小模型应理解工具副作用：EffectWorld 基准与训练方法
原标题：Small agents should know what their tools do
AI 摘要
研究者提出 EffectWorld，一个带有副作用类型（纯、读、写、破坏性）和幂等性声明的可执行工具世界，用于训练和评测小型开源语言模型在工具调用中的副作用安全与调用效率。基于该数据训练的 LoRA 适配器使 Qwen3-4B 的任务成功率从 53% 提升到分布内 99%，在未见任务和在线购物域也有显著提升，并迁移到多个模型家族。研究还发现公开工具调用语料中的拒绝模板会削弱模型调用工具的意愿，过滤后可修复 BFCL 准确率。
正文节选
Small Agents Should Know What Their Tools Do Abstract Small open language models can call tools, but they do not reason about what a call does: they delete without asking, repeat writes, re-read state they already hold, and issue one call per turn where a short program would do. We introduce EffectWorld, an executable world of effect-typed tools in which every tool declares its effect (pure, read, write or destructive), whether it is idempotent, and which resources it reads and writes. These dec