The adjustable values inside a model.
THE DEFINITION
One adjustable value inside the model, set during training.
A model with seven billion parameters has seven billion such values.
WHAT SIZE CORRELATES WITH
Generally, capability. Larger models handle more complex tasks and know more.
WHAT SIZE DOES NOT GUARANTEE
Accuracy. Larger models still hallucinate. Suitability. A smaller model may be better for a narrow task. Speed. Larger models are slower and more expensive.
THE TRADE-OFFS
Large: more capable, slower, costlier
Small: faster, cheaper, can run locally, less capable
WHAT MATTERS BESIDES SIZE
Training data quality, training technique, and how the model was aligned.
A well-trained smaller model can beat a poorly trained larger one.
WHAT TO ACTUALLY CARE ABOUT
Whether it does your task well. Test rather than comparing numbers.
RELATED TERMS
Model, training, fine-tuning.
