Pretty good for RP
This is excellent model for Roleplay if you do not need extreme violence or ERP. All abliterated models I tried in the past were bad (losing intelligence, creativity, personality), but this one works and seems to keep the intelligence and personality of L3.1. For me currently the best local model for RP (better than L3.1 finetunes I tried so far).
It still has some positive bias but can play evil characters. It is hesitant in extreme cases, but that I think is natural. And with heavy stirring/prompting you can get it there too.
It does not rush but will advance plot/scenes on its own (very rarely gets stuck but then you can easily advance yourself and it will go on from there).
It is suitable for a story to run for a long time. With some user help even consistent stories with hundreds of messages.
It can keep track of several characters - eg three main characters (+user) is usually no problem. With some help (summarizing, author notes, subtle hints in replies) it can keep track of more.
It does not need many rerolls/edits. But it is still LLM so it needs a bit of help from you. Eg. most LLM's want to go fetch or research some ancient artifacts all the time. So when it is crucial moment where new quest/project is decided, it is good place to reroll until you get something sensible/interesting/different.
Occasionally it produces reply that is too inconsistent/illogical so that is good place to reroll or edit too.
It is not best if you want something especially dark and gloom with no hope at all. If things go very badly for you this model tries to help somehow.
For sampler I only use Minp: 0.02 and DRY default 0.8/2/1.75. Everything else neutral/disabled.
I run it locally, tried GGUF from IQ4_XS to IQ2_XS. IQ3_S and above are great. IQ3_XXS still pretty good. IQ2_M and IQ2_XS are usable but there is visible drop in performance.
Hey thanks for your feedback, I'm adding a link to your post in the model card :)