Realtime Research Walkthrough: Parenthesis Balancing in 1L Toy Language Model (Part 2)
Автор: Neel Nanda
Загружено: 2023-08-22
Просмотров: 815
A research walkthrough of reverse-engineering how a 1L language model (trained on code + internet text) can balance parentheses, and trying to understand its neurons. Done for my MATS Scholars, but hopefully of broader interest!
Part 1: • Realtime Research Walkthrough: Parenthesis...
Доступные форматы для скачивания:
Скачать видео mp4
-
Информация по загрузке: