moondream

1.

Moondream is a lightweight, open-source vision language model available on GitHub. It boasts two versions: a larger 2 billion parameter model and a smaller 500 million parameter model optimized for resource-constrained devices. The model excels at image understanding tasks such as captioning and question answering. Users can interact with Moondream via Python or Node.js client libraries, or through a Hugging Face Transformers integration offering GPU support. The project's GitHub repository includes code, documentation, and example applications.

Link: https://github.com/vikhyat/moondream

By erdal on Dec. 9, 2024, 8:07 a.m.