---
title: "Discussion on \"Reranking for RAG: Using Cross-Encoders to Improve Top Results\" | Hashnode"
description: "Discussion on \"Reranking for RAG: Using Cross-Encoders to Improve Top Results\". A beginner-friendly look at why first-pass retrieval is rough, how cross-encoder rerankers refine the list, and how to balance candidate count vs latency."
url: "https://hashnode.com/posts/reranking-for-rag-using-cross-encoders-to-improve-top-results/6af9a8b82b7173cdce112615"
---

[Feed](https://hashnode.com/feed) /

Discussion

# [Reranking for RAG: Using Cross-Encoders to Improve Top Results](https://whoai.hashnode.dev/reranking-for-rag-using-cross-encoders-to-improve-top-results?utm_source=hashnode&utm_medium=feed)

A beginner-friendly look at why first-pass retrieval is rough, how cross-encoder rerankers refine the list, and how to balance candidate count vs latency.

2d ago · 4 min read · whoai.hashnode.dev  
Read the article

-   [#ai](https://hashnode.com/tag/ai)
-   [#search](https://hashnode.com/tag/search)
-   [#rag](https://hashnode.com/tag/rag)
-   [#reranking](https://hashnode.com/tag/reranking)

## Written by

[Hu Jin](https://hashnode.com/@whoai)

@whoai

  
[View profile](https://hashnode.com/@whoai)

Software engineer writing about AI coding tools, RAG, and Azure AI Search.

Software engineer writing about AI coding tools, RAG, and Azure AI Search.

## Responses 1

Markdown supported · ⌘↵ to post

[Ahmet Özel](https://hashnode.com/@ahmetozel)

[17h ago](https://hashnode.com/posts/reranking-for-rag-using-cross-encoders-to-improve-top-results/6af9a8b82b7173cdce112615/comment/6afb74f9ba8669b3bad273fc)

The question-and-chunk pair has a token budget of its own, which is worth showing beside the 20 to 100 candidate starting range. A candidate can enter the shortlist and still lose its answer-bearing tail when the cross-encoder tokenizer truncates the pair.

I would log the tokenized pair length and whether truncation occurred, then include a fixture whose relevant sentence lies just beyond that limit. That would distinguish a ranking failure from an input-construction failure before increasing candidate count or blaming the reranker model's relevance judgment.
