Skip to main content

Module Introspection

A model is a tree. Every node is a module, and every leaf is a parameter.

Key Insight​

nn.Module is PyTorch's base class for all neural network components. When you assign self.linear = nn.Linear(...) inside __init__, the parent class notices and registers it. named_modules() walks this tree recursively, yielding every sub-module and its dot-separated path, so you can inspect any model without modifying it.

Why This Matters​

Knowing every layer's name, type, and parameter count is the foundation for debugging, profiling, and targeted fine-tuning. It is also the first step toward weight surgery: you must know the exact key names in a model before you can load or remap them.