Skip to content

Pull requests: NVIDIA/Megatron-LM

New pull request New
918 Open 4,431 Closed
918 Open 4,431 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: AUT-2254 keep mHC MoE layers out of local CUDA graphs Run selective unit tests
#7276 opened Sep 12, 2026 by svcnemo-autobot Collaborator • Draft
3
Add a user guide for post-training quantization and distillation community-request
#7274 opened Sep 12, 2026 by adityaghai07 Contributor • Draft
1
fix(tests): stop the PP-agnostic model from shadowing SwigluFactoryModel community-request
#7273 opened Sep 12, 2026 by Anai-Guo Contributor Loading…
2
chore: AUT-2257 nightly sync main into dev (12_09_2026) Run CICD Run full unit tests Run the full H100 unit-test suite instead of PR impact selection Run functional tests
#7272 opened Sep 12, 2026 by svcnemo-autobot Collaborator Loading…
2
Add QuantizedDBuffer for MXFP8 plane storage complexity: medium Run functional tests
#7265 opened Sep 11, 2026 by wujingyue Contributor Loading…
5
Fix FlashInfer NVLS routing buffer race complexity: low Run functional tests
#7263 opened Sep 11, 2026 by wdykas Contributor Loading…
2
Add mcore-determinism-debug skill
#7262 opened Sep 11, 2026 by ZhiyuLi-Nvidia Contributor • Draft
6 tasks
2
Rename DBuffer.get_local_tensor to get_tensor_view complexity: low Run tests
#7261 opened Sep 11, 2026 by wujingyue Contributor Loading…
2
Add recipe coverage checks against determinism evidence
#7260 opened Sep 11, 2026 by Connor-XY Contributor • Draft
1
Add paired determinism performance benchmarks
#7259 opened Sep 11, 2026 by Connor-XY Contributor • Draft
1
Separate DBuffer layout construction from allocation complexity: low Run tests
#7257 opened Sep 11, 2026 by wujingyue Contributor Loading…
3
Harden dynamic inference request lifecycle handling complexity: high Final Review PR is in the "final review" stage
#7256 opened Sep 11, 2026 by lmcafee-nvidia Contributor Loading…
3
Wire packed metadata and Super models into multimodal SFT
#7255 opened Sep 11, 2026 by tylerpoon Contributor • Draft
1 task
1
Add the Energon multimodal SFT data pipeline
#7254 opened Sep 11, 2026 by tylerpoon Contributor • Draft
1 task
1
Support packed multimodal inputs in Super hybrid models
#7253 opened Sep 11, 2026 by tylerpoon Contributor • Draft
1 task
2
Track finite packed SFT datasets and metrics
#7252 opened Sep 11, 2026 by tylerpoon Contributor • Draft
1 task
2
Initialize multimodal models from language checkpoints
#7251 opened Sep 11, 2026 by tylerpoon Contributor • Draft
1 task
3
Support multimodal metadata and vision context parallelism
#7250 opened Sep 11, 2026 by tylerpoon Contributor • Draft
1 task
3
Add multimodal vision token utilities complexity: medium
#7249 opened Sep 11, 2026 by tylerpoon Contributor Loading…
1 task done
2
Add multimodal tokenizer conversation formats complexity: high
#7248 opened Sep 11, 2026 by tylerpoon Contributor Loading…
1 task
2
Handle NVRx installs without package version complexity: low
#7247 opened Sep 11, 2026 by tylerpoon Contributor Loading…
1 task done
2
chore(deps): Update transformer engine to 2.19 in pyproject.toml and uv.lock Run functional tests Run MBridge tests Attach this for testing this PR against MBridge main
#7246 opened Sep 11, 2026 by balasaajay Contributor • Draft
6 tasks
3
Add speculative decoding support for GDP
#7245 opened Sep 11, 2026 by santhnm2 Contributor • Draft
6 tasks
1
Fix/dist ckpt get torch version nameerror community-request
#7242 opened Sep 11, 2026 by A-transformer Contributor Loading…
6 tasks
2
Fix/cuda graphs weakref arg nameerror community-request
#7241 opened Sep 11, 2026 by A-transformer Contributor Loading…
6 tasks
2
ProTip! Exclude everything labeled bug with -label:bug.