From 7f4b440598285977a950847141cf40c848c4f8ba Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?Kacper=20=C5=81ukawski?= Date: Tue, 20 Jan 2026 15:44:11 +0100 Subject: [PATCH] Improvements in pooling techniques --- .../multi-vector-search/module-3/pooling-techniques.md | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-) diff --git a/qdrant-landing/content/course/multi-vector-search/module-3/pooling-techniques.md b/qdrant-landing/content/course/multi-vector-search/module-3/pooling-techniques.md index ff60dc3fd..64cfba805 100644 --- a/qdrant-landing/content/course/multi-vector-search/module-3/pooling-techniques.md +++ b/qdrant-landing/content/course/multi-vector-search/module-3/pooling-techniques.md @@ -84,8 +84,8 @@ That's a **32× reduction** in vector count and memory footprint. **Trade-offs to consider:** - **Loss of fine-grained resolution**: Small details that span partial rows may blend together -- **Row pooling** works well for Western text documents where reading flows horizontally -- **Column pooling** better captures vertical structures like tables, sidebars, or Asian language text +- **Row pooling** may work better for horizontally-oriented content, like text +- **Column pooling** may better capture vertical structures like tables, sidebars, or vertically-oriented text - You can combine both (64 vectors) for a balanced approach ```python @@ -142,7 +142,7 @@ This approach adapts to the content itself. For a document with dense text and s You've learned two complementary strategies for reducing the number of vectors per document: -- **Row/column pooling**: Exploits spatial structure in image embeddings for a fixed 32× reduction +- **Row/column pooling**: Exploits spatial structure in image embeddings for a fixed reduction (32x for ColPali) - **Hierarchical pooling**: Content-adaptive clustering that works for any multi-vector representation Combined with quantization from the previous lesson, you can achieve dramatic memory savings: