跨模型记忆迁移：目标侧读取器适配的关键作用

AI 摘要

该研究探讨了跨模型冻结记忆迁移的有效性，发现仅冻结记忆内容不足以实现迁移，目标侧读取器的适配至关重要。实验表明，通过训练轻量级读取器，冻结记忆可显著提升不同目标模型的性能，相对困惑度降低最高达15.7%。研究提出Engram式哈希记忆可作为可重用的外部知识工件，但需目标模型具备兼容的读取接口。

正文节选

Cross-Model Memory Transfer via Target-Side Reader Adaptation Abstract Methods for improving knowledge use in large language models typically fall into two regimes. Non-parametric retrieval offers flexible access to external knowledge, but adds retrieval latency, context overhead, and only shallow integration with the backbone. Parametric adaptation is efficient at inference time, but entangles knowledge with model weights and can be hard to update, audit, or transfer. Engram-style hashed memory

发布时间：2026-08-19 12:00

抓取时间：2026-08-19 12:09

来源机构：arXiv