On the evening of February 21st Beijing time, US technology giant Google announced the launch of the world's most powerful and lightweight open-source model series, Gemma.
It is reported that the large model is available in two sizes: 2B (2 billion parameters) and 7B (7 billion), with the 2B version even running directly on a laptop.
Google stated that Gemma adopts the same research and technology as the Gemini model, developed by Google DeepMind and other Google teams, specifically for responsible AI development.
Google claims that out of 18 key benchmark tests for language comprehension, reasoning, mathematics, and more, 11 of the Gemma model scored higher than open source models with larger parameters such as Meta Llama-2.