The extension is intended to provide a kind of prompt chaining for SillyTavern. The general idea behind it is to force an LLM to generate a character's thoughts (emotions, plans - whatever you wish) before running the regular prompt generation. This increases the overall waiting time for a response in favor of the quality of that response.
Just open the Extensions
menu in SillyTavern, then click on the Install Extension
button.
After that, you may paste the url of this repository to install the extension: https://github.com/cierru/st-stepped-thinking
You can find the extension settings in the Stepped Thinking
section of the SillyTavern Extensions
menu.
There are also several settings in data/default-user/settings.json
of SillyTavern that are not presented in the UI.
You
probably don't want to change them, however there is a list of the settings just in case:
The extension has been tested on SillyTavern 1.12.6 but should also work in other versions.
You are always welcome to contribute to this project. You may create issues, pull requests, and describe them in English or Russian - whatever is more convenient for you. However, if you want to propose a feature that is not in the "Planned" section of the README file, please describe your request and reasoning in as much detail as possible.
Special credits to the anon from the /llama thread who described the idea and wrote a PoC of the approach that formed the basis of this extension. You can familiarize yourself with the results of his efforts here: https://rentry.co/LLMCrutches
MIT License
Copyright (c) 2024 cierru