arguments

CI:L2Run doctests, unit tests, functional tests, and convergence testsRun doctests, unit tests, functional tests, and convergence tests

PerformanceRelated to improving performanceRelated to improving performance

super-v3

Status: Open (in progress).

NVIDIA-NeMo/RLnumber 1944

#1944 In NVIDIA-NeMo/RL;

· guyueh1 opened on Feb 14, 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

gpt-oss sft megatron does not support sequence packing

bugSomething isn't workingSomething isn't working

community-request

waiting-on-maintainersWaiting on maintainers to respondWaiting on maintainers to respond

x-mini

x-sf

Status: Open.

#1685 In NVIDIA-NeMo/RL;

· jordane95 opened on Dec 22, 2025