Nils Schüler

📢 𝗦𝘄𝗶𝘁𝘇𝗲𝗿𝗹𝗮𝗻𝗱 𝗶𝘀 𝗯𝘂𝗶𝗹𝗱𝗶𝗻𝗴 𝘁𝗵𝗲 𝗼𝗽𝗲𝗻 𝗔𝗜 𝗺𝗼𝗱𝗲𝗹 𝗕𝗶𝗴 𝗧𝗲𝗰𝗵 𝘄𝗼𝗻’𝘁. ETH Zürich, EPFL, and the Swiss National Supercomputing Centre (CSCS) will release a fully open Large Language Model (LLM) – trained on 𝗔𝗹𝗽𝘀, one of Europe’s fastest AI supercomputers. 𝗪𝗵𝗮𝘁 𝘀𝗲𝘁𝘀 𝗶𝘁 𝗮𝗽𝗮𝗿𝘁? ✅ 𝗠𝘂𝗹𝘁𝗶𝗹𝗶𝗻𝗴𝘂𝗮𝗹 𝗯𝘆 𝗱𝗲𝘀𝗶𝗴𝗻 Trained on 1,500+ languages – with strong capabilities in underrepresented communities. ✅ 𝗙𝘂𝗹𝗹𝘆 𝗼𝗽𝗲𝗻 (unlike commercial models) Weights, code, and training data will be public. Licensed under Apache 2.0 for unrestricted reuse – commercial, scientific, or regulatory. ✅ 𝗚𝗹𝗼𝗯𝗮𝗹𝗹𝘆 𝗰𝗼𝗺𝗽𝗲𝘁𝗶𝘁𝗶𝘃𝗲 Available in 8B and 70B parameter sizes – trained on 15 trillion high-quality tokens. ✅ 𝗧𝗿𝘂𝘀𝘁𝗲𝗱 𝗯𝘆 𝗱𝗲𝘀𝗶𝗴𝗻 Developed entirely on Swiss public infrastructure – aligned with Swiss law and the EU AI Act. ✅ 𝗣𝗼𝘄𝗲𝗿𝗲𝗱 𝗯𝘆 𝗔𝗹𝗽𝘀 – 𝗴𝗿𝗲𝗲𝗻, 𝘀𝗼𝘃𝗲𝗿𝗲𝗶𝗴𝗻, 𝗮𝗻𝗱 𝗯𝘂𝗶𝗹𝘁 𝘁𝗼 𝘀𝗰𝗮𝗹𝗲 Trained on Alps, using 100% carbon-neutral electricity and 10,000+ NVIDIA Grace Hopper Superchips – enabled by a 15+ year collaboration between NVIDIA, Hewlett Packard Enterprise, and CSCS. ✅ 𝗪𝗶𝘁𝗵 𝘀𝘁𝗿𝗼𝗻𝗴 𝗿𝗼𝗼𝘁𝘀 𝗶𝗻 𝘁𝗵𝗲 𝗚𝗿𝗲𝗮𝘁𝗲𝗿 𝗭𝘂𝗿𝗶𝗰𝗵 𝗔𝗿𝗲𝗮 😎 Where world-class universities, sovereign infrastructure, public funding, and a dense AI talent pool create one of Europe’s most trusted ecosystems for building technology at scale.🚀 👏 𝗟𝗲𝗱 𝗯𝘆 𝘄𝗼𝗿𝗹𝗱-𝗰𝗹𝗮𝘀𝘀 𝗔𝗜 𝗺𝗶𝗻𝗱𝘀: ETH AI Center’s Imanol Schlag, EPFL AI Center’s Antoine Bosselut and Martin Jaggi. Supported by the Swiss AI Initiative and ELLIS - European Laboratory for Learning and Intelligent Systems 🔗 𝗗𝗶𝘀𝗰𝗼𝘃𝗲𝗿 𝗵𝗼𝘄 𝗚𝗿𝗲𝗮𝘁𝗲𝗿 𝗭𝘂𝗿𝗶𝗰𝗵 𝘀𝘂𝗽𝗽𝗼𝗿𝘁𝘀 𝗴𝗹𝗼𝗯𝗮𝗹 𝗔𝗜 𝗯𝘂𝗶𝗹𝗱𝗲𝗿𝘀: https://okt.to/NCI68o More about the Swiss LLM: https://okt.to/BR2HIV

