Weights
Also known as: model weights · parameters
Weights are the adjustable numbers inside a neural network. During training, the model tweaks them billions of times until its answers get good. Once trained, those final numbers are the model — all its “knowledge” lives in their values.
This is why:
- Size — a model with 7 billion weights is called a “7B” model. More weights = more capacity (and more to store and run).
- Files — when you download a model, you’re downloading its weights (often a several-GB file).
- Sharing — “open weights” just means the numbers are public for anyone to download and run.
Mental model: weights are a model's *mind*. Same code, different weights = different model, different behavior. That's the whole trick.