Eliciting In-context Retrieval and Reasoning for Long-context Large Language Models

TL;DR

This paper introduces a new benchmark, ICR^2, to evaluate long-context language models in realistic retrieval and reasoning tasks, and proposes methods that significantly improve their performance, even surpassing GPT-4-Turbo on some tasks.

Contribution

The paper presents a new benchmark ICR^2 for realistic evaluation of LCLMs and introduces three methods to enhance their retrieval and reasoning capabilities.

Findings

Significant performance improvements on LOFT and ICR^2 benchmarks.

Best method outperforms GPT-4-Turbo on most tasks.

Methods yield +15 to +17 points in Exact Match scores.

Abstract

Recent advancements in long-context language models (LCLMs) promise to transform Retrieval-Augmented Generation (RAG) by simplifying pipelines. With their expanded context windows, LCLMs can process entire knowledge bases and perform retrieval and reasoning directly -- a capability we define as In-Context Retrieval and Reasoning (ICR^2). However, existing benchmarks like LOFT often overestimate LCLM performance by providing overly simplified contexts. To address this, we introduce ICR^2, a benchmark that evaluates LCLMs in more realistic scenarios by including confounding passages retrieved with strong retrievers. We then propose three methods to enhance LCLM performance: (1) retrieve-then-generate fine-tuning, (2) retrieval-attention-probing, which uses attention heads to filter and de-noise long contexts during decoding, and (3) joint retrieval head training alongside the generation head. Our evaluation of five well-known LCLMs on LOFT and ICR^2 demonstrates significant gains with our best approach applied to Mistral-7B: +17 and +15 points by Exact Match on LOFT, and +13 and +2 points on ICR^2, compared to vanilla RAG and supervised fine-tuning, respectively. It even outperforms GPT-4-Turbo on most tasks despite being a much smaller model.