-
Notifications
You must be signed in to change notification settings - Fork 3.4k
Pull requests: NVIDIA/Megatron-LM
Author
Label
Projects
Milestones
Reviews
Assignee
Sort
Pull requests list
docs: fix broken images, links, and typos across documentation
docs-only
documentation only (docs or docstrings)
Expert Review
Apply this label to indicate that your PR is ready for expert review.
module: documentation
#2794
opened Jan 2, 2026 by
sbhavani
Loading…
1 of 6 tasks
[Dev] Support bf16 pricision optimizer store bf16 ckeckpoint
community-request
dev branch
Dev branch related issues and development
#2790
opened Dec 31, 2025 by
Baidu-AIAK
Loading…
[Dev] Fix precision issues when resuming training from a checkpoint with BF16 and optimizer offload enabled
community-request
dev branch
Dev branch related issues and development
#2789
opened Dec 31, 2025 by
Baidu-AIAK
Loading…
[Dev] Support optimizer offload when enable --fp8-param-gather
community-request
dev branch
Dev branch related issues and development
#2788
opened Dec 31, 2025 by
Baidu-AIAK
Loading…
Move kitchen extension file to private kitchen repository
#2779
opened Dec 30, 2025 by
kwyss-nvidia
Loading…
6 tasks
Fix clip_qk for virtual pipeline size > 1
dev branch
Dev branch related issues and development
#2776
opened Dec 30, 2025 by
juntaowww
Loading…
6 tasks
Make default for rerun_mode=disabled not terminate with non-fatal rer…
complexity: low
#2773
opened Dec 29, 2025 by
kwyss-nvidia
Loading…
6 tasks
Align gpt-oss window-size with 128-token sliding window
community-request
#2771
opened Dec 29, 2025 by
returnL
Loading…
1 of 6 tasks
Fix: Perform sigmoid calculation in fp32 for aux loss stability
community-request
#2765
opened Dec 26, 2025 by
CodersAcademy006
Loading…
[MoE]Enable Casting-Free FP8-Flow-MoE Blockwise FP8 Dataflow
community-request
complexity: high
#2764
opened Dec 26, 2025 by
xiaoxi-wangfj
Loading…
1 of 6 tasks
Fuse permute+pad and unpermute+unpad ops for FP8 optimization
community-request
#2763
opened Dec 26, 2025 by
xiaoxi-wangfj
Loading…
1 of 6 tasks
Replaces ModuleSpec with Protocols for some of the inputs to SelfAttention/CrossAttention
community-request
complexity: medium
#2761
opened Dec 25, 2025 by
nschank
Loading…
2 of 6 tasks
[Dev] Optimizer State and Master Weight Offloading
dev branch
Dev branch related issues and development
[DEV][WIP][DO NOT MERGE][REFACTOR] Introduce ContextParallelHandler for Unified Context Parallelism Abstraction
community-request
#2749
opened Dec 24, 2025 by
littsk
Loading…
5 of 6 tasks
Add retry logic with configurable delays to checkpoint write operations
complexity: medium
#2746
opened Dec 24, 2025 by
apaithankar
Loading…
6 tasks
Add RL support for MOEs
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.
Run functional tests
Run tests
Previous Next
ProTip!
Adding no:label will show everything without a label.