What is Weights?
They are numerical values learned by artificial intelligence models during training that shape their decisions.
Overview
They are massive tables of numbers representing the strength of synaptic connections in an artificial intelligence model's brain. While the model is being trained, these numbers are continuously updated, and it is through this that the importance of each piece of information is determined. The actual accumulated knowledge that enables a model to take an input and produce a correct output is stored in these values.
How it works
As the model is trained with millions of data points, these numerical values are adjusted by small amounts every time it makes a mistake. When the training is finished, these values are fixed, and the model now responds to newly incoming questions based on these optimized numbers.
Where it is used
In files of large language models·that is, LLMs·that can be downloaded and run on a computer, such as GGUF formats, these numerical data are directly included.
Commonly confused with
It is confused with the model architecture. The architecture is like the floor plan of the rooms in an empty building; the weights are the items placed inside those rooms that make the building livable.
Questions and answers
Can we manually change these numerical values?
It is theoretically possible, but because there are billions of parameters, instead of doing this manually, it is preferred for computers to train them automatically.
Are these values shared in open-source models?
Yes, in open-weights models, these numerical values are made publicly available for anyone to download and use.
Related terms
This explanation was written in plain language for TreScout with AI and machine-translated from the Turkish original · the Turkish version prevails. If something looks wrong or missing, write to hello@trescout.com. Read in Turkish →