Gemma provides an open-weights Large Language Model (LLM) library from Google DeepMind, built on Gemini research. It enables users to utilize and fine-tune Gemma models, offering a JAX-based implementation for researchers and developers working with advanced AI models.

How It Works

Gemma is implemented as a JAX library, allowing for efficient computation on CPUs, GPUs, and TPUs. The library provides pre-trained model checkpoints and tools for sampling, including multi-turn and multi-modal conversations. Its design leverages JAX's automatic differentiation and hardware acceleration capabilities for both inference and fine-tuning.

Quick Start & Requirements

• Install via pip: pip install gemma
• Requires JAX installation (CPU, GPU, or TPU). Refer to JAX documentation for setup.
• GPU RAM: 8GB+ for 2B checkpoint, 24GB+ for 7B checkpoint.
• Documentation: https://github.com/google-deepmind/gemma
• Technical Reports: Gemma 1, Gemma 2, Gemma 3

Highlighted Details

• Supports multi-turn and multi-modal conversations.
• Includes examples for fine-tuning and LoRA.
• Based on Google DeepMind's Gemini research.

Maintenance & Community

This project is not an official Google product. Contributions are welcome via contributing guidelines.

Licensing & Compatibility

The repository does not explicitly state a license in the provided README. Compatibility for commercial use or closed-source linking is not specified.

Limitations & Caveats

The README does not specify the license, which may impact commercial use. Model weights need to be downloaded separately as per documentation.