Try tinygrad’s webgpu backend on windows: github.com/tinygrad/tinyg…

end to end will win neural network frameworks just like it is winning self driving cars. if you have a Conv3D guy, you are going to lose. he's the new cone guy.

Our MI350X machines are here, thanks AMD! They are just two racks down from their MI300X friends.

tinygradのコードとかドキュメントとか参考にしながらやってるんだけど、これマジですごいな。読めば読むほど感動が生えてくる。 洗練された設計は読んでいて楽しいな。

MI350X draws 250W at idle?!

x.com/i/article/1943…

之前在安装 tinygrad 的时候，选择了错误的 cuda 版本，导致一直 generated code ptx 版本有问题，不断反复安装都没有解决，让 claude-code 接手之后，他找了到 cache 成功解决了问题

Got my company laptop and the first thing I did after convincing IT to give me WSL..? Install cuda and ran beautiful_mnist.py the tiny corp

All of tinygrad, including non existent dependencies, fits in the context window. Seed AI type vibes.

America has once again put up roadblocks to us getting a visa. And I'm sure it's not just us. Every decision made like this slowly chips away at America's lead in AI. Most Americans have never lived in a country that wasn't rich, but this is the road to that.

tinygrad is a bet against irreducible complexity. When you look at Triton and similar, so many things are GPU specific. The task in reverse engineering is ignoring the labels, what is that really? Forget the abstractions, search for the perfect program that runs on the hardware.

This is tinygrad's description of the tensor cores of all the major GPUs. No per GPU dialects, just a spec for what they each are.

Check out our complete zero dep 139 line Llama implementation. github.com/tinygrad/tinyg…

Every week we have a meeting open to all in our Discord, and it's meeting time now!

Python is an amazing language. It needs more work on its type system, but no language is close for the ability to express raw ideas with minimal overhead.

Who thinks tinygrad is going to win?