Favorite Model
I have searched and searched and have yet to find a model I like more. Even in the 30-70B range. Most models are you extremely negative bias or cater to my every whim. This model has a fantastic balance and is great at following instructions. It handled my deep psychology cards well. It can be deep, but I long for something deeper. That is my only negative and why I continue to search.
Another 24B merge (and possibly 12B) using the scream method is being tested. Not sure how it will compare but you can try it.
I have searched and searched and have yet to find a model I like more. Even in the 30-70B range.
I started playing around with models back like 18-24 months ago. Many smaller models were pretty weak in roleplay and creative purposes, you had to hit the 70B or higher to start getting good outputs, but VRam is a big bottleneck. You did need to push the 30B in order to do decent roleplaying in my mind; the threshold/bar being a roleplay within a roleplay if it can keep thing separated or do both. Easiest example is D&D around a table, then becoming the character in the story, then backing out after a fight scene or something.
Recently the Gemma 4, Qwen3.6 and other models have really been a huge step forward. Better optimization? Or just better base models to finetune? Not sure, probably a mixture of the two with a little sprinkled RNG.
Most models are you extremely negative bias or cater to my every whim. This model has a fantastic balance and is great at following instructions.
Mhmm... Sounds promising. I've only seen 1 model push back with a justification of why it did something rather than saying 'oh you are totally right' even when i was potentially wrong.
I'll have to give this one a try then, a model that isn't a pushover would make things ore fun. But there's so many flavors of LLMs.