Public
  1. Public

onnxruntime

Public

Commits

Tianlei WuGitHub
7c05f7bab10
Tianlei Wu authored and GitHub committed 742edec5e88
[CUDA] Add PackedMultiHeadAttention operator (#16779)

### Description
Add new operator for MultiHeadAttention with inputs removed padding.
This only supports packed QKV format.