NullFoundry (@nullfoundry) profile picture

software maker. building weird things with AI, local LLMs and backend systems. shipping, breaking, benchmarking, repeat.

somewhere
Joined October 2019
Pokemon origami-style video generated. Btw 3090 minimax-h3 + ComfyUI
118
Replying to @Gemini
@Gemini could you fix your shit? This image was already created, but it wasn't loading. In fact, none of them are loading. thanks
1
62
hey @thsottiaux your button didn't work
29
playing with my own AI engine to run mimo flash on 32GB RAM ddr4 + 2x3090s trying to improve it as much as I can
1
47
I don’t regret spending the money and building my AI rig. I’m so happy I did this!
28
stupid questions are a pretty good model test playing with @TypeSafeAI (jev) and this is more fun than it should be napoleon's white horse → 87% white car wash 50m away 52% drive 40% walk the married looking at unmarried one actually broke it 59% cannot determine 37% yes it's yes. always yes. which one would your favorite model fail?
33
Watching @AnthropicAI and @OpenAI right now: unbelievable.
1
1
2
42
I'm not getting good results with dflash2. MTP is still better.
54
I’ve been hacking on a Qdrant plugin for Hermes Agent. If you’ve used Roo Code’s codebase indexing, it’s the same general idea: keep a searchable index of a repo so the agent can quickly figure out where to look. This one indexes each project incrementally, uses dense and lexical search depending on the query, and returns a small shortlist of files for Hermes to open. I’ve been testing it on a repo with around 3,000 files. The code is here: GitHub - github.com/brunocasado/herme…
2
262
interesting article build.nvidia.com/rtx/multi-g… I bought a couple of 3090s a few months ago, but I haven’t had enough time to try building a rig with both of them connected. So, I’m looking for a way to do this right without spending money on the wrong stuff. It's interesting that this article mentions it's wrong to mix different graphics cards, but I'm seeing a bunch of guys doing the opposite. Nvidia claims that using the same video card model can almost double performance. The image below shows something interesting. I was struggling to decide if I should go for 16x / 4x or search for a motherboard that's at least 8x/8x capable, and the answer is crystal clear. What do you think, people?
21