13 comments

  • woctordho2 hours ago
    Petals is from 2022. Nowadays intelligence of smaller models, quantization techs, and optimizations to run models faster on consumer GPUs have improved a lot.<p>For distributed inference of smaller LLMs and diffusion models that fits in one consumer GPU rather than splits on multiple machines, there are already pretty good solutions such as AI Horde (formerly Stable Horde) [0]. Notably, it&#x27;s the default provider that powers SillyTavern. It also has an interesting economy model of kudos.<p>[0] <a href="https:&#x2F;&#x2F;stablehorde.net&#x2F;" rel="nofollow">https:&#x2F;&#x2F;stablehorde.net&#x2F;</a>
  • maccam9122 hours ago
    The recently discussed <a href="https:&#x2F;&#x2F;meshllm.cloud&#x2F;" rel="nofollow">https:&#x2F;&#x2F;meshllm.cloud&#x2F;</a> is the one I&#x27;ve been playing with but don&#x27;t have the hardware to try with a model split between nodes, which apparently is supported and just not part of the public demo.
    • atum472 hours ago
      Can&#x27;t even scroll the website on mobile, too much going on
  • N_Lens2 hours ago
    It&#x27;s an interesting concept but the timing is probably too early. If more people had reliable low latency gigabit or ideally 10gbit throughput, then it might start to approach feasibility. There&#x27;s other blockers too, but that springs to mind immediately.<p>It&#x27;s cool to imagine a planet wide neural network interconnected with fiber - the nervous system of a planetary intelligence. But perhaps mushrooms do that already (Alpha Centauri ever relevant).
    • c7b6 minutes ago
      Not a neural network though ;) <a href="https:&#x2F;&#x2F;biologyinsights.com&#x2F;do-fungi-have-a-nervous-system-the-biology-explained&#x2F;" rel="nofollow">https:&#x2F;&#x2F;biologyinsights.com&#x2F;do-fungi-have-a-nervous-system-t...</a>
    • woctordho2 hours ago
      Relevant: Why Switzerland has 25 Gbit internet and America doesn&#x27;t <a href="https:&#x2F;&#x2F;news.ycombinator.com&#x2F;item?id=47652400">https:&#x2F;&#x2F;news.ycombinator.com&#x2F;item?id=47652400</a>
    • hahahaa2 hours ago
      Too early too in that later (hopefully) each node can serve a full model so it is then a simpler share tit for tat of complete models.
    • behnamoh1 hour ago
      Too early? the project has been going for years now...
  • wxw2 hours ago
    Hm, looks like this is an old project (2022) associated with HuggingFace (<a href="https:&#x2F;&#x2F;huggingface.co&#x2F;bigscience" rel="nofollow">https:&#x2F;&#x2F;huggingface.co&#x2F;bigscience</a>).<p>Doesn&#x27;t look like it&#x27;s very active nowadays: <a href="https:&#x2F;&#x2F;github.com&#x2F;bigscience-workshop&#x2F;petals" rel="nofollow">https:&#x2F;&#x2F;github.com&#x2F;bigscience-workshop&#x2F;petals</a><p>Article with more details: <a href="https:&#x2F;&#x2F;techcrunch.com&#x2F;2022&#x2F;12&#x2F;20&#x2F;petals-is-creating-a-free-distributed-network-for-running-text-generating-ai&#x2F;" rel="nofollow">https:&#x2F;&#x2F;techcrunch.com&#x2F;2022&#x2F;12&#x2F;20&#x2F;petals-is-creating-a-free-...</a><p>&gt; [...] volunteers can donate their hardware power to tackle a portion of a text-generating workload and team up others to complete larger tasks, similar to Folding@home and other distributed compute setups.
  • dang1 hour ago
    Related. Others?<p><i>Run LLMs at home, BitTorrent‑style</i> - <a href="https:&#x2F;&#x2F;news.ycombinator.com&#x2F;item?id=37546810">https:&#x2F;&#x2F;news.ycombinator.com&#x2F;item?id=37546810</a> - Sept 2023 (125 comments)
  • matheusmoreira1 hour ago
    This is awesome. Now we also need distributed <i>training</i> of models!
    • woctordho37 minutes ago
      Distributed training is much harder than distributed inference but not impossible. See the recent development of DiLoCo at Nous Research and Prime Intellect.
