Google DeepMind Releases EmbeddingGemma 2, a 740M Open Multimodal Embedding Model Built on Gemma 4
https://www.marktechpost.com/2026/10/06/google-deepmind-releases-embeddinggemma-2-a-740m-open-multimodal-embedding-model-built-on-gemma-4/📌 740M 參數打通五種模態，Google DeepMind 開源可離線跑的嵌入模型
TL;DR：EmbeddingGemma 2 用單一向量空間統一文字、程式碼、圖片、影片與音訊，Apache 2.0 授權，手機上就能跑 RAG。
一支手機能同時「看懂」文字、照片、影片和語音備忘錄嗎？Google DeepMind 剛發布的 EmbeddingGemma 2，把這五種模態全部塞進同一個 768 維向量空間，而且模型只有 740M 參數，今天就能下載部署。
🤔 為什麼需要跨模態 embedding
Embedding 模型把內容轉成一組數字向量，用來捕捉語意，意思相近的內容在向量空間裡會彼此靠近，方便搜尋與比對。在 RAG（Retrieval-Augmented Generation）流程中，這些向量讓 LLM 能檢索自己訓練時沒見過的最新資訊。把 embedding 生成放在裝置本地執行，資料不出裝置、延遲更低，也能離線運作，這正是 EmbeddingGemma 2 瞄準的場景：裝置端搜尋、分類與重視隱私的 RAG。
🧩 建立在 Gemma 4 架構上的模組化設計
EmbeddingGemma 2 以 Gemma 4 架構為基礎，支援文字查詢檢索圖片、語音備忘錄檢索影片，甚至把文字、圖片、示範影片交錯混合的商品列表，編碼成單一向量。模型設計是模組化的，分成三個部分，開發者可依需求只載入需要的版本：270M（純文字）、440M（文字＋視覺）、570M（文字＋音訊），或 740M（全模態）。所有版本共享同一個向量空間，也就是說用純文字版本編碼的查詢，依然能比對出全模態版本編碼的文件。
此外，context window 來到 8,192 tokens，是上一版的 4 倍，足以容納約 29 張圖片、58 個影片畫格，或 5.5 分鐘音訊。
📊 跑分與資源消耗
Google 研究團隊表示，在 sub-1B 參數規模的多模態 embedding 模型中，EmbeddingGemma 2 在 MTEB Code 與 MAEB 上拿下領先成績；在 768 維、全精度下，程式碼檢索（code retrieval）提升 9.92 分，約 14%，多語言文字品質則維持穩定。不過體量更大的模型在部分排行榜仍領先，例如 Qwen3-VL-Embedding-2B 在自家 MMEB-V2 跑分報 73.2，參數量約為 EmbeddingGemma 2 的 2.7 倍，且不支援音訊。
在資源消耗方面，於 Pixel 11 Pro 上搭配量化，純文字權重的 active RAM 約 191MB，全模態模型約 567MB；量化感知訓練（quantization-aware training）把權重壓縮到 INT4／INT8。Google AI Edge 團隊在 MacBook M5 Pro GPU 上測得每張圖片 37.3 ms（70-token 視覺預算）。
透過 Matryoshka Representation Learning（MRL），開發者可以把向量截斷到 512、256 或 128 維：從 768 維降到 128 維，儲存空間最多省 6 倍。截到 256 維時，MTEB 多語言分數從 61.36 僅略降到 60.41；但到 128 維，MMEB 分數會掉到 45.65，因此 Google 建議 128 維主要用於純文字場景。
💡 部署生態已經鋪好
EmbeddingGemma 2 目前可在 sentence-transformers v6.1.0+、Transformers、vLLM、SGLang、MLX、llama.cpp、Ollama、LM Studio、LiteRT 與 MediaPipe 上運作；向量儲存可搭配 Qdrant，微調可用 Unsloth。針對 Android 的 ML Kit 整合（含 NPU 加速）也將在數週內推出。權重已上架 Hugging Face 與 Kaggle，想快速試用的話，執行 ollama pull embeddinggemma-2 即可，tag 從 270m（378MB）到 740m（1.3GB）都有，Demo 可在 Google AI Edge Gallery 體驗。
🎯 實務啟示
對要做隱私優先 RAG 或裝置端多模態搜尋的工程師，EmbeddingGemma 2 的模組化設計值得關注：可以先用 270M 純文字版上線，之後再無縫升級到全模態版本而不必重建索引，因為所有子模型共享同一個向量空間。128 維截斷特別適合純文字的大規模向量資料庫，能大幅壓低儲存成本。
🔗 來源
- 標題：Google DeepMind Releases EmbeddingGemma 2, a 740M Open Multimodal Embedding Model Built on Gemma 4
- 作者／機構：Asif Razzaq
- 連結：https://www.marktechpost.com/2026/10/06/google-deepmind-releases-embeddinggemma-2-a-740m-open-multimodal-embedding-model-built-on-gemma-4/
#EmbeddingGemma #GoogleDeepMind #Gemma4 #MultimodalAI #RAG #OnDeviceAI #OpenSource #VectorEmbeddings #MachineLearning #AIEngineering
原始資料 MarkTechPost · 收集於 2026-10-06
摘要原文
Google DeepMind has released EmbeddingGemma 2 , an open model that embeds text, code, images, video and audio into one 768-dimensional space. It has 740M parameters, an 8K token context window and an Apache 2.0 license. It targets on-device search, classification and privacy-first RAG. This article analyzes, compares and showcase how EmbeddingGemma 2 fits in the space. Deployable today? Yes. Weights are live on Hugging Face and Kaggle , with Ollama , llama.cpp GGUF and LiteRT builds available now. An embedding model converts content into a vector of numbers that captures meaning. Similar