Skip to content

Pull requests: NVIDIA/Megatron-LM

New pull request New
426 Open 2,765 Closed
426 Open 2,765 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

Fix CLI arg name mismatch for inference max sequence length community-request
#4836 opened May 17, 2026 by Liamu-Lin Loading…
5 tasks
2
Add minimal DBuffer prototype
#4835 opened May 17, 2026 by wujingyue Contributor • Draft
1
[Dev][fix] FSDP EP-overlap CUDA-graph guard uses post-refactor API
#4834 opened May 17, 2026 by buptzyb Contributor • Draft
3 tasks
3
Fix tokenizers bug in nightly
#4833 opened May 16, 2026 by Phlip79 Member • Draft
1
4
varlendataset for thd e2e and benchmark
#4832 opened May 16, 2026 by xiaoyao0115 Contributor Loading…
5 tasks
@xiaoyao0115
1
chore: nightly sync main into dev (16_05_2026) Run functional tests Run MBridge tests Attach this for testing this PR against MBridge main
#4831 opened May 16, 2026 by svcnvidia-nemo-ci • Draft
7
fix(mtp): bypass save_for_backward for non-tensor kwargs in _checkpointed_forward (Fixes #3643) community-request
#4830 opened May 16, 2026 by PavelPaha Loading…
4 of 5 tasks
1
2
[1/4] Hyperparameter Transfer: add scaling policy infrastructure community-request
#4829 opened May 16, 2026 by plugyawn Contributor Loading…
3 of 5 tasks
3
Fix CP RNG tracker semantics across dropout and checkpoint loading community-request
#4828 opened May 16, 2026 by ewan0x79 Loading…
1 of 5 tasks
1
Avoid GPU0 CUDA context growth from checkpoint common-step broadcast community-request Final Review PR is in the "final review" stage
#4827 opened May 16, 2026 by yhgalaxy Loading…
2 of 5 tasks
4
Add opt-in MXFP8 LM-head output projection complexity: medium
#4825 opened May 15, 2026 by gdengk Contributor Loading…
3 of 5 tasks
2
Allow YAML MoE configs to use model specs community-request
#4822 opened May 15, 2026 by chawkins-nvidia Loading…
2
Add Auto Quantize in ModelOpt quantize example
#4821 opened May 15, 2026 by jenchen13 Contributor • Draft
5 tasks
2
Move bert and t5 pretrain files complexity: low
#4820 opened May 15, 2026 by Phlip79 Member Loading…
2
Move *_builders.py into megatron/training/builders complexity: medium
#4819 opened May 15, 2026 by Phlip79 Member Loading…
1
[MXFP8] Mirror fixes in Mbridge for mxfp8 param gather dev branch Dev branch related issues and development
#4818 opened May 15, 2026 by zhongbozhu Contributor Loading…
5 tasks
1
[Dev][opt] Optimize e_proj and h_proj TP communication for MTP with mHC community-request
#4817 opened May 15, 2026 by Baibaifan Loading…
5
[Dev] add support for deepep/hybridep dispatcher under thd format training dev branch Dev branch related issues and development Expert Review [deprecated] Apply this label to indicate that your PR is ready for expert review. module: moe
#4816 opened May 15, 2026 by HaochenYuan Contributor Loading…
5 tasks done
2
chore: Bump nvrx to 0.6.0 Run functional tests
#4814 opened May 15, 2026 by chtruong814 Contributor Loading…
5 tasks
ci: declare workflow-level contents: read on 7 read-only workflows community-request
#4811 opened May 14, 2026 by arpitjain099 • Draft
3 tasks done
2
Fix CUDA IMA in fsdp_double_buffer when an FSDP unit's bucket doesn't fit the pool complexity: low Run tests
#4810 opened May 14, 2026 by wujingyue Contributor Loading…
1 of 2 tasks
3
Modernize post-training modelopt example scripts complexity: high
#4807 opened May 14, 2026 by kevalmorabia97 Contributor Loading…
5 tasks done
12
test: add inference performance test harness for GPT 583M, hybrid 2B,… complexity: low Run functional tests
#4806 opened May 14, 2026 by shanmugamr1992 Contributor Loading…
5 tasks
5
Fix bug with Megatron-FSDP zero counter not working with decoupled gradients. complexity: low Final Review PR is in the "final review" stage nemotron
#4802 opened May 14, 2026 by cspades Member Loading…
5 tasks
@cspades
2
Enhance MimoOptimizer and tests for distributed checkpointing support
#4801 opened May 14, 2026 by kamran-nvidia • Draft
2 of 5 tasks
4
ProTip! no:milestone will show everything without a milestone.