  • brainless2 hours ago
    I have been thinking of something on these lines but with much smaller models. The entire model has to fit on a single computer. Host owner would choose the model they prefer, perhaps because they already use it. Then it is more about utilizing the GPU for LLM requests.<p>Peer to peer, consumers get to route their request to a host with compatible model. Consumers have to also contribute GPU but it does not have to be equal - I have not thought through the fairness part. Perhaps initially it starts with &quot;create your friends group and have access to all the host nodes&quot;.
    • inigyou2 hours ago
      Can&#x27;t wait to run a Sybil farm of subsidized proxy nodes and log all the juicy prompts.
      • woctordho2 hours ago
        AI Horde has some measures to prevent Sybil attack that returns wrong results, but not enforce zero data retention. Prompts belong to the whole open source community. For example <a href="https:&#x2F;&#x2F;huggingface.co&#x2F;datasets&#x2F;la-ji&#x2F;sd-prompt-in-the-wild" rel="nofollow">https:&#x2F;&#x2F;huggingface.co&#x2F;datasets&#x2F;la-ji&#x2F;sd-prompt-in-the-wild</a>
      • brainless2 hours ago
        I do not know what&#x2F;who Sybil is but yes, there is a lot of plumbing in order to make sure that host nodes are ZDR* compliant and much more. Zero visibility of source prompt.<p>Also, this is the reason I want to start with trust based groups only - invite people you already know. I have friends who have Macs with 32GB or 48GB unified memory but sending prompts to them is not a out of the box thing.<p>* typo
        • fragmede1 hour ago
          &gt; A Sybil attack is a type of attack on a computer network service in which an attacker subverts the service&#x27;s reputation system by creating a large number of pseudonymous identities and uses them to gain a disproportionately large influence.<p><a href="https:&#x2F;&#x2F;en.wikipedia.org&#x2F;wiki&#x2F;Sybil_attack" rel="nofollow">https:&#x2F;&#x2F;en.wikipedia.org&#x2F;wiki&#x2F;Sybil_attack</a>
        • mcpindex-ai2 hours ago
          [flagged]
  • ProofHouse10 minutes ago
    I did a triple take. Many years old but an early ground breaking project. Has some exploratoy ideas a year or two ago last I looked. Crazy to see Petals at 16 frontage (at time of writing)
  • htrp1 hour ago
    Any updates here to make this work on more modern llms?
  • wsc9812 hours ago
    Seems a bit like GNUS [0].<p>---<p>[0]: <a href="https:&#x2F;&#x2F;www.gnus.ai&#x2F;" rel="nofollow">https:&#x2F;&#x2F;www.gnus.ai&#x2F;</a>
  • petterroea1 hour ago
    This is super cool technology, but will be abused to death
  • crowcroft2 hours ago
    Imagine if there was some kind of way to cryptographically &#x27;prove&#x27; your GPU is doing some kind of &#x27;work&#x27; here and contributing to the network.<p>You could even hand out some kind of digital &#x27;currency&#x27; to the people proportional to the amount of work their doing!
    • schoen1 hour ago
      I&#x27;ve occasionally written comments about the difficult of achieving proof-of-useful-work mechanisms and I should probably find or write a standard post about this question.<p>In order to be as decentralized as Bitcoin and related PoW mechanisms, a proof-of-work mechanism should have<p>* the ability to create an unbounded number of instances of the problem deterministically from numeric seeds<p>* the ability to scale difficulty up and down using a difficulty parameter<p>* in a way where anyone can easily confirm that a given problem instance corresponds to a given seed, and anyone can easily confirm that a given solution correctly solves a given problem instance<p>I&#x27;m not aware of a proof-of-useful-work mechanism that meets these criteria. (I&#x27;m also not aware of an argument that it&#x27;s impossible to have one. Zero-knowledge proofs might actually go far toward making it possible in the future. but it&#x27;s kind of complicated.)
    • bravoetch2 hours ago
      As we&#x27;ve seen, that&#x27;s a solution looking for a problem.
      • goda901 hour ago
        But what if the work is useful beyond just crunching numbers for the sake of proving it?
      • gverrilla1 hour ago
        More like a problem looking for more problem.
  • lostmsu2 hours ago
    Not gonna work, will be starved by inter node bandwidth.
    • gatuamgb1 hour ago
      You can easily build a localized, access-controlled compute grid using Apple Silicon hardware. Haven&#x27;t tested but as far as I can see, the key to making it performant would be keeping full models resident on individual nodes rather than attempting to distribute the tensor processing across the network.