micrograd CUDA Teaching myself basic CUDA by building GPU-accelerated tensor-based autodiff from the ground up, inspired by Andrej's micrograd. No dependencies other than Python's standard library and CUDA. Compiling To compile the CUDA kernels: nvcc -shared -o liboperations.so micrograd_cuda/operations.cu -Xcompiler -fPIC Usage import random import time from micrograd_cuda.mlp import MLP from micrograd_cuda.tensor import Tensor from micrograd_cuda.operations import Operations # Model model = ML...