Feedback.
Feedback thread.
Q5_K_S-imat
Great stuff. This model really is unhinged. Honestly it's one of the best models at just generating the wildest stuff without anything moralizing or telling you it's wrong. She just does it all with a simple compliant prompt without holding back anytime.
Does well with formatting in my Alpaca extended use, formats greatly actually, for example characters with Stats/very specific reply structures, writes well enough too, just needs some salting with RepPen and a bit of Temperature.
So far so good. Will need to take her for a longer spin to check coherence more.
Based/10
Love to hear that you have been having good results with it so far!
Is she going on any memerboard or are you concerned it might not perform too well? Just get a point of comparison.
@Lewdiculous Could you please share your settings with temp and reppen? I have some difficulties with it
@woofwolfy These:
But, added RepPen 1.15, RepRange 1024, Temperature 1.15, MinP 0.075*. Just play with Temp/RepPen until you're satisfied.
@Lewdiculous I liked this model but I didn't like that it was repeated in the answers and was a bit strange because of that, but now everything is better. Thank you!
@Nitral-AI She's got good initial feedback it seems, at least she's cute, and funny. That's always good to see. She's surprisingly good, the only "issue" I see is proneness for similar outputs that can be overcome with more temperature and RepPen as needed. It's a 7B, swiping is fast anyways.
Appreciate it as usual my dude! Been trying to keep the unhinged habits while pumping it with 3.05 smarts but im failing atm.
@woofwolfy I meant to type MinP 0.075 xD
Oof
@Lewdiculous Yeah, I already thought of that, haha. I would also like to test on exllamav2 if will be possible
EXL2 👀 🤢
JK 🤗