---
title: "AI4AI测试时框架：通过脚手架实现强到弱能力迁移"
source: "arXiv AI/CL/LG（论文）"
source_tier: "t2"
url: "https://arxiv.org/abs/2608.12307v1"
permalink: "https://feed.astralor.com/items/itm_6d6bf8549af65828e5fb103a"
published: "2026-08-12T17:53:18.000Z"
score: 76
channel: "ai"
tags: ["arXiv","强到弱迁移","测试时脚手架","能力蒸馏"]
---

# AI4AI测试时框架：通过脚手架实现强到弱能力迁移

## AI 点评

> 这篇论文最值得看的地方在于把“脚手架”从系统工程词变成可量化的迁移机制，而且收益接近翻倍，让人觉得训练时蒸馏的很多成本或许可以省下来了。对做模型部署和推理优化的工程师来说，这是必须跟踪的方向；对研究蒸馏的学者，它也会提醒你测试时设计可能比参数更新更便宜。

## 速览

该论文提出一种测试时的能力迁移方法，由强模型构建推理脚手架帮助弱模型提升任务表现，无需更新弱模型参数。在四个心智理论基准上，平均目标模型性能从0.49提升至0.91，接近翻倍。分析表明，收益主要源于将不稳定的推理外化为确定性代码、任务路由和严格格式约束，而非鼓励模型更深入推理或广泛采样。研究发现构建者模型的推理努力单调提升脚手架质量，平台效应相对较小，且弱模型受益最大。这项工作表明测试时脚手架设计是传统蒸馏的重要补充。

## 关键点

- 无需更新弱模型参数，通过强模型构建推理脚手架实现测试时能力迁移。
- 在四个心智理论基准上，平均性能从0.49提升至0.91，接近翻倍。
- 收益主要来自确定性代码外化、任务路由和严格输出格式约束。
- 弱模型受益最大，且构建者推理努力单调提升脚手架质量。

## 全文

Recent work on distillation transfers the capabilities of large models to smaller ones often by updating the latter's parameters, through teacher forcing, on-policy distillation, and related training-time methods. In this paper, we ask whether such transfer can instead occur at test time. We study strong-to-weak scaffolding: whether a stronger builder model can construct inference-time harnesses that help a weaker target model solve tasks more reliably without any parameter updates. Using four representative Theory-of-Mind benchmarks, each builder model uses 5% of the data as a validation set to iteratively refine its harness over multiple rounds, after which the finalized harness is evaluated on the full test set. Empirically, this form of test-time capability transfer is highly effective, nearly doubling average target-model performance from 0.49 to 0.91. Our analysis shows that the gains come primarily from offloading unstable model reasoning into deterministic code, benchmark-specific routing, and strict answer-format enforcement, rather than from encouraging the target model to reason more extensively or sample more broadly. We further find that builder-model reasoning effort improves harness quality monotonically, platform effects are modest relative to the builder model's own capability, and weaker target models receive the largest gains. These results suggest that inference-time harness design is an important complement to conventional training-time distillation, enabling strong models to transfer cognitive structure to weaker models without retraining.

---

引用前请核对原文;本文由星际流动整理:https://feed.astralor.com/items/itm_6d6bf8549af65828e5fb103a