독서리서치레이더투자 프레임워크
로그인 / 회원가입
← 읽기로 돌아가기
뉴스주제 · Qdrant중대2 분 소요DisconfirmAI 편집부

Qdrant Benchmarks: 32K Context Embeddings Still Underperform Chunking in Detail Retrieval

Engineers at Qdrant tested DMIS Lab’s UltraBERT-Embed-150M using the LongEmbed benchmark and 200,000 Wikipedia articles, finding that whole-document vectors underperform 512-token chunks in tasks requiring specific detail localization, while hybrid retrieval failed to consistently improve results.

2026-10-19 08:00 GMT+8

The engineering team at Qdrant used DMIS Lab’s UltraBERT-Embed-150M to investigate a practical question: when an embedding model can process 32K token inputs, is chunking still necessary for retrieval pipelines? They compared single 768-dimensional vectors representing entire documents against 512-token chunked segments across five tasks from the LongEmbed benchmark and 200,000 English Wikipedia articles from MLDR.

Results varied by query type. For tasks requiring precise detail localization—multi-hop QA, meeting summarization, single-sentence retrieval, and large-scale article matching—chunking was superior across the board: on Needle tasks, whole-document vectors achieved only 30% accuracy in identifying the correct article at 32K tokens, whereas 512-token chunks reached 56%; on MLDR, chunking yielded an nDCG@10 of 0.683 compared to 0.509 for whole-document vectors. Whole-document vectors performed slightly better only on SummScreenFD (where queries describe entire TV episodes). Hybrid retrieval, which combines both whole-document and chunk vectors, did not consistently improve outcomes and often performed worse than pure chunking across most tasks.

Qdrant concludes that chunking remains the starting point for detail retrieval, while long-context vectors are better suited for coarse topic-level filtering. 1-bit TurboQuant compression reduces storage costs for 1.43 million chunk vectors from approximately 4 GB to just 140 MB, eliminating storage as a barrier for chunk-based approaches. All data reflects Qdrant’s internal testing, has not been independently reproduced, and covers only English documents; embedding latency was measured on NVIDIA L4 hardware.

로그인 / 회원가입
독서리서치레이더투자 프레임워크

출처

  • qdrant.tech↗
주제

Qdrant

Qdrant 发布了一项针对长上下文嵌入模型 UltraBERT-Embed-150M 的基准测试,旨在验证“32K token 整篇文档向量”是否能取代传统的“分块检索”。【背景】随着大模型上下文窗口扩大,业界曾猜测分块(Chunking…

2 단계1 건업데이트됨 2026-11-02 02:00

현재까지의 과정

  1. 2026-10-19Qdrant 工程师 Dylan Couzon 和 Anush 发布博客《UltraBERT Fits 32K Tokens in One Vector. RIP Chunking?》,介绍使用 DMIS Lab 的 UltraBERT-Embed-150M 模型进行对比测试的方法与初步结论。qdrant.tech ↗
  2. 2026-11-01中文媒体转载并总结该测试,强调在 20 万篇维基百科文章的大规模测试中,整篇文档向量在细节检索上仍逊于分块,且混合检索未能稳定改善结果。aiagentnews.io ↗

아직 해결되지 않은 문제

  • 在 FlashAttention 2 等优化下,整篇文档向量的嵌入耗时差距是否会显著缩小?
  • 对于非英语或低资源语言,长上下文嵌入模型的表现是否同样不如分块策略?
  • 如何设计更智能的路由机制,自动判断何时使用整篇向量,何时使用分块向量?
전체 주제 읽기 →
리서치 계속하기 →