-
Notifications
You must be signed in to change notification settings - Fork 4.4k
Pull requests: NVIDIA/Megatron-LM
Author
Label
Projects
Milestones
Reviews
Assignee
Sort
Pull requests list
[Dev] Support device-initiated grouped linear without TE op fuser
#6847
opened Aug 25, 2026 by
lhb8125
Contributor
Loading…
Support MIMO input projections on language ranks
Run functional tests
#6843
opened Aug 25, 2026 by
yashaswikarnati
Contributor
•
Draft
[dev](feature): Native implementation for Attention Residuals (AttnRes) with Block AttnRes, pipeline parallelism and MTP support
#6840
opened Aug 25, 2026 by
jingqiny-99
Contributor
•
Draft
[dev] moe(fix): Fix MLA shared K rope grad for TP>1 and SP=False.
complexity: low
#6839
opened Aug 25, 2026 by
yuzhongw-nvidia
Contributor
Loading…
6 tasks
[dev] moe(perf): MLA Latent Context Parallel
#6829
opened Aug 25, 2026 by
yuzhongw-nvidia
Contributor
•
Draft
6 tasks
[Main] Support paged stash with device-initiated GroupedLinear
complexity: medium
#6828
opened Aug 25, 2026 by
lhb8125
Contributor
Loading…
4 of 6 tasks
Reduce CUDA graph side-stream allocator fragmentation
complexity: medium
nemotron
#6827
opened Aug 25, 2026 by
JF-D
Contributor
Loading…
4 of 6 tasks
Assert the runtime-CP group contract both ways
complexity: low
Final Review
PR is in the "final review" stage
#6825
opened Aug 24, 2026 by
ilml
Contributor
Loading…
Fix order local refit copies after source updates
community-request
Final Review
PR is in the "final review" stage
#6823
opened Aug 24, 2026 by
Sunt-ing
Loading…
3 of 6 tasks
Keep RerunDataIterator wrapping on the hybrid CP data iterator
complexity: low
#6822
opened Aug 24, 2026 by
ilml
Contributor
Loading…
Fix runtime CP group handling in attention for hybrid CP
complexity: low
Final Review
PR is in the "final review" stage
#6821
opened Aug 24, 2026 by
ilml
Contributor
Loading…
Previous Next
ProTip!
Adding no:label will show everything without a label.