Google 发布 EmbeddingGemma 2：统一多模态嵌入空间
原标题：EmbeddingGemma 2 Expands Beyond Text to Support Code, Images, Video, and Audio
AI 摘要
Google 发布 EmbeddingGemma 2，这是一个基于 Gemma 4、采用 Apache 2.0 许可的紧凑型开放权重多模态嵌入模型，可在手机和笔记本等消费级硬件上运行。其核心特性是将文本、代码、图像、视频和音频映射到统一的 768 维嵌入空间，使开发者无需为不同模态维护互不兼容的嵌入空间。模型支持按需加载编码器，参数量从 2.7 亿（文本和代码）扩展到 7.4 亿（完整多模态），并支持 Matryoshka 表示学习以截断向量、降低存储。模型已在 Hugging Face 上发布。
正文节选
Based on Gemma 4 and released under the Apache 2.0 license, Google's EmbeddingGemma 2 is a compact, open-weight, multimodal embedding model designed to make semantic search and retrieval-augmented generation (RAG) more efficient while optimizing performance for on-device deployment. Designed to run on consumer hardware such as mobile devices and laptops, EmbeddingGemma 2 delivers low-latency semantic representations for on-device applications, like search, retrieval-augmented generation (RAG), c