google-deepmind/gemma

Gemma open-weight LLM library, from Google DeepMind

What it solves

It provides a JAX-based implementation for using and fine-tuning the Gemma family of open-weights Large Language Models (LLMs), which are built on Gemini research and technology.

How it works

The project is delivered as a PyPI package (gemma) that leverages JAX to run on CPUs, GPUs, or TPUs. It includes a ChatSampler API for handling multi-turn and multi-modal conversations, as well as tools for loading model checkpoints and performing fine-tuning.

Who it’s for

Developers and researchers who want to integrate Gemma models into their applications or fine-tune them for specific tasks using the JAX ecosystem.

Highlights

  • Multi-modal capabilities: Supports multi-turn conversations involving both text and images.
  • Hardware flexibility: Compatible with CPU, GPU, and TPU hardware.
  • ** uma JAX library**: Specifically designed for high-performance ML operations via JAX.
  • Fine-tuning support: Includes built-in support for fine-tuning and LoRA (Low-Rank Adaptation).

Related

  • Project
    AI-Hypercomputer/maxtext

    A high-performance, scalable LLM library and reference implementation written in JAX for training and post-training models on Google Cloud TPUs and GPUs.

  • Project
    google/tunix

    A JAX-based post-training library for LLMs that provides scalable support for supervised fine-tuning and reinforcement learning, specifically optimized for TPU performance.

  • Dispatch
    Google Gemma Open LLM Release

    Google has released Gemma, a family of open-access large language models based on Gemini, available in 2B and 7B parameter sizes with base and instruction-tuned variants.

  • Dispatch
    Gemma 3 Release Notes / What's New

    Google has released Gemma 3, a multimodal, multilingual open-weight LLM family ranging from 1B to 27B parameters with context windows up to 128k tokens.

  • Project
    instadeepai/jumanji

    A diverse suite of scalable, hardware-accelerated reinforcement learning environments written in JAX for high-speed RL research and experimentation.