atai-gemma3-tool

atai-gemma3-tool

atai-gemma3-tool is a command-line interface (CLI) tool that uses Google's Gemma 3 model to generate descriptive text from local image files. It leverages the power of a state-of-the-art multimodal model to process images and stream textual outputs in real time.

Features

• Multimodal Processing: Accepts image input and produces text output.
• Real-Time Streaming: Generates and streams tokens as they are produced.
• Customizable Prompt: Allows users to define a custom prompt.
• Easy Installation: Installable via pip with all dependencies handled.
• Asynchronous Generation: Utilizes asynchronous token streaming for quick response times.

Installation

Clone the repository and install the package in editable mode:

pip install git+https://github.com/huggingface/transformers@v4.49.0-Gemma-3

pip install atai-gemma3-tool

Usage

Run the CLI tool from your terminal by specifying the path to your image file and an optional custom prompt:

atai-gemma3-tool "path/to/your/local_image.jpg" --prompt "Describe this image in detail."

atai-gemma3-tool https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/p-blog/candy.JPG

Command Line Arguments

• image_path: The path to your local image file or a image url.
• --prompt: (Optional) Custom prompt for text generation.
  Default: "Describe this image in detail."

The tool will load the image, process it using the Gemma 3 model, and output the generated text to your console in real time.

License

This project is licensed under the MIT License. See the LICENSE file for details.

Acknowledgements

• Google DeepMind: For the Gemma 3 model.
• Hugging Face: For the Transformers library and supporting tools.