Phase 4 · Deep learning · Tooling
PyTorch autograd and modules
Let PyTorch compute gradients; build models with nn.Module.
Open this lesson Create a free account
Why it matters
You've built backpropagation by hand and as a tiny engine. PyTorch's autograd is the same engine for tensors, and nn.Module is how every PyTorch model, from a two-layer classifier to Llama, organizes its weights. After this lesson you can read the model code in most open-source AI projects, and you're one step away from the standard training loop.
What you'll be able to do
- Get gradients from PyTorch with requires_grad and backward()
- Control the graph: accumulation and zero_grad, no_grad, detach
- Build a model as an nn.Module from nn.Linear layers and inspect its parameters
- Write one manual gradient-descent step on a model
Before this lesson
Next in this phase: GPUs and compute