研究人员在代理语言模型的多教师同策略蒸馏中发现了一个“行为杠杆失衡”问题。这种失衡会导致模型过度调用工具，即使在直接回答更合适的情况下也是如此，而这个问题无法通过聚合损失指标来检测。为了解决这个问题，提出了一种名为Soft Clamp的新方法，该方法对每令牌发散进行校准以减轻极端信号。实验表明，Soft Clamp显著减少了过度调用，并提高了APIGen-MT等基准测试的决策准确性。 AI
影响 这项研究可能有助于开发更可靠的代理语言模型，使其能更好地理解何时使用工具以及何时提供直接答案。
排序理由 该集群包含一篇详细介绍新AI模型训练方法的学术论文。
AI 生成摘要 · Google Gemini · 来自 3 个来源。 我们如何撰写摘要 →
arXiv:2607.07050v1 Announce Type: new Abstract: Agentic language models must learn when to call tools, when to consume tool responses, and when to answer directly. This makes multi-teacher on-policy distillation a natural training strategy: one teacher can specialize in tool call…
Agentic language models must learn when to call tools, when to consume tool responses, and when to answer directly. This makes multi-teacher on-policy distillation a natural training strategy: one teacher can specialize in tool calls, another in direct responses, and the student …
Agentic language models must learn when to call tools, when to consume tool responses, and when to answer directly. This makes multi-teacher on-policy distillation a natural training strategy: one teacher can specialize in tool calls, another in direct responses, and the student …