Skip to content

Pull requests: NVIDIA/TransformerEngine

New pull request New
184 Open 2,517 Closed
184 Open 2,517 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

[PyTorch] Use torch's register_custom_class API for opaque quantizers when available community-contribution PRs from external contributor outside the core maintainers, representing community-driven work.
#3495 opened Sep 8, 2026 by mmarcinkiewicz Contributor Loading…
2 of 13 tasks
@pggPL
2
[Docs] Add Mixture of Experts guide documentation Improvements or additions to documentation
#3494 opened Sep 7, 2026 by pggPL Collaborator Loading…
2
[PyTorch] [torch.compile] Prepare LayerNormLinear and LayerNormMLP for torch.compile
#3493 opened Sep 7, 2026 by pggPL Collaborator • Draft
6 of 8 tasks
[PyT] Disable FA3 for training when head_dim_qk != head_dim_v community-contribution PRs from external contributor outside the core maintainers, representing community-driven work.
#3490 opened Sep 7, 2026 by yuweih205 Loading…
6 tasks done
1
2
Add native transport for dynamic context parallelism community-contribution PRs from external contributor outside the core maintainers, representing community-driven work.
#3489 opened Sep 6, 2026 by xiaoyao0115 • Draft
Support row-only MXFP8 distributed master-weight casts community-contribution PRs from external contributor outside the core maintainers, representing community-driven work.
#3488 opened Sep 6, 2026 by xiuhu17 Contributor Loading…
5 tasks done
2
fix(jax): interpolate the tensor-sequence-parallelism warning community-contribution PRs from external contributor outside the core maintainers, representing community-driven work.
#3486 opened Sep 5, 2026 by Anai-Guo Loading…
1
[PyTorch] Fix NaN expert weight gradients at num_groups == 1 with SReLU community-contribution PRs from external contributor outside the core maintainers, representing community-driven work.
#3485 opened Sep 5, 2026 by GarlGuo Loading…
6 of 13 tasks
5
[Common] Fix Grouped MXFP8 work mapping and TMA synchronization
#3483 opened Sep 4, 2026 by Oleg-Goncharov Collaborator Loading…
3 of 13 tasks
3
[PyTorch] Split FusedAttnFunc into single-argument forward/backward helpers 2.20
#3480 opened Sep 4, 2026 by pggPL Collaborator Loading…
6 of 13 tasks
2
Drop the unsupported window_size argument from attention backend queries community-contribution PRs from external contributor outside the core maintainers, representing community-driven work.
#3479 opened Sep 4, 2026 by Anai-Guo Loading…
1
[PyT] Linear Attention API 2.20
#3477 opened Sep 4, 2026 by KshitijLakhani Collaborator • Draft
13 tasks
@KshitijLakhani
[PyTorch] Enable fused activation recompute for ScaledTanhSReLU community-contribution PRs from external contributor outside the core maintainers, representing community-driven work.
#3473 opened Sep 3, 2026 by wanyingw Contributor • Draft
13 tasks
5
[PyTorch] torch.compile support for FusedAttention
#3472 opened Sep 3, 2026 by pggPL Collaborator • Draft
7 of 13 tasks
[PyTorch] DeepSeekV3Layer: full MoE transformer layer (MLA + DeepSeek MoE)
#3471 opened Sep 3, 2026 by pggPL Collaborator • Draft
7 of 13 tasks
[All] Guard THD learnable dSink on older cuDNN attention bug Something isn't working
#3470 opened Sep 3, 2026 by KshitijLakhani Collaborator Loading…
2 of 13 tasks
1
@KshitijLakhani
6
[PyTorch] Fix FP8 illegal memory access in single-process multi-GPU execution community-contribution PRs from external contributor outside the core maintainers, representing community-driven work.
#3469 opened Sep 3, 2026 by SuperGoodGame Loading…
1
2
Automatically omit unused columnwise primary weights for backward overrides community-contribution PRs from external contributor outside the core maintainers, representing community-driven work.
#3468 opened Sep 3, 2026 by xiuhu17 Contributor Loading…
6
[Common] row-scaled nvfp4 path: add single-launch group fused amax community-contribution PRs from external contributor outside the core maintainers, representing community-driven work.
#3467 opened Sep 3, 2026 by cael-ling Contributor Loading…
2 of 13 tasks
3
[JAX] Make MoEBlock aware of Dense TP axes
#3462 opened Sep 2, 2026 by jberchtold-nvidia Collaborator • Draft
8 of 13 tasks
1
Add non-TMA MXFP8 cast-only kernels for specialized rowwise-only and row+colwise
#3459 opened Sep 1, 2026 by tdophung Collaborator • Draft
13 tasks
1
[Common] Group NVFP4 Quantize Kernels
#3458 opened Sep 1, 2026 by Oleg-Goncharov Collaborator Loading…
9 of 13 tasks
4
[JAX] Add SiTU-GLU support to JAX
#3457 opened Sep 1, 2026 by jberchtold-nvidia Collaborator • Draft
8 of 13 tasks
2
[PyTorch] Schedule delayed-scaling updates after backward
#3456 opened Sep 1, 2026 by pggPL Collaborator Loading…
9 of 13 tasks
1
[Common] row-scaled nvfp4 path: fuse row/col amax into a single TMA-tiled kernel community-contribution PRs from external contributor outside the core maintainers, representing community-driven work.
#3454 opened Sep 1, 2026 by cael-ling Contributor Loading…
2 of 13 tasks
2
ProTip! Find all pull requests that aren't related to any open issues with -linked:issue.