Pacha - A TUI Frontend for llama.cpp

[mounta11n]

I share with you my little app 'Pacha' and hope that we have something that stays lightweight and terminal based like llama.cpp, but still can provide a minimum of comfort. At some point I just found it annoying to have to type or copy a whole command again for every little difference in parameter value I wanted to test, etc.

There are ready to use binaries for windows, linux and macOS (Intel). Just put it into the same folder as llama.cpp and there you go!

https://github.com/mounta11n/Pacha

This frontend is not meant to be a chat UI or to replace anything, but rather a tool to quickly test a model, a prompt style and/or certain parameters. I think this might be a good first stop to test new models.

The top bar changes its color based on the current cpu usage.

Here is an asciinema demonstration:

The app is currently buggy in some places, but I'm working on it. However, I felt that it is now functional enough that it can be released without any problems.

There are more features planned... like integrating bert.ggml for semantic context and a workspace for training baby-llamas from scratch.

To smart people who are familiar with javascript: Please look over my code and tell me how I can improve the corresponding buggy parts. For example, I just don't figure out why there is a line break after the first chunk in the output. Tried for ages to understand and fix it, but ... Idk. And I don't dare ask GPT-4 anymore. First, I'll be busy debugging GPT's f+cking mistakes more than half the time. And besides, I'm pretty sure my wife will kill me as soon as the next OpenAI bill comes.