12 comments

  • embedding-shape1 hour ago
    &gt; Hugging Face is the bottleneck, not your link.<p>README could clearly make use of a cleanup, seems to be more like a session log dump now than a good introduction to the project for a new user. Maybe try something like &quot;Remove anything from the README.md that wouldn&#x27;t be helpful to someone who sees this project with zero context, for the first time. Rewrite all paragraphs and sections to be concise and remove all fluff, leave only important details new users must know before using the project&quot;.
    • Eufrat1 hour ago
      I hate this AI style writing because since it doesn’t really understand flow, it’s being inserted in irrelevant places and it is extremely irritating to read.
      • carloslfu1 hour ago
        I feel you! fix incomming
        • Eufrat1 hour ago
          For what it’s worth, this comment was not targeted at you, but rather the model kinda forcing it. I get the sense that Anthropic did not think much of this, but it seems to have gotten worse with recent models and it really comes off as a kind of nails on the chalkboard writing style.<p>I have to image whatever style of writing this was trained on is a lot more pleasant to read and I feel bad for whoever writes like this now being associated as bad AI writing.
    • carloslfu1 hour ago
      thanks! I&#x27;ll do!
    • xlayn1 hour ago
      [flagged]
      • embedding-shape1 hour ago
        &gt; if someone completely new comes and read the current page... isn&#x27;t that piece of information something they want to know?<p>Again, the README is (was?) written in way that it was clear it was a conversation between an agent and a human, with literal replies in the README. I&#x27;m in no way whining the way you are or demonstrate, and I even give OP helpful advice to fix it. What more can you want from me?
  • atif0891 hour ago
    As someone who is just looking at the theoretical benchmarks of each of these models I&#x27;m curious if anyone could share what are the problems (maybe around code) that flash-next was able to solve which 27b was not able to
    • carloslfu53 minutes ago
      This is the best I could find: <a href="https:&#x2F;&#x2F;huggingface.co&#x2F;Qwen&#x2F;Qwen3.8-Flash-Next?utm_source=chatgpt.com#language" rel="nofollow">https:&#x2F;&#x2F;huggingface.co&#x2F;Qwen&#x2F;Qwen3.8-Flash-Next?utm_source=ch...</a><p>About the specifics, I have only anecdotal evidence, but I guess this info can be found somewhere
  • whartung1 hour ago
    I&#x27;m hoping to see progress in this space.<p>Folks talking about how 32G is not enough for local use, but then there&#x27;s been work like this to empower it.<p>My hope is that the new 32G M6 will be &quot;useful&quot; locally, possibly because of work like this.
    • carloslfu1 hour ago
      yes! I&#x27;m bullish on this. there is a lot of work to do. I&#x27;ve been experimenting with pruning, distillation, and retraining too. I&#x27;m sure your 32gb m6 will run a badass local model!
  • prometheus19921 hour ago
    It&#x27;s hard to believe 16GB unified memory will give you 5 tok&#x2F;sec unless you are ignoring the thermal warnings. I am running Qwen3.6-35B-A3B on my 16GB M3 and get 7-8 tokens&#x2F;sec with all the optimizations while keeping the peak memory and thermal warnings at check. <a href="https:&#x2F;&#x2F;github.com&#x2F;deepanwadhwa&#x2F;samosa-chat" rel="nofollow">https:&#x2F;&#x2F;github.com&#x2F;deepanwadhwa&#x2F;samosa-chat</a>
    • Balooga24 minutes ago
      Now I&#x27;m feeling pretty good about getting 10-11 tokens&#x2F;sec running Qwopus 3.6-35B-A3B Q6_K on an old Mac Pro 2013 (trashcan) with 128GB RAM (DDR3), 12 core Xeon, dual D700s. Arch Linux and llama.cpp.
    • carloslfu51 minutes ago
      interesting! Yes, thermal is important. Pretty cool project man! Starred and checking it out!
  • ErenayDev2 hours ago
    how much energy does it consume?
    • carloslfu1 hour ago
      Good one! I haven&#x27;t measured this. I&#x27;ll include it!
  • karmakaze2 hours ago
    It seems we could use a new kind of memory that streams the weight data in, like GDDR in reverse.
    • 0x4571 hour ago
      High Bandwidth Flash? <a href="https:&#x2F;&#x2F;www.sandisk.com&#x2F;company&#x2F;newsroom&#x2F;blogs&#x2F;2025&#x2F;scaling-beyond-the-wall-inside-sandisks-high-bandwidth-flash-for-ai" rel="nofollow">https:&#x2F;&#x2F;www.sandisk.com&#x2F;company&#x2F;newsroom&#x2F;blogs&#x2F;2025&#x2F;scaling-...</a>
    • carloslfu1 hour ago
      yes! I guess future hardware designs will have something like that!
  • jonplackett1 hour ago
    Is this going to destroy my SSD?
    • egorfine1 hour ago
      no it&#x27;s reading, not writing
    • cromka1 hour ago
      By reading it?
      • carloslfu1 hour ago
        I don&#x27;t know actually. I&#x27;ll check haha. My best guess is it isn&#x27;t.
        • carloslfu1 hour ago
          I hope not! this is a new macbook lol!
