stripe_flash_attn_varlen_func

zhuzilin / ring-flash-attention

Ring attention implementation with flash attention

MIT License

571 stars 45 forks source link

Closed leo6022 closed 6 months ago

leo6022 commented 6 months ago

@zhuzilin 请教一下楼主，为什么没有stripe_flash_attn_varlen_func的实现？

zhuzilin commented 6 months ago

hmm... because it seems to be really tricky to write that... and I'm afraid the performance (of the impl in my mind) won't be good...