DisCo：将 GitHub 仓库蒸馏为 AI 技能，显著提升研究代理性能

AI 摘要

Hugging Face 每日论文页面介绍了 DisCo，一个将 GitHub 仓库中的操作知识提炼为可复用技能的研究代理。通过任务无关和任务导向两种蒸馏方式，DisCo 从 1000 个广泛使用的 ML 仓库中提炼出 5000 多个验证技能，形成 AREX-Skill 库。在固定 GPT-5.5 骨干和预算下，配备技能的代理在 MLE-bench、PaperBench、FrontierCS 和 PassNet 上分别提升了 134.3%、34.4%、9.2% 和 14.0% 的性能。

正文节选

Repo-To-Skill: Distilling GitHub Repositories Into AI4AI Skills Abstract DisCo is a research agent that distills operational knowledge into reusable skills, significantly improving autonomous ML research performance across benchmarks. Autonomous agents are beginning to carry out machine-learning (ML) research end to end. These agents combine a model backbone with a harness for planning, execution, memory, and verification, but this architecture still leaves domain-specific know-how outside the a

发布时间：—

抓取时间：2026-09-03 12:27

来源机构：Hugging Face