Q4_K_S versus Q3_K_XL
#3
by
segmond
- opened
This is more a generic question, but does anyone know which is better? I haven't seen data that compares both. I have been running the prior kimi's in Q3_K_XL but I think I might be able to squeeze in Q4_K_S. Trying to figure out how kl divergence compares between both.
Bigger is usually always better. It really depends on your hardware limit. In this case Q4 is better
I am able to load the Q3XL, and sometimes it has given me python code that does not compile because of syntax errors ( like in a if it skiped using the " == " for checking equality). so if I could load a Q4 I would use that.