Google Research today released Gemma, a new family of lightweight, open-source language models designed for developers and researchers. Inspired by the company's powerful Gemini models, Gemma offers state-of-the-art performance at smaller sizes, making it accessible for more users and applications.
Key Features:
- Performance: Available in two sizes (2B and 7B parameters), Gemma models achieve best-in-class performance for their size compared to other open models. They can even surpass significantly larger models on key benchmarks while adhering to Google's rigorous safety and responsibility standards.
- Accessibility: Gemma models are optimized to run on diverse hardware, including laptops, workstations, and Google Cloud, making them accessible to a broader range of users.
- Open-Source: Google is committed to responsible AI development, and Gemma reflects this commitment. The models are open-source and come with a new Responsible Generative AI Toolkit to help developers build safe and ethical AI applications. This toolkit includes:
- Safety classification: A novel method for building robust safety classifiers with minimal data.
- Debugging: A tool to help investigate Gemma's behavior and address potential issues.
- Guidance: Best practices for building responsible AI models based on Google's experience.
- Flexibility: Gemma is designed to be adaptable to specific needs. Models can be fine-tuned on custom data for tasks like summarization or text generation. Additionally, Gemma supports various frameworks and tools, including TensorFlow, PyTorch, JAX, and Hugging Face Transformers.
- Partnership and Optimization: Google has partnered with NVIDIA to optimize Gemma for NVIDIA GPUs, ensuring industry-leading performance and integration with cutting-edge technology. Similarly, Gemma is optimized for Google Cloud, offering seamless deployment and advanced customization options.
Free Access and Getting Started:
Google is committed to supporting the open-source community. Researchers and developers can get started with Gemma for free through various resources:
- Kaggle: Free access to Gemma models.
- Colab notebooks: Free tier for experimentation.
- Google Cloud: $300 credits for first-time users and up to $500,000 for research projects.
Looking Ahead:
Google promises to expand the Gemma family with new variants for diverse applications, and welcomes the community to connect, learn, and build with Gemma. Visit ai.google.dev/gemma for more information and quickstart guides.
Overall, Gemma represents a significant step forward in making powerful and responsible language models accessible to a wider audience. Its combination of performance, open-source nature, and built-in responsible AI tools positions it to play a major role in advancing the field of natural language processing.
