Testing it for a while - quite a good model
#6
by
cmp-nct
- opened
I tested many of the latest models, mostly up to 32B.
This one here is quite good, smart responses mostly, good knowledge.
The only issue I found was with longer generation, for testing purposes I tried to let it write "a book", just to see how it does.
Despite many attempts, it never managed to really build a larger chapter and after 15k context length it started to get repetitive.
It did not get into full loops but it also did not make much sense anymore.
Overall a nice model