Skip to content

Pull requests: NVIDIA/Megatron-LM

New pull request New
705 Open 3,724 Closed
705 Open 3,724 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 script-transfer procedure to GPT-to-Hybrid migration skill community-request docs-only documentation only (docs or docstrings)
#6251 opened Aug 4, 2026 by guihong-nv Contributor Loading…
2 of 6 tasks
2
refactor(ci): migrate triage consumer to Cerno Approved All necessary approvals have been made community-request complexity: low
#6250 opened Aug 4, 2026 by thomasdhc Contributor Loading…
1 of 6 tasks
2
Enable GTP for heterogeneous MIMO training community-request
#6249 opened Aug 4, 2026 by yashaswikarnati Contributor • Draft
1
[Dev] Release distributed optimizer initialization memory community-request
#6248 opened Aug 4, 2026 by lhb8125 Contributor • Draft
1
avoid circular dependency usage between mcore and modelopt- #5644 community-request complexity: medium Expert Review [deprecated] Apply this label to indicate that your PR is ready for expert review. Run functional tests
#6247 opened Aug 4, 2026 by dimapihtar Contributor Loading…
1 of 6 tasks
2
[dev] Optimize MTP contiguous packed-CP rolls community-request complexity: high
#6246 opened Aug 4, 2026 by hxbai Contributor Loading…
6 tasks
9
Add gigatoken support [not ready for review]
#6245 opened Aug 4, 2026 by dimapihtar Contributor • Draft
6 tasks
1
[dev] Add chunked optimizer-state and master-weight offload community-request complexity: high
#6244 opened Aug 4, 2026 by hxbai Contributor Loading…
1 task
12
[dev] docs(lite): announce Agent Compose transition docs-only documentation only (docs or docstrings)
#6243 opened Aug 4, 2026 by ISEEKYAN Contributor Loading…
1
[feat] GTP+MTP complexity: low Final Review PR is in the "final review" stage nemotron
#6242 opened Aug 4, 2026 by fanshiqing Member Loading…
6 tasks done
1
@fanshiqing
4
Let MLA skip the thd V pad on the packed path community-request
#6241 opened Aug 4, 2026 by bzantium Loading…
5 of 6 tasks
1
Update Nemotron 3 Super GB200 release config
#6239 opened Aug 4, 2026 by Phlip79 Member • Draft
4 of 5 tasks
1
fix: reset loss logging after checkpoint resume community-request
#6238 opened Aug 4, 2026 by Burceoneo12345 Loading…
3
[DO NOT MERGE] Heterogeneous Hybrid Model
#6235 opened Aug 4, 2026 by Phlip79 Member • Draft
2 of 5 tasks
1
[GTP] Cleanup usage of pg_collection in gtp complexity: medium nemotron
#6234 opened Aug 4, 2026 by fanshiqing Member Loading…
6 tasks done
@fanshiqing
16
[main] moe(perf): Refactor CP layout for better long context training performance
#6233 opened Aug 4, 2026 by yuzhongw-nvidia Contributor • Draft
6 tasks
1
Fix GTP prefetch parameter readiness nemotron
#6232 opened Aug 4, 2026 by JF-D Contributor Loading…
1 task done
@JF-D
3
tests: update golden values for inference_optimized_swiglu and moe_resume_gb200 (MCORE-610)
#6231 opened Aug 4, 2026 by kajalj22 Contributor • Draft
3 tasks
2
Release MFSDP storage after model deletion complexity: low Final Review PR is in the "final review" stage Run functional tests
#6230 opened Aug 3, 2026 by wujingyue Contributor Loading…
3
[NCCL EP] Support eager mode and overflow handling complexity: medium
#6229 opened Aug 3, 2026 by YangFei1990 Contributor Loading…
6 tasks done
5
Support TE single-grouped MoE weights in Megatron-FSDP v2
#6228 opened Aug 3, 2026 by wujingyue Contributor • Draft
1
route collectives through nccl2 community-request
#6227 opened Aug 3, 2026 by tushar00jain • Draft
1
Streaming tool call deltas support complexity: medium
#6226 opened Aug 3, 2026 by nvcsathe Contributor Loading…
6 tasks
1
Materialize MFSDP main gradients on reduce-scatter stream complexity: low Run functional tests
#6224 opened Aug 3, 2026 by wujingyue Contributor Loading…
1
Inference: Move tokenizer to a separate thread and scale frontend replicas with DP-size
#6223 opened Aug 3, 2026 by sidsingh-nvidia Contributor • Draft
6 tasks
4
ProTip! Filter pull requests by the default branch with base:main.