Skip to content

Pull requests: NVIDIA/TransformerEngine

New pull request New
205 Open 2,554 Closed
205 Open 2,554 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

Treat total_num_pages as a physical KV cache budget, not the page-table size community-contribution PRs from external contributor outside the core maintainers, representing community-driven work.
#3557 opened Sep 21, 2026 by 0z5a Loading…
2
Add an opt-in batch-invariant BF16 GEMM community-contribution PRs from external contributor outside the core maintainers, representing community-driven work.
#3556 opened Sep 21, 2026 by 0z5a Loading…
4
Add an opt-in float32 reduction dtype for the row-parallel all-reduce community-contribution PRs from external contributor outside the core maintainers, representing community-driven work.
#3555 opened Sep 21, 2026 by 0z5a Loading…
5
Reuse the host split sizes GroupedLinear's forward already computed community-contribution PRs from external contributor outside the core maintainers, representing community-driven work.
#3554 opened Sep 21, 2026 by 0z5a Loading…
1
Isolate FlashAttention 2/3 imports so a broken install cannot break TE community-contribution PRs from external contributor outside the core maintainers, representing community-driven work.
#3553 opened Sep 21, 2026 by 0z5a Loading…
1
2
Accumulate delayed wgrads into shared (tied) parameters community-contribution PRs from external contributor outside the core maintainers, representing community-driven work.
#3552 opened Sep 21, 2026 by 0z5a Loading…
1
3
Support learnable softmax (attention sink) with cp_comm_type="p2p" community-contribution PRs from external contributor outside the core maintainers, representing community-driven work.
#3551 opened Sep 20, 2026 by 0z5a Loading…
5
Fix FusedSGD zero_grad argument forwarding community-contribution PRs from external contributor outside the core maintainers, representing community-driven work.
#3549 opened Sep 19, 2026 by wujingyue Contributor Loading…
4 tasks done
2
KDA attention 2.20
#3548 opened Sep 18, 2026 by ksivaman Member Loading…
8 of 13 tasks
3
[PyTorch] Compile Linear and Bias operations with forward fusion
#3547 opened Sep 18, 2026 by pggPL Collaborator • Draft
6 of 9 tasks
[PyTorch] Enable LayerNormLinear UB AG overlap under no_grad community-contribution PRs from external contributor outside the core maintainers, representing community-driven work.
#3546 opened Sep 18, 2026 by ravimajeti Loading…
7 of 8 tasks
1
1
[PyTorch] Keep DistributedWeight objects on ctx across saved-tensor hooks community-contribution PRs from external contributor outside the core maintainers, representing community-driven work.
#3545 opened Sep 18, 2026 by xrennvidia Collaborator Loading…
1 of 13 tasks
2
feat: vmm slot for local cuda graph offload community-contribution PRs from external contributor outside the core maintainers, representing community-driven work.
#3544 opened Sep 18, 2026 by Xdydy Loading…
1 of 13 tasks
6
Use PyTorch CUTLASS for BF16 grouped GEMM on SM100 community-contribution PRs from external contributor outside the core maintainers, representing community-driven work.
#3542 opened Sep 18, 2026 by wujingyue Contributor Loading…
5 of 7 tasks
9
[PyTorch][CP] Support compact metadata with FA4
#3540 opened Sep 17, 2026 by sudhakarsingh27 Member Loading…
13 tasks
3
fix: add missing NVTE_BHSD labels to fused_attn to_string community-contribution PRs from external contributor outside the core maintainers, representing community-driven work.
#3538 opened Sep 17, 2026 by andrewwhitecdw Contributor Loading…
1
Prototype of green context + VMM localization community-contribution PRs from external contributor outside the core maintainers, representing community-driven work.
#3537 opened Sep 17, 2026 by WanZzzzzz Contributor • Draft
13 tasks
test(attention): anchor the context-parallel suite to an independent reference community-contribution PRs from external contributor outside the core maintainers, representing community-driven work.
#3536 opened Sep 17, 2026 by nvegesna-netizen Contributor Loading…
2
[Docs] Refresh README
#3535 opened Sep 17, 2026 by sbhavani Collaborator Loading…
5 of 13 tasks
1
[PyTorch] [torch.compile] torch.compile support for LayerNormLinear and LayerNormMLP
#3534 opened Sep 17, 2026 by pggPL Collaborator Loading…
2
[PyTorch] Retain NVFP4 RNG tensors through quantization community-contribution PRs from external contributor outside the core maintainers, representing community-driven work.
#3533 opened Sep 17, 2026 by Connor-XY Loading…
1
fix(attention): stop handing FlashAttention 4 the -1 window sentinel community-contribution PRs from external contributor outside the core maintainers, representing community-driven work.
#3532 opened Sep 17, 2026 by nvegesna-netizen Contributor Loading…
1
4
[PyTorch] Extend no-load-balance CP to the a2a comm type community-contribution PRs from external contributor outside the core maintainers, representing community-driven work.
#3530 opened Sep 16, 2026 by Rudin6 Loading…
13 tasks
@sudhakarsingh27
2
feat(attention): cuDNN FROST attention backend for head_dim in (256, 512], with context parallelism 2.21.0 community-contribution PRs from external contributor outside the core maintainers, representing community-driven work.
#3527 opened Sep 16, 2026 by nvegesna-netizen Contributor Loading…
22
[Pytorch][NCCL EP] Create tokens_per_expert on pinned CPU memory when running eager
#3526 opened Sep 16, 2026 by YangFei1990 Collaborator Loading…
8 of 13 tasks
4
ProTip! Updated in the last three days: updated:>2026-09-18.