Google has introduced Gemma, a new open-source AI tool designed to assist developers in creating innovative applications safely. This release comes just a week after the announcement of Gemini models, showcasing the tech giant’s commitment to advancing AI technology.
The Gemma models, known as lightweight open-source models, include Gemma 2B and Gemma 7B. These models are inspired by Gemini and cater to different computational constraints, applications, and developer requirements. Gemma 7B boasts seven billion parameters for efficient deployment on GPUs and TPUs, while Gemma 2B, with two billion parameters, is suitable for CPU and on-device applications.
In addition to the new models, Google has introduced a responsible generative AI toolkit to provide guidance and essential tools for creating safer AI applications with Gemma. The company also highlighted the performance of Gemma models, showcasing their superiority over similar-sized open models on various text-based tasks and academic benchmarks for language understanding, reasoning, and safety.
Developers can now access Gemma through integration with JAX, PyTorch, and TensorFlow via native Keras 3.0. Google is also offering ready-to-use Colab and Kaggle notebooks, as well as integrations with Hugging Face, MaxText, and Nvidia’s NeMo. These pre-trained and fine-tuned models can be run across different platforms once optimized for specific use cases.
Moreover, Google Cloud AI Vice President Burak Gokturk emphasized the benefits of using Vertex AI to streamline operational processes and customize Gemma for specific needs. While Gemma provides a contained environment for testing products, Google stresses the importance of rigorous safety testing before widespread deployment.
To ensure the reliability of Gemma pre-trained models, Google employed automated techniques to filter out personal information and sensitive data from training sets. Extensive fine-tuning and reinforcement learning from human feedback were also utilized to align the models with responsible behaviors.
In an effort to encourage developers and researchers to leverage Gemma, Google is offering free access via Kaggle or Colab and providing $300 in credits for first-time Google Cloud users. Additionally, researchers can apply for grants of up to $500,000 for their projects.
Overall, Gemma reflects Google’s commitment to upholding its AI Principles and fostering responsible AI innovation. With a focus on safety, reliability, and accessibility, Gemma aims to empower developers and researchers to create cutting-edge applications while adhering to ethical standards and best practices in AI development.