Last Updated: 1/11/2025
There's some kind of gguf tokenization issue where its not properly handling the <|im_end|> EOS token and instead breaks it into multiple smaller ones which degrades performance for gguf users. Need to submit an issue ticket. Unsure if this impacts other quantization formats.
I see a lot of folks looking into this model now so I want to make sure folks understand, this is a branch where I upload random test runs. Its publically available but I make no guarantees the model you download will function. I'm doing some wacky stuff. Working on getting a v0.1 out so folks can have something more "stable" to mess with.
A public repo where I'll put my latest KTO results for people to mess around with. Resulting models will be chatML. Obviously experimental and not release ready, but feel free to mess around with whatever is here.
The goal of this model is to make something that feels different from other models out there. It has seen very very little synthetic data and without RL is unusable. I think with some patience, you can enjoy a conversation with this model. I've used the "Simple-1" preset and that seems to work well enough, feel free to experiment.
Temp: 0.85 TopK: 40 TopP: 0.9 RepPen: 1.05
Formatting Settings for Silly Tavern. It should end up as ChatML with username/character names.
