News for hackers. Stuff that matters.

A comment on Attention Sinks in LLMs for endless fluency.

Re:Attention Sinks in LLMs for endless fluency by bpiche on Monday October 09, 2023 @01:16PM (#37822718)
Appears that this is an open source implementation of the same "Efficient Streaming Language Models with Attention Sinks" paper from MIT, linked here 7 days ago. Published on Sept 29, 2023.

https://news.ycombinator.com/item?id=37740932

[ Parent | on HN ]
Re:Attention Sinks in LLMs for endless fluency by cubie on Monday October 09, 2023 @04:54PM (#37825374)
That is exactly correct
[ Parent | on HN ]