论文：用蒸馏 4B 决策模型替换 Jev，agent 成本降 14.2% 至 89%
教你怎么给 agent 省钱：拿 Jev 蒸馏个 4B 小模型接进 harness，最省能砍掉 89% 成本还几乎不掉成功率。
这篇论文提出 JevAgentHarness，把 agent 的常规决策交给一个从 Jev 蒸馏而来的 4B 开源决策模型。决策模型只在置信度超过基于 agent 自身轨迹拟合的阈值时接管，低于阈值则由 LLM 正常决策。该级联方案在 500 个任务上结果不变，成本降低 14.2%，而固定 0.5 阈值会损失 21.5 分。训练出的 AgentDecision-4B 在陌生任务上让 agent 达到 91.5% 成功率，高于直接用 Jev 的 82.0%；在熟悉任务上成本降低 81.1% 到 89.0%，速度提升最高 2.5 倍，成功率下降不超过 3 分。
Recommended paper integrating decision models like Jev into harnesses. DAIR.AI @dair_ai Must-read paper integrating decision models like Jev into harnesses. They distilled an open 4B decision model, mostly from Jev's answers on the agent's own states, that beats Jev inside that agent and is much cheaper to run. Their harness, JevAgentHarness, sends each routine choice to the decision model and uses its answer only when the confidence clears a threshold fitted on the agent's own trajectories. Below the threshold, the LLM decides as usual. With the fitted threshold, tool and argument cascades cut cost by 14.2% without changing the outcome of any of 500 tasks. A fixed threshold of 0.5 lost 21.5 points. They also train AgentDecision-4B on states the agent reaches after its own mistakes, with Jev's answers as labels. On tasks unlike its training data, the agent solved 91.5% with it and 82.0% with Jev. On familiar tasks, 4B models trained this way made agents 81.1 to 89.0% cheaper and up to 2.5 times faster while staying within 3 points of success. Paper: academy.dair.ai/papers/jevagen… 🔗 View Quoted Tweet 💬 0 🔄 1 ❤️ 1 👀 1648 📊 1 ⚡
