Beyond a Million Tokens: Benchmarking and Enhancing Long-Term Memory in LLMs

TL;DR

This paper introduces BEAM, a new benchmark for long-term memory in LLMs, and LIGHT, a framework inspired by human cognition that enhances models with multiple memory systems, significantly improving their performance on long conversations.

Contribution

The paper presents a novel benchmark (BEAM) for evaluating long-term memory in LLMs and proposes LIGHT, a memory-augmented framework that improves model performance on long dialogues.

Findings

LIGHT improves LLM performance by up to 12.69%.

Models with 1M token context struggle with long dialogues.

Ablation confirms each memory component's effectiveness.

Abstract

Evaluating the abilities of large language models (LLMs) for tasks that require long-term memory and thus long-context reasoning, for example in conversational settings, is hampered by the existing benchmarks, which often lack narrative coherence, cover narrow domains, and only test simple recall-oriented tasks. This paper introduces a comprehensive solution to these challenges. First, we present a novel framework for automatically generating long (up to 10M tokens), coherent, and topically diverse conversations, accompanied by probing questions targeting a wide range of memory abilities. From this, we construct BEAM, a new benchmark comprising 100 conversations and 2,000 validated questions. Second, to enhance model performance, we propose LIGHT-a framework inspired by human cognition that equips LLMs with three complementary memory systems: a long-term episodic memory, a short-term working memory, and a scratchpad for accumulating salient facts. Our experiments on BEAM reveal that even LLMs with 1M token context windows (with and without retrieval-augmentation) struggle as dialogues lengthen. In contrast, LIGHT consistently improves performance across various models, achieving an average improvement of 3.5%-12.69% over the strongest baselines, depending on the backbone LLM. An ablation study further confirms the contribution of each memory component.