Minimal neural network in ~600 lines of Zig. GEMM-batched training on Apple Silicon via Accelerate. - View it on GitHub