openai / consistency_models

Official repo for consistency models.

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

implement a non-Cuda flash attention module

by321 opened this issue · comments

commented

The current flash attention module by Hazy Research is Cuda-only, so this limits this repo to Cuda-only too. I suggest writing a separate flash attention module for computers without Nvidia video card. This current module can still be used if Nvidia card is present.

A couple of people have raised this issue with Hazy Research, but they said they're focused on Cuda only, and are not interested in writing a non-Cuda version.

Added a pr #37 to avoid using flash Attention