What settings to use?
#1
by
Alendar
- opened
hlep me
Hi! This merge seems slightly worse than Lunaris itself to me, so I don't play with it myself. For tests, I used the same settings I use for Lunaris: LLama 3 instruct suffixes/prefixes and temp 1.25, smoothing factor 1.25, rep pen 1.05 (1024 tokens range), and minP 0.075 as sampler params. It's not like those settings are always optimal, though. You might want to decrease the temp and/or increase the smoothing if the output seems chaotic. Or the other way around for more diverse responses.