---
kind: concept
slug: passage-reranking
title: Passage reranking
summary: >-
  Re-scores a short list of retrieved passages with a second model before one is used as context. One card so far: In-Context RALM, where a frozen LM scores the candidates with no training, or a trained RoBERTa reranker picks one. Zero-shot reranking is reported on five corpora, the trained reranker on WikiText-103 only, and the trained variant needs 300,000 examples and a second model.
tags: [retrieval, reranking, language-modeling]
papers: ["2302.00083"]
related: [retrieval-augmented-generation, dense-retrieval, learned-sparse-retrieval]
status: draft
updated_at: 2026-10-19
---

## Overview

A retriever returns several candidate passages for a query. A reranker then
scores those candidates again, and only the best one goes on to the reader. In
the one card so far, the candidates are the top 16 BM25 passages for the text
written so far, and the chosen passage is placed in front of a frozen language
model [arxiv:2302.00083 §6] [arxiv:2302.00083 §3.1]. The language model's
weights do not change; the reranker only decides which passage is used
[arxiv:2302.00083 §6.2].

The paper's numbers are perplexity on language-modelling sets, reported by the
authors for their own setups [arxiv:2302.00083 tab. 1]. Nothing here shows
reranking on question answering.

## Variants

### Zero-shot LM scoring

No training. Each candidate passage is placed before the prefix, and the
language model's log-probability of the last `s'` prefix tokens is computed
under that passage. The candidate with the highest score is kept
[arxiv:2302.00083 §6.1]. The scoring model does not have to be the generating
model [arxiv:2302.00083 §6.1]. The scoring length `s'` was set to 16 after a
search over {4, 8, 16, 32} on WikiText-103 dev [arxiv:2302.00083 §6.1]. Card:
[In-Context Retrieval-Augmented Language Models](/papers/2302.00083).

With BM25 and GPT-2 Small, word-level perplexity on WikiText-103 falls from
29.6 to 28.6 when zero-shot reranking is added. The four other corpora move in
the same direction [arxiv:2302.00083 tab. 1].

### Trained predictive reranker

A RoBERTa-base classifier gives each of the `k = 16` candidates a relevance
score against the prefix. The scores are softmax-normalised over the candidates,
and the best one is chosen [arxiv:2302.00083 §6.2]. Training uses a loss built
from the frozen language model's likelihood of the next stride under each
candidate, so only the reranker is updated [arxiv:2302.00083 §6.2]. The
training set has 300,000 examples from WikiText-103 train, scored with GPT-2
110M. The paper reports 10,000 steps, a peak learning rate of 1e-5 and a batch
size of 32; the schedule, optimizer and dropout are not stated
[arxiv:2302.00083 §6.2].

With BM25 and GPT-2 Small, WikiText-103 word-level perplexity is 26.8, against
28.6 for zero-shot scoring and 29.6 without reranking. With GPT-2 XL, it is 15.4,
against 16.1 and 16.6 [arxiv:2302.00083 tab. 1]. The paper reports this variant
on WikiText-103 only [arxiv:2302.00083 tab. 1].

## When to use which

| Situation | Use | Why |
|---|---|---|
| You can only call the LM, and you have no training data or budget | Zero-shot LM scoring | Needs no training. Costs one scoring pass per candidate, so `k = 16` passes per retrieval [arxiv:2302.00083 §6.1]. |
| You can build a training set from your own text and can run a second, small encoder | Trained predictive reranker | Lower WikiText-103 perplexity than zero-shot in the paper's setup [arxiv:2302.00083 tab. 1]. Needs 300,000 examples, a RoBERTa-base model and training [arxiv:2302.00083 §6.2]. |
| The goal is answer accuracy on open-domain QA | No evidence on this page | The paper's QA table uses the top 2 DPR passages without a reranker [arxiv:2302.00083 tab. 4]. |

## Implementation sketch

The two variants share the same loop: score each candidate against the prefix,
then pick one. They differ in where the score comes from.

```pseudocode
# Zero-shot reranking of k candidate passages  [arxiv:2302.00083 §6.1]
# Input:  prefix tokens, k passages, scoring LM, scoring length s'
# Output: index of the chosen passage

def rerank_zero_shot(prefix, passages, lm, s_prime):
    # the target is the last s' prefix tokens, known at test time  [§6.1]
    target = prefix[-s_prime:]
    context = prefix[:-s_prime]
    # score each passage placed before the context  [§6.1]
    scores = [log_prob(lm, target, p + context) for p in passages]
    return argmax(scores)
```

```pseudocode
# Loss for one training example of the predictive reranker  [arxiv:2302.00083 §6.2]
# Input:  prefix, k passages, next-stride tokens y, reranker f, frozen LM
# Output: scalar loss

def predictive_loss(prefix, passages, y, reranker, lm):
    # relevance scores, softmax-normalised over the k passages  [§6.2]
    weights = softmax([reranker(prefix, p) for p in passages])
    # LM likelihood of the next stride given each passage  [§6.2]
    likes = [log_prob(lm, y, p + prefix) for p in passages]
    # -log of a weighted sum, computed in log space  [§6.2]
    return -logsumexp([log(w) + ll for w, ll in zip(weights, likes)])
```

## Contradictions

None known. Only one card covers this concept, so the page has no
cross-paper disagreement to report, and the evidence rests on one paper's
setup.

## Papers

- [In-Context Retrieval-Augmented Language Models](/papers/2302.00083): applies zero-shot LM scoring and a trained RoBERTa reranker to the top 16 BM25 passages for a frozen GPT-2 family model.