uetuluk/code-interpreter-lite

The main application of the Code Interpreter Lite stack.

Code Interpreter Lite

An experimental client that uses LLMs with docker-python-sandbox-supervisor and docker-python-sandbox to run python code in a sandboxed environment.

While the current version only works with GPT API, the goal is to make it work with any LLMs, especially local ones like LLaMA2.

Showcase

Project Structure

Running

Requirements

1. Docker

Setting Up

1. Clone the repo
2. Download/build all the required images. Check the docker-python-sandbox-supervisor and docker-python-sandbox for guides on how to build those images. The Code Interpreter Lite application can be built by running the following command.

    build-production-image

3. After you have all the images ready, you can run the following command to set up the Code Interpreter Lite stack.

   make full-detached

The application will be available at http://localhost:7860.

Development

Requirements

1. Docker
2. Python 3.10 or higher (Tested on 3.10.4 and 3.10.12)

Setting Up

1. Clone the repo
2. Run the following command to set up the development environment.

   make build-development-environment

3. Run the following command to start the development environment.

   make develop

The application will be available at http://localhost:7860.

TODO

• More reliable handling of files - Currently, the LLMs get confused about how to access the files
• Add support for local LLMs - LLaMa does not use the correct format, therefore the code cannot be run.
• Add support for Claude from Antrophic - Currently Claude gets stuck running the code again and again until it times out.