When a model is called seven B or seventy B, that B means billion — as in billions of parameters. Parameters are simply the numbers inside the model. In a real sense, they are the model.
Picture a giant wall covered in tiny dials. Training is the long, slow process of nudging every single dial to just the right setting. A seven B model has seven billion of them, all tuned.
Each dial on its own is meaningless. But turned together, just so, they store everything the model knows — grammar, facts, style, reasoning. Knowledge, smeared across billions of little settings.
More dials means more room to learn, so bigger models are often smarter. The catch: every extra dial is more memory to hold and more maths per word, so it's heavier and slower to run.
So parameters are the model's billions of tuned dials — more of them means more capacity, and more VRAM. That number in the name is the size of the brain. That's parameters. Off the Cloud.