Google推出Gemma 3 AI模型,用于单一的GPU,在多语种文本和图像分析方面表现优异。 Google unveils Gemma 3 AI models for single GPUs, excelling in multilingual text and image analysis.

Google推出了Gemma 3,一套AI模型,目的是在单一的GPU和TPU上有效发挥功能。 Google has launched Gemma 3, a series of AI models designed for efficient performance on single GPUs and TPUs. Gemma 3支持超过35种语言, 并擅长分析文字、图像及短片。 Available in four sizes, Gemma 3 supports over 35 languages and excels in analyzing text, images, and short videos. 这些模型具有128k代币的大背景窗口, 在人类偏好测试中超越了Llama-405B和o3-mini等竞争对手. The models feature a large 128k-token context window and outperform competitors like Llama-405B and o3-mini in human preference tests. Google还引入了HeldGemma 2, 这是一种安全工具, 将 AI 生成的图像分类, 用于内容过滤 。 Google also introduced ShieldGemma 2, a safety tool that categorizes AI-generated images for content filtering. Gemma 3可在各种平台上进入,供开发商试验和部署AI应用程序。 Gemma 3 is accessible on various platforms for developers to experiment and deploy AI applications.