Lattice：8MB 静态检索模型，7 分钟嵌入维基百科
原标题：Lattice: an 8 MB static retriever that embeds Wikipedia in 7 minutes
AI 摘要
Hugging Face 博客发布了一篇关于静态检索模型 Lattice 的文章。作者 ErikKaum 训练了 lattice-retrieval 模型，在 660M 精选查询/文档对上训练，BEIR 基准上 NDCG@10 达到 0.4749，优于参考模型。该模型仅 8MB，可在 8 核 M2 MacBook Air 上 7 分 26 秒嵌入整个英文维基百科。文章探讨了静态模型的量化、训练效率和性能极限。
正文节选
- I trained lattice-retrieval , a static embedding model, on 660M curated query/document pairs. It scores 0.4581 NDCG@10 on decontaminated BEIR before fine-tuning and 0.4749 after fine-tuning, compared with 0.4334 forsentence-transformers/static-retrieval-mrl-en-v1 . - Static models are unusually forgiving quantization targets. The best quality/size trade-off I found is int4-row at 512 dimensions: a 7.94 MB weight file that scores 0.4697, effectively the same as fp32 at the same dimension. - I b