Parameter
파라미터
Also known as: weights
The numbers a model learns. Their count is what people call model size.
Parameters are the weights of a neural network. Sizes are quoted as counts: seven billion, seventy billion.
Bigger is usually better, but not always. A smaller model tuned on domain data often wins, and it is faster and cheaper.
Do not confuse them with settings. Parameters are learned; the values a person sets are hyperparameters.
- Notation7B means seven billion.
- ChoosingWeigh speed and cost alongside quality.