Great model
This is a great model. I've been using various Miqu models which are nice, but they have a lot of -ism's that have gotten annoying. This one though has that nice balance I've been looking for. Only thing that I notice with this model, and pretty much any Mixtral model in my experience, is that is will try to wrap up a story early. Adding another scenario helps put it back on track though.
Hey @Gesard , sincerely appreciate your feedback and giving my model a try!
Also noticed that the word choice seems to avoid the typical catchphrases and cliches we see thrown around a lot, glad that has been your experience as well. Wanted to take a chance on Mixtral 8x7b for this one since I didn't see a whole lot of experiments using model stock for them.
As for the issue about the premature wrap-up, I've experienced that with Mixtrals too. I found it can be usually be mitigated with some preemptive steps. If you're already using the Advanced context + instruct prompts I can also suggest including a brief author's note at depth 0. Something like [Ensure you stick to the System Prompt above all else] worked pretty well in some quick tests up to 10-12K context. Hope this helps, and thanks again for reaching out.
Thanks for the suggestions, I'll try it out.