🏆🇨🇭 Hey there awesome participant 🫕🌻
We are a collaboration of research groups commited to advancing the state of the art of Large Language Models with powerful hardware and talented people as part of the Swiss AI Initiative (swiss-ai.org).
This page is for participants and organizers of hackathons, some of which we are supporting directly, others which are self-organized, or can just benefit from open resources.
Introducing Apertus | Announcing Apertus 1.5
If you’ve never heard of it, no problem: we had our first public release just over a year ago, and the improved 1.5 update was deployed in the summer. Already, there are several great ways to access Apertus.
Let’s dig in! Find out how to get started with your hackathon project ⬇️
Go local
Apertus is completely free (Apache 2.0 licensed) to download and use on your local or cloud machine. The 8B variant runs well on top laptops, and we even have the Apertus Mini builds for even the most modest machines. For the adventurous with plenty of disk space, the User Guides cover free and open source harnesses like Jan, Ollama, LM Studio and Open Web UI.
Screenshot of LM Studio from my blog post
New: visit our User Guides for other approaches to installing Apertus.
Think sovereign
If you don’t have the hardware, or can’t be bothered to download multiple gigabytes of model weights, you can get started with Apertus for free at providers like Infomaniak using their AI Services page. They are a popular choice for many startups right now, with a climate-friendly new data centre in Geneva. We hear you can get a million tokens just for signing up!
Public AI is a non-profit which runs hosting for select open models. This is a demo service right now with relatively low rate limits, completely free for testing Apertus. It is quick and easy to get started with a personal API key at https://platform.publicai.co/docs
You can find further providers highlighted at –> Get Started
Connecting academia
The Apertus team is contributing to the development of the Swiss AI Research Platform, where CSCS offers direct access to early versions of Apertus as well as other open-source models to collaborators in the national academic community. Read more here
The ETHZ HPC service has GPU clusters available. For students and staff of the EPFL, we recommmend using the RCP AI as a Service platform. The costs are generally covered by teaching or research grants, and are generally only accessible through the university network or official VPN.
Renkulab from the Swiss Data Science Centre is recommended for members of all Swiss universities.
Act global
The Inference Endpoints of Swisscom’s Swiss AI Platform, as well as Hugging Face Inference Endpoints, were sponsored at hackathons during the launch of Apertus. They are both an excellent option for developers of demanding applications.
Featherless.ai launched the Open world campaign, hosting both Apertus models on their platform. They have generously provided starter keys for our recent hackathons at the EPFL and ETHZ - let us know if you need more access for your hack.
As a model compatible with open weight LLM standards, with community builds in GGUF, MLX and other formats, you should have no trouble running Apertus on the AI platforms of various commercial clouds. Tip: it is usually easier if they support loading models directly from Hugging Face.
You can find some recipes in the links in –> Get Started
Always be opening
So as you can see.. our schools, many companies, and even NGOs are invested into the idea of a truly sovereign and fully model as the foundation of great projects and investigations. But this vision depends on many of us making tough choices, and demanding fairness in AI.
More providers are on their way, and the list on our web page will be updated over time. Drop us a line on the hackathon platform if you need help with this. Can’t wait to hear about your results!
Visit https://apertus-ai.org/subscribe to stay tuned.