Skip to content

Pull requests: NVIDIA/Megatron-LM

New pull request New
273 Open 1,353 Closed
273 Open 1,353 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

Update moe github test cases. Run functional tests
#3077 opened Jan 26, 2026 by Victarry Loading…
6 tasks
Core 0.16
@Victarry
2
[Placeholder] Fix Linear-Cross-Entropy Convergence Issue
#3076 opened Jan 26, 2026 by shjwudp • Draft
6 tasks
1
Cherry-pick PR#2581 and PR#2776 to main
#3075 opened Jan 26, 2026 by BestJuly Loading…
6 tasks
Core 0.16
(Draft)[Main] fix cg missing wgrad hook bug Something isn't working
#3074 opened Jan 26, 2026 by Wohox Loading…
6 tasks
1
debug: Unit test failures
#3073 opened Jan 25, 2026 by chtruong814 Loading…
6 tasks
Core 0.16
Change final review logic for PRs complexity: medium
#3072 opened Jan 25, 2026 by Phlip79 Loading…
6 tasks
Core 0.16
2
[Megatron-FSDP] Add dtype customization to Megatron-FSDP. Expert Review Apply this label to indicate that your PR is ready for expert review. module: megatron-fsdp
#3067 opened Jan 24, 2026 by cspades Loading…
3 of 6 tasks
Core 0.16
@cspades
Draft: Harden GRPO functional tests Run functional tests
#3065 opened Jan 24, 2026 by jon-barker Loading…
2 of 6 tasks
Core 0.16
@jon-barker
13
Add block hash tracking for prefix caching
#3063 opened Jan 23, 2026 by lmcafee-nvidia • Draft
6 tasks
1
Inference functional tests: Write outputs to INFERENCE_OUTPUT_PATH instead of TENSORBOARD_PATH
#3061 opened Jan 23, 2026 by mathemakitten Loading…
6 tasks
Core 0.16
Sajadn/extract embd
#3060 opened Jan 23, 2026 by sajadn Loading…
1
μP: Maximal Update Parameterization community-request
#3058 opened Jan 23, 2026 by plugyawn Loading…
3 of 6 tasks
5
Reapply "Various CUDA graph improvements on capture time, replay time, memory footprint (#2572)" Run functional tests
#3056 opened Jan 23, 2026 by jiemingz Loading…
6 tasks
Core 0.16
@jiemingz
2
docs: Release docs
#3055 opened Jan 23, 2026 by ko3n1g Loading…
6 tasks
Core 0.16
1
Siddharth/inference optimized moe layer module: inference module: moe
#3054 opened Jan 23, 2026 by sidsingh-nvidia • Draft
6 tasks
Core 0.16
@sidsingh-nvidia
1
fix(fsdp): add CLI argument for outer_dp_sharding_strategy community-request Final Review Apply this label to indicate that your PR is ready for final review. module: megatron-fsdp needs-follow-up Issue needs follow-up
#3053 opened Jan 23, 2026 by liuyun7345 Loading…
4 tasks done
Core 0.16
1
5
Added --ft-num-warmup-iters option. complexity: low Final Review Apply this label to indicate that your PR is ready for final review. module: training
#3052 opened Jan 23, 2026 by hexinw-nvidia Loading… Core 0.16
2
added vllm fakequant export support
#3050 opened Jan 22, 2026 by kinjalpatel27 Loading…
6 tasks
1
Refactor rl_offload_kv_cache_during_training to offload KV cache to CPU while retaining fixed virtual address
#3048 opened Jan 22, 2026 by mathemakitten Loading…
1 of 6 tasks
Core 0.16
3
[dev] pull main 260122
#3045 opened Jan 22, 2026 by FDecaYed Loading…
6 tasks
Core 0.16
13
Add absorbed-mla
#3044 opened Jan 22, 2026 by kunlunl Loading…
6 tasks
1
[fix] Bug fix for offloading in evaluate() Expert Review Apply this label to indicate that your PR is ready for expert review. needs-follow-up Issue needs follow-up
#3043 opened Jan 22, 2026 by lhb8125 Loading…
6 tasks
Core 0.16
@lhb8125
2
[main] perf(moe): Refine gated delta net implementation
#3042 opened Jan 22, 2026 by yuzhongw-nvidia • Draft
6 tasks
1
[dev] perf(moe): Refine gated delta net implementation
#3040 opened Jan 22, 2026 by yuzhongw-nvidia Loading…
6 tasks
Core 0.16
3
Fuse MLA DOWN projection GEMMs community-request complexity: medium dev2main: mbridge dev to main: this PR is needed in main for mbridge Expert Review Apply this label to indicate that your PR is ready for expert review.
#3039 opened Jan 22, 2026 by cjld Loading…
6 tasks
Core 0.16
4
ProTip! Updated in the last three days: updated:>2026-01-23.