---
id: 20261031-T0-18
title: "Artifact Distillation：从前沿模型到小语言模型的免训练迁移"
title_en: "Artifact Distillation Enables Training-Free Transfer from Frontier Models to Small Language Models"
url: https://ai.daily.yangsir.net/daily/20261031-T0-18
issue_date: 2026-10-31
publish_date: 2026-10-30T04:00:00.000Z
source_name: "arXiv cs.CL (NLP)"
source_url: https://arxiv.org/abs/2610.37680
---

# Artifact Distillation：从前沿模型到小语言模型的免训练迁移

arXiv:2610.37680v1 公告类型：新论文。摘要：缩小前沿模型与较小语言模型之间的差距通常通过蒸馏来实现：在前沿模型（教师）的输出上微调小模型（学生），这使迁移受限于训练算力，并使得迁移的知识……（摘要原文在此处截断，后续内容缺失）。

## English Version

**Artifact Distillation Enables Training-Free Transfer from Frontier Models to Small Language Models**

arXiv:2610.37680v1 (new paper). Narrowing the gap between frontier models and smaller language models is typically achieved through distillation: fine-tuning a small model (student) on the outputs of a frontier model (teacher). This approach ties transfer to training compute and limits the knowledge that can be transferred. The paper introduces Artifact Distillation, a training-free method for transferring capabilities from frontier models to small language models. Note: the published abstract is truncated at this point, so further details of the method and results are not available in the source text.

---

**来源**：[arXiv cs.CL (NLP)](https://arxiv.org/abs/2610.37680)

**详情页**：https://ai.daily.yangsir.net/daily/20261031-T0-18

---

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