$\delta$-mem: Efficient Online Memory for Large Language Models

TL;DR

$4mem4 is a lightweight online memory mechanism that enhances large language models' ability to utilize historical information efficiently without full fine-tuning.

Contribution

It introduces $4mem, a compact online associative memory that improves long-term information retention in language models with minimal overhead.

Findings

$4mem$ improves model scores by 1.10x on average over the backbone.

It achieves 1.31x improvement on MemoryAgentBench.

Effective memory can be integrated without full fine-tuning or context extension.

Abstract

Large language models increasingly need to accumulate and reuse historical information in long-term assistants and agent systems. Simply expanding the context window is costly and often fails to ensure effective context utilization. We propose $\delta$-mem, a lightweight memory mechanism that augments a frozen full-attention backbone with a compact online state of associative memory. $\delta$-mem compresses past information into a fixed-size state matrix updated by delta-rule learning, and uses its readout to generate low-rank corrections to the backbone's attention computation during generation. With only an $8\times8$ online memory state, $\delta$-mem improves the average score to $1.10\times$ that of the frozen backbone and $1.15\times$ that of the strongest non-$\delta$-mem memory baseline. It achieves larger gains on memory-heavy benchmarks, reaching $1.31\times$ on MemoryAgentBench and $1.20\times$ on LoCoMo, while largely preserving general capabilities. These results show that effective memory can be realized through a compact online state directly coupled with attention computation, without full fine-tuning, backbone replacement, or explicit context extension.