Excellent model
I've been playing with the 3.35 on dual 3090's and am feeling very spoiled by these models. The writing style is excellent, really good with descriptions and prose. Playing with "Yes, My Liege" the conversations feel very much in the genre. Complex political situations requiring a cool head and intelligent response. The characters are on point in how they act towards a "liege". I'm also running the Lorebook and the 12k context here really shines. When asking for character descriptions I have had the model ignore some odd race info on the character card/start text for the extras, like switching a fox-folk Domestic Advisor to an elf. But otherwise the character descriptions were very good and matched their roles well. Instruction following is also good. If I'm doing an image generation prompt and ask for specific details like "period appropriate clothing and weapons" it follows instructions. I haven't really don't long conversations so I can't tell how good the model is with remembering prior items, especially in the longer context. I will also say the model isn't too aggressive, especially in this playbook. I feel like that's part of the model being good at following instructions.
I haven't really messed with the other settings. Pretty much the defaults. The model just seems solid overall without much tweaking needed, aside from model settings/drivers needed to shove as much context size as possible into ram.
Hey, thanks for the input and I'm glad the model is working well for you! Have you ever used just Euryale 1.3? If so, how does this one compare, considering that it contains Euryale 1.3?
I'm just getting back into the scene after several months, where the prior best general coherent roleplay models would've been maybe Vicuna 13B 1.1, Alpaca 30B or even stock LLaMA 65B v1 when that was able to be ran on dual 3090's at a decent speed(exllama v1). But from the perspective of being new to the recent models, Venus pretty much "just works" really well without any fiddling.