TLDR AI · ENGINEERING & RESOURCES

CUSTOMIZED FLASH ATTENTION

2024-11-27github.comGitHub repository

Summary from the TLDR AI newsletter

New flash attention fork that can have ragged Q/V matrix sizes.

Read the original article ↗View the archived TLDR AI issue