Skip to content

Pull requests: NVIDIA/TransformerEngine

New pull request New
128 Open 1,896 Closed
128 Open 1,896 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

Add MXFP8 attention
#2719 opened Mar 1, 2026 by cyanguwa • Draft
13 tasks
10
pass params_dtype to qk_norm creation
#2718 opened Feb 28, 2026 by pstjohn Loading…
4
Hongbinl/offload activation cuda graph mxfp8 offload fix
#2716 opened Feb 27, 2026 by lhb8125 • Draft
13 tasks
6
[PyTorch] Remove is_first_microbatch setting after cudagraph warmup
#2715 opened Feb 27, 2026 by buptzyb Loading…
13 tasks
4
[JAX] CGEMM with Shardy
#2714 opened Feb 27, 2026 by phu0ngng Loading…
8 of 13 tasks
7
Add DCP compatibility for FSDP2-TP sharding in TransformerEngine.
#2713 opened Feb 26, 2026 by cspades • Draft
13 tasks
Enable dequantization from MXFP8 tensor with only columnwise data
#2712 opened Feb 26, 2026 by ptrendx Loading…
13 tasks
6
[JAX] Support calling MOE router kernels from JAX side
#2711 opened Feb 26, 2026 by tdophung Loading…
1 of 13 tasks
23
[Common][PyTorch] Add z_loss_weight and log_sum_exp output to parallel_cross_entropy
#2707 opened Feb 26, 2026 by bassoy • Draft
8 tasks done
[Draft] Newton-Schulz via cuSOLVERMp
#2706 opened Feb 25, 2026 by vcherepanov-nv Loading…
6 of 13 tasks
17
[All] Added better error messages
#2705 opened Feb 25, 2026 by ptrendx Loading…
7
Fix Flash Attention 3 API compatibility for window size parameters 2.14.0
#2704 opened Feb 25, 2026 by jhvmhg Loading…
3 of 13 tasks
11
[JAX] Deprecate GSPMD: remove infer_sharding_from_operands and GSPMD tests
#2702 opened Feb 24, 2026 by phu0ngng Loading…
8 of 13 tasks
7
[Draft][PyTorch] torch.compile support for TE Linear
#2701 opened Feb 24, 2026 by pggPL • Draft
13 tasks
Add fused_adam, quantized_model_init, and fsdp2 example
#2698 opened Feb 22, 2026 by pstjohn Loading…
1
30
[PyTorch] Zero-initialize learnable softmax_offset in DotProductAttention
#2694 opened Feb 20, 2026 by fjosw Loading…
7 of 13 tasks
4
Enable sm120 support for fused attn if cuDNN is 9.18.1+
#2693 opened Feb 20, 2026 by KshitijLakhani • Draft
13 tasks
[JAX] Fix get_seqlens_and_offsets() to accept vmapped seg ids and non vmapped seg offsets 2.14.0
#2692 opened Feb 19, 2026 by KshitijLakhani Loading…
7 of 13 tasks
1
@KshitijLakhani
16
NVFP4 primary weight support
#2691 opened Feb 19, 2026 by WanZzzzzz Loading…
10 of 13 tasks
50
[PyTorch] Error out if constructing LayerNormLinear with row tensor parallelism bug Something isn't working
#2688 opened Feb 17, 2026 by timmoon10 Loading…
6 of 13 tasks
7
[PyTorch] torch.compile support for permutation functions
#2686 opened Feb 17, 2026 by pggPL Loading…
9 of 13 tasks
19
[JAX] Integrate BF16 Grouped GEMM with on-device group sizes
#2680 opened Feb 13, 2026 by jberchtold-nvidia Loading…
8 of 13 tasks
1
16
[PyTorch][Fused Attn] Add support for cuDNN to return Softmax Stats always and Max when return_max_logit=True
#2677 opened Feb 12, 2026 by sudhakarsingh27 Loading…
7 of 13 tasks
27
[PyTorch] Add dtype information to QuantizedTensorStorage class
#2676 opened Feb 12, 2026 by ptrendx Loading…
1 of 13 tasks
14
[Common] MOE Split dBias cpu_overhead enhancement New feature or request MoE
#2674 opened Feb 11, 2026 by Oleg-Goncharov Loading…
8 of 13 tasks
19
ProTip! Mix and match filters to narrow down what you’re looking for.