A Adept Blog company blog

FlashAttention: Fast Transformer training with long sequences

adept.ai 6 min read

‍Transformers have grown deeper and wider, but training them on long sequences remains difficult. The attention layer at their heart is the compute and memory bottleneck: doubling the sequence length would quadruple the runtime and memory requirements.

Points

148

HN comments

10

Submissions

1

Cited by

0 commenters

AI & Machine LearningSystems & Performance2023

Related posts from the same era

Bing: “I will not harm you unless you harm me first”

Simon Willison20233,363 points1,673 commentscited by 6 details

GPT-4o

OpenAI Blog20243,138 points2,366 commentscited by 16 details

Open source AI is the path forward

Meta Newsroom20242,360 points887 commentscited by 5 details

Building a Virtual Machine Inside ChatGPT

Jonas Degrave20222,029 points908 commentscited by 16 details

The man who killed Google Search?

Ed Zitron20241,884 points878 commentscited by 47 details

ChatGPT Plugins

OpenAI Blog20231,875 points1,106 commentscited by 14 details