NVIDIA-NeMo/Automodel

View on GitHub

🚀 Pytorch Distributed native training library for LLMs/VLMs with OOTB Hugging Face support

Stars: 960
Forks: 316
Open beginner issues: 0
Indexed issues: 192
Avg merge: 3d 20h
Merged PRs (30d): 143

Dominant language: Python
License: Apache-2.0
Last GitHub push: Sep 18, 2026
Latest indexed: Sep 19, 2026
Contributing guide: Contributing guide
Code of conduct: No code of conduct
Beginner labels: good first issue

Beginner friendly Everything

Open All

192 open issues indexed

qwen3_omni_moe adapter_config target_modules miss the thinker. namespace, so peft also adapts the talker Open

bug community-request

NVIDIA-NeMo/Automodel#3944 · 1 assignee · Opened Sep 19, 2026
GLM-5.3-Flash: which Transformers version produced the published HF logit parity? Open

community-request

NVIDIA-NeMo/Automodel#3940 · 1 assignee · Opened Sep 18, 2026
default_collater reshapes the caller's example tensors from [S] to [1, S] Open

community-request

NVIDIA-NeMo/Automodel#3933 · 1 assignee · Opened Sep 17, 2026
VR200 container + tests run Open

PoR

Difficulty 4/5 3-5 days Newbie friendliness 35/100

NVIDIA-NeMo/Automodel#3930 · Opened Sep 17, 2026
Configure checkpoint saving behavior through YAML Open

community-request enhancement

Difficulty 4/5 3-5 days Newbie friendliness 55/100

NVIDIA-NeMo/Automodel#3928 · 2 comments · Opened Sep 17, 2026
Masked CE losses overwrite the caller's labels in place, zeroing a reused batch's loss Open

community-request waiting-on-maintainers

NVIDIA-NeMo/Automodel#3917 · 1 assignee · Opened Sep 16, 2026
Unify PEFT export naming and require real HF reload tests Open

Difficulty 5/5 Over a week Newbie friendliness 35/100

NVIDIA-NeMo/Automodel#3867 · 4 comments · Opened Sep 11, 2026
refactor(distributed): establish shared native TP replica ownership Open

Difficulty 5/5 Over a week Newbie friendliness 25/100

NVIDIA-NeMo/Automodel#3857 · Opened Sep 10, 2026
[TEST] Continuous day-0 release validation Open

enhancement PoR testing

Difficulty 1/5 Under an hour Newbie friendliness 20/100

NVIDIA-NeMo/Automodel#3836 · Opened Sep 8, 2026
grad clipping computes a different "global" norm on every EP group with TE experts Open

bug community-request waiting-on-maintainers

NVIDIA-NeMo/Automodel#3816 · 1 assignee · Opened Sep 5, 2026
model deprecation page in docs Open

NVIDIA-NeMo/Automodel#3807 · 1 assignee · Opened Sep 4, 2026
ChunkedCrossEntropy returns num_chunks x the loss for reduction='mean' and wrong-shaped output for 'none' Open

community-request waiting-on-maintainers

NVIDIA-NeMo/Automodel#3803 · 1 assignee · Opened Sep 3, 2026
Fully-masked batch yields NaN loss in 3 of 4 losses; only MaskedCrossEntropy guards it Open

community-request waiting-on-maintainers

NVIDIA-NeMo/Automodel#3796 · 1 assignee · Opened Sep 3, 2026
Mistral3ForConditionalGeneration: forward does not declare logits_to_keep, so fused losses are silently replaced Open

community-request waiting-on-maintainers

Difficulty 3/5 1-2 days Newbie friendliness 76/100

NVIDIA-NeMo/Automodel#3778 · Opened Sep 1, 2026
--nproc-per-node is overridden by the local GPU count Open

community-request waiting-on-maintainers

NVIDIA-NeMo/Automodel#3774 · 1 assignee · Opened Sep 1, 2026
max_steps is silently ignored when num_epochs is left unset Open

community-request waiting-on-maintainers

Difficulty 2/5 1-3 hours Newbie friendliness 86/100

NVIDIA-NeMo/Automodel#3767 · Opened Aug 31, 2026
group_by_length silently does nothing for lazily-tokenized datasets Open

community-request waiting-on-maintainers

Difficulty 3/5 1-2 days Newbie friendliness 76/100

NVIDIA-NeMo/Automodel#3755 · Opened Aug 30, 2026
[ALERT] Access to NVIDIA's Self-Hosted Runners is Expiring Open

nvidia-runners

Difficulty 3/5 1-2 days Newbie friendliness 25/100

NVIDIA-NeMo/Automodel#3754 · Opened Aug 30, 2026
Loading Qwen/Qwen3.8-27B failed with NeMoAutoModelForCausalLM(OOM), but successfully load with Qwen3_5ForCausalLM Open

bug community-request waiting-on-maintainers

NVIDIA-NeMo/Automodel#3745 · 4 comments · 1 assignee · Opened Aug 29, 2026
Qwen3.5 4B TE packed seq Open

enhancement

NVIDIA-NeMo/Automodel#3735 · 1 assignee · Opened Aug 28, 2026
Benchmark ultraep + moonep Open

NVIDIA-NeMo/Automodel#3727 · 1 assignee · Opened Aug 27, 2026
Hunyuan-Image3 image Open

