Weights

What it is

Weights are the numerical values within a neural network that determine how information flows between neurons. Each connection between neurons has a weight. During training, these weights are adjusted to minimise errors. A trained model's knowledge is encoded entirely in its weights.

Why it matters

When you download a model or see model files, you're getting weights. When people talk about "the model weights leaking," they mean the actual learned parameters. Weights are what make each model unique—two architectures with different weights will behave completely differently.