Owner

NVIDIA-NeMo

1 indexed repository · View on GitHub

RL
Scalable toolkit for efficient model reinforcement

Python · 2025 stars

Open All

CP+SP (sequence parallel) failed in DTensor worker Open

bug t-pytdensor

NVIDIA-NeMo/RL#659 · 3 comments · 0 reactions · 1 assignee · Indexed Sep 17, 2026
Directly supporting Megatron Core checkpoints Open

t-mcore

NVIDIA-NeMo/RL#671 · 2 comments · 9 reactions · 0 assignees · Indexed Sep 17, 2026
Issues encountered in custom multi-turn example Open

external x-google

NVIDIA-NeMo/RL#682 · 1 comment · 0 reactions · 0 assignees · Indexed Sep 17, 2026
[suggest] please refactor data module Open

external x-shailab

NVIDIA-NeMo/RL#688 · 4 comments · 0 reactions · 0 assignees · Indexed Sep 17, 2026
[meta] Before posting an issue Open

NVIDIA-NeMo/RL#701 · 0 comments · 0 reactions · 0 assignees · Indexed Sep 17, 2026
Break up grpo_train to be more composable Open

enhancement external x-futurehouse x-google

NVIDIA-NeMo/RL#706 · 0 comments · 0 reactions · 0 assignees · Indexed Sep 17, 2026
Sequence Packing (and Dynamic Batching) support in DPO Open

t-mcore t-pytdensor

NVIDIA-NeMo/RL#719 · 1 comment · 0 reactions · 1 assignee · Indexed Sep 17, 2026
Reward Model Training on Megatron backend Open

t-mcore

NVIDIA-NeMo/RL#720 · 8 comments · 0 reactions · 0 assignees · Indexed Sep 17, 2026
Add option for sequence padding with packed sequences Open

enhancement t-pytdensor

NVIDIA-NeMo/RL#740 · 1 comment · 0 reactions · 1 assignee · Indexed Sep 17, 2026
[PyT DTensor]Qwen/Qwen3-30B-A3B fails due to assertion error Open

bug t-pytdensor

NVIDIA-NeMo/RL#758 · 1 comment · 0 reactions · 0 assignees · Indexed Sep 17, 2026
PyT DTensor Path - Llama 70B with 4k seq gives OOM with sequence packing enabled Open

bug memory issue t-pytdensor

NVIDIA-NeMo/RL#769 · 4 comments · 0 reactions · 1 assignee · Indexed Sep 17, 2026
RuntimeError: The size of tensor a (8192) must match the size of tensor b (8189) at non-singleton dimension 2 Open

bug i-dgx

NVIDIA-NeMo/RL#775 · 1 comment · 0 reactions · 1 assignee · Indexed Sep 17, 2026
SFT training fails when using sequence_packing and TP=2 on 16K sequences Open

bug community-request external t-pytdensor x-google

NVIDIA-NeMo/RL#786 · 3 comments · 0 reactions · 1 assignee · Indexed Sep 17, 2026
[mcore] Qwen3-30B-A3B performance on long context (40k) Tracker Open

bug t-mcore

NVIDIA-NeMo/RL#788 · 3 comments · 0 reactions · 1 assignee · Indexed Sep 17, 2026
[multimodal dtensor] Inconsistent logprobs for multimodal models Open

bug t-pytdensor

NVIDIA-NeMo/RL#793 · 2 comments · 0 reactions · 1 assignee · Indexed Sep 17, 2026
Indifferent prompt/generation sequence length Open

bug

NVIDIA-NeMo/RL#796 · 0 comments · 1 reaction · 1 assignee · Indexed Sep 17, 2026
Scaling up non-colocated generation cluster Open

asyncRL

NVIDIA-NeMo/RL#813 · 1 comment · 0 reactions · 1 assignee · Indexed Sep 17, 2026
Set timeout and RAM limit inside code environment Open

external x-google

NVIDIA-NeMo/RL#816 · 0 comments · 0 reactions · 0 assignees · Indexed Sep 17, 2026
Add Qwen 235B recipe Open

t-mcore

NVIDIA-NeMo/RL#819 · 2 comments · 0 reactions · 1 assignee · Indexed Sep 17, 2026
[RFC] Heterogeneous GPU types for training and generation Open

needs-design

NVIDIA-NeMo/RL#824 · 1 comment · 0 reactions · 0 assignees · Indexed Sep 17, 2026
[Feature Request] Add Closed-Loop Critic Refinement and Alignment Open

algorithm external x-abeja

NVIDIA-NeMo/RL#828 · 0 comments · 0 reactions · 0 assignees · Indexed Sep 17, 2026
Migrate Llama-Nemotron reasoning tutorial to Nemo RL Open

NVIDIA-NeMo/RL#831 · 1 comment · 0 reactions · 1 assignee · Indexed Sep 17, 2026
Reinforce++ Open

algorithm r0.6.0

NVIDIA-NeMo/RL#832 · 1 comment · 0 reactions · 1 assignee · Indexed Sep 17, 2026
gpt-oss tools support Open

NVIDIA-NeMo/RL#842 · 1 comment · 0 reactions · 0 assignees · Indexed Sep 17, 2026
Have an example showing Reward/Critic runs in an independnent rayVirtualWorker Open

external x-SBIntuitions

NVIDIA-NeMo/RL#853 · 1 comment · 0 reactions · 0 assignees · Indexed Sep 17, 2026
[Proporsal]Stop injecting the chat template into `messages_log_batch[i]["content"]`; store raw text instead Open

external x-SBIntuitions

NVIDIA-NeMo/RL#854 · 4 comments · 0 reactions · 1 assignee · Indexed Sep 17, 2026
Examples for code environment and/or tool use. Open

