Skip to content

Pull requests: NVIDIA/Megatron-LM

New pull request New
250 Open 1,016 Closed
250 Open 1,016 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

Add moe layer perf UT.
#2673 opened Dec 16, 2025 by Victarry • Draft
6 tasks
1
docs: improve documentation organization and add additional guides docs-only documentation only (docs or docstrings) Expert Review Apply this label to indicate that your PR is ready for expert review.
#2671 opened Dec 16, 2025 by sbhavani Loading…
1 of 6 tasks
1
[dev] feat(moe): Support MLA CP exchanging latent KV
#2670 opened Dec 16, 2025 by yuzhongw-nvidia • Draft
6 tasks
1
[docs] Add ability to disable autodoc2 for local builds docs-only documentation only (docs or docstrings)
#2669 opened Dec 15, 2025 by Phlip79 Loading…
6 tasks
1
2
Prep work for migrating to types from ModuleSpec community-request
#2668 opened Dec 15, 2025 by nschank Loading…
2 of 6 tasks
1
fix hang issue for pp>1 vpp>1 and variable_seq_lengths==True Expert Review Apply this label to indicate that your PR is ready for expert review.
#2664 opened Dec 15, 2025 by shifangx Loading…
6 tasks
1
add all_gather process-group for overlapping in fsdp disributed training
#2663 opened Dec 15, 2025 by jeffnvidia • Draft
6 tasks
1
feat: manual registration mode for nccl-ub option when using megatron-fsdp
#2661 opened Dec 15, 2025 by youngeunkwon0405 Loading…
6 tasks
Core 0.16
@youngeunkwon0405
11
Add example decoupled configs for functional tests.
#2660 opened Dec 15, 2025 by Victarry • Draft
6 tasks
1
feat: Add VPP training simulation framework for performance profiling community-request
#2659 opened Dec 14, 2025 by lsy643 Loading…
6 tasks
1
Update RNG sharding to include EP rank
#2658 opened Dec 13, 2025 by paul-gibbons Loading…
6 tasks
Core 0.16
3
Add check for full_iteration scope before instantiating CudaGraphManager
#2657 opened Dec 13, 2025 by vasunvidia Loading…
6 tasks
Core 0.16
2
DRAFT:Mr4 add flashinfer
#2656 opened Dec 13, 2025 by kanz-nv Loading…
6 tasks
1
[MAIN][NVFP4][MOE] 128 Zero Padding for Grouped Quantization kernels and Cuda Graph Support
#2655 opened Dec 13, 2025 by zhongbozhu Loading…
6 tasks
1
[DEV][NVFP4][MOE] 128 Zero Padding for Grouped Quantization kernels and Cuda Graph Support
#2654 opened Dec 13, 2025 by zhongbozhu Loading…
6 tasks
Core 0.16
2
Move full model init to cuda stream to avoid race condition leading to empty parameters in DDP
#2652 opened Dec 12, 2025 by jstjohn Loading…
6 tasks
1
[main] feat(moe): Support packed sequence for gated delta net (GDN)
#2645 opened Dec 12, 2025 by yuzhongw-nvidia • Draft
6 tasks
1
[dev] feat(moe): Support packed sequence for gated delta net (GDN)
#2644 opened Dec 12, 2025 by yuzhongw-nvidia Loading…
6 tasks
1
[Main] Feat(moe): Gated delta net context parallel (CP)
#2642 opened Dec 12, 2025 by yuzhongw-nvidia • Draft
6 tasks
1
Fix CUDA RNG Tracker
#2641 opened Dec 12, 2025 by buptzyb Loading…
6 tasks
Core 0.16
@buptzyb
3
[Dev] Fix CUDA RNG Tracker core_dev_r0.15.0 dev branch Dev branch related issues and development Expert Review Apply this label to indicate that your PR is ready for expert review. Final Review Apply this label to indicate that your PR is ready for final review.
#2640 opened Dec 12, 2025 by buptzyb Loading…
6 tasks
Core 0.16
@buptzyb
4
Print more verbose error message about incorrect model_parallel_size. community-request
#2639 opened Dec 12, 2025 by rj42 Loading…
6 tasks
1
Quantization type debug logging.
#2636 opened Dec 11, 2025 by kwyss-nvidia Loading…
6 tasks
3
revert: 7968d5f98f8457297d4a73f96d8a086d84a8fa67
#2635 opened Dec 11, 2025 by ko3n1g Loading…
6 tasks
Core 0.16
RL: Rollouts should be distributed over the regular data parallel group
#2634 opened Dec 11, 2025 by sidsingh-nvidia Loading…
6 tasks
Core 0.16
@sidsingh-nvidia
1
ProTip! Filter pull requests by the default branch with base:main.