Google claims that the research and technology used to develop its Gemini AI models was also utilized to construct the open AI model known as Gemma. Gemma, according to the firm, is their gift to the public and is designed to assist developers “in building AI responsibly.” Therefore, along with Gemma, it also introduced the Responsible Generative AI Toolkit. It includes an AI development best practices guide based on Google’s experience, along with a debugging tool.
Gemma is now offered by the firm in two distinct sizes, Gemma 2B and Gemma 7B. Both sizes provide pre-trained and instruction-tuned variants, and both are sufficiently lightweight to be used directly on a desktop or laptop computer used by developers. According to Google, Gemma performs better than other open models in both model sizes and beats considerably larger models on important benchmarks.
In addition to being powerful, the Gemma models were trained to be safe. Google used automated techniques to strip personal information from the data it used to train the models, and it used reinforcement learning based on human feedback to ensure Gemma’s instruction-tuned variants show responsible behaviors. Companies and independent developers could use Gemma to create AI-powered applications, especially if none of the currently available open models are powerful enough for what they want to build.
Google has plans to introduce even more Gemma variants in the future for an even more diverse range of applications. That said, those who want to start working with the models right now can access them through data science platform Kaggle, the company’s Colab notebooks or through Google Cloud.
