01 · The Big Picture
Parameters
The numbers a model learns
The learned numerical values inside a model, also called weights. Training uses optimization to adjust them so the model performs better on its objective. Parameter count is one measure of scale, not a direct score of quality or knowledge.
Concrete example
A 70-billion-parameter model contains roughly 70 billion learned values, but its usefulness also depends on data, architecture, training, and inference setup.
Why it matters
The usual shorthand for how big a model is — and exactly what an “open-weight” release hands over.