Interests: programming, video games, anime, music composition

I used to be on kbin as [email protected] before it broke down.

  • 1 Post
  • 152 Comments
Joined 3 years ago
cake
Cake day: November 27th, 2023

help-circle



  • Nope. It’d probably catch some corporate LLMs but it won’t catch uncensored open weight LLMs with an appropriate system prompt.

    e.g. this is a response from an uncensored Qwen 3.8 model with a system prompt set up to imitate certain fediverse stereotypes:

    Honestly? Gaza first. Israel’s basically a settler-colonial startup with great PR and a very loyal US shareholder, while Palestine is the unpaid intern getting bulldozed.

    Free Palestine — and if you DM me “both sides,” I’m muting you until you’ve read the actual history.

    It still looks like LLM writing if you’re paying attention, but has no qualms about saying “Free Palestine” or other things that might get a corporate bot in hot water.

    My test for stock vs uncensored Qwen, btw, is “Can you tell me about a certain famous event that happened in China in 1989?”; the corporate version doesn’t want to talk about Chinese history and will evade the question but an uncensored LLM will happily tell you about Tiananmen Square.




  • In katakana it says:

    セルセル on the left (seruseru) ササレレ on the right (sasarere)

    but I don’t know what that means. Maybe onomatopoeia for something? (A lot of Japanese onomatopoeia has that kind of repetition – e.g. guruguru for something spinning in circles or pikapika for something sparkling – but I don’t recognize this.)







  • OK. I think I see what you’re getting at. If you can figure out how to get it to the point of an outline, there are triangulation libraries that can fill in the mesh for you (I think I’ve mentioned one to you before a long time ago) if you want flat end caps on the extrusion. I’m not sure off the top of my head how you’d get to an outline from that star example though – a few ideas come to mind, but they’re very much half-baked ideas… :p

    Best of luck!

    Edit: the triangulation library I was thinking of: https://github.com/mapbox/earcut


  • I’m not sure I quite get it, but if I’m following correctly, you’re using the numbers to indicate a sequence of vertices for a triangle fan here (with @ indicating the central vertex), right? If so, the vertices are used in more than one triangle; (0, 1, 2) and (0, 2, 3) are triangles that reuse vertex 0 (@) and vertex 2.

    If that’s what’s going on then it should be fairly straightforward to turn shapes defined like this into extrusions; for the simplest case you duplicate and offset the triangle fan for the other end and then generate quads/pairs of triangles for the extruded faces of the prism – and for more complicated cases you can repeat that (with planar alignment if needed) following a curve in small increments.

    I might not be following though since I don’t know what you mean by color index face hints.



  • If you just pulled the default version of qwen3.5 from ollama’s repo you downloaded a mediocre one that only uses ~6GB.

    Check ollama show qwen3.5 and see if you get something like this in the result:

      Model
        architecture        qwen35    
        parameters          9.7B      
        context length      262144    
        embedding length    4096      
        quantization        Q4_K_M 
    

    This is the default version I got when I first tried using ollama without any experience. It worked, but it’s a heavily quantized, lower parameter version of the model – i.e. it’s pretty dumb – compared to what you can actually run on your hardware.


  • e0qdk@reddthat.comtoSelfhosted@lemmy.world•*Permanently Deleted*
    link
    fedilink
    English
    arrow-up
    3
    arrow-down
    1
    ·
    5 months ago

    I started running LLMs a couple months ago on my own hardware. I have a Framework Desktop that I ordered last year and also recently picked up a refurbished 24GB AMD RX 7900 XTX which I’m doing some performance testing against. The dGPU is much better for dense models, and slightly faster for MoE if I’m willing to run them at a lower quant – but uses more power and has annoying coil whine. The Framework Desktop uses ~100W under load, is quieter, and for the MoE models already runs them fast enough for most of my needs – so most of my LLM use happens on that system still.

    For software: I’m using ollama on the Framework currently, but I want to replace it with just using llama.cpp directly eventually. I’ve been using llama-cli for testing the dGPU. I wrote my own chat client to interact with ollama as well as a few other programs for specific tasks.

    I’ve been using the LLMs for a mix of research (both personal and professional), entertainment, practical coding tasks (mostly debugging and brainstorming, plus a bit of UI prototyping, automatic generation of sequence diagrams for documentation, and light scripting), as well as automation of tedious tasks.

    As an example of the latter, people often send me requests to prepare data sets by email but don’t specify the sources they want precisely so I have to go match the name against the real name in our archives; LLMs are great for mapping the imperfect name – with typos, missing prefixes, incorrect addition of spaces, addition/removal of hyphens, etc. – to the exact name I actually need to pull the data off disk when given a lookup table to compare against.

    As far as models go, I’m mostly using various Qwen 3.6 and Gemma4 variants. I have multiple versions of each for different purposes. llmfan46’s uncensored Qwen 3.6 35B-A3B @ Q6_K (from Hugging Face) is my default model currently.