Models & Architecture
Weights
The individual strengths of the connections in a neural network, which are what actually change as a model learns.
A weight is a number attached to a connection between two units in a network, controlling how strongly a signal passes along it. Training is the process of nudging millions or billions of these weights until the model produces good outputs. When people talk about open weights, they mean a model whose trained weight values have been released so anyone can run or build on it.
Related terms
Parameters
The adjustable numbers inside a model that get tuned during training and store what the model has learned.
Neural Network
A model loosely inspired by the brain, built from layers of simple units that each pass signals to the next.
Backpropagation
The algorithm that figures out how to adjust each weight in a network to reduce its errors.
Gradient Descent
The optimization method that gradually tweaks a model to lower its error, like walking downhill to the lowest point.