Perplexity AI Releases pplx-embed-v2-late: A 0.6B Edge Model and a 9B Model Scoring 92.4% on MADQA
https://www.marktechpost.com/2026/10/07/perplexity-ai-releases-pplx-embed-v2-late-a-0-6b-edge-model-and-a-9b-model-scoring-92-4-on-madqa/📌 128 維逐 token 嵌入：Perplexity 開源多模態檢索模型
TL;DR：Perplexity 開源 ColBERT 式多模態嵌入模型，9B 版本 MADQA 拿下 92.4%。
多數嵌入模型把一整份文件壓成一個向量，查詢時難免丟掉細節。Perplexity 這次反過來，為每個 token 都保留一個向量，而且文字、圖片、PDF 頁面全部共用同一個嵌入空間。
🤔 一個模型吃下文字、圖片與 PDF
Perplexity 釋出 pplx-embed-v2-late，是一對 ColBERT 風格的多模態嵌入模型，提供兩種尺寸：0.6B 版本主打快速、低成本查詢，9B 版本主打最高品質。兩者都能檢索文字、圖片以及渲染後的 PDF 頁面，並共用同一個嵌入空間，意味著可以用小模型查詢、大模型建立的索引，或反過來混搭使用。
🧩 用 MaxSim 取代單一向量壓縮
一般 dense 嵌入模型會把整份文件壓縮成 1 個向量，pplx-embed-v2-late 則是為每個 token 保留一個 128 維向量，比對時用 MaxSim 計分：每個查詢 token 找出文件中最相似的那個 token，再把所有最大值加總。頁面是以圖片形式編碼，因此不需要額外的 OCR 步驟。兩個模型都是從一個 18B 的教師模型蒸餾而來，採用 LEAF 風格的 token 層級訓練，正是這套訓練方式讓文字與影像共用同一個嵌入空間成為可能。
📊 9B 模型在 MADQA 拿下 92.4%
依照 Perplexity 官方公告的數據：
- 最強成績：9B 模型在 MADQA 拿下 92.4%。
- 最大領先幅度：在 BrowseComp+ 上領先最佳 dense 模型 8.7 個百分點。
- 最弱成績：0.6B 模型在 ViDoRe v3 Markdown 上為 61.2%，但在該項目中仍是第二好的分數。
- 影像檢索是明顯弱項：Gemini Embedding 2 在 MIRACL-Vision 上勝過 9B 模型，在 PPLX-Q2I 上也領先 2 個百分點。
- 混搭尺寸反而更有效率：用 9B 建索引、0.6B 查詢，在 ViDoRe v3 影像檢索上拿到 63.5%，比起 0.6B 建索引、0.6B 查詢的 62.3% 還高，但查詢成本維持在小模型等級。
💡 部署門檻與環境需求
兩個模型都已上架 Hugging Face，採 MIT 授權，代表可以自行下載部署；官方雖有計畫推出託管 API 端點，但目前尚未上線。模型卡顯示兩者都需要 CUDA GPU，並依賴 sentence-transformers 6.0.0 以上版本與 transformers 5.4.0 以上版本。公開的 checkpoint 以 F32 格式儲存，下載體積會是半精度的兩倍；若只看權重本身，以每參數 2 bytes 估算記憶體用量（此為估算值，非官方數字），0.6B 版本適合邊緣裝置或低延遲查詢場景。
⚠️ 影像檢索仍有落差
雖然在文字問答與跨模態檢索的多項測試中表現亮眼，pplx-embed-v2-late 在純影像檢索（MIRACL-Vision、PPLX-Q2I）上仍落後 Gemini Embedding 2，說明這套方案的強項目前集中在文字為主、圖文混合檢索的場景，而非純視覺比對。
🎯 實務啟示
對於需要同時檢索文字、圖片與 PDF 的 RAG 系統，pplx-embed-v2-late 提供了一個可自行部署、MIT 授權的選項，而且 9B 大索引配 0.6B 小模型查詢的混搭模式值得注意，能在不犧牲太多準確率的前提下壓低查詢端的運算成本。若場景以純影像比對為主，仍建議評估 Gemini Embedding 2 等競品。
🔗 來源
- 標題：Perplexity AI Releases pplx-embed-v2-late: A 0.6B Edge Model and a 9B Model Scoring 92.4% on MADQA
- 作者／機構：Asif Razzaq（MarkTechPost）
- 連結：https://www.marktechpost.com/2026/10/07/perplexity-ai-releases-pplx-embed-v2-late-a-0-6b-edge-model-and-a-9b-model-scoring-92-4-on-madqa/
#Perplexity #EmbeddingModel #MultimodalAI #ColBERT #RAG #OpenSource #HuggingFace #InformationRetrieval #VectorSearch #EdgeAI
原始資料 MarkTechPost · 收集於 2026-10-08
摘要原文
Perplexity has released pplx-embed-v2-late , a pair of ColBERT-style multimodal embedding models. They come in 2 sizes: 0.6B for fast, cheap queries and 9B for maximum quality. Both models retrieve text, images and rendered PDF pages, and they share one embedding space. Is it deployable? Yes, if you host it yourself. Both models are on Hugging Face under the MIT license. A hosted Perplexity API endpoint is planned but not live. Perplexity designed the 0.6B model as a lightweight query encoder that can also run on edge devices. Both model cards show CUDA GPU usage. They need sentence-transformers >= 6.0.0 and transformers >= 5.4.0 . The memory figures are our estimate at 2 bytes per parameter, for weights only. The published checkpoints are stored in F32, which doubles the download size. All numbers below are from Perplexity’s announcement : The strongest result: 92.4% on MADQA, set by the 9B model. The biggest margin is on BrowseComp+, at 8.7pp over the best dense model. The weakest result: 61.2% on ViDoRe v3 Markdown, from the 0.6B model. It is still the 2nd-best score on that benchmark. Image search is the real gap: Gemini Embedding 2 beats the 9B model on MIRACL-Vision and by 2pp on PPLX-Q2I. Mixing sizes: a 9B index queried by the 0.6B model scored 63.5% on ViDoRe v3 image retrieval. That beats 62.3% with 0.6B on both sides, at the same query cost. Dense models compress a document into 1 vector. pplx-embed-v2-late instead keeps a 128-dim vector for every token. It scores with MaxSim: each query token finds its best document token, and those maxima are summed. Pages are encoded as images, so no OCR step is needed. Perplexity distilled both models from an 18B teacher using LEAF -style token-level training. That training is what creates the shared space. 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 Perplexity AI Releases pplx-embed-v2-late: A 0.6B Edge Model and a 9B Model Scoring 92.4% on MADQA appeared first on MarkTechPost .
由 claude-code/sonnet 自動生成