  • drcongo1 hour ago
    &quot;Disk is the gate that bites first&quot;<p>AI;DR
    • thirtygeo52 minutes ago
      Ha! AI;DR is a great phrase. Had not seen that before
    • drums87871 hour ago
      The never ending gate bites.<p>How I have come to detest certain phrases.
      • bogzz1 hour ago
        Load bearing gate bites.
        • drcongo47 minutes ago
          ...the seam.
          • brailsafe10 minutes ago
            Interesting! You&#x27;re absolutely right, I&#x27;ll check
  • AmazingTurtle2 hours ago
    There are already a handful of repos doing essentially exactly this: `mlx-moe-offload`, `streamlx`, `mlx-moe`, `mlx-flash`, and `deepseek-v4-flash-mlx` - i.e. keep the resident parts of an MoE in unified memory and page&#x2F;stream routed experts from SSD on Apple Silicon.<p>At this point I&#x27;d much rather see people collaborate on one of these implementations, benchmark against them, or upstream the useful bits into MLX&#x2F;MLX-LM instead of producing yet another near-identical repo.<p>The local-LLM ecosystem really does not need every implementation idea rediscovered five times and wrapped in a new README. AI-assisted coding makes producing a new repo cheap; maintaining, benchmarking, and integrating one is the actually valuable part.
    • brailsafe7 minutes ago
      This is one of the aspects of this year that I&#x27;ve been finding very grating and wasteful. Collaboration still happens among people with the ability to do so and the technical skills, but everyone else is taking their own helicopter to the top of the mountain, &quot;putting it out there&quot;, and there&#x27;s just a ton of redundant projects that do the same thing.
    • carloslfu2 hours ago
      I see your point. As an oss defender myself, I agree, however, the spirit of this is to see how fast I can make it. I&#x27;m sharing this with the community, which I think is aligned with the original oss spirit.<p>It&#x27;s an experiment for myself but I am committing to maintain it. I&#x27;ve been an oss person for a loooong time, way before AI was a thing. Think about it as a new, from-scratch take at it, not as a re-reproduction.
      • xlayn1 hour ago
        Hey carloslfu, kudos from the other side of the internet, don&#x27;t get down on people nitpicking everything here, experimenting and discovering is part of learning so keep going!, remember this is the place that said dropbox was dumb and could be replaced by a script.
    • Barbing2 hours ago
      Vouched especially since OP might have a perspective on this. And readers may want to look up those other repos and compare for themselves.
      • carloslfu2 hours ago
        Thanks for the feedback! I&#x27;ll create a section with a benchmark and comparisons. This will hold the project accountable and speed things up imo
    • genxy2 hours ago
      Why should they do that? For you? You could merge those projects and see if they get traction.
    • kzrdude2 hours ago
      And there are `Mference` and `SwiftLM` too, I think they are doing the same use case.
    • dofm2 hours ago
      AI NIH
      • carloslfu2 hours ago
        Sorry, I don&#x27;t get &quot;NIH&quot;. what&#x27;s that?
        • noir_lord2 hours ago
          Not Invented Here.
          • carloslfu1 hour ago
            Ah! Yeah, I didn&#x27;t invent anything (yet!). The goal is to see how far I can take it in terms of speed without consuming that much RAM.
            • dofm1 hour ago
              I&#x27;m only joking anyway — it&#x27;s more a comment on the whole AI-accelerated trend of everyone having their own version of a thing.<p>I do agree that, ultimately, combining your efforts with others working in this whole area is probably really worth it, but I can see how there&#x27;s an ease of pushing forward on your own these days.<p>I do not have fast internet so I am not sure when I&#x27;ll really be able to download the weights but I do have an M1 Max to try this on, so I will at some point!
              • carloslfu1 hour ago
                I see! yes, downloading the weights part is painful. I tried a couple fixes and it is as fast as it can get downloading from HuggingFace. I think the field is heading toward smaller, more capable models soon, so you won&#x27;t have to wait that long!
    • api2 hours ago
      &gt; every implementation idea rediscovered five times and wrapped in a new README<p>That&#x27;s open source since forever, unfortunately.
      • carloslfu2 hours ago
        I agree with the sentiment, but have you seen those videos in which all men say other men are gay? This feels like the same, so much AI paranoia!<p>I genuinely want to contribute. And hey! I was doing oss this since 2014 so waay before AI was cool.
      • docheinestages2 hours ago
        It&#x27;s what happens when you don&#x27;t do market research.
        • carloslfu2 hours ago
          I&#x27;m sorry this makes it seem like I didn&#x27;t do my research. I did a TON. To fix it I&#x27;ll add a benchmark&#x2F;comparison table. Also, I wouldn&#x27;t call it market research since this is not commercial AT ALL.
          • EyMaddis2 hours ago
            Hey Carlos, thanks for sharing with the community! Appreciated
        • genxy1 hour ago
          Does a painter check to make sure that a portrait hasn&#x27;t been painted? What a dismissive comment.
        • oceanplexian1 hour ago
          Half the people on here are using Ollama. No one is doing market research.
  • bewareofscams1 hour ago
    [dead]
  • bewareofscams1 hour ago
    [dead]
  • aislopnogo1 hour ago
    [dead]