AI Gemma 3 - the latest version in the lightweight, open-source artificial intelligence model family, capable of operating on a wide range of devices from phones to workstations, and supporting over 35 languages.
Google says this is a lightweight, modern open-source model built on the same research and technology as Gemini 2.0. Gemma 3 was released more than a year after the company first announced the open-source Gemma model.
This model has 27 billion parameters (27B) that run on an Nvidia H100 GPU, while smaller versions are compatible with devices such as smartphones or regular laptops.
In a blog post on X, Google CEO Sundar Pichai stated: "You need at least 10 times the computing power to achieve the same performance as other models."
Google also highlighted Gemma 3's superiority over Meta Platforms Llama-405B, DeepSeek V3, and OpenAI's o3-mini in initial reviews on LMArena. In this ranking, only DeepSeek's R1 model scored higher than Google's new AI model.
The Gemma 3 toolkit is now released with multiple model scaling options, including versions such as 1B, 4B, 12B, and 27B, giving users flexibility to choose according to their hardware capabilities and performance requirements.
Gemma 3 is available in 35 languages and supports trained languages in over 140 languages.
Google also launched ShieldGemma 2, a 4B image safety checker built on the Gemma 3 platform. ShieldGemma 2 provides a safety solution for existing images, labeling them into three categories: dangerous content, pornography, and violence.
Source: https://kinhtedothi.vn/google-ra-mat-ai-gemma-3.html
Comment (0)