GGUF (ollama) version are far from API version
#20
by
papipsycho
- opened
Hello Guys,
I'm testing the GGUF version with ollama, but for now i'm getting result far from API version,
i try to tweak different kind setting such like temperature, top_p, i also try different version of quantize 5 to 8, but i'm not able to get the quality of the API version
do you have any idea what should i change to increase the quality ?
It's just spewing nonsense after some tokens.