---
id: 20260908-T0-02
title: "SharedSAE：跨语言模型共享特征字典，减少重复训练"
title_en: "SharedSAE: One Feature Dictionary Across Multiple Language Models"
url: https://ai.daily.yangsir.net/daily/20260908-T0-02
issue_date: 2026-09-08
publish_date: 2026-09-07T04:00:00.000Z
category: research
source_name: "arXiv cs.LG (ML)"
source_url: https://arxiv.org/abs/2609.04344
---

# SharedSAE：跨语言模型共享特征字典，减少重复训练

稀疏自编码器（SAE）通常需要为每个语言模型单独训练和标注，成本高昂。arXiv新论文提出SharedSAE，证明一个共享的SAE可以替代多个模型各自的SAE集合。这一发现可大幅降低模型可解释性研究的计算成本，加速多模型分析对比。

## English Version

**SharedSAE: One Feature Dictionary Across Multiple Language Models**

SAEs typically require separate training for each language model. New arXiv paper introduces SharedSAE, showing a single shared SAE can replace per-model SAEs, significantly cutting computational cost for interpretability research and enabling easier multi-model comparisons.

---

**来源**：[arXiv cs.LG (ML)](https://arxiv.org/abs/2609.04344)

**详情页**：https://ai.daily.yangsir.net/daily/20260908-T0-02

---

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