NVIDIA-NeMo/RL#858 · 0 comments · 6 reactions · 0 assignees · Indexed Sep 17, 2026
Environments currently return role as part of content Open

NVIDIA-NeMo/RL#862 · 1 comment · 0 reactions · 0 assignees · Indexed Sep 17, 2026
Qwen2.5VL-7B-Instruct model loss starts with nan Open

bug t-pytdensor

NVIDIA-NeMo/RL#874 · 1 comment · 1 reaction · 0 assignees · Indexed Sep 17, 2026
Nemotron-H doesn't support sequence parallel Open

t-mcore

NVIDIA-NeMo/RL#881 · 0 comments · 0 reactions · 0 assignees · Indexed Sep 17, 2026
Non-colocated with vllm tp8pp2 has abnormal token_mult_prob_error in FSDP2 backend Open

bug t-pytdensor

NVIDIA-NeMo/RL#882 · 0 comments · 0 reactions · 0 assignees · Indexed Sep 17, 2026
Enable GRPO Qwen 3 32B with 128k context length Open

enhancement memory issue Performance research t-mcore training

NVIDIA-NeMo/RL#883 · 2 comments · 0 reactions · 1 assignee · Indexed Sep 17, 2026
GRPO OOM testing script Open

research t-mcore t-pytdensor

NVIDIA-NeMo/RL#884 · 1 comment · 0 reactions · 0 assignees · Indexed Sep 17, 2026
Long sequence length (1M+ seq len) memory efficiency Open

research t-mcore

NVIDIA-NeMo/RL#885 · 1 comment · 0 reactions · 0 assignees · Indexed Sep 17, 2026
Qwen3-Coder-480B-A35B-Instruct and large MoE support Open

research t-mcore

NVIDIA-NeMo/RL#886 · 1 comment · 0 reactions · 0 assignees · Indexed Sep 17, 2026
SFT runtime efficiency is slower than expected Open

research t-mcore

NVIDIA-NeMo/RL#888 · 2 comments · 0 reactions · 0 assignees · Indexed Sep 17, 2026
SFT accuracy verification and CI Open

research

NVIDIA-NeMo/RL#889 · 2 comments · 0 reactions · 0 assignees · Indexed Sep 17, 2026
GRPO accuracy regression and CI Open

NVIDIA-NeMo/RL#891 · 3 comments · 0 reactions · 1 assignee · Indexed Sep 17, 2026
Nemo-rl OOM issue for 1.5b sft training Open

research t-mcore

NVIDIA-NeMo/RL#903 · 0 comments · 0 reactions · 0 assignees · Indexed Sep 17, 2026
[docs] inaccurate / misleading blog entry Open

bug Documentation

NVIDIA-NeMo/RL#904 · 0 comments · 0 reactions · 0 assignees · Indexed Sep 17, 2026
gemma-3-4b-it has abnormal behavior with load_checkpoint Open

bug t-pytdensor

NVIDIA-NeMo/RL#907 · 0 comments · 0 reactions · 0 assignees · Indexed Sep 17, 2026
Refactor dataset module for train Open

data module UX

NVIDIA-NeMo/RL#909 · 0 comments · 0 reactions · 2 assignees · Indexed Sep 17, 2026
[FSDP2] set expandable_segments:True to reduce memory fragmentation Open

bug external t-pytdensor x-pyt

NVIDIA-NeMo/RL#919 · 4 comments · 0 reactions · 1 assignee · Indexed Sep 17, 2026
When batch size is bigger than dataset size, the SFT finishes without error, but not doing any grad updates Open

bug

NVIDIA-NeMo/RL#921 · 0 comments · 0 reactions · 0 assignees · Indexed Sep 17, 2026
nvtx markers for async llm engine Open

Performance

NVIDIA-NeMo/RL#922 · 2 comments · 0 reactions · 1 assignee · Indexed Sep 17, 2026
Coverage report for `nemo_rl/distributed/model_utils.py` is incorrect Open

bug CI Infra testing

NVIDIA-NeMo/RL#928 · 0 comments · 0 reactions · 0 assignees · Indexed Sep 17, 2026
Performance issues for SFT on Qwen 30B-a3b Open

Performance research t-mcore

NVIDIA-NeMo/RL#937 · 1 comment · 1 reaction · 1 assignee · Indexed Sep 17, 2026
Rollout observability: GRPO uploads 1-2 rollouts per step to W&B Open

research

NVIDIA-NeMo/RL#938 · 0 comments · 0 reactions · 0 assignees · Indexed Sep 17, 2026
Expand Verifier Type to Add Support for Multiple Heuristics Open

algorithm external x-google

NVIDIA-NeMo/RL#940 · 0 comments · 0 reactions · 0 assignees · Indexed Sep 17, 2026
ray ports exhausted when using async vllm on some clusters Open

bug

NVIDIA-NeMo/RL#951 · 1 comment · 0 reactions · 0 assignees · Indexed Sep 17, 2026

Showing the newest 100

This page lists what was indexed most recently. Advanced filter has the whole inventory, narrowed by language, difficulty and how long a task takes.

Open advanced filter

NVIDIA-NeMo

Showing the newest 100

Get new issues in your inbox