Which ones? I think fine-tunes are where I see most of this (I'll just call it) "model spam", but the base models don't seem to exhibit this behavior. I do see

Which ones? I think fine-tunes are where I see most of this (I'll just call it) "model spam", but the base models don't seem to exhibit this behavior. I do see some models perform way below the curve compared to their benchmark performance, though (Phi family being the most famous).