flash-attention icon indicating copy to clipboard operation
flash-attention copied to clipboard

Published 20 hours ago • verified publisher icon Dao-AILab

install the flash-attention

Open SISTMrL opened this issue 1 year ago • 3 comments

hello, i run the pip install flash-attn, but it fails. the gpu i used is tesla m40, can the flash attn be installed on m40?

SISTMrL avatar Aug 01 '23 02:08 SISTMrL

No, m40 is of Maxwell generation (sm_52). FlashAttention right now supports Ampere (sm_80) and after.

tridao avatar Aug 01 '23 03:08 tridao

@tridao

Can you explain why it does not support maxwell? Are there any technical limitations?

twmht avatar Apr 20 '24 02:04 twmht

@tridao

Can you explain why it does not support maxwell? Are there any technical limitations?

Because it takes work. Maxwell does not have tensor cores. Someone would need to implement a version that doesn't use tensor cores.

tridao avatar Apr 20 '24 03:04 tridao