Show HN: TurboGPT: train 22KiB transformer in 13s (github.com)
27 points by lostmsu 3 hours ago
Inspired by minGPT for home experiments.
Requires CUDA 13.4; build script is Windows only.
fleshmonad 7 minutes ago
Who cares. This will not get you hired. Why do these submissions get any points at all. The same people who upvote this are the ones seething AI cannot create anything new. I don't see it. Not to shit on the work that went into this.
kadoban a minute ago
Jesus. Would hate to see the work you _do_ want to shit on.
vjsrinivas 5 minutes ago
Interesting trend of labeling models after how big they are on disk vs how many learnable parameters they have.
jey 2 hours ago
At that scale it seems like you could just solve the KKT conditions directly. Exaggerating but only a little