Adebayodamilola20/autograd-engine ? reverse-engineered prompt
Reverse engineered prompt
Build me a Python project called nabla that does automatic differentiation from scratch, with no PyTorch, TensorFlow, or JAX. I want to be able to create values, combine them with normal math, call backward, and get correct gradients through the whole computation graph. Please include a small neural network library on top of it, with layers, activations, losses, and a training loop, plus a simple MNIST example that can reach about 97 percent accuracy.
Also add a tiny web demo where I can draw a digit and see the prediction and how the hidden layers are behaving, and include a text generation example that can train on a plain text file and generate new text from a prompt. Make it clean, well tested, and easy to read, since the main goal is to understand how backpropagation works as much as to use it. If you need to check the latest docs for anything, feel free to look them up online.
Are you gonna build this?
make sure you review the code using coderabbit