Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

No Parameter Left Behind: How Distillation and Model Size Affect Zero-Shot Retrieval

Guilherme Moraes Rosa, Luiz Bonifacio, Vitor Jeronymo, Hugo Abonizio, Marzieh Fadaee, Roberto Lotufo, Rodrigo Nogueira

TL;DR

The paper investigates how distillation and parameter count affect zero-shot retrieval. It compares BM25, MiniLM-L6, and monoT5 variants across MS MARCO fine-tuning and BEIR zero-shot tasks. It finds that larger parameter counts improve zero-shot generalization, and large rerankers achieve state-of-the-art BEIR performance, outperforming dense models of similar size. It argues that in-domain effectiveness is not a reliable proxy for zero-shot transfer and highlights the limitations of dense retrieval in cross-domain scenarios.

Abstract

Recent work has shown that small distilled language models are strong competitors to models that are orders of magnitude larger and slower in a wide range of information retrieval tasks. This has made distilled and dense models, due to latency constraints, the go-to choice for deployment in real-world retrieval applications. In this work, we question this practice by showing that the number of parameters and early query-document interaction play a significant role in the generalization ability of retrieval models. Our experiments show that increasing model size results in marginal gains on in-domain test sets, but much larger gains in new domains never seen during fine-tuning. Furthermore, we show that rerankers largely outperform dense ones of similar size in several tasks. Our largest reranker reaches the state of the art in 12 of the 18 datasets of the Benchmark-IR (BEIR) and surpasses the previous state of the art by 3 average points. Finally, we confirm that in-domain effectiveness is not a good indicator of zero-shot effectiveness. Code is available at https://github.com/guilhermemr04/scaling-zero-shot-retrieval.git

No Parameter Left Behind: How Distillation and Model Size Affect Zero-Shot Retrieval

TL;DR

The paper investigates how distillation and parameter count affect zero-shot retrieval. It compares BM25, MiniLM-L6, and monoT5 variants across MS MARCO fine-tuning and BEIR zero-shot tasks. It finds that larger parameter counts improve zero-shot generalization, and large rerankers achieve state-of-the-art BEIR performance, outperforming dense models of similar size. It argues that in-domain effectiveness is not a reliable proxy for zero-shot transfer and highlights the limitations of dense retrieval in cross-domain scenarios.

Abstract

Recent work has shown that small distilled language models are strong competitors to models that are orders of magnitude larger and slower in a wide range of information retrieval tasks. This has made distilled and dense models, due to latency constraints, the go-to choice for deployment in real-world retrieval applications. In this work, we question this practice by showing that the number of parameters and early query-document interaction play a significant role in the generalization ability of retrieval models. Our experiments show that increasing model size results in marginal gains on in-domain test sets, but much larger gains in new domains never seen during fine-tuning. Furthermore, we show that rerankers largely outperform dense ones of similar size in several tasks. Our largest reranker reaches the state of the art in 12 of the 18 datasets of the Benchmark-IR (BEIR) and surpasses the previous state of the art by 3 average points. Finally, we confirm that in-domain effectiveness is not a good indicator of zero-shot effectiveness. Code is available at https://github.com/guilhermemr04/scaling-zero-shot-retrieval.git
Paper Structure (7 sections, 2 figures, 2 tables)

This paper contains 7 sections, 2 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Experiments
  4. Results
  5. Latency
  6. Conclusion
  7. Appendix

Figures (2)

  • Figure 1: Distilled rerankers such as MiniLM have similar in-domain effectiveness to models 100 times larger (top). However, large rerankers such as monoT5-3B outperform distilled ones and dense models of equivalent size on zero-shot tasks (bottom).
  • Figure 2: Model size vs effectiveness on in-domain (MS MARCO) vs out-of-domain (others) data. Effectiveness increase with respect to the number of model parameters only on out-of-domain datasets.