>>105878133 (OP)Fallen Command A (111 billion parameters I think). Install SillyTavern + Oobabooga (or LM-Studio if you're an Easy! Button brainlet), download Fallen Command A from HuggingFace (refer to DontPlanToEnd's UGI Leaderboard on HuggingFace for a good table of models and test scores).
Without any system prompts, Fallen Command A v1 comes right out the gate with sassy insults and backhanded comments, and is a very intelligent model.
You need a Raptor Lake or newer and 128-256GB RAM, or a Mac Studio 128-256GB Unified Memory if you want GPU-like performance on the cheap (relative to an nvidia multicard setup, which is a third option I think the Mac supercedes for inference purposes).