6 comments

  • altano27 minutes ago
    How much memory does this use of the test machine's (M3 Max) 128 GB unified memory?
    • WASDx5 minutes ago
      The model is only 0.3B params so probably not much.
  • PaulRobinson2 hours ago
    Local LLMs are the future, and one of the reasons I think the data centre furore is just going to end in a market crash.<p>LLMs that can reliably be used for control problems are the future, and I think classic&#x2F;deep RL has generally been overlooked for years for a whole host of problems by wider industry because it felt inaccessible. The first thing I thought of when I saw Jev (and then Laya), was &quot;this might move the needle in a really, really interesting way&quot;.<p>Local LLMs that can reliably be used for control problems smash through a lot of barriers I&#x27;m interested in, and this intrigues me a lot. Guess I&#x27;m about to become a big Laya fan if it can run on this kind of hardware to this performance.
    • ipsi19 minutes ago
      The future for whom? The general public? Not a chance, no way, not unless it&#x27;s able to run on a phone (anywhere from 20-40% of internet users, world-wide, are phone-only).<p>For companies? I think that&#x27;s a lot more plausible, as that&#x27;s mostly just a question of money - is it cheaper to run <i>and administrate</i> our own models, or outsource that?<p>For technically inclined users? I think that&#x27;s unlikely unless they&#x27;re able to operate on relatively cheap hardware while still being just as good as the hosted models. And by that I don&#x27;t mean &quot;a mac studio,&quot; that&#x27;s far more money than I think is reasonable. A single RTX 5080, maybe, once memory prices start to drop.
    • frag2 hours ago
      that&#x27;s not a local LLM. If it&#x27;s local, it doesn&#x27;t matter in this case. Laya is a System 1 &quot;AI&quot;, namely works like a classifier, given a state and questions, it shoots probabilities for each. I publish an episode tomorrow about Laya and Typesafe AI on <a href="https:&#x2F;&#x2F;www.youtube.com&#x2F;@DataScienceatHome" rel="nofollow">https:&#x2F;&#x2F;www.youtube.com&#x2F;@DataScienceatHome</a><p>Stay tuned ;)
    • bigyabai1 hour ago
      It won&#x27;t. Laya is a finetuned version of Google&#x27;s BeRT model, which is almost 10 years old right now.<p>If BeRT had any potential to disrupt the datacenter buildout, it already would have.
  • frag2 hours ago
    Great job! Did you finetune your own Laya for the snake game or what?
  • tentacleuno2 hours ago
    This looks like a local AI model playing Snake -- is that correct? The article offers no explanation.
    • putna3 minutes ago
      Correct, the cli commands are just copy paste to run on your machine.
    • ryuuseijin2 hours ago
      The linked github project [1] contains more information.<p>[1]: <a href="https:&#x2F;&#x2F;github.com&#x2F;mizorewww&#x2F;laya-coreml" rel="nofollow">https:&#x2F;&#x2F;github.com&#x2F;mizorewww&#x2F;laya-coreml</a>
  • oezi2 hours ago
    [dead]
  • aidiveyt1 hour ago
    [dead]