Reproducing and Comparing Distillation Techniques for Cross-Encoders

TL;DR

This paper systematically compares various distillation techniques for cross-encoders across different transformer backbones, revealing that relative comparison objectives significantly improve effectiveness, often matching the gains from larger models.

Contribution

It provides a comprehensive reproduction and comparison of distillation strategies for cross-encoders across multiple architectures and datasets, highlighting the importance of objective choice.

Findings

Relative comparison objectives outperform pointwise baselines.

Objective choice can match the benefits of larger backbone architectures.

Distillation strategies are effective across diverse transformer models.

Abstract

Recent advances in Information Retrieval have established transformer-based cross-encoders as a keystone in IR. Recent studies have focused on knowledge distillation and showed that, with the right strategy, traditional cross-encoders could reach the level of effectiveness of LLM re-rankers. Yet, comparisons with previous training strategies, including distillation from strong cross-encoder teachers, remain unclear. In addition, few studies cover a similar range of backbone encoders, while substantial improvements have been made in this area since BERT. This lack of comprehensive studies in controlled environments makes it difficult to identify robust design choices. In this work, we reproduce \citet{schlattRankDistiLLMClosingEffectiveness2025} LLM-based distillation strategy and compare it to \citet{hofstatterImprovingEfficientNeural2020} approach based on an ensemble of cross-encoder teachers, as well as other supervised objectives, to fine-tune a large range of cross-encoders, from the original BERT and its follow-ups RoBERTa, ELECTRA and DeBERTa-v3, to the more recent ModernBERT. We evaluate all models on both in-domain (TREC-DL and MS~MARCO dev) and out-of-domain datasets (BEIR, LoTTE, and Robust04). Our results show that objectives emphasizing relative comparisons -- pairwise MarginMSE and listwise InfoNCE -- consistently outperform pointwise baselines across all backbones and evaluation settings, and that objective choice can yield gains comparable to scaling the backbone architecture.