Skip to content

Pull requests: NVIDIA/Megatron-LM

New pull request New
727 Open 3,751 Closed
727 Open 3,751 Closed
Author
Filter by author
Loading
Label
Filter by label
Loading
Use alt + click/return to exclude labels
or ⇧ + click/return for logical OR
Projects
Filter by project
Loading
Milestones
Filter by milestone
Loading
Reviews
Filter by reviews
No reviews Review required Approved review Changes requested
Assignee
Filter by who’s assigned
Assigned to nobody Loading
Sort
Sort by
Newest Oldest Most commented Least commented Recently updated Least recently updated Best match
Most reactions
👍 👎 😄 🎉 😕 ❤️ 🚀 👀

Pull requests list

Require an explicit process-group collection in LanguageModule
#6303 opened Aug 5, 2026 by Connor-XY Contributor • Draft
5 of 6 tasks
1
Revert "avoid circular dependency usage between mcore and modelopt- #5644" complexity: medium
#6302 opened Aug 5, 2026 by Phlip79 Member Loading…
2
Reject unsupported heterogeneous hybrid inference
#6301 opened Aug 5, 2026 by Phlip79 Member • Draft
1
Add Nemotron direct hybrid architecture examples
#6300 opened Aug 5, 2026 by Phlip79 Member • Draft
1
Integrate direct hybrid architectures with training
#6299 opened Aug 5, 2026 by Phlip79 Member • Draft
2
Support direct hybrid MTP layer specs
#6298 opened Aug 5, 2026 by Phlip79 Member • Draft
2
Execute direct hybrid layers across PP and VPP
#6297 opened Aug 5, 2026 by Phlip79 Member • Draft
2
fix(packaging): AUT-1323 include megatron.post_training in package discovery (dev Run tests
#6296 opened Aug 5, 2026 by svcnemo-autobot Collaborator Loading…
2
Add direct hybrid architecture descriptors
#6295 opened Aug 5, 2026 by Phlip79 Member • Draft
17
Remove global process-group reads from megatron/core (207 -> 155)
#6293 opened Aug 5, 2026 by Connor-XY Contributor • Draft
5 of 6 tasks
1
Revert "Add Agent Compose package and runtime interface skeleton" complexity: medium
#6292 opened Aug 5, 2026 by Phlip79 Member • Queued
3
fix(packaging): AUT-1323 include megatron.post_training in package disco Run tests
#6290 opened Aug 5, 2026 by svcnemo-autobot Collaborator • Draft
4
[DEV] fix(multimodal_dev): derive mock samples from the sample index
#6287 opened Aug 5, 2026 by xuwchen Contributor • Draft
6 tasks
1
Apply the layerwise param layout on the ModelBuilder DDP path Run functional tests
#6286 opened Aug 5, 2026 by deepakn94 Contributor • Draft
2 tasks
2
Support MXFP8 parameter gather in MIMO training
#6284 opened Aug 5, 2026 by yashaswikarnati Contributor • Draft
3 of 6 tasks
1
cp: Update DEEPEP_COMMIT in Dockerfile.ci.dev to commit de0dd1… (6261) into core_r0.19.0 cherry-pick Run CICD
#6283 opened Aug 5, 2026 by ko3n1g Contributor Loading…
2
save tokenizer assets [not ready for review]
#6280 opened Aug 5, 2026 by dimapihtar Contributor • Draft
1 of 6 tasks
1
[Dev] Add TileLang DSA and indexer weights projection precision controls
#6279 opened Aug 5, 2026 by buptzyb Contributor • Draft
1
Fix latent-MoE expert fc2 initialization community-request
#6278 opened Aug 5, 2026 by bo-ke Loading…
6 tasks
2
[DEV] fix(multimodal_dev): make --record-memory-history work on the model_provider entry complexity: low
#6277 opened Aug 5, 2026 by xuwchen Contributor Loading…
6 tasks
1
Print params L2 norm before the start of training complexity: low
#6276 opened Aug 5, 2026 by gautham-kollu Contributor Loading…
6 tasks
3
[Dev] Add MoonEP balanced expert-parallel token dispatcher
#6274 opened Aug 5, 2026 by Wohox Contributor • Draft
1
TE Layernorm dtype guard with fp32 residuals
#6272 opened Aug 5, 2026 by mkhona-nvidia Contributor • Draft
1 of 6 tasks
2
Fix fused MLA and MTP checkpointing for Megatron-FSDP. complexity: medium
#6271 opened Aug 4, 2026 by rapatel Contributor Loading…
6 tasks done
@rapatel @gautham-kollu @cspades
9
Make CPU-only model conversion CUDA-free complexity: low Run tests
#6270 opened Aug 4, 2026 by cuichenx Contributor • Draft
23
ProTip! no:milestone will show everything without a milestone.