Google Launches Two New Lightweight Open Source Models for Developers

Google has announced the launch of Gemma, a family of cutting-edge open language models tailored to assist developers and researchers in building AI applications. Gemma models share technical and infrastructure components with Gemini, Google's largest and most capable AI model.

Key Points About Gemma

• Optimized Sizes: Gemma offers two model sizes – the compact Gemma 2B and the larger Gemma 7B – each with pre-trained and instruction-tuned variations for flexibility.
• Responsible AI Focus: Google reinforces its commitment to safe AI with the Responsible Generative AI Toolkit, aiding developers using Gemma.
• Wide Framework Compatibility: Gemma supports JAX, PyTorch, and TensorFlow via Keras 3.0, plus seamless integration with Hugging Face, MaxText, NVIDIA NeMo, and TensorRT-LLM for streamlined development.
• Accessibility: Colab and Kaggle notebooks make experimentation with Gemma simple. Deployment options on Google Cloud via Vertex AI and GKE further enhance accessibility.
• Impressive Performance: Despite its smaller size, Gemma achieves class-leading results, surpassing even larger models on crucial benchmarks while ensuring safe and responsible outputs.

Safety and Reliability Built-in

Google's AI Principles are deeply ingrained in Gemma's creation. By filtering sensitive data from training sets, Google ensures the models are reliable and promote responsible use. Gemma is available worldwide, starting today.