Skip to content

Pull requests: NVIDIA/Megatron-LM

New pull request New
174 Open 647 Closed
174 Open 647 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

[Dev] add fp4 when get align size in HybridepManager
#2140 opened Nov 5, 2025 by qijiaxing Loading…
6 tasks
1
Delete redundant import in yaml_arguments.py
#2139 opened Nov 5, 2025 by wplf Loading…
6 tasks
1
Fix Megatron-FSDP checkpoint save failure
#2138 opened Nov 5, 2025 by shjwudp Loading…
6 tasks
1
add device and dtype to empty inv_dt init bug Something isn't working dev branch Dev branch related issues and development Expert Review Apply this label to indicate that your PR is ready for expert review.
#2137 opened Nov 5, 2025 by maanug-nv Loading…
1 of 6 tasks
Core 0.16
@maanug-nv
feat(moe): Support placing MTP layers into standalone stages Expert Review Apply this label to indicate that your PR is ready for expert review. module: moe
#2136 opened Nov 5, 2025 by BestJuly Loading…
6 tasks
Core 0.16
@BestJuly @Wohox @shifangx
1
Add CP + Sequence Packing support for Mimo
#2135 opened Nov 4, 2025 by mehraakash Loading…
5 of 6 tasks
@mehraakash
1
ci: LTS container
#2133 opened Nov 4, 2025 by ko3n1g Loading…
6 tasks
Core 0.16
[Dev] Add more tests for LayerwiseDistOpt with dist_ckpt Expert Review Apply this label to indicate that your PR is ready for expert review. Run tests
#2132 opened Nov 4, 2025 by BoxiangW Loading…
6 tasks
Core 0.15
@BoxiangW
9
Improve ModelOpt paths & add more Nemotron/hybrid model support
#2131 opened Nov 4, 2025 by jenchen13 Loading…
1 of 6 tasks
14
Add debugging hint for case when cudagraphs are created but no matching runner is found
#2129 opened Nov 4, 2025 by mathemakitten Loading…
1 of 6 tasks
Core 0.16
Fix runaway Etpt in straggler detector by resetting FLOPs accumulator
#2128 opened Nov 4, 2025 by sbhavani Loading…
1
Allow DTensors in sharded state dict
#2127 opened Nov 4, 2025 by dimapihtar • Draft
6 tasks
1
remove flattened_range code paths
#2126 opened Nov 4, 2025 by dimapihtar • Draft
6 tasks
1
[Dev] fix(offloading): Accuracy mismatch when offloading and recomputing same module
#2123 opened Nov 4, 2025 by lhb8125 Loading…
6 tasks
Core 0.16
4
[Dev] Remove calculation of padding token in moe routing loss dev branch Dev branch related issues and development module: moe
#2121 opened Nov 4, 2025 by HaochenYuan Loading…
6 tasks
Core 0.16
14
remove training dependency from megatron core for fsdp checkpoint with EP core_r0.15.0 Expert Review Apply this label to indicate that your PR is ready for expert review.
#2113 opened Nov 3, 2025 by ananthsub Loading…
3 of 6 tasks
Core 0.15
3
Update README.md
#2111 opened Nov 3, 2025 by mvirts Loading…
6 tasks
1
Tensorize dynamic inference mixed sampling Expert Review Apply this label to indicate that your PR is ready for expert review. Run functional tests Trains for 50-100 steps and tests against golden values Run tests
#2105 opened Nov 3, 2025 by tdene Loading…
6 tasks done
Core 0.16
12
multi thread read full parallel save ckpt
#2104 opened Nov 3, 2025 by 861482002 Loading…
6 tasks done
1
Add router replay for MoE models module: moe
#2101 opened Nov 3, 2025 by litianjian Loading…
6 tasks
1
ci: Run functional tests Run functional tests Trains for 50-100 steps and tests against golden values
#2100 opened Nov 3, 2025 by ko3n1g Loading…
6 tasks
Core 0.16
Ko3n1g/chore/update dev release settings
#2099 opened Nov 3, 2025 by ko3n1g Loading…
6 tasks
Core 0.16
Remove redundant reduce in aux_loss logging
#2095 opened Nov 3, 2025 by BestJuly Loading…
6 tasks
Core 0.16
@BestJuly
[Dev] Remove redundant reduce in aux_loss logging Expert Review Apply this label to indicate that your PR is ready for expert review. module: moe
#2094 opened Nov 3, 2025 by BestJuly Loading…
6 tasks
Core 0.16
@BestJuly
chore: Merge main into dev
#2093 opened Nov 3, 2025 by chtruong814 Loading…
6 tasks
Core 0.16
5
ProTip! Follow long discussions with comments:>50.