mrkn1

Cool project too. If you are looking for a 500MB instead of gigabytes, with evals on Jevbench that you can run fast on CPU check out gutsy.

https://github.com/kouhxp/gutsy

thanks!

yapsnap has 10 languages

love seeing more sub-20MB, CPU-first models. if anyone wants a CLI built on the same ethos (no GPU, no cloud), been using yapsnap streaming Zipformer ASR, plus diarization and timestamps all on CPU! It supports 10 languages. Unlimited transcription for free.

Why not have a CPU-first decision model for free? check out gutsy

If you rather run your decision model on your CPU, check gutsy [0]

[0] - https://news.ycombinator.com/item?id=49976996

You can transcribe in seconds on CPU for free. Check out yapsnap repo, its free and open source

That's true. I would definitely smoke test your use case before any serious work. And if it faceplants, you can always fine tune to your taste. If the false negative/positive is cheap, I think it's a mighty tool.

I made Gutsy to see what quality of inferencing I could get for a Jev style model on laptops and cheap VMs. The result is pretty good for triaging, relevance scoring and yes no type questions. I'd love to hear what latency others are getting. For me it was subsecond inference time for short queries.

> And, seriously, there are so many verified accounts that just post AI-generated replies to popular posts, and we all just ignore them. It's the most insane, Kafkaesque thing. Absolutely nobody likes it. It's an eyesore. And we're supposed to be living in the most powerful age for AI. You're telling me that nobody at SpaceX/Xai/Grok/SuperGrok/whatever can write a classifier to ban these accounts? I can press a button on any post and be given a full explanation by Grok, a button which I've pressed 2 dozen times on accident and never on purpose. You can't run a decision model for $0.0000001 cents whenever someone hits "Post" and realize that a reply makes no sense?

Funny you mention this. I've been building exactly that. gutsy [0] is a small (775MB) local decision model (like Jev). No API calls, runs on CPU. Classifying "does this reply make sense" is basically the use case. Platforms not doing this already isn't a capability gap at all.

[0] https://github.com/kouhxp/gutsy

gutsy 0.8B, runs on CPU, and is Llama.cpp compatible [0]

[0] - https://huggingface.co/kouhxp/gutsy

if you'd rather skip the API key and keep audio local, yapsnap transcribes any video URL or file on CPU, no GPU needed. github.com/kouhxp/yapsnap

If you want something even lighter than Jev to compare against, there's also gutsy (https://github.com/kouhxp/gutsy) runs on CPU

That makes sense, I agree. I still think there might be use cases when people don't want to use the Jev API, and end on a different trade-off.

thank you

For smaller scale decision model that runs on CPU, check https://news.ycombinator.com/item?id=49923223

While reading your post, I kept thinking of K&R, and the beards of the 70s. Isn't this a return to early Unix? Users writing their own small tools in C, the system shipping building blocks... walls between users rather than programs?

I'm sure there's a simple answer to this, and I'm probably just missing it, but what happened with the supergravity one?

And how many runs did it take before this one? They say "in one shot" with nothing more than "keep going." But we only see the successful run, reported by the people who ran it.

https://github.com/kouhxp/yapsnap free cpu first and open source

fwiw kimchi-derived probiotic bacterium (Leuconostoc mesenteroides CBA3656) was shown to bind nanoplastics and help mice excrete more of them. But it’s not yet proven that eating kimchi removes microplastics from humans

Has anyone tried the others listed? Any feedback?

This runs for free on CPU https://github.com/kouhxp/textsnap

The key moments feat is neat. Been working on a free opensource offline transcriber that runs fast on CPU and does diarization too

https://github.com/kouhxp/yapsnap

No rigorous eval, and I love Tesseract. Here's the example that motivated me to build textsnap (which is in the github's README), parsed with Tesseract:

https://imgur.com/a/i2eQra8

No reason other than their Q4 model working reasonably well and fast on my CPU laptop. Should work with any ONNX VLM model

109 languages, including other alphabets.

Just ran

  textsnap "https://i.ytimg.com/vi/LBNDfxjEYlA/maxresdefault.jpg"

and got this

  $('.count').each(function () {
  $('this').prop('Counter', 0).animate({
    Counter: $('this').text()
  }, {
      duration: 4000,
      easing: 'swing',
      step: 'function (now) {
          $('this").text(Math.ceil(now));
      }
    }); 
  });

thank you!

thank you! what is it about?