Dave " " Buchhofer

@DavidKPiano sweet thanks, I was just trying out xstate last week and this seems to have made some solid suggestions on where to push next to improve my machine.

Released the XState v5 skill 🚀 The goal isn't just "know the API" but rather "help agents make really good state machines" Tuned & tested against real examples, common anti-patterns, and the mistakes* LLMs tend to make with XState npx skills add statelyai/skills

@dmarcos @__tinygrad__ A lot of that researchy/mesh gen stuff is very heavily tied to cuda, and isn’t always a one-to-one port to get working

@vsaitoo @__tinygrad__ Nice! Thanks! You do 3d graphics. How does it work on the Mac for blender workflows, training gaussian splats, do you run any 3d meshes generative, world models locally?

Qwen 3.5 27B getting 18.5 tok/s on Mac Mini with external 7900XTX. It should be able to be 3x faster than this with work, SSM stuff is still in PR. Hopefully Mac eGPU support brings in devs.

@dmarcos @__tinygrad__ I initially started playing with it with the intent to do splats. I have a local pipeline that is a mixture of brush and tinygrad that kind of works, but it's been a learning experience. Nothing‘s going to be out of the box yet. It's not a general driver, it's a backend for TG

@dmarcos @__tinygrad__ ADT-UT3G is the one the one I am using that works. (same one from the picture I believe) should be other comparable options though.

@__tinygrad__ Any eGPU enclosure recommended or it doesn’t matter?

me, scoffing in the back row of SIGGRAPH: ok I guess this is good if I ever need to render a teapot I guess??

@tosolini @XGRIDS_OFFICIAL @XRarchitect Always a classic effect

Explore a 3D Gaussian Splatting model via webcam. Features include face tracking for off-axis parallax and hand gestures for rotation and scaling. Developed with Gemini using @XGRIDS_OFFICIAL PortalCam captures. Inspired by @XRarchitect and Johnny Lee. gemini.google.com/share/8ff5e93b…

@keenanisalive Thanks, looks like a handy replacement/enhancement on the 'gltf-transform inspect' and https//gltf.report drag-and-drop options I tend to fall into currently depending on which use case I'm in at any moment.

@RadianceFields @lichtfeldstudio @SplatK1ng Fancy giveaway SplatKing / @SplatK1ng and quality engagement hackery sir, /salute. Will check it out.

I'm giving away a NVIDIA RTX PRO 6000, but you only have three days left to enter. Also my capture app, @SplatK1ng, is now available in the EU! Thank you to NVIDIA for providing the GPU and hosting me at GTC.

@__tinygrad__ Yep, set up a 3090 on my desk a couple weekends back, it's been an interesting experiment

Mac Mini + eGPU. Both NVIDIA and AMD supported.

your biggest enemy is still complexity. it's also your agent's biggest enemy. but it has no holistic view of your code base, so it keeps adding complexity. and you think that's how it's supposed to be, because the clanker shat it out, and you don't know the stack. glhf!

when you think of feature A your job is to recognize that it's actually a special case of feature X and if you implement feature X it unlocks feature A B C people are struggling with this lately because they can't think bigger than what their agent can do

neat, 38tok/s is better than I expected. Especially given I'm not sure how many terrible sins have been generated while getting it to this point.

It's dangerous now that it’s easy-ish to run a wild ass experiment just to learn? can I run the new qwen model on osx locally via my old 3090 in tinygrad vulkan -> llama.cpp -> egpu osx egpu worked, it's ~3x faster than the same on metal for stuff that can stay GPU resident?

Unsurprisingly, it's choking under the sheer weight of the vibes before it's functional, but I still enjoy this learning by osmosis/fail style on occasion. Back to things I KNOW how to do for a while, see where this ends up next weekend maybe!

it loaded! 🎉 ✅ Qwen3.5-35B-A3B loaded successfully! Architecture: qwen35moe Blocks: 40 GatedDeltaNet: 30, Transformer: 10 Experts: 256 total, 8 active That's realize=False though — and a whole lot of random ass changes. Guess we'll try more tomorrow after all?

Anyway, no one should take anything from this as fact. Seems obviously something that I'm just a little too early on, and I'm looking forward to where it goes when my comma has these new compute powers (!)

just half pushing on it until it explodes while walking the dog is a wild place to be?!

The VRAM problem seems architectural in tinygrad’s load path? need a Q4_0Linear that stores quantized blocks and dequants in the forward pass (like Int8Linear already does)?

The actual quantized files have mixed quant types all over. Had a script that requantized 60 tensors (Q5_0 -> Q4_0) In bartowski Q4_0. After that Pass 1 (realize=True): KeyError: ‘qwen35moe.feed_forward_length’ • Pass 2 (realize=False): MemoryError at ~18.71 GB allocation

New ad by the Norwegian Consumer Council: "A Day in the Life of an Ensh*ttificator"

> Status: GREEN. The eGPU stack works for LLMs on macOS. that’s a little bit of a stretch at one token per second for the 30 B model, but I wasn’t really expecting you to work at all. I just wanted to know where it broke. 3090 24gb via egpu on osx? Silly.