PoR

Difficulty 5/5 Over a week Newbie friendliness 15/100

NVIDIA-NeMo/Automodel#3726 · Opened Aug 27, 2026
Self-Flow support in Automodel Open

PoR

Difficulty 5/5 Over a week Newbie friendliness 20/100

NVIDIA-NeMo/Automodel#3725 · Opened Aug 27, 2026
REPA Support in Diffusion training Open

PoR

Difficulty 5/5 Over a week Newbie friendliness 20/100

NVIDIA-NeMo/Automodel#3724 · Opened Aug 27, 2026
Swe-bench for MoE Open

PoR

NVIDIA-NeMo/Automodel#3723 · 1 comment · 1 assignee · Opened Aug 27, 2026
CP for dLLM Open

NVIDIA-NeMo/Automodel#3722 · 1 assignee · Opened Aug 27, 2026
Validate async tensor parallelism for the main SFT path Open

NVIDIA-NeMo/Automodel#3721 · 1 assignee · Opened Aug 27, 2026
dLLM RL with AM as backend in Nemo-RL Open

NVIDIA-NeMo/Automodel#3720 · 1 assignee · Opened Aug 27, 2026
TP + LoRA perf improvement Open

PoR

NVIDIA-NeMo/Automodel#3718 · 1 assignee · Opened Aug 27, 2026
DiffusionGemma CP Support Open

PoR

NVIDIA-NeMo/Automodel#3717 · 1 assignee · Opened Aug 27, 2026
Upgrade deepep 673 Open

PoR

NVIDIA-NeMo/Automodel#3716 · 1 assignee · Opened Aug 27, 2026
NVFP4 for Automodel Open

NVIDIA-NeMo/Automodel#3714 · 1 assignee · Opened Aug 27, 2026
Model Capability Registry Phase 2 Open

PoR

NVIDIA-NeMo/Automodel#3713 · 1 assignee · Opened Aug 27, 2026
Add a VLM parity test for Qwen3.5-MoE Open

NVIDIA-NeMo/Automodel#3707 · 1 assignee · Opened Aug 27, 2026
RL + dLLM Open

dLLM PoR

NVIDIA-NeMo/Automodel#3706 · 1 assignee · Opened Aug 27, 2026
Centralize MoE runtime backend capability truth for checkpoint adapters Open

NVIDIA-NeMo/Automodel#3689 · 1 assignee · Opened Aug 26, 2026
Support grouped expert bias in checkpoint split/merge adapters Open

NVIDIA-NeMo/Automodel#3688 · 1 assignee · Opened Aug 26, 2026
Profile consolidated checkpoint export storage overhead Open

automodel ckpt enhancement Performance

NVIDIA-NeMo/Automodel#3670 · 1 assignee · Opened Aug 25, 2026
Clean up and take ownership of HF/DCP checkpoint storage Open

automodel ckpt enhancement Performance

NVIDIA-NeMo/Automodel#3669 · 1 assignee · Opened Aug 25, 2026
nemotron_omni peft save skips all key conversion, adapters only work inside automodel Open

bug community-request waiting-on-maintainers

NVIDIA-NeMo/Automodel#3656 · 1 assignee · Opened Aug 25, 2026
Audit and align MoE router precision with reference implementations Open

Difficulty 5/5 Over a week Newbie friendliness 25/100

NVIDIA-NeMo/Automodel#3650 · 5 comments · Opened Aug 24, 2026
Add public architecture registration and entry-point discovery Open

Difficulty 4/5 3-5 days Newbie friendliness 20/100

NVIDIA-NeMo/Automodel#3646 · Opened Aug 24, 2026
[Tracking] checkpoint I/O architecture, correctness, performance, and memory Open

ckpt enhancement Performance PoR

NVIDIA-NeMo/Automodel#3576 · 1 comment · 1 assignee · Opened Aug 18, 2026
Add job-to-Linear issue table to nemo-ci failure reports Open

Automation ci

Difficulty 4/5 3-5 days Newbie friendliness 58/100

NVIDIA-NeMo/Automodel#3568 · Opened Aug 17, 2026
Support TE rope fusion Open

PoR

NVIDIA-NeMo/Automodel#3565 · 1 assignee · Opened Aug 17, 2026
Fix mtp+cp support for existing models Open

bug

NVIDIA-NeMo/Automodel#3564 · 1 assignee · Opened Aug 17, 2026
WAN2.2-Animate-2 Support Open

NVIDIA-NeMo/Automodel#3519 · 1 reaction · 1 assignee · Opened Aug 12, 2026
Gemma4 26b moe source AM v/s HF logit parity investigation Open

NVIDIA-NeMo/Automodel#3465 · 1 assignee · Opened Aug 8, 2026
Discuss PEFT broad-target semantics: match_all_linear vs target_modules ["*"] Open

enhancement

NVIDIA-NeMo/Automodel#3462 · 2 comments · 1 assignee · Opened Aug 7, 2026
Gemma 4 31B random-init pretraining shows transient loss spikes in both AutoModel and Megatron Bridge Open

NVIDIA-NeMo/Automodel#3443 · 1 assignee · Opened Aug 6, 2026

NVIDIA-NeMo/Automodel

Get new issues in your inbox