The performance is lower than the base model?
#4
by
Spico
- opened
It seems the model does not surpasses its base model (mlabonne/Marcoro14-7B-slerp).
Am I missing something?
You're right, you can also see it here: https://huggingface.co/spaces/mlabonne/Yet_Another_LLM_Leaderboard
Every expert in Beyonder ranks significantly lower than the base model. In addition, the RP and code models probably decrease its performance on this benchmark since there's no code or storytelling involved.