Skip to content

Pull requests: NVIDIA/TransformerEngine

New pull request New
161 Open 2,393 Closed
161 Open 2,393 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

mxfp8: add swizzled-scale fast path for cast-only quantization community-contribution PRs from external contributor outside the core maintainers, representing community-driven work.
#3338 opened Aug 10, 2026 by WanZzzzzz Contributor Loading…
13 tasks
3
[common] Improved performance of Group MXFP8 kernels
#3337 opened Aug 10, 2026 by Oleg-Goncharov Collaborator Loading…
6 of 13 tasks
1
[PyTorch] Fine-grained recipe docs documentation Improvements or additions to documentation
#3336 opened Aug 10, 2026 by negvet Collaborator Loading…
13 tasks
2
[PyTorch] Add torch_stable compatibility layer for incremental migration to the torch stable ABI
#3335 opened Aug 10, 2026 by pggPL Collaborator • Draft
7 of 13 tasks
Add a backwards linear function to be used with the fused mla q up-proj community-contribution PRs from external contributor outside the core maintainers, representing community-driven work.
#3330 opened Aug 7, 2026 by chaseblock Contributor Loading…
13 tasks
3
[PyTorch] Fix deferred initialization in fusible ops
#3327 opened Aug 7, 2026 by denera Collaborator Loading…
8 of 13 tasks
1
@denera
9
Refactor GroupedLinear quantization dispatch
#3326 opened Aug 7, 2026 by negvet Collaborator Loading…
13 tasks
10
Prototype NVFP4 with FP8 UE5M3 block scales enhancement New feature or request
#3325 opened Aug 7, 2026 by timmoon10 Member • Draft
5 of 13 tasks
1
[PyTorch] Enable NVFP4 row-scaled (per-token) backward for GroupedLinear community-contribution PRs from external contributor outside the core maintainers, representing community-driven work.
#3324 opened Aug 7, 2026 by cael-ling Contributor Loading…
1 of 13 tasks
1
[Pytorch] Enable TE Op to consume extra_outputs from a previously run Op in TE Sequential
#3320 opened Aug 5, 2026 by vthumbe1503 Collaborator Loading…
13 tasks
12
[PyTorch] Advance FusedAdam step counter for empty param groups community-contribution PRs from external contributor outside the core maintainers, representing community-driven work.
#3318 opened Aug 5, 2026 by adityasingh2400 Loading…
1
@ptrendx
5
[Common][PyTorch] Fuse the RHT into grouped NVFP4 quantize on non-SM100 architectures community-contribution PRs from external contributor outside the core maintainers, representing community-driven work.
#3317 opened Aug 5, 2026 by davidkny22 Contributor Loading…
6 of 13 tasks
3
[Common/PyTorch] Grouped weighted-SwiGLU MXFP8 kernel community-contribution PRs from external contributor outside the core maintainers, representing community-driven work.
#3315 opened Aug 4, 2026 by cael-ling Contributor Loading…
3 of 13 tasks
12
Log when thd with dropout falls to the composite cuDNN engine community-contribution PRs from external contributor outside the core maintainers, representing community-driven work.
#3313 opened Aug 4, 2026 by bzantium Loading…
@KshitijLakhani
8
[CI] Publish GB200 aarch64 wheel artifacts community-contribution PRs from external contributor outside the core maintainers, representing community-driven work.
#3311 opened Aug 4, 2026 by bvolpato Loading…
5 of 13 tasks
3
nvrtc MXFP8 kernels community-contribution PRs from external contributor outside the core maintainers, representing community-driven work.
#3302 opened Aug 3, 2026 by CarlosGomes98 Contributor • Draft
13 tasks
NVRTC NVFP4 quantization kernels community-contribution PRs from external contributor outside the core maintainers, representing community-driven work.
#3301 opened Aug 3, 2026 by CarlosGomes98 Contributor Loading…
8 of 13 tasks
18
Add NVFP4 RHT Support for SM120 and SM121 community-contribution PRs from external contributor outside the core maintainers, representing community-driven work.
#3300 opened Aug 3, 2026 by new-TonyWang Loading…
12
[PyTorch] Scope the quantized-param caching flag to its own graph capture community-contribution PRs from external contributor outside the core maintainers, representing community-driven work.
#3298 opened Aug 1, 2026 by xiuhu17 Contributor Loading…
6 tasks done
1
[Pytorch][Common] Option to Disable 2nd level Scale in NVFP4
#3297 opened Jul 31, 2026 by vthumbe1503 Collaborator Loading…
13 tasks
3
[PTQ] Store FP32 global scaling factors (absmax or scale_inv) for all quantized activations and weights. org-contribution
#3296 opened Jul 31, 2026 by cspades Member Loading…
7 of 13 tasks
5
[Common] Fix pointer arithmatic to generate correct LDS/STS instructions
#3291 opened Jul 30, 2026 by kainzhong Collaborator Loading…
8 of 13 tasks
10
Fix cuDNN SDPA score_mod cache collision, sm_12x arch gates, per-device plan cache community-contribution PRs from external contributor outside the core maintainers, representing community-driven work.
#3289 opened Jul 30, 2026 by YangXu1990uiuc Loading…
1
[PyTorch][torch.compile] Support for DotProductAttention on flash and unfused backends
#3286 opened Jul 30, 2026 by pggPL Collaborator Loading…
9 tasks done
7
[JAX] Allow input tensors to be sharded with TP in EP dispatch/combine ops
#3283 opened Jul 30, 2026 by phu0ngng Collaborator • Draft
13 tasks
ProTip! Adding no:label will show everything without a label.