让模型自动生成并优化推理策略，提升复杂任务解决能力。
Scaffold-Mediated Post-Training: Co-Evolving Model Parameters and Procedural Scaffold Graphs
- 将推理流程构建成可演化图谱，与模型参数共同优化
- 自动发现技能使通过率提升8.1个百分点，蒸馏后仍保持27.7%成功率
- 适合需要复杂逻辑推理的模型增强场景
大语言模型的后训练仅优化参数，而推理时使用的程序化框架通常独立设计，导致难以自动获取和内化复杂策略。本文提出架构建模后训练：将程序化框架组织为可演化的图结构，与模型参数通过发现、蒸馏和动态重组协同进化。我们将其具体实现为技能训练（Skill Training）。在FeatureBench上，自动发现的技能使通过率提升8.1个百分点；经过渐进式蒸馏后，模型无需外部框架仍能达到27.7%的通过率（蒸馏保留率为85.2%，定义为蒸馏后/有技能时通过率），显著优于同数据集上的标准监督微调（SFT）。
原文摘要 · Abstract (English)
Post-training of large language models optimizes only parameters, while inference-time procedural scaffolds are typically designed independently of parameter training. This disconnect makes it difficult to automatically acquire and internalize complex strategies. We propose scaffold-mediated post-training: procedural scaffolds are organized into an evolvable graph structure that co-evolves with model parameters through discovery, distillation, and dynamic recompilation. We instantiate this paradigm as Skill Training. On FeatureBench, automatically discovered skills improve the passed rate by 8.1pp, and after progressive distillation the model still achieves a 27.7% passed rate without any external scaffold (distillation retention rate 85.2%, defined as post-distillation / with-skill passed rate), significantly outperforming standard SFT on the same data.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。