items land close together, so they are easy to search and compare. In a RAG pipeline, these vectors let an LLM retrieve fresh information it was not trained on. Generating embeddings locally keeps data on the device, cuts latency and works offline. EmbeddingGemma 2 is built on the Gemma 4 architecture. A text query can retrieve a photo. A voice memo can retrieve a video clip. Interleaved inputs, like a product listing with text, images and a demo video, produce a single embedding. The design is modular. It has three parts: Developers load only what they need: 270M for text, 440M for text and vision, 570M for text and audio, or 740M for everything. All setups share one vector space. A query embedded with the text-only setup can match documents embedded by the full model. The context window is 8,192 tokens, 4x larger than version 1. That fits about 29 images, 58 video frames or 5.5 minutes of audio. Google research team reports leading scores among sub-1B multimodal embedders on MTEB Code and MAEB. Full-precision results at 768 dimensions: Source: EmbeddingGemma 2 model card Code retrieval gains 9.92 points, roughly 14%. Multilingual text quality holds steady. Bigger models still lead some boards. Qwen3-VL-Embedding-2B reports 73.2 on its own MMEB-V2 run, with about 2.7x the parameters and no audio support. With quantization on a Pixel 11 Pro, active RAM is about 191MB for text-only weights. The full multimodal model needs about 567MB. Quantization-aware training compresses weights to INT4 and INT8. The Google AI Edge team measured 37.3 ms per image on a MacBook M5 Pro GPU, using a 70-token vision budget. Matryoshka Representation Learning (MRL) lets developers truncate vectors to 512, 256 or 128 dimensions. Moving from 768 to 128 dimensions cuts storage up to 6x. At 256 dimensions, MTEB multilingual only slips from 61.36 to 60.41. At 128 dimensions, MMEB drops to 45.65, so Google recommends 128d mainly for text-only workloads. It runs on sentence-transformers v6.1.0+, Transformers, vLLM, SGLang, MLX, llama.cpp, Ollama, LM Studio, LiteRT and MediaPipe. Qdrant covers vector storage and Unsloth covers fine-tuning. ML Kit support for Android, with NPU acceleration, is coming within weeks. On Ollama, run ollama pull embeddinggemma-2 . Tags range from 270m (378MB) to 740m (1.3GB). Demos live in Google AI Edge Gallery . See the developer guide for more. Check out the Model Weights on HF and Technical details . All credit goes to the researcher of this project. Also, feel free to follow us on Twitter and don’t forget to join our 150k+ML SubReddit and Subscribe to our Newsletter . Wait! are you on telegram? now you can join us on telegram as well. [Sponsored] The web is the one API most agents are missing. Databases, calendars and repos have APIs. The open web mostly doesn’t. The TinyFish MCP server gives any MCP client four tools: TinySearch, TinyFetch (full pages as markdown, JavaScript included), TinyBrowser for logins and forms, and TinyAgent for multi-step jobs. Search and Fetch are free. The post Google DeepMind Releases EmbeddingGemma 2, a 740M Open Multimodal Embedding Model Built on Gemma 4 appeared first on MarkTechPost .
由 claude-code/sonnet 自動生成