Apertus – Open Foundation Model for Sovereign AI
摘要
Apertus 由瑞士 AI Initiative 推动，参与方包括 EPFL、ETH Zurich 和 CSCS，定位为面向 Sovereign AI（主权 AI）的完全开放基础模型。项目强调开放训练数据、代码、模型权重、训练方法和对齐原则，且过程可复现。官方称模型在训练中遵循 EU AI Act 相关要求，包括尊重数据退出（opt-out）、移除个人身份信息（PII）以及降低记忆化风险。性能方面，Apertus 宣称其 8B 和 70B 参数版本可与同规模领先开源模型竞争，并从一开始支持 1000 多种语言。页面还列出近期进展，包括发布 16 个用于展示蒸馏与量化技术的 Apertus Mini 模型、ACL 2026 接收技术报告，以及为提契诺地区内部 AI 翻译系统提供微调模型。
荐读理由
开源权重与全部训练数据代码方法对齐原理完全可迁移到独立 AI 工程项目，Swiss AI Initiative 直接提供 EU AI Act 合规（opt-out/PII 移除/去记忆）框架，8B-70B 规模多语言（1000+种）训练后端能直接复制到自家产品
原文
Fully Open Foundation Model for Sovereign AI
Developed by the Swiss AI Initiative as a collaborative effort between EPFL, ETH Zurich, and CSCS. Open weights, open data, open science.
Get StartedTechnical Information
Fully Open
Training data, code, weights, methods, and alignment principles — all documented and reproducible. Apertus is to AI as Open is to Source.
Compliant at Scale
Built to meet EU AI Act requirements: the model respects opt-outs, removes PII, prevents memorization. A global foundation to build on.
Run for Performance
Competitive with top open models at an equivalent scale of 8B and 70B parameters. Multilingual from day one — trained on 1000+ languages.
Swisscom is a Strategic Partner of the Swiss AI Initiative.
Jun 15
Apertus Mini
A set of 16 small language models released to demonstrate distillation and quantization techniques.
Apr 25
Apertus paper at ACL 2026
Our technical report has been accepted at a leading conference for AI & NLP.
Mar 17
Apertus for Ticino
Fine-tuned model powers in-house AI translation
这条对你有帮助吗？