Repository navigation

XNNPACK: delegate bf16 softmax, computing in fp32 - #24805

Merged

JakeStevens merged 2 commits into

pytorch:mainfrom

JakeStevens:xnnpack-softmax-bf16

Nov 9, 2026

Merged

XNNPACK: delegate bf16 softmax, computing in fp32#24805
JakeStevens merged 2 commits into
pytorch:mainfrom
JakeStevens:xnnpack-softmax-bf16

Commits on Nov 9, 2026
- XNNPACK: delegate bf16 softmax, computing in fp32
  JakeStevens
  committed
  3a83656
- Share the bf16 fp32-value and convert helpers between matmul and softmax
  JakeStevens
  committed
  61f0cab