Apertus is the open source 8b and 70b LLM from swiss-ai. They've published both the base and the instruct sft models. Very cool that projects like this exist.
Is it any good?
I haven't tried it for anything myself yet. The paper provides several benchmarks. The emphasis during training was on multi-language support (over 1800 languages are represented in its pre-training data, which is 40% non-English) and non-copyrighted training data... and the benchmarks seem to suffer for it.
2023, but deadlines less than a month ago? Seems to be been updated continiously so (2023) doesn't really fit here.
I propose every Linux post should be tagged (1991) from now on
(2023) Little said at the time (4 points, 1 comment) https://news.ycombinator.com/item?id=38529956
Related 2023 discussion (22 comments): <https://news.ycombinator.com/item?id=38523736>
Why is this not written in German, I'm afraid to ask?
Most researchers in Switzerland are non-Swiss, and many institutes have English as language of business
Why is it not written in French? Or Italian? Or Romansh? Because Switzerland has four official languages and English makes it easier for everyone
english is the lingua franca