How is it working?
#1
by
jackboot
- opened
Compared to original instruct. I am disappointed badly with most tunes for the base.
Also having the best time on instruct with the dreaded chat-ml and good system prompt.
Am a bit scared to download another 30gb.
Seems to work okay with the LimaRP prompt format and temp 1.5/min-p 0.05 so far - didn't get seem to trigger any looping with that. Haven't spent much time testing it though, mostly working on speculative decoding models.