Tensorizing Micrograd in Rust
Taking the scalar autograd engine from the last post and making it hold matrices instead of single numbers, then watching it fall apart on MNIST until I fixed the weight init.
Taking the scalar autograd engine from the last post and making it hold matrices instead of single numbers, then watching it fall apart on MNIST until I fixed the weight init.
Building a small scalar autograd engine in Rust from scratch, then using it to train a tiny neural network. Based on Karpathy’s micrograd.