Gemma: Open Models Based on Gemini Research and Technology

ABSTRACT

We introduce Gemma, a family of lightweight, state-of-the-art open models built from the same research and technology used to create the Gemini models. Gemma models are offered in two sizes: a 7 billion parameter model suited for efficient deployment on consumer GPUs and TPUs, and a 2 billion parameter model designed for CPU and on-device applications. Both models demonstrate strong performance across text generation, reasoning, and coding benchmarks, outperforming similarly sized open models like Llama 2. We release the pre-trained weights, instruction-tuned checkpoints, and a comprehensive responsible generative AI toolkit to foster innovation and safe deployment within the open-source community.