Skip to content

Pull requests: NVIDIA/Megatron-LM

New pull request New
306 Open 1,918 Closed
306 Open 1,918 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

RL support for nanov3 sft checkpoint
#3741 opened Mar 6, 2026 by jon-barker Loading…
5 tasks
@jon-barker
4
fix(offload): reset activation offload manager after eval as well as …
#3739 opened Mar 6, 2026 by rapatel Loading…
5 tasks
3
Continue emerging optimizer refactoring
#3737 opened Mar 6, 2026 by skyw • Draft
6 tasks
1
revert of #2658 complexity: low Expert Review Apply this label to indicate that your PR is ready for expert review.
#3736 opened Mar 6, 2026 by dimapihtar Loading…
5 tasks
Core 0.16
2
[dev] fix(dev): Customized attention mask for TE backend core attention.
#3732 opened Mar 6, 2026 by yuzhongw-nvidia • Draft
5 tasks
5
Add TP2 FSDP test Run functional tests
#3731 opened Mar 6, 2026 by gautham-kollu Loading…
5 tasks
Core 0.16
6
[FSDP] don't sync param when use no_shard
#3730 opened Mar 6, 2026 by wplf • Draft
3
Prefix caching mamba v3 ssm initial states
#3726 opened Mar 5, 2026 by lmcafee-nvidia • Draft
5 tasks
@lmcafee-nvidia
1
[TEST][DO NOT REVIEW] Mirror of #3373 RL: Hybrid MoE training cudagraphs and fix training <-> inference transition
#3725 opened Mar 5, 2026 by mathemakitten Loading…
5 tasks
Core 0.16
3
Add parser arguments to RL inference
#3722 opened Mar 5, 2026 by ArEsKay3 • Draft
6 tasks
1
Add SPECIAL_999 to the list of available pad tokens to support Nemotron
#3721 opened Mar 5, 2026 by ArEsKay3 • Draft
6 tasks
1
Enhance optimizer state loading with runtime overrides community-request
#3720 opened Mar 5, 2026 by yhgalaxy Loading…
6 tasks
8
chore: Switch to ruff for linting
#3719 opened Mar 5, 2026 by ahmadki • Draft
6 tasks
2
rename hybrid-cp to dynamic-cp and fix for rope when enabling THD + Dynamic-CP
#3717 opened Mar 5, 2026 by xiaoyao0115 Loading…
1
add warning for cp > 1 and not per token loss dev
#3716 opened Mar 5, 2026 by wplf Loading…
1
Scaling fixes for MuP over Muon optimizer. community-request Expert Review Apply this label to indicate that your PR is ready for expert review. Final Review PR is in the "final review" stage
#3715 opened Mar 5, 2026 by plugyawn Loading…
4 of 6 tasks
@BoxiangW
6
Add warning for cp_size > 1 and no per token loss
#3714 opened Mar 5, 2026 by wplf Loading…
1
Fix: work around to enable symmetric registration for HSDP
#3713 opened Mar 5, 2026 by youngeunkwon0405 Loading…
6 tasks
Core 0.16
3
Normalize tool_calls and gate parser tool-calls to tool-enabled requests complexity: low
#3710 opened Mar 4, 2026 by i-riyad Loading…
6 tasks
2
Extract the changes from Jorge's branch
#3701 opened Mar 4, 2026 by tdene • Draft
6 tasks
1
find optimal number of workers complexity: medium Expert Review Apply this label to indicate that your PR is ready for expert review.
#3699 opened Mar 4, 2026 by dimapihtar Loading…
6 tasks
Core 0.16
5
Fix config.softmax_scale not being considered Final Review PR is in the "final review" stage
#3698 opened Mar 4, 2026 by janEbert Loading… Core 0.16
3
[Main][feat] Support CUDA Graph capture offloading modules complexity: medium enhancement New feature or request
#3697 opened Mar 4, 2026 by lhb8125 Loading…
6 tasks
Core 0.16
@lhb8125
8
fix ddp bug when --overlap-grad-reduce and --num-distributed-optimi for dev
#3694 opened Mar 4, 2026 by wplf Loading…
6 tasks
3
fix ddp bug when --overlap-grad-reduce and --num-optim > 1 Final Review PR is in the "final review" stage
#3693 opened Mar 4, 2026 by wplf Loading…
3
ProTip! no:milestone will show everything without a milestone.