Skip to content

Pull requests: NVIDIA/Megatron-LM

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
Assignee
Filter by who’s assigned
Assigned to nobody Loading
Sort

Pull requests list

[CI] Fix gpt_dynamic_inference_tp2_pp2_ep2_gptoss_20b_swa tests complexity: low
#5527 opened Jun 28, 2026 by asolergi-nv Contributor Loading…
1 of 6 tasks
Preserve DSA output across fused inverse RoPE complexity: low
#5526 opened Jun 28, 2026 by kunlunl Contributor Loading…
6 tasks
[codex] Fix tensor-parallel label smoothing complexity: low Final Review PR is in the "final review" stage
#5522 opened Jun 27, 2026 by ilml Contributor Loading…
Handle HybridEP packed padding masks
#5515 opened Jun 26, 2026 by seonjinn Contributor Draft
6 tasks
Add CI duties to oncall docs-only documentation only (docs or docstrings)
#5510 opened Jun 26, 2026 by Phlip79 Member Loading…
1 task done
Huvu/gemma4 e4b tp sp
#5508 opened Jun 26, 2026 by huvunvidia Contributor Draft
6 tasks
fix(megatron-fsdp): support A2A overlap with partial CUDA graph
#5505 opened Jun 26, 2026 by xuwchen Contributor Draft
6 tasks
chore: nightly sync main into dev (25_06_2026) complexity: high Run functional tests Run MBridge tests Attach this for testing this PR against MBridge main
#5503 opened Jun 25, 2026 by svcnvidia-nemo-ci Loading…
Hybrid prefix caching fixes complexity: medium
#5502 opened Jun 25, 2026 by santhnm2 Contributor Loading…
1 of 6 tasks
add safe version of numpy.load complexity: low Expert Review [deprecated] Apply this label to indicate that your PR is ready for expert review. Final Review PR is in the "final review" stage
#5500 opened Jun 25, 2026 by dimapihtar Contributor Loading…
1 of 6 tasks
Add cos/sin width guard to fused MLA RoPE kernels community-request Final Review PR is in the "final review" stage waiting-on-maintainers Waiting on maintainers to respond
#5497 opened Jun 25, 2026 by ShauryaaSharma Loading…
4 of 6 tasks
Document stacked dependent PR handling in split PR skill docs-only documentation only (docs or docstrings)
#5496 opened Jun 25, 2026 by wujingyue Contributor Loading…
Inference: Use FA4 for prefill and FA2 for decode
#5494 opened Jun 25, 2026 by sidsingh-nvidia Contributor Draft
1 of 6 tasks
Add statistics logging for params and activations
#5492 opened Jun 25, 2026 by philipcmonk Draft
4 of 6 tasks
Refactor RL rollout pipeline complexity: medium
#5491 opened Jun 25, 2026 by lauradang Contributor Loading…
6 tasks
ProTip! Adding no:label will show everything without a label.