Lune

ACL Findings2026

MemTR: Enhancing Tool-Calling Reliability via Uncertainty-Triggered FFN-Space Retracing

Hongtao Duan, Lu Jiang, Minying Zhang, Xiaobing Zhu, Tianpeng Bu, Hao Jiang, Xinyu Wei, Lulu Hu

摘要

Tool calling requires Large Language Models (LLMs) to generate structured decisions including tool names and schema-constrained arguments, where small decoding mistakes can cause hard failures.Existing methods either rely on costly tool-use training data or only constrain syntax, leaving tool selection and argument value errors largely unsolved.We analyze tool calling failures through a Where-When lens: (Where) failures correlate with persistent uncertainty in late transformer layers, (When) uncertainty concentrates on content-bearing tokens (tool names and argument values) rather than schema tokens.Based on this, and motivated by evidence that transformer Feed Forward Networks (FFNs) act as key-value style memories that store and retrieve factual or associative mappings, we propose Memory Space Tool Retracing (MemTR), a weight-free decoding-time method that retrieves relevant tool evidence from the tool library and mixes it into the FFN-output at the uncertain layer, treating FFNs as key-value memories.Through extensive experiments on various model families (Qwen, Llama, and xLAM) and benchmarks (BFCL, ACEBench, APIBank), MemTR reduces tool calling failures by 2%-9% with only 1%-2% runtime overhead, without any fine-tuning or additional tool-use training data.

问 Lune

几分钟读透这篇 Paper,不用再花一下午。

Lune 能读完全文,也能读引用它的顶会 Paper,讲清它的结论、证据和局限,每一点都注明出自哪篇 Paper。不用硬啃,也不靠猜。

可以从这些问题问起

Lune 可以查阅

全文 · 相关的 Paper

获取答案

免费开始,无需绑卡

脉络

  • Reflexion: language agents with verbal reinforcement learning 引用了 Toolformer: Language Models Can Teach Themselves to Use Tools
  • ToolLLM: Facilitating Large Language Models to Master 16000+ Real-world APIs 引用了 Reflexion: language agents with verbal reinforcement learning
  • ToolLLM: Facilitating Large Language Models to Master 16000+ Real-world APIs 引用了 Toolformer: Language Models Can Teach Themselves to Use Tools
它引用的引用它的从左到右按年份排列,引用越多,位置越高。

它引用的 Paper17

  1. Toolformer: Language Models Can Teach Themselves to Use ToolsTimo Schick, Jane Dwivedi-Yu, Roberto Dessì, Roberta Raileanu 等NeurIPS 2023 · 5,989 次引用
  2. Reflexion: language agents with verbal reinforcement learningNoah Shinn, Federico Cassano, Ashwin Gopinath, Karthik Narasimhan 等NeurIPS 2023 · 5,828 次引用
  3. The Curious Case of Neural Text DegenerationAri Holtzman, Jan Buys, Li Du, Maxwell Forbes 等ICLR 2020 · 4,112 次引用
  4. ToolLLM: Facilitating Large Language Models to Master 16000+ Real-world APIsYujia Qin, Shihao Liang, Yining Ye, Kunlun Zhu 等ICLR 2024 · 1,469 次引用
  5. ReTool: Reinforcement Learning for Strategic Tool Use in LLMsJiazhan Feng, Shijue Huang, Xingwei Qu, Ge Zhang 等ICLR 2026 · 406 次引用

相关 Paper

  1. Self-Guided Function Calling in Large Language Models via Stepwise Experience RecallSijia Cui, Aiyao He, Shuai Xu, Hongming Zhang 等EMNLP Findings 2025 · 2 次引用
  2. MemoryLLM: Plug-n-Play Interpretable Feed-Forward Memory for TransformersAjay Jaiswal, Lauren Hannah, Han-Byul Kim, Duc Hoang 等ICML 2026 · 3 次引用
  3. Dynamic Tool Dependency Retrieval for Lightweight Function CallingBhrij Patel, Davide Belli, Amir Jalalirad, Maximilian Arnold 等ACL Findings 2026
  4. Memory-augmented Query Reconstruction for LLM-based Knowledge Graph ReasoningMufan Xu, Gewen Liang, Kehai Chen, Wei Wang 等ACL Findings 2025 · 2 次引用
  5. ToolPRM: Fine-Grained Inference Scaling of Structured Outputs for Function CallingJianghao Lin, Yuanyuan Shi, Xin Peng, Renjie Ding 等ACL 2026 · 3 次引用
黄昏的海面,两侧是细线勾勒的悬崖