Extremely impressive
Below comments reference usage for largely SFW roleplay, with some question - answer usage mixed in.
I've used several of your models in the past but this one is a big step better then any others. Extremely impressive, exceptional instruct following capability mixed with good creativity. It consistently showed very good understanding of the character card. The dialog it produces is engaging and quite natural, I find myself doing far fewer regens then I'm used to. It's not naturally as verbose as most models but it picks up well on any mention that the character is talkative or something similar.
It does have a strange reluctance to take actions, strongly preferring to talk rather then act which is frustrating at times, but that's my only complaint. It's capable of it but needs a lot of prodding and remains abrupt in that area. One I'll happily overlook for the dialog it produces.
I've been testing it off and on for the last two weeks, and it's slowly been becoming the default across most of my character cards.
Good old Fimbulvetr V.2 has finally been dethroned and I've entirely replaced it with Honey Yuzu. A long overdue replacement given how long it's been my default in such a rapidly evolving space.
Comes at an awkward time with Nemo models slowly starting to proliferate and those seem like they have a ton of potential, but none of the Nemo derivatives thus far seem like their even as good as the launch Instruct model, filled with a lot of slop that accomplishes nothing more then dumbing down the original. I'm cautiously optimistic that future Nemo finetunes can be a real step forward over anything else.... but that'll take time, and for today Honey Yuzu is clearly the best LLM for RP usage out there imo.
Oh, wow!! Thank you so much for the encouragement and helpful feedback! I'm so glad you liked Honey Yuzu, and that it's working well for you!
Interesting observation about the actions. I'll keep an eye out for that in the future to hopefully improve my models. It sounds like you've already done a lot of prompt testing, but if you do find a fix for it, let me know! I'll even add it to the README and credit you!
Once again, thank you very much! Have a great day :D