Ministral 14b
#12
by
ElvisM
- opened
I would really like to know how you would tackle this model. Based on my experience, it’s very creative. It likes to write very long replies and often draws parallels and thinks outside of the box. It often makes me wonder whether it has completely gone off the rails before pivoting back and showing me that it hasn’t, that it was actually being very clever and laying out a good explanation for why something is the way it is.
The problem, though, is that I feel my settings make it a little unreliable. What templates and sampling parameters should I be using with this model?