Google has released the open-source AI model Gemma, designed for artificial intelligence developers and researchers. Gemma is created by DeepMind and other Google teams, third-party AI developers can build their own models based on it, and it is based on the same technologies and research as Gemini.

Gemma comes in two sizes: the lightweight one, 2B, consists of a neural network of 2 billion adjustable variables called parameters, and the other, more productive, 7B, consists of a neural network of 7 billion parameters. The Gemma 2B and 7B models can run directly on a laptop or in the cloud, but Google hasn’t made Gemma’s source code completely open-source, so the company can still set the terms of use.

Google claims that both variants of the Gemma have the best performance compared to other open models in their class. Gemma can be used in commercial projects, regardless of their size. The company emphasized that the development was optimized for NVIDIA GPUs; integrations with popular tools like Hugging Face, MaxText, NVIDIA NeMo, and TensorRT-LLM are also available. The name Gemma comes from Latin and means “precious stone”.

Літературний редактор.

Commentary