跳到正文
arXiv cs.AI· Fanyu Zhao, Ruike Cao, Liang Dong, Fugen Yao, Jian Xu, Guanjun Jiang, Yifei Zhao, Yinsheng Li, Han Zhang·· 4 小时前AI 评分38

RPMem:为 LLM 智能体学习跨会话的长期循环参数化记忆

RPMem: Learning Long-Term Recurrent Parametric Memory Across Sessions for LLM Agents

AI 导读

RPMem 是一个两阶段架构,将每次会话通过前向计算编译为与模型无关的潜在记忆,再用任务训练的循环门控选择性整合,最终映射为骨干模型专属的 LoRA 参数,使记忆在更换骨干后仍可复用。在 Qwen3-8B 上于 PERMA 基准达到 85.52%,分别超出最强参数化基线和文本基线 5.32 与 12.98 个百分点,并在三个长期记忆基准、五种骨干上验证了近恒定的更新成本与内存占用。

正文 · 原文

View PDF HTML (experimental)

Abstract:Long-running LLM agents require memory that persists and evolves across sessions. Text-based memory retrieves and reconstructs past interactions at every query, making long-horizon performance increasingly dependent on retrieval quality and contextual reasoning as histories grow. Parametric memory encodes experience directly into model computation, but existing approaches provide limited support for cross-session memory evolution. Their coupling to a specific backbone further restricts memory reuse after model replacement. We introduce RPMem, a two-stage architecture that compiles each session into a model-independent latent memory through forward computation and selectively integrates it with retained memory via a task-trained recurrent gate. The consolidated memory is then mapped to backbone-specific low-rank adaptation (LoRA) parameters, allowing the encoding capability to transfer when the backbone is replaced. Evaluation across three long-term memory benchmarks and five diverse backbones demonstrates broad generalization with near-constant update cost and memory footprint. With Qwen3-8B on PERMA, RPMem reaches 85.52%, outperforming the strongest parametric and text-based baselines by 5.32 and 12.98 percentage points, respectively. Ablations validate the complementary roles of session compilation and cross-session consolidation, while dynamics analyses reveal that the gate acquires task-specific memory integration strategies. These results establish RPMem as a lifecycle-independent parametric memory framework that maintains evolving cross-session memory that remains reusable across backbone replacements. Our code is available at this https URL.
Comments: 45 pages, 7 figures. Code: this https URL
Subjects: Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
Cite as: arXiv:2609.23466 [cs.CL]
  (or arXiv:2609.23466v3 [cs.CL] for this version)
  https://doi.org/10.48550/arXiv.2609.23466

arXiv-issued DOI via DataCite

Submission history

From: Fanyu Zhao [view email]
[v1] Sun, 20 Sep 2026 08:51:10 UTC (1,434 KB)
[v2] Tue, 22 Sep 2026 10:17:56 UTC (1,388 KB)
[v3] Tue, 17 Nov 2026 07:22:20 UTC (1,427 KB)

来源:arXiv cs.AI · arxiv.org