LLM Settings
#1
by
webboty
- opened
Hi, thanks for the work.
What temperatur, top-p and top-k are recommended for this model?
I did not try anything other than defaults.
As recently reported, this quant might have issues with tool calls, so use it for chat only and should be fine--it was optimized for chat interaction. I will post in a bit a new quant of the VCoder with just 8-hi, that seems to be cleaner, and the size does not change by much
Temp=1 seems to be most efficient, from what I heard