Most recently, Google presented its new product in the field of artificial intelligence – LLM Gemini. Not stopping there, the corporation recently introduced the world to Gemma, an innovative family of open, lightweight language models. Based on the same science and technology as Gemini, the latest developments reflect Google's commitment to advancing the ethical approach to AI. Gemma's models are designed to empower developers to create safe and responsible artificial intelligence.
The first models in the line are Gemma 2B and Gemma 7B. They are offered as ready-made solutions, adaptable to specific tasks. They can be freely used for both commercial and research purposes. Access to them is already open, and it is completely free. These models also come with a specially designed Responsible Generative AI Toolkit designed to make working with AI easier.
Despite their rather modest size, Gemma 2B and 7B are distinguished by high performance. This was achieved by leveraging the same technology and infrastructure solutions that underpin Gemini, Google's most powerful AI model. Representatives of the corporation claim that in MMLU tests, the new models outperformed well-known competitors – Mistral 7B and Llama 13B.
One of the most significant advantages of the models is their availability for running on personal computers, laptops, as well as in the Google cloud infrastructure. This makes them more attractive compared to their competitors. Another serious advantage is that these models are optimized for NVIDIA GPUs. In addition, they provide integration with services and tools such as Colab, Kaggle, Hugging Face, MaxText, and TensorRT-LLM.
Tris Warkentin, director of product at Google DeepMind, highlighted the significant growth in quality content generation seen throughout 2023. Features that were previously only available in the largest models are now available in much smaller ones. Configuration on local machines with RTX GPUs or on GCP cloud platforms with TPUs significantly expands the boundaries for developers. Overall, LLMs like Gemma open up new perspectives for creating AI products.