---
id: 20260509-T0-05
title: "多个小模型协作替代单个大模型，SAT训练法保证每次更新都比上一次强"
title_en: "SAT Coordinates Multiple Small LLMs to Match Large Models with Guaranteed Improvement"
url: https://ai.daily.yangsir.net/daily/20260509-T0-05
issue_date: 2026-05-09
publish_date: 2026-05-08T04:00:00.000Z
source_name: "arXiv cs.LG (ML)"
source_url: https://arxiv.org/abs/2605.05216
---

# 多个小模型协作替代单个大模型，SAT训练法保证每次更新都比上一次强

部署超大参数LLM成本高昂。替代方案是用多个小模型协作，但多模型联合训练很难保证每个模型持续进步而不退化。SAT（Sequential Agent Tuning）让多个LLM在无中心协调器的情况下依次训练，并给出单调提升的理论保证——每轮训练后整体性能不会回退。实验显示，小模型团队可以匹敌甚至超过单个大模型，且部署成本大幅降低。适合资源有限但需要强推理能力的团队。

## English Version

**SAT Coordinates Multiple Small LLMs to Match Large Models with Guaranteed Improvement**

Deploying massive LLMs is prohibitively expensive. SAT (Sequential Agent Tuning) trains teams of smaller LLMs without a central coordinator, with theoretical guarantees of monotonic improvement—performance never regresses after each round. Experiments show small model teams can match or exceed single large models at a fraction of deployment cost.

---

**来源**：[arXiv cs.LG (ML)](https://arxiv.org/abs/2605.05216)

**详情页**：https://ai.daily.yangsir.net/daily/20260509-T0-05

---

*智语观潮 · Daily — https://ai.daily.yangsir.net/llms.txt*