LLaM-e

LLaM-e is a HIGHLY EXPERIMENTAL but beautiful cross-platform chat client for LLaMA-like models, built in Rust 🦀 and TypeScript.

This program is based on the llama-rs library, which is a Rust rewrite of llama.cpp.

The goal is to create a simple interface for interacting with LLaMA-like models, in a similar fashion to OpenAI's Playground.

Supported (Tested) Models

LLaMA 7B (Q4)
LLaMA 13B (Q4)
Alpaca 7B (Q4)
Alpaca 13B (Q4)

If you have had success with other models, please add them to the list.

Demo

In the demo, Alpaca 7B (Q4) is running on a MacBook Pro with a 10 core M1 Pro CPU and 16GB of RAM.

With this hardware, 13B models can be run withouth any problems, though a bit slower.

Architecture

LLaM-e is built with the TAURI framework, which is a cross-platform framework for creating lightweight desktop applications with web technologies and Rust. (aka. Electron, but better)

The frontend uses SolidJS and the HopeUI component library.

Performance

You can expect the same performance as the original llama-rs codebase.

Note that the performance of the model is highly dependent on the hardware you are running it on, and it degrades significantly when the program is ran in debug mode.

Compiling

Setup your system for compiling a TAURI application. See the docs.
Install NodeJS and NPM.
Run npm install to install the dependencies.
Run npm run tauri build to create the package for your platform.

Contributing

There is a lot of work to be done, both in this repository and in the llama-rs library, so if you want to contribute, please feel free to do so. 🙏🏼

Roadmap

Improve model loading system:
- Open load model dialog when the user attempts to submit a prompt without a model loaded. (now you have to click on the box icon on the top right)
- Unify model selection dialog with loading progress dialog.
- Properly implement progress calculation for model loading.
Implement session management:
- Allow session clearing.
- Add session export/import capabilities.

About the Model

LLaMA is a language model released by Meta. Due to licensing restrictions, the model is not included in this repository.

For more information, refer to the llama.cpp repository, which includes the tools to quantize the models.

I have noticed that Alpaca, which is a model released by Stanford (which is based on LLaMA), produces better results than LLaMA itself.

This project shall not be used for any commercial purposes.

sonodima / llam-e