arXiv cs.CL· Pengfeng Song, Yunkai Zhuang, Liye Shi·· 9 小时前AI 评分31
LMMR:面向小语言模型智能体的学习式多粒度记忆检索
LMMR: Learned Multigranular Memory Retrieval for Small Language Model Agents
AI 导读
LMMR 是一种学习式多粒度记忆检索方法,可同时选择下一个信息来源及读取长度,让冻结的 2B 答案模型在 400-token 预算下于 HotpotQA、2WikiMultihopQA 和 MuSiQue 分别达到 61.47、43.26、30.58 F1。
正文 · 原文
Abstract:Small language model agents have limited space to read the memory needed for multi-hop questions. Relevant facts are often scattered across records whose surrounding text consumes much of that space. We propose LMMR, a learned multigranular memory retrieval method that selects both the next source and how much of it to read. LMMR links paragraphs and sentence fragments to their original sources and pairs views that supply the same missing support during training. Relevant distractors and repeated information provide contrasting examples. At inference, the reranker scores what each candidate adds to an anchor passage, and a selector chooses the evidence within a token budget. On 22,398 official validation questions from HotpotQA, 2WikiMultihopQA, and MuSiQue, a frozen 2B answer model reaches 61.47, 43.26, and 30.58 F1 under a 400-token budget. The gains over query expansion are 3.06, 2.14, and 4.55 points, while memory input decreases by 23.1-26.9%. LMMR also uses 23.5-27.7% fewer input tokens than a coarse retriever trained on the same questions. Against that retriever followed by LLMLingua-2 at comparable input lengths, F1 gains reach 3.99 points and are significant on HotpotQA and 2WikiMultihopQA after Holm correction. All eight small answer models evaluated on HotpotQA improve over query expansion and MMR in both F1 and exact match.
| Comments: | 18 pages, 11 figures. Code and experimental results: this https URL |
| Subjects: | Computation and Language (cs.CL) |
| Cite as: | arXiv:2611.22425 [cs.CL] |
| (or arXiv:2611.22425v1 [cs.CL] for this version) | |
| https://doi.org/10.48550/arXiv.2611.22425 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Pengfeng Song [view email]
[v1]
Tue, 17 Nov 2026 09:00:31 UTC (579 KB)
来源:arXiv cs.CL · arxiv.org