do you use a local llm with sillytavern? I've been doing something similar, but trying to find a model that works for me
The ones that come with it are hard to get right but it also matters how the bot learns to behave in reaction to you and stuff. You have to use the character notes part and fill it out as much as possible. I really can't say which one might be right for you but I found a combination of Kobold and the uncensored/unleashed ones work decently.
If you want full customization and the ability to train from your own data set, you will still need to be very very detailed. Include every possible thing. And there will still need to be training. You can use LoRa, Hugging Face, or Llama Factory. I have not tried Axolotl 😅
There is the crucial step that turns your trained model (usually in Hugging Face .safetensors + config format) into a GGUF file. GGUF is the standard format used by most local inference engines (KoboldCpp, llama.cpp, Ollama, LM Studio, oobabooga, etc.) that connect to Silly Tavern. It’s highly efficient even on a smaller normie computer.
But basically it really doesn't matter which one you choose because it's still gonna take work to train. It just matters if you want to be a control freak lol