Issues · NVIDIA/TransformerEngine · GitHub

Labels Milestones

Multi-GPU FP8 (accelerate.dispatch_model + DelayedScaling HYBRID) — illegal memory access on H200 sm_90

#3124

· roir opened

on Jun 12, 2026

Vendor-neutral reference: NVFP4 vs OCP MXFP4 block-structure parameters (request for confirmation)

#3105

· gHashTag opened

on Jun 8, 2026

`test_comm_gemm_overlap.py::test_multi_layer_with_overlap_bf16` fails on A100

#3097

· francesco-bertolotti opened

on Jun 5, 2026

Low-overhead stateless functional quantize/GEMM API (plain tensors in → out)

#3087

· cael-ling opened

on Jun 5, 2026

Fused attention should distinguish bias input from dBias request

#3082

· KshitijLakhani opened

on Jun 4, 2026

NVFP4 default recipe (RHT + stochastic rounding) crashes on sm_120: fused kernel exceeds 101376-byte shared-memory opt-in cap

#3062

· Infatoshi opened

on May 29, 2026

[Common] Move heavy kernel compilation to NVRTC

#3054

· CarlosGomes98 opened

on May 28, 2026

[JAX] score_mod attention to support "enforce_precompiled" flag in cuDNN graph deserialization

#3046

· jberchtold-nvidia opened

on May 26, 2026

[PyTorch] Integrate cuDNN GQA + DSA backend into DotProductAttention

#3028

· nvMelissa opened

on May 21, 2026

·

[Common + PyTorch] Support Dispatch FP8

#3026

· phu0ngng opened

on May 21, 2026

[JAX] Dispatch/Combine for BF16 Tensor

#3025

· phu0ngng opened

on May 21, 2026

[PyTorch] Dispatch/Combine for BF16 tensor + Zero-Copy

#3024

· phu0ngng opened

on May 21, 2026