Latest Version: v0.1.3.post3

• Fixed absolute path for libsentencepiece.0.0.0.dylib
• Interface changes due to updates in gemma.cpp.
• Enhanced user experience for ease of use 🙏. Give it a try!

gemma-cpp-python provides Python bindings for gemma, a high-performance C++ library, bridging advanced computational functionalities to Python.

Special thanks to the creators and contributors of gemma.cpp for their foundational work.

Prerequisites: Ensure Python 3.8+ and pip are installed.

System requirements: For now, I only tested it on the Unix-like Platforms and the MacOS. Please visit the gemma.cpp installation for more details.

Models: pygemma supported 2b-it-sfp model for now, to install model, please visit here

For a quick setup, install directly from PyPI:

pip install pygemma==0.1.3

To install the latest version or for development purposes:

1. Clone the repo and enter the directory:

git clone https://github.com/namtranase/gemma-cpp-python.git
cd gemma-cpp-python

2. Install Python dependencies and pygemma:

pip install .

To acctually run the model, you need to install the model followed on the gemma.cpp repo

For usage examples, refer to tests/test_chat.py. Here's a quick start:

from pygemma import Gemma
gemma = Gemma()
gemma.show_help()
gemma.show_config()
gemma.load_model("/path/to/tokenizer", "/path/to/compressed_weight/", "model_type")
gemma.completion("Write a poem")

Contributions are welcome. Please clone the repository, push your changes to a new branch, and submit a pull request.

gemma-cpp-python is MIT licensed. See the LICENSE file for details.