Live data from Hacker News

Viewing profile — Tepix

Tepix

HN member
Joined
Mon, Aug 19, 2013, 9:32 AM UTC
HN karma
13,892
Public activity
6,145 items

About Tepix

German sw dev, admin and former founder interested in privacy, sovereign GenAI, decentralisation, sustainability, security, open source, 360° media, travel …

I do not use this alias anywhere else.

Got something cool? Get in touch at username at altmails.com

Recent public activity

  1. comment
    Comment #49219270

    An infinite number of data centers even!

  2. comment
    Comment #49219247

    Are you talking about S or XS? S is too large for a 3090 at 118b parameters.

  3. comment
    Comment #49206748

    The license is clear. Your post merely adds to the FUD.

  4. comment
    Comment #49201144

    That‘s the same price as a Keyboardio Model 100 which is split, wooden and has sculpted RGB backlit keys. Whoah. That said, big fan of that volume button!

  5. comment
    Comment #49183264

    Privacy is often a requirement.

  6. comment
    Comment #49177848

    Addendum: I was wrong, you‘ll need two of these cards.

  7. comment
    Comment #49173856

    Yes. It's called REAP and from what I've seen, results aren't stellar.

  8. comment
    Comment #49173833

    For sure if you want to properly utilize the model with several users in parallel and large context you'll want two MI350P.

  9. comment
    Comment #49168718

    If you have 2x DGX Spark it will run quite nicely. They cost only $8000 or so and use less power so you may be able to rent them cheaper than the MI300X. I found an offer to rent t…

  10. comment
    Comment #49168702

    Unfortunately, the MI300X is an OAM module. The MI350P is the one you want: It's a PCIe card, but it has less memory: 144GB. Luckily, DeepSeek V4 Flash will run in 144GB too becaus…

  11. comment
    Comment #49168581

    I share the author's sentiment. But I have another hatred: Monospace fonts in blogs. Monospace makes text harder to read and ugly. It was a necessity. It no longer is. Get over it.…

  12. comment
    Comment #49165179

    Latent space is calling the whole subject "inference engineering" in their episode https://www.latent.space/p/inference-eng

  13. comment
  14. comment
    Comment #49165169

    Well, INT4 is good enough apparently and it's very fast.

  15. comment
    Comment #49165167

    The latest Latent.Space episode https://www.latent.space/p/inference-eng

  16. comment
    Comment #49165158

    vLLM tested with Llama-3.3-70B-Instruct, Qwen3-30B-A3B-Instruct-2507, Qwen3-30B-A3B-Thinking-2507, and Qwen3.5-27B. I'm wondering if it needs to be tested with every other model or…

  17. comment
    Comment #49164421

    It uses javascript. Still a static file. Lets you play chess

  18. comment
    Comment #49155674

    Put it on a boat, navigate to an area with a current away from any land and then slowly release it

  19. comment
    Comment #49153946

    Why is there no orange? The entire colour palette is a bit depressing. Also my terminal (Ubuntu in WSL2) had an ugly colour palette defined by default with several identical colour…

  20. comment
    Comment #49151745

    I consider less than 4bit to be too poor quality usually. Got any benchmarks to measure quality?

  21. comment
    Comment #49151736

    Workaround: reduce font size on iPhone SE

  22. comment
    Comment #49122933

    Yes, the Dual DGX Spark looks like the sweet spot for this model for now. Good preprocessing speed. Lots of context. Fast enough for 1-5 devs perhaps. Around 8200€ as of today (use…

  23. comment
    Comment #49122870

    DS4Flash has 284B weights. 64GB? No go.

  24. comment
    Comment #49122853

    Do they do other things with the data, like selling it?

  25. comment
    Comment #49121629

    Why mention it? Just download the weights when they become available.