---
title: "Comment by Ahmet Özel on \"Reranking for RAG: Using Cross-Encoders to Improve Top Results\" | Hashnode"
description: "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 short"
url: "https://hashnode.com/posts/reranking-for-rag-using-cross-encoders-to-improve-top-results/6af9a8b82b7173cdce112615/comment/6afb74f9ba8669b3bad273fc"
---

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

Comment [Reranking for RAG: Using Cross-Encoders to Improve Top Results](https://hashnode.com/posts/reranking-for-rag-using-cross-encoders-to-improve-top-results/6af9a8b82b7173cdce112615)

[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.
