Weights
Weights are the billions of learned numerical values stored inside a model file—such as a .gguf or .safetensors download—that represent everything the AI knows.
During training, an AI reads massive amounts of text and code, constantly tweaking these numbers until it gets good at predicting the next word or pixel. Once training finishes, those numbers are frozen and saved to your hard drive as the model file. Unlike traditional software built out of if/else rules written by programmers, an AI model is essentially a giant lookup table of math driven by its weights.
Before you can type a single prompt, your local inference software has to copy all of those weights from your SSD into your GPU's VRAM (or Apple Unified Memory). Because the graphics card has to read through billions of weights every single time it generates a word, both the total size of the weights and the bandwidth of your memory dictate what hardware you can run.
Looking for verified hardware rigs?
Check our real-time £/GB VRAM index to find systems optimized for local model inference.