Best I've seen so far
#2
by
Utochi
- opened
So my system can only handle the Q2_K version of this. using Faraday at 32k max tokens with a complex character card of over 2700 tokens, this one is intelligent and obedient. i do feel a little bit of subtlety and i have to encourage it to be creative but this still blows every other model I've been able to use out of the water which is most pleasing to me.
For now, this is the top of the line for sub 25 GB large models that i can find.