EPFL, ETH Zurich and the Swiss National Supercomputing Centre (CSCS) released Apertus today, Switzerland’s first large-scale, open, multilingual language model — a milestone in generative AI for transparency and diversity.

all 4 comments

sorted by: hot top controversial new old
[–] 11 points 1 year ago* (1 child)

Nice. This is the one that supposedly comes with open data sets, training data and everything, and it's a true "open-source" model. Seems it's avalable in 7B and 70B.

  • source
  • hideshow 1 child comment
  • [–] 7 points 1 year ago (1 child)
  • [–] 6 points 1 year ago* (last edited 1 year ago)

    For the questions that I asked it, it was pretty useless. Those are not representative and you should definitely try it yourself. This model probably has it's strengths in other aspects. It's just two random things that have been on my mind and I wanted a quick response to so that I can then do a deeper dive myself.

    1. Technical question looking for software that can provide a certain functionality had multiple false results and didn't mention the most common solution (which you can find in the top google result, a reddit thread from 2022 about the topic). Other AIs all responded with info which matched that reddit thread.

    2. Question about biomechanics of plants and why the use of human feces as fertilizer is bad (which was done historically and still done in some places in the world. My family abroad still does it because of tradition and I want to explain to them why they should stop). It didn't want to answer because the question "is inappropriate". After being trivially convinced that the question is in fact not inappropriate, it kept saying "it's bad, not recommended, don't do that. Instead do this." but it could never explain WHY it is bad. Further followups just resulted in an error. Other AIs immediately gave a list of stuff that is in human feces and how it contaminates the soil and which ones can be absorbed into the plant and cause harm for consumers. (Which I then used to find relevant articles on health and studies that measure plant absorbtion of those things)

  • source
  • parent
  • 8 comments in c/localllama@sh.itjust.works view that post
    [–] 5 points 11 months ago (7 children)

    Did someone try it? What's your experience?

    I think I like it. But, I'm having massive issues with it becoming very repetetive after a while. And I'm not sure if it's the model or my sampler settings.

  • source
  • hideshow 7 child comments
  • [–] 4 points 11 months ago (2 children)

    It's a dense model so it's unfortunately much slower than the newer MoE models.

    I also was a little bit disappointed with its translation abilities, especially considering that its a Swiss model but cannot properly do German <-> English translations.

  • source
  • parent
  • hideshow 2 child comments
  • [–] 2 points 11 months ago (1 child)

    I'm only tinkering with the 8B variant, so speed is alright. I hadn't noticed yet. But yes, seems English to German leads to weird results. German to English seems to be fine, though. At least for the 2 texts I put in.

  • source
  • parent
  • hideshow 1 child comment
  • [–] 3 points 11 months ago

    I tested the 70b model at Q8_K_XL quantization on Strix Halo and it's not too slow to use for short queries but definitely much slower than something like gpt-oss-120b.

    English to German leads to weird results. German to English seems to be fine, though. At least for the 2 texts I put in.

    Didn't get to German to English before giving up, English to German was just too awful. Cases were constantly wrong, very weird word choices and incorrect grammatical genders.

  • source
  • parent
  • [–] 4 points 11 months ago* (last edited 11 months ago)

    Found it not very compliant, asked it about french politics and it fared worse than other 70B models I tested through openrouter.

    Mistral 24B gave better answers.

    I like the group that is behind though. EPFL and ETH are some of the best research institutes in Europe. Catching up with big corporate models is hard, and it is nice they are giving it a shot. Don't expect them to be there yet on the first try though.

  • source
  • parent
  • [–] 2 points 11 months ago (2 children)

    How are you running it? Would you be able to post your run arguments?

  • source
  • parent
  • hideshow 2 child comments
  • [–] 3 points 11 months ago (1 child)

    I'm running the 8B version with KoboldCPP and pretty much the default settings of the Min-P sampler. That tends to work very well with almost all other models and I'm not aware of any specific recommendations for Apertus...

  • source
  • parent
  • hideshow 1 child comment
  • [–] 3 points 11 months ago

    We recommend setting temperature=0.8 and top_p=0.9 in the sampling parameters.

    Try that. I believe those params are available I Kobold. Id that doesn't work, send me a sample of what you're doing and I'll try it out

  • source
  • parent
  • 5 comments in c/opensource@programming.dev view that post

    Seems like it evaluates on par with llama3.1. Haven't had many reasons to use LLMs, but this looks like a good option.

  • source
  • [–] 6 points 11 months ago (3 children)

    Good that it's open source, but is it ethical?

  • source
  • hideshow 3 child comments
  • [–] 4 points 11 months ago* (last edited 11 months ago) (1 child)

    They're fairly transparent with everything. You could call it open-source. And it's supposed to be the first large model which complies with the EU AI regulations. They try to make an effort not to include too much material from people who objected to AI use, and there is a way to opt out. They did not deliberately pirate books like Meta did. But with that said, it's still AI. Training needs a lot of water and energy. Though I think this Alps supercomputing center tries to be carbon-neutral and use Swiss hydropower. Whatever that means in practice. Opt-out is probably the best thing we can do but it's not exactly consent from the authors of the training material. And I don't think there is a way to compensate them. And AI can of course be problematic once used, so that depends on what people do with it.

    I'd call it more ethical (than other models). But I don't see how it'd be strictly ethical in absolute terms. Looks to me like an effort to improve, maybe substantially on what other people did. But there's still a lot of problematic aspects of AI which scientists and society hasn't addressed yet.

  • source
  • parent
  • hideshow 1 child comment