#embedding

8 stories · 1 in 7 days · 3 in 30 days

RAG 3 向量检索 3 Qdrant 2 向量搜索 2 arXiv论文 1 BGE 1 ColBERT 1 Hugging Face 1

Stories, products and papers tagged "embedding", newest first.

Nov 11

How-to BuildersOfficial22:49

Chinese contentToken Pooling-Aware Training：给多向量嵌入模型瘦身

ColBERT 这类多向量检索很好用，但每个 token 都存一个向量，成本吃不消。这篇博客讲怎么在训练时就学会合并 token 向量，少存很多还不掉点。自己在搭 RAG 检索的工程师可以看看。

#ColBERT #embedding #RAG #向量检索

Primary sourceHugging Face: BlogSource

Nov 4

How-to BuildersSelected00:40

Milvus 3.0 Walkthrough: Swapping Embedding Models with Zero Downtime

The non-obvious insight is that embedding model swaps are a data-migration problem: old vectors must stay searchable until the new index is fully backfilled and validated, or search quality collapses mid-rollout. Teams running RAG or semantic search in production can follow the walkthrough to plan migrations with built-in evaluation and rollback points, comparing their own Recall numbers against the SciFact baseline before committing.

#Milvus #BGE #RAG #vector database

Milvus@milvusioSource

Oct 22

Papers Builders09:39

Chinese contentHyperspherical Simplex 表征：让模型升级后检索免重建索引

如果你维护检索系统，模型一升级就得重做全量索引，这篇论文给了个不用重索引的思路，连旧模型都不用依赖，很实用。

#向后兼容 #向量检索 #embedding #arXiv论文

arXiv cs.LG@Simone Ricci et al. (4)Source

Oct 13

How-to Builders23:35

Chinese contentWeaviate 发文讲解如何为任务选择 embedding 模型

Weaviate 这篇教你按自己的数据选 embedding 模型，还拿 OpenAI 大小模型实测对比，方法可直接照抄。

#embedding #Weaviate #OpenAI #RAG

Weaviate@weaviate_ioSource

Sep 29

Models BuildersSelected14:15

Qdrant's Nano and Zero Embeddings Work With Stella Indexes

Embedding compatibility across model sizes means teams can swap between a fast, cheap query encoder and a higher-quality one without re-embedding their corpus, which cuts both cost and maintenance work. Engineers running RAG or search pipelines with tight latency budgets should evaluate where Nano and Zero land against their current setup.

#Qdrant #Stella #embeddings #RAG