매개변수와 층의 모양 살펴보기

Once a model exists, you can peek at every weight it holds. Inspecting the parameters builds trust before you train.

Call model.parameters() to iterate over every learnable tensor. The optimizer uses this exact same list.

Use named_parameters() to see each tensor paired with its name, so you know which layer it belongs to.

A linear layer holds a weight matrix and a bias vector. Their shapes follow directly from the in and out features.

Sum the elements of every tensor to get a total parameter count, a quick measure of model size.

Just printing the model shows its layers and their sizes in order. It is the fastest overview you can get.

Pass a sample batch through and print the result's shape to confirm dimensions line up end to end.

The requires_grad flag tells you which tensors get updated. Frozen layers show it as False.

The state_dict maps every parameter name to its tensor. It is what you save and load to checkpoint a model.

Most beginner errors are shape mismatches between layers. Printing shapes pinpoints exactly where the chain breaks.

Inspecting parameters and shapes turns a black box into something you fully understand. Confidence comes from looking. 🔍

Use parameters, named_parameters, and printing the model to inspect weights and shapes. This habit makes debugging fast and painless. 🎯