a from-scratch deep learning engine
about this specimen
Autograd, layers, optimizers, and a threaded vectorized C core — 512³ matmul at 8.6 GFLOPS, 34× faster than naive C on a phone. The engine behind cryotorch.
fig. 2 — one-line installanywhere corros runs
$ curl -fsSL https://raw.githubusercontent.com/CocoCopi/cryoquench/main/install.sh | sh