Giters

open-compass / MixtralKit

A toolkit for inference and evaluation of 'mixtral-8x7b-32kseqlen' from Mistral AI

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

why the performance for the GSM8K and MATH lower than the original mixtral blog?

runzeer opened this issue · comments

commented

I noticed that the performance for the math reasoning is lower than the official blog. Is it due to the zero-shot setting compared the official 5-shot setting?