# Hugging Face explains pooling-aware training for compressed multi-vector retrieval

> Original title: Token Pooling-Aware Training for Multi-Vector Embedding Models

- Source: Hugging Face Blog
- Published: 2026-11-11T00:00:00.000Z
- Bad Influence: https://badinfluence.news/items/b3rno34w0h3ibh53brzl8vws2
- Original: https://huggingface.co/blog/token-pooling

## Summary

Hugging Face explains how token pooling shrinks multi-vector embedding stores and how pooling-aware training preserves retrieval quality at the same vector budget. The post says a normally trained model retains 94.6% of unpooled NanoBEIR nDCG@10 at pool_factor=4, while pooling-aware training raises retention to 97.8%.