L-CiteEval: Do Long-Context Models Truly Leverage Context for Responding?

TL;DR

This paper introduces L-CiteEval, a comprehensive benchmark for assessing long-context models' understanding and faithfulness across diverse tasks, revealing current models' limitations in citation accuracy and the benefits of retrieval-augmented generation.

Contribution

L-CiteEval provides a multi-task, fully automated benchmark for evaluating long-context models' understanding and faithfulness, highlighting the gap between open-source and closed-source models.

Findings

Open-source models lag behind closed-source in citation accuracy.

RAG improves faithfulness but slightly reduces generation quality.

Attention mechanisms correlate with citation generation processes.

Abstract

Long-context models (LCMs) have made remarkable strides in recent years, offering users great convenience for handling tasks that involve long context, such as document summarization. As the community increasingly prioritizes the faithfulness of generated results, merely ensuring the accuracy of LCM outputs is insufficient, as it is quite challenging for humans to verify the results from the extremely lengthy context. Yet, although some efforts have been made to assess whether LCMs respond truly based on the context, these works either are limited to specific tasks or heavily rely on external evaluation resources like GPT4.In this work, we introduce L-CiteEval, a comprehensive multi-task benchmark for long-context understanding with citations, aiming to evaluate both the understanding capability and faithfulness of LCMs. L-CiteEval covers 11 tasks from diverse domains, spanning context lengths from 8K to 48K, and provides a fully automated evaluation suite. Through testing with 11 cutting-edge closed-source and open-source LCMs, we find that although these models show minor differences in their generated results, open-source models substantially trail behind their closed-source counterparts in terms of citation accuracy and recall. This suggests that current open-source LCMs are prone to responding based on their inherent knowledge rather than the given context, posing a significant risk to the user experience in practical applications. We also evaluate the RAG approach and observe that RAG can significantly improve the faithfulness of LCMs, albeit with a slight decrease in the generation quality. Furthermore, we discover a correlation between the attention mechanisms of LCMs and the citation generation process.