MeMo: Memory as a Model

2026-05-14

AI Summary

This work addresses the challenge that large language models, once pretrained, have fixed parameters and struggle to efficiently incorporate real-time, domain-specific knowledge. To overcome this limitation, the authors propose MeMo, a framework that encodes new information in a dedicated memory module and integrates it into the inference process without requiring access to model weights or logits, thereby avoiding fine-tuning. MeMo supports modeling of complex cross-document relationships, exhibits robustness to retrieval noise, and prevents catastrophic forgetting. Notably, its retrieval overhead during inference is independent of the corpus size. Empirical evaluations on BrowseComp-Plus, NarrativeQA, and MuSiQue demonstrate that MeMo consistently achieves significant performance gains over existing methods.

Abstract

Large language models (LLMs) achieve strong performance across a wide range of tasks, but remain frozen after pretraining until subsequent updates. Many real-world applications require timely, domain-specific information, motivating the need for efficient mechanisms to incorporate new knowledge. In this paper, we introduce MeMo (Memory as a Model), a modular framework that encodes new knowledge into a dedicated memory model while keeping the LLM parameters unchanged. Compared to existing methods, MeMo offers several advantages: (a) it captures complex cross-document relationships, (b) it is robust to retrieval noise, (c) it avoids catastrophic forgetting in the LLM, (d) it does not require access to the LLM's weights or output logits, enabling plug-and-play integration with both open and proprietary closed-source LLMs, and (e) its retrieval cost is independent of corpus size at inference time. Our experimental results on three benchmarks, BrowseComp-Plus, NarrativeQA, and MuSiQue, show that MeMo achieves strong performance compared to existing methods across diverse settings.

Problem

• large language models
• knowledge updating
• memory integration
• domain-specific information
• frozen parameters

Innovation

• MeMo
• memory-augmented LLM
• parameter-free adaptation
• retrieval robustness
• plug-and-play integration