Issues · NVIDIA/TransformerEngine · GitHub

Labels Milestones New issue

Expose Batch Invariant Kernels

#2986

· wdykas opened

on May 13, 2026

TE MoE TEOpsSequential with MXFP8 AttributeError: 'MXFP8TensorStorage' object has no attribute 'reshape'

#2973

· faradawn opened

on May 10, 2026

TE EP MXFP8 fails with last_dim % MXFP8_BLOCK_SIZE == 0

#2966

· faradawn opened

on May 7, 2026

Expose max_logit (and softmax aux statistics) from JAX fused attention higher-level APIs

#2945

· zeryx opened

on Apr 30, 2026

[BUG] multi_tensor_apply: int32 overflow in TensorListMetadata::sizes causes illegal memory access for tensors with numel > INT_MAX

#2918

· yezhengmao1 opened

on Apr 23, 2026

TransformerLayer implementation is overly complex and difficult to follow

#2903

· Hukongtao opened

on Apr 20, 2026

ub_overlap_ag_fprop disabled during activation recomputation due to is_grad_enabled check

#2902

· SophusDavid opened

on Apr 20, 2026

[BUG] `CrossEntropyFunction.forward()` modifies input in-place without `ctx.mark_dirty()`, causing GPU memory leak

#2899

· shen-jiabin opened

on Apr 17, 2026

[PyTorch] Reduce CPU overhead in grouped MLP block

#2897

· timmoon10 opened

on Apr 16, 2026

transformer-engine-cu12 PyPI wheels have incorrect WHEEL tag (cp310 hardcoded), causing reinstall loops

#2896

· leoleoasd opened

on Apr 16, 2026

[JAX] Create initial MoE Block

#2895

· tdophung opened

on Apr 16, 2026

[BUG] SIGSEGV in fused_attn_fwd when cu_seqlens_q != cu_seqlens_q_padded with FP8 blockwise

#2892

· NoonePauseferg opened

on Apr 16, 2026