This repository contains experiments and example code for using LangChain with Llamafile, running local LLMs in Python.

• Python 3 project with Pipenv for easy dependency management
• All required libraries for LangChain and Llamafile are specified in the included Pipfile
• Example code for interacting with a local LLM via LangChain in the src directory
• Ready-to-use VSCode workspace for convenient development
• Easily switch between different local LLMs downloaded from HuggingFace, e.g. Mozilla/Llama-3.2-1B-Instruct-llamafile

• Python 3.12 (or compatible version)
• pipenv (recommended, but you can use any Python environment manager)

1. Clone the repository:

   git clone https://github.com/brakmic/langchain-experiments.git
   cd langchain-experiments

2. Install dependencies:

   pipenv install

   Or, if you prefer another environment manager, use the Pipfile as a reference.

3. Download a Llamafile model:

   • Visit HuggingFace or another LLM provider.
   • Download the .llamafile model and run it locally (see the model's README for instructions).

4. Set the Llamafile API URL (optional):

   • By default, the code expects the Llamafile server at http://localhost:8080.
   • You can override this by setting the LLAMAFILE_URL environment variable.

• Source code is in the src directory. For example, to run a simple prompt:

  python src/llamafile.py

• You can open the project in VSCode using the included workspace file:

  code langchain.code-workspace

• The .venv and local/ directories are excluded via .gitignore.
• You can use any Python 3 environment manager if you prefer not to use Pipenv.
• Example models and prompts are provided, but you can adapt them for your own experiments.

MIT