Google DeepMind jumps back into open source AI race with new model Gemma

Google DeepMind has introduced Gemma, a set of 2B and 7B open-source models created using the same technology as the recent Gemini models. The Gemma models are accompanied by pre-trained and instruction-tuned variants and will be available under a permissive commercial license along with a new Responsible Generative AI toolkit. Google is offering toolchains for inference and supervised fine-tuning across major frameworks like JAX, PyTorch, and TensorFlow via Keras 3.0. Users can access Colab and Kaggle notebooks, and Gemma is integrated with platforms like Hugging Face, MaxText, and NVIDIA NeMo. The pre-trained Gemma models can run on various platforms, including laptops, workstations, Google Cloud, Vertex AI, and Google Kubernetes Engine.

Nvidia has collaborated with Google to optimize Gemma performance across its AI platforms, including local RTX AI PCs. Google’s Jeanine Banks highlighted Gemma’s continuity with the company’s history of open-sourcing AI technologies and emphasized the potential synergy between open models and APIs in developers’ workflows. Tris Warkentin mentioned plans to release benchmarks comparing Gemma with other models, which will be publicly available on OpenLLM leaderboards. He noted transparency in benchmarking collaborations with Nvidia and Hugging Face, underscoring Gemma’s safety features and extensive evaluations.

Google DeepMind’s blog post emphasized Gemma’s alignment with the company’s AI principles, focusing on automated techniques to filter sensitive data, fine-tuning with human feedback, and rigorous evaluations to mitigate risks. Warkentin stressed the importance of engaging diverse perspectives globally to enhance safety measures, highlighting the community’s role in providing valuable feedback for responsible AI development.

The introduction of Gemma underscores Google’s commitment to advancing AI technology while adhering to ethical principles and ensuring model safety. The collaborative efforts with industry partners and the emphasis on transparency and community engagement reflect Google’s dedication to fostering a responsible and inclusive AI ecosystem.