Google unveiled Gemma 2, their latest state-of-the-art open models, in parameter sizes of 9 billion (9B) and 27 billion (27B), in June. The 27B model outperformed popular versions more than twice its size in real discussions and soon rose to the top of the LMSYS Chatbot Arena leaderboard after its launch.
However, Gemma is more than just a performer. It prioritizes accessibility and safety and is based on responsible AI. To fulfill this promise, Google is thrilled to present three brand-new Gemma 2 members:
Gemma 2 2B, a fully redesigned version of Google’s popular 2 billion (2B) parameter model, provides a powerful combination of performance and efficiency, along with integrated safety enhancements. It can be obtained from websites like Google’s AI Studio toolkit, Kaggle, and the Vertex AI model library. It is licensed for specific research and commercial uses.
Also Read: Google Cloud Partners with Mistral AI to Boost Vertex AI with Codestral Code Generation
ShieldGemma is a collection of safety content classifier models that attempt to detect toxicity like hate speech, harassment, and sexually explicit content filters. It is based on Gemma 2, which can be used to filter prompts to an AI model input and output to protect users.
Also Read: Google Unveils Gemini 1.5 Flash: Advanced AI Chatbot Now Available in 40 Languages
A brand-new tool for model interpretability called Gemma Scope offers unparalleled access to the inner workings of Google’s models. In a blog post, Google explains it as follows: “[Gemma Scope is composed of] specialized neural networks that assist us in deconstructing the intricate, dense data that Gemma 2 processes and transforming it into a more comprehensible and readable format. Researchers may learn a great deal about how Gemma 2 recognizes patterns, interprets data, and eventually makes predictions by examining these enlarged views.
With these enhancements, researchers and developers can now produce safer user experiences, acquire never-before-seen insights into Google’s models, and safely and responsibly use strong AI directly on devices, opening up new avenues for innovation.
Shortly after endorsing open AI models in a preliminary report, the U.S. Commerce Department released the new Gemma 2 models. As per the report, open models make generative AI more accessible to smaller enterprises, researchers, nonprofits, and individual developers. However, they also underscore the necessity of having the ability to keep an eye on these models for any possible hazards.
Also Read: 2024 Olympics: Google Becomes Team USA’s Official AI Sponsor
