google / google/praxis

[Feature Request] Need Matmul Attention layer instead of Einsum to support GPU running

Open
#46 0 comments 0 reactions 0 assignees View on GitHub
Dominant language
Python
Stars
200
Forks
45
PR merge metrics
No merged PRs in 30d

Description

Einsum kernel couldn't' be lowered to cudnn GEMM. The computing performance is seriously affected.
Can you believe it? JAX(Flax or Praxis) attention layers are even slower than Tensorflow version (not Keras)!

Contributor guide

Open the contributing guide

Assessment

This issue has not been assessed yet.

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.