This project allows you to convert Llama model weights to the Hugging Face format, making it easier to work with Llama models in various applications.
Before you begin, ensure you have the following installed:
- Python 3.9 or higher
- pip (Python package installer)
- curl (for testing the application)
Visit the Meta website to download the desired Llama model. Save the model files in the models/meta-llama directory. Your directory structure should look something like this:
project-root/
├── convert_llama_weights_to_hf.py
├── LICENSE.txt
├── llama_model_list.txt
├── Makefile
├── models
│ ├── meta-llama
│ │ ├── Llama3.2-3B-Instruct # example model directory
│ │ │ ├── checklist.chk
│ │ │ ├── consolidated.00.pth
│ │ │ ├── params.json
│ │ │ ├── tokenizer.model
│ └── transformers
└── ...
Open the Makefile and replace the LLAMA_MODEL variable with the name of the model you downloaded. For example:
LLAMA_MODEL ?= Llama3.2-1B-Instruct
MODEL_SIZE ?= 1B
LLAMA_VERSION ?= 3.2
Run the following command in your terminal to set up the environment and convert the model:
make install
This command will create a virtual environment, install the required packages, and convert the Llama model weights.
To run the model, execute:
make run
You can test the application with a sample POST request using curl:
make test
For more information on the model and prompt formats, visit:
This project uses the Llama 3.2 model, which is licensed under the Llama 3.2 Community License, Copyright © Meta Platforms, Inc. All Rights Reserved.
You can find the terms of the Llama 3.2 Community License at: Llama 3.2 Community License
This project also includes code and resources that are licensed under the Apache License, Version 2.0, January 2004.
You can find the terms of the Apache License at: Apache License 2.0