Skip to content

Pull requests: NVIDIA/TransformerEngine

New pull request New
147 Open 2,308 Closed
147 Open 2,308 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

[Common] Experimental CuTeDSL MXFP4 backend community-contribution PRs from external contributor outside the core maintainers, representing community-driven work.
#3223 opened Jul 20, 2026 by janekb04 Collaborator • Draft
1 of 13 tasks
[Common] Migrate NCCL EP submodule to NVIDIA/nccl-extensions
#3222 opened Jul 20, 2026 by phu0ngng Collaborator Loading…
6 of 13 tasks
1
[PyTorch] Add experimental packed-contiguous THD all-gather CP
#3221 opened Jul 18, 2026 by sudhakarsingh27 Member • Draft
Fix SM120 FP8 output ordering before DelayedScaling amax update community-contribution PRs from external contributor outside the core maintainers, representing community-driven work.
#3215 opened Jul 15, 2026 by AlbertYang514 Loading…
3
Explore error correction for pertoken recipe community-contribution PRs from external contributor outside the core maintainers, representing community-driven work.
#3214 opened Jul 15, 2026 by YigongQin Contributor • Draft
13 tasks
Fix duplicate FP8 state updates during activation checkpoint recomputation community-contribution PRs from external contributor outside the core maintainers, representing community-driven work.
#3213 opened Jul 15, 2026 by AlbertYang514 Loading…
7 of 13 tasks
4
[common] [NVFP4] Per-token (row-scaled transpose) cast + CUTLASS GEMM kernels (1/3, split from #3045) community-contribution PRs from external contributor outside the core maintainers, representing community-driven work.
#3206 opened Jul 13, 2026 by cael-ling Contributor Loading…
13 tasks
6
[PyTorch] Support padded THD sequences with FlashAttention 4 community-contribution PRs from external contributor outside the core maintainers, representing community-driven work.
#3205 opened Jul 13, 2026 by xiaoyao0115 • Draft
7 tasks done
[PyTorch][torch.compile] Support for UnfusedDotProductAttention
#3201 opened Jul 10, 2026 by pggPL Collaborator Loading…
7 of 13 tasks
2
[PyTorch][torch.compile] DotProductAttention: declarative packed QKV/KV inputs
#3200 opened Jul 10, 2026 by pggPL Collaborator Loading…
8 of 13 tasks
7
Developer documentation
#3198 opened Jul 9, 2026 by ptrendx Member • Draft
1 of 13 tasks
Add transformer_engine.test() installation smoke check community-contribution PRs from external contributor outside the core maintainers, representing community-driven work.
#3197 opened Jul 9, 2026 by CodersAcademy006 Loading…
1
5
docs: add datatype and hardware support matrix community-contribution PRs from external contributor outside the core maintainers, representing community-driven work.
#3195 opened Jul 9, 2026 by CodersAcademy006 Loading…
2
1
adam: separate param/grad dtypes in AdamCapturableFunctor community-contribution PRs from external contributor outside the core maintainers, representing community-driven work.
#3192 opened Jul 8, 2026 by rapatel Loading…
5 of 13 tasks
2
[PyTorch][torch.compile] Make get_attention_backend traceable without graph breaks
#3189 opened Jul 8, 2026 by pggPL Collaborator Loading…
8 of 13 tasks
5
NCCL EP zero copy additional support community-contribution PRs from external contributor outside the core maintainers, representing community-driven work.
#3187 opened Jul 7, 2026 by YangFei1990 Contributor Loading…
7 tasks done
4
Optimize single-group grouped MLP quantization community-contribution PRs from external contributor outside the core maintainers, representing community-driven work.
#3184 opened Jul 7, 2026 by sraman-rgb Contributor Loading…
13 tasks
7
Prefer Python CUDA libs over implicit system paths build Build system installation
#3182 opened Jul 7, 2026 by sbhavani Collaborator Loading…
5 of 12 tasks
1
4
grouped linear single param fixes community-contribution PRs from external contributor outside the core maintainers, representing community-driven work.
#3178 opened Jul 6, 2026 by CarlosGomes98 Contributor Loading…
3 of 13 tasks
17
Add Graph-safe NVFP4 CUTLASS Group GEMM to unified entry points community-contribution PRs from external contributor outside the core maintainers, representing community-driven work.
#3175 opened Jul 4, 2026 by cael-ling Contributor Loading…
8 tasks done
16
[PyTorch] Enable fused FP8 block-scaling path in GroupedLinear module 2.18 FP8 MoE
#3171 opened Jul 2, 2026 by denera Collaborator Loading…
8 of 13 tasks
1
@denera
18
Reverse MXFP8 quantization row raster community-contribution PRs from external contributor outside the core maintainers, representing community-driven work.
#3170 opened Jul 2, 2026 by sraman-rgb Contributor Loading…
13 tasks
6
Add fused multi-tensor kernel for 1D blockwise FP8 quantization community-contribution PRs from external contributor outside the core maintainers, representing community-driven work.
#3168 opened Jul 2, 2026 by shangxiaokang • Draft
13 tasks
3
Bump transformers from 4.57.0 to 5.3.0 in /docs/examples/te_llama community-contribution PRs from external contributor outside the core maintainers, representing community-driven work. dependencies Pull requests that update a dependency file python Pull requests that update python code
#3167 opened Jul 2, 2026 by dependabot Bot Loading…
@sudhakarsingh27
Bump transformers from 4.55.0 to 5.3.0 in /docs/examples/te_gemma community-contribution PRs from external contributor outside the core maintainers, representing community-driven work. dependencies Pull requests that update a dependency file python Pull requests that update python code
#3166 opened Jul 2, 2026 by dependabot Bot Loading…
@sudhakarsingh27
ProTip! Adding no:label will show everything without a label.