Skip to content

Pull requests: NVIDIA/Megatron-LM

New pull request New
247 Open 947 Closed
247 Open 947 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

Fix: don't enter branch if mtp_num_layers == 0 community-request
#2581 opened Dec 6, 2025 by rj42 Loading…
6 tasks
1
Various quality-of-life improvements in training loop
#2580 opened Dec 6, 2025 by deepakn94 • Queued Core 0.16
1
Ignore log level in functional test
#2579 opened Dec 5, 2025 by kwyss-nvidia Loading…
6 tasks
1
Synchronize total block count across pipeline parallel ranks
#2578 opened Dec 5, 2025 by santhnm2 Loading…
6 tasks
2
Fix FA3 import Run functional tests
#2577 opened Dec 5, 2025 by santhnm2 • Queued
6 tasks
Core 0.16
3
fix: ckpt loading failed because of padding metadata in dist optimizer Expert Review Apply this label to indicate that your PR is ready for expert review.
#2576 opened Dec 5, 2025 by yaoyu-33 Loading…
6 tasks
@yaoyu-33
1
[Megatron-FSDP] Support both old and new DeviceMesh APIs. Expert Review Apply this label to indicate that your PR is ready for expert review.
#2575 opened Dec 5, 2025 by cspades Loading…
3 of 6 tasks
Core 0.16
@shjwudp @cspades
Add initial support for Kimi Delta Attention (Feature Request #2446) community-request
#2573 opened Dec 5, 2025 by CodersAcademy006 Loading…
1
14
draft: partial cudagraph scopes and improvements for training
#2572 opened Dec 5, 2025 by jiemingz Loading…
6 tasks
@jiemingz
1
Fix world size mismatch causing distributed init deadlock (Issue #2458) community-request
#2571 opened Dec 5, 2025 by CodersAcademy006 Loading…
1
3
Fix: Ensure token IDs respect vocab_size in dataset, embeddings, and … community-request
#2570 opened Dec 5, 2025 by CodersAcademy006 Loading…
1
6
[Dev] Improve MoE Logging
#2569 opened Dec 5, 2025 by yanring • Draft
6 tasks
Core 0.16
2
[Dev] Optimize TE CUDA Graph _get_sample_arguments() Time
#2568 opened Dec 5, 2025 by buptzyb Loading…
6 tasks
Core 0.16
@buptzyb
2
Add offset method for slow tokenizer community-request
#2567 opened Dec 5, 2025 by cael-ling Loading…
6 tasks
1
Support TP greater than num_kv_heads by supporting QKV activation sub-sharding
#2565 opened Dec 5, 2025 by deepakn94 • Queued Core 0.16
3
feat: Api compat add decorator dev
#2545 opened Dec 4, 2025 by pablo-garay Loading…
6 tasks
1
Use autodoc2 and remove automodule
#2542 opened Dec 4, 2025 by Phlip79 • Draft
6 tasks
1
feat: m4 leftover changes
#2506 opened Dec 4, 2025 by yaoyu-33 Loading…
6 tasks
4
[Draft] Modaltiy bridge for Modality-Decoupled Parallelism(MDP)
#2500 opened Dec 4, 2025 by shifangx Loading…
6 tasks
2
fix: Assign tokenizer to Encoder.tokenizer in legacy mode community-request
#2498 opened Dec 4, 2025 by iuyo5678 Loading…
1 of 6 tasks
Core 0.16
3
Log softmax decomposition
#2497 opened Dec 4, 2025 by shanmugamr1992 Loading…
6 tasks
1
remove fp16 assert in moe_grouped_gemm & EP
#2495 opened Dec 4, 2025 by HaochenYuan Loading…
6 tasks
Core 0.16
[Dev] remove fp16 assert in moe_grouped_gemm & EP
#2494 opened Dec 4, 2025 by HaochenYuan Loading…
6 tasks
Core 0.16
3
Reopen qwen3next functional test in lightweight mode
#2493 opened Dec 4, 2025 by yuzhongw-nvidia Loading…
6 tasks
Core 0.16
2
Refactoring of PP implementation for hybrid models to support virtual pipeline parallelism
#2492 opened Dec 4, 2025 by deepakn94 Loading… Core 0.16
1
ProTip! no:milestone will show everything without a milestone.