Compact ai model improves long-term memory writing across frameworks
Hy-Memory-30B-A3B: Attributed Reinforcement Learning for Long-Term Memory Writing
Computation and Language
Summary
Language models have a limited memory window, so they rely on external memory systems to remember important information over time. The authors present a new model called Hy-Memory-30B-A3B that learns to decide what to remember and how to store it effectively across different memory systems. This model is trained using reinforcement learning and performs better than some existing large models on multiple tests. Their approach helps keep the model’s training aligned with how it is used in practice.
What this means in practice
- •For natural language processing engineers: Enhance language models with improved persistent memory management across different memory frameworks to boost long-document understanding.
- •For chatbot development teams: Build chatbots that retain relevant conversation history more efficiently over long interactions by integrating the Hy-Memory-30B-A3B writer.
- •For enterprise knowledge management teams: Improve automated long-term document summarization and retrieval systems by applying reinforcement-learned memory writing policies.$Commercial implications: Enables sale of advanced memory-enhanced document assistants that outperform current tools in storing and updating enterprise knowledge.
Authors
Tianxiang Fei, Jiazheng Kang, Mao Zheng, Xingyan He, Xiang Yu
Abstract
External-memory systems extend language-model applications beyond a context window, but their quality depends critically on the model executing the write path: deciding what to retain, how precisely to represent it, and how new evidence should update persistent state. Existing frameworks typically place a general-purpose large language model behind framework-specific prompts. This raises serving cost, and a writer specialized for one protocol does not transfer directly to another; a unified model that can operate across established memory frameworks remains largely unexplored. We present Hy-Memory-30B-A3B, a compact long-term-memory writer trained across Mem0, MemOS, and Hy-Memory. Memory On-Policy Distillation teaches framework-specific write protocols on student-generated states; Memory RL then optimizes downstream answerability using dense per-chunk probes, attributed cross-chunk rewards, over-storage controls, and chunk-level GRPO. Carrying policy-written banks across windows keeps training aligned with deployment. On fixed 20% benchmark subsamples, Hy-Memory-30B-A3B ranks first on all four evaluated harnesses, including held-out EverMemOS, with a 55.49% macro average versus 54.62% for Claude Opus 5.