Trajectory-Level Speculation with Semantic Verification for Tool-Using Agents
作者: Zilin Wang, Yinhong Liu, Chao Chen, Qun Liu, Zhijiang Guo
分类: cs.CL, cs.AI, cs.LG
发布日期: 2026-10-15
💡 一句话要点
提出SpecAgent以优化工具使用代理的任务执行效率
🎯 匹配领域: 支柱九：具身大模型 (Embodied Foundation Models)
关键词: 工具使用代理 模型审核 任务执行优化 资源效率 协作协议
📋 核心要点
- 现有方法依赖强模型生成每一步，导致计算资源浪费和延迟增加。
- 提出SpecAgent，通过将强模型转变为审核者，优化弱模型的草稿验证过程。
- 在多个基准测试中，SpecAgent显著降低了内存流量和计算时间，同时保持了高成功率。
📝 摘要（中文）
大型语言模型代理通过长序列的工具调用来解决任务，现有方法通常由单一强模型生成每一步，导致资源浪费。本文提出SpecAgent，一个协作协议，将强模型从每一步的作者转变为弱模型草稿的审核者。弱模型在草拟过程中执行只读工具调用，确保每一步基于真实观察。强模型通过单次预填充验证草拟段落，承诺其判断为语义有效的最长前缀，并修复第一个被拒绝的步骤。实验结果表明，SpecAgent在多个基准测试中显著降低了内存流量和延迟，同时保持了高质量的输出。
🔬 方法详解
问题定义：本文旨在解决大型语言模型代理在任务执行中对计算资源的高需求，现有方法通常由强模型生成每一步，导致效率低下和延迟增加。
核心思路：提出SpecAgent，通过将强模型转变为审核者，允许弱模型草拟步骤并进行验证，从而减少强模型的负担。这样的设计使得每一步都基于真实的观察，确保了输出的有效性。
技术框架：SpecAgent的整体架构包括两个主要模块：弱模型负责草拟步骤并执行只读工具调用，强模型则对草拟的段落进行验证。验证过程通过单次预填充完成，强模型仅需解码短的判决和偶尔的修复。
关键创新：最重要的创新在于将强模型的角色从生成者转变为审核者，这一转变使得验证过程更高效，避免了自回归解码的复杂性。
关键设计：在设计中，弱模型执行只读工具调用，确保每一步的草拟基于真实观察。强模型在验证时，承诺最长的语义有效前缀，并修复第一个被拒绝的步骤，从而优化了整体性能。实验中，SpecAgent在多个基准测试中表现出色，成功率接近强模型。
🖼️ 关键图片
📊 实验亮点
SpecAgent在$τ^2$-Bench上成功率接近强模型，同时内存流量和计算时间减少约50%。在AutomationBench和VitaBench上，SpecAgent恢复了44-75%的质量差距，并在MCP-Atlas边缘云部署中缩小了53-75%的弱强质量差距，强模型生成的token数量减少至一半。
🎯 应用场景
该研究的潜在应用领域包括智能助手、自动化工具和机器人系统等，能够显著提高工具使用代理的任务执行效率，降低计算资源消耗。未来，SpecAgent的设计理念可能会被广泛应用于其他需要高效决策的AI系统中。
📄 摘要（原文）
Large language model agents solve tasks through long sequences of tool calls, and the prevailing design has a single strong model generate every step, although many steps are routine enough for a much smaller model. Existing ways of sharing this work keep the strong model as the generator: routers and collaborative agents decide which model handles each query or step, and speculative methods check a small model's drafts one step at a time, usually against the strong model's own action. We introduce SpecAgent, a collaboration protocol that turns the strong model from the author of each step into a reviewer of steps drafted by a weak model. The weak model executes read-only tool calls as it drafts, so that every step is grounded in real observations, and holds the first action that would modify the environment. The strong model then verifies the drafted segment in a single prefill pass, commits the longest prefix it judges semantically valid, and repairs the first rejected step. Because verification replaces autoregressive decoding with parallel prefill, the strong model decodes only short verdicts and occasional repairs rather than every step. On $τ^2$-Bench, SpecAgent comes within one point of the strong model's success rate while roughly halving memory traffic and wall time, and on AutomationBench and VitaBench it recovers 44-75% of the quality gap while lowering both. In an edge-cloud deployment on MCP-Atlas with three cloud models, it closes 53-75% of the weak-strong quality gap while the strong model generates at most half as many tokens. These results suggest that a strong model need not author every step of an agent: reviewing grounded drafts retains most of its quality while cutting both latency and memory traffic.