jquesnelle / yarn

YaRN: Efficient Context Window Extension of Large Language Models
MIT License
1.25k stars 110 forks source link

A hardcore-mode multiple passkey evaluation #30

Closed honglu2875 closed 9 months ago

honglu2875 commented 9 months ago

It is much harder than the original passkey in two ways:

The performance is quite interesting rather than ~100% everything for our paper. But I know this might be out-of-scope for our paper. But in any case I'm leaving this single-file script here in case we want to do anything with it.