[DRAFT][Tracking] NeMo AutoModel 26.12 roadmap
Maintainers usually reply within 1 day
Nobody has claimed this yet.
Assessment
- Difficulty
- 5/5
- Estimated time
- Over a week
- Newbie friendliness
- 5/100
Research direction
This is a draft roadmap tracking issue, not a single task: it links many sub-issues (#4604, #3713, #4607 and its grouped annotation issues, among others) and has no assignees or comments. Start by reading the grouped issues it references and the Linear PoR project for owners. Done means the listed sub-issues are individually scoped, owned and closed or deferred, which is maintainer planning work rather than a first contribution.
Written by the indexing model from the issue text.
Description
Note: this is a WIP draft.
Track planned features, performance improvements, and engineering work for NeMo AutoModel 26.12.
Proposed schedule: code freeze December 1; release December 17, 2026. Inclusion below reflects current PoR tracking, not a guarantee of delivery.
Linear project: NeMo AutoModel 26.12 PoR
Performance and distributed training
- #4604 — NCCL-EP integration and validation.
- #4605 — Unify performance and convergence recipes.
- #4606 — Complete MeshContext migration.
- #3718 — TP + LoRA performance improvements.
- #3565 — TE RoPE fusion support.
- #1439 — Fix RoPE fusion correctness with CP for MoEs.
Model capabilities
- #3725 — Self-Flow support.
- #3724 — REPA support in diffusion training.
- #3717 — DiffusionGemma context parallelism.
- #1108 — Orpheus speech-model support.
- #3713 — Model Capability Registry, phase 2: standardized capability tests and model-card visibility.
Correctness and maintainability
- #3723 — MoE SWE-bench convergence coverage: exercise EP and PP combinations.
- #2475 — Explicit FP32 storage for full-parameter Adam/AdamW training.
- #3219 — Unify multimodal module discovery and policy configuration.
- #4096 — Checkpoint I/O follow-ups: prioritize reproduced correctness problems and measured bottlenecks.
- #4607 — Extend component type annotations, delivered through six groups:
| Group | Tracking issue |
|---|---|
| Model execution and adapters | #4622 |
| Data, evaluation, and objectives | #4614 |
| Distributed execution and CUDA graphs | #4610 |
| Checkpointing and serialization | #4608 |
| Training lifecycle and optimizers | #4626 |
| Configuration, launchers, and utilities | #4612 |
Annotation work must establish truthful contracts without Any, casts, suppressions, or checker exclusions used as escape hatches. Detailed requirements live in the parent and grouped issues.
Hardware enablement — stretch goal
- #3930 — VR200 container and functional testing. Proposed 26.12 scope: ARM/Hecate engineering preview, per the October 29 discussion. The issue's older 26.10 description needs updating.
Tracking cleanup before finalizing scope
- Reconcile #3957 (Kimi K3 performance/correctness): its description says all inventoried PRs were merged. Identify remaining release work or close the tracker.
- Exclude #3836 (release-validation prototype) from delivery commitments: its description explicitly identifies it as a disposable sample issue.
- Confirm owners and acceptance criteria for title-only or template-only issues before code freeze.
Completion requires the linked issue's acceptance criteria and appropriate validation evidence; merged code alone does not establish performance, convergence, or hardware readiness.
- Dominant language
- Python
- Stars
- 1.1k
- Forks
- 366
- Avg merge
- 1d 20h
- Merged PRs (30d)
- 281
Getting set up
- No Dockerfile or Docker Compose file
- Has a pull request template
- Read the contributing guide
First steps
- Read the whole issue, then the project's contributing guide.
- Comment on the issue to say you are picking it up — it saves two people doing the same work.
- Fork the repository and make your change on a branch.
- Open a pull request that references the issue number.
More from NVIDIA-NeMo/Automodel
-
community-request waiting-on-maintainers
Difficulty 2/5 1-3 hours Newbie friendliness 86/100
NVIDIA-NeMo/Automodel#3767 ·
Maintainers usually reply within 1 day
-
Add sound type annotations: training lifecycle and optimizersPossibly taken @akoumpa claimed this 4 days ago. Opengood first issue help wanted PoR
NVIDIA-NeMo/Automodel#4626 · 1 comment · 1 assignee ·
Maintainers usually reply within 1 day
-
Add sound type annotations: model execution and adaptersPossibly taken A pull request linked to this issue is open or already merged. Opengood first issue help wanted PoR
Difficulty 4/5 3-5 days Newbie friendliness 35/100
NVIDIA-NeMo/Automodel#4622 · 3 comments ·
Maintainers usually reply within 1 day
-
Add sound type annotations: data, evaluation, and objectivesPossibly taken @Alan-donk claimed this 4 days ago. Opengood first issue help wanted PoR
Difficulty 3/5 Half a day Newbie friendliness 55/100
NVIDIA-NeMo/Automodel#4614 · 1 comment ·
Maintainers usually reply within 1 day
-
Add sound type annotations: configuration, launchers, and utilitiesPossibly taken @zahidSkywalker claimed this 4 days ago. Opengood first issue help wanted PoR
Difficulty 3/5 1-2 days Newbie friendliness 48/100
NVIDIA-NeMo/Automodel#4612 · 3 comments ·
Maintainers usually reply within 1 day
All issues in NVIDIA-NeMo/Automodel
Similar issues
-
binding:python bug
Difficulty 2/5 1-3 hours Newbie friendliness 82/100
Maintainers usually reply within 1 day
-
The Browser tab fails with a Wayland error when Open Terminal runs from a Wayland desktop sessionOpen
Difficulty 2/5 1-3 hours Newbie friendliness 78/100
open-webui/open-terminal#210 ·
-
triage/confirmed
Difficulty 2/5 1-3 hours Newbie friendliness 78/100
agentscope-ai/agentscope#3807 ·
Maintainers usually reply within 2 days
-
Difficulty 2/5 1-3 hours Newbie friendliness 78/100
NousResearch/hermes-agent#150542 ·
Maintainers usually reply within 1 day
-
Difficulty 1/5 Under an hour Newbie friendliness 91/100
triton-inference-server/server#9069 ·
Maintainers usually reply within 1 day