903

16 Kommentare

I find Switzerland’s launch of "Apertus" a national LLM built by EPFL, ETH Zürich, and CSCS truly inspiring. Multilingual by design: Trained on 15 trillion tokens across 1,000+ languages, with 40% non-English, including underrepresented ones like Swiss German and Romansh. Scalable: Comes in 8B and 70B parameter versions. Fully open: Model weights, data, and training recipes are all accessible ensuring transparency and compliance with Swiss & EU laws. Developer ready: Deployment supported through platforms like Transformers, vLLM, SGLang and MLX Permissive open-source license: For research, education, and even commercial applications. Could such an open approach accelerate local innovation and ensure AI inclusivity for millions who don’t interact in English? "Apertus" is a bold step not just for Switzerland but for the global AI community. It shows what’s possible when openness, transparency, and inclusivity are put at the center of AI development. Would love to see India take a similar path. https://lnkd.in/gRGX74NU #AI #OpenSource #LLM #India #Switzerland #MultilingualAI #EthicalAI

164

3 Kommentare

🚀 Switzerland’s first large-scale, open, multilingual language model is here: Apertus   EPFL, ETH Zürich, and CSCS have just released #Apertus, a milestone in generative AI built with transparency, diversity, and accessibility at its core.   So, what makes Apertus unique❓ 🔓 Fully open & transparent:  Apertus is Latin for “open”, highlighting that its architecture, training data, and model weights are fully accessible and documented. 🌍 Multilingual by design: Trained on 15 trillion tokens across 1,000+ languages, with 40% non-English data. This includes underrepresented languages such as Swiss German and Romansh. 📈 Scalable access: Available in two versions (8 billion and 70 billion parameters) via Hugging Face and Swisscom’s sovereign Swiss AI platform. 🤝 Built for the public good: Open-source licensing enables use in research, education, and commercial applications. 🌐 Global availability: People outside Switzerland can also access Apertus through the Public AI Inference Utility, making it part of a worldwide movement for public, open AI.   With Apertus, Switzerland positions itself at the forefront of open and responsible AI, turning cutting-edge research into a foundation for innovation across society, academia, and industry.   Read more in this article: https://lnkd.in/drnMUjSt   Learn how to access Apertus: https://lnkd.in/ey6mypX9   #WhySwitzerland #OpenSourceAI EPFL AI Center ETH AI Center Swiss {ai} Weeks Presence Switzerland swisstech Deep Tech Nation Switzerland Swissnex Basel Area Business & Innovation Greater Geneva Bern area (GGBa) Greater Zurich Area St.GallenBodenseeArea Lucerne Business Economic Promotion Valais Standort Promotion in Obwalden Standortförderung Kanton Aargau Brigitte Fäh Martin Jaggi Thomas Schulthess Imanol Schlag Daniel Dobos Joshua Tan

97

3 Kommentare

As several media articles recently pointed out, large AI models and the data centres that power them consume a lot of energy. How serious are these concerns, and what can we do about them? We asked Ana Klimovic, Assistant Professor at ETH Zürich and specialist in sustainable machine learning frameworks, for her view. She explains why #AI workloads often underutilize precious #GPU resources and how we can improve their efficiency. Discover more: https://lnkd.in/d7uherYt

142

1 Kommentar

A project close to my heart: The EPFL AI Center just opened the second phase of the Geneva AI Initiative! Phase 1 made one thing clear: there’s incredible ambition in International Geneva to put AI to work for good. So we asked ourselves, how can we connect missions and technology in a way that works and scales? That’s what this next phase is all about. We've redesigned the program completely, partnering with the Swiss Data Science Center (SDSC) for their well-known capacity building and project implementation expertise. Here's what the new phase looks like: 📚 Up to 25 project leaders will go through DASHI, a 4-day intensive Data Science for Humanitarian Impact course (Jan 22-23 & 29-30) 🤝 You'll refine your project idea with AI experts at the AMLD Intelligence Summit in February 2026 ✍️ You'll get one-on-one support developing a strong proposal 💰 The best projects can get funded for up to 12 months I hope this becomes the start of something bigger. In a world that's fragmenting, we need such institutions more than ever. Deadline: November 7, 17:00 CET Apply: https://lnkd.in/eUcuhaED

228

11 Kommentare