CodeLlama 70B was released yesterday, and we found the chat/instruct version to be the best scoring model in HumanEval (not counting models evaluated externally). But how does the prompt format work? This is always painful, so let's take a look together. ⤵️
— Pedro Cuenca (@pcuenq) Jan 30, 2024
from Twitter https://twitter.com/pcuenq
January 30, 2024 at 10:42AM
via IFTTT
No comments:
Post a Comment