handstand tag - reasoning
#2
by SizyKing - opened
handstand
It seems like it doesn't understand some tags at all. Like the handstand tag, which works perfectly on 0.6. But not on 4B. Furthermore, it doesn't encode in other languages. It would seem the advantage of such a clip is precisely that it could be multilingual.
I'm not being judgmental, just my thoughts. It seems to me that, besides everything else, the model is operating in reasoning mode, which ruins everything. Because it takes a very long time to generate the embedding, even compared to other 4B models, as if it's reasoning.