Using Reasoning Models

You can use reasoning models in Yozakura if you want. They might do better at certain tasks, like memory generation or next-speaker selection, and you can use different models on a per-prompt basis if you want.

But you will need to increase the default max_tokens settings. You can do that in LLM Settings. Go through all of those settings (or at least the ones you intend to use a reasoning model for) and either remove the max_tokens to make it unlimited, or increase it by a couple thousand.

The reason you have to do this is because reasoning tokens count towards the max_tokens budget, making it much easier to hit the default max values if you're using a reasoning model.