An Empirical Study of Uniform-Architecture Knowledge Distillation in Document Ranking

TL;DR

This study investigates how different loss functions affect the effectiveness of uniform-architecture knowledge distillation for BERT-based document ranking models, highlighting the importance of pairwise loss in training smaller models.

Contribution

It provides empirical insights into the optimal distillation strategies for cross-encoder ranking models, emphasizing the role of pairwise loss functions.

Findings

Pairwise loss of hard labels is crucial for training student models.

Intermediate Transformer layer distillation may reduce performance.

Optimal distillation strategies differ from those in general NLP tasks.

Abstract

Although BERT-based ranking models have been commonly used in commercial search engines, they are usually time-consuming for online ranking tasks. Knowledge distillation, which aims at learning a smaller model with comparable performance to a larger model, is a common strategy for reducing the online inference latency. In this paper, we investigate the effect of different loss functions for uniform-architecture distillation of BERT-based ranking models. Here "uniform-architecture" denotes that both teacher and student models are in cross-encoder architecture, while the student models include small-scaled pre-trained language models. Our experimental results reveal that the optimal distillation configuration for ranking tasks is much different than general natural language processing tasks. Specifically, when the student models are in cross-encoder architecture, a pairwise loss of hard labels is critical for training student models, whereas the distillation objectives of intermediate Transformer layers may hurt performance. These findings emphasize the necessity of carefully designing a distillation strategy (for cross-encoder student models) tailored for document ranking with pairwise training samples.