Model weights (the numbers a model learns in training)
Japanese 重み(ウェイト)
Model weights are the many numbers an AI learns during training and multiplies with its input to work out an answer. What the model has learned is stored in them, so the terms for sharing them matter.
Model weights are the many numbers stored inside an AI model. While training, the AI looks at a lot of text and images and nudges these numbers little by little. When training ends, the adjusted numbers stay. Think of a chef's feel for flavor after years of practice: what the model has learned lives in this bundle of numbers. When the AI answers, it multiplies the numbers with the input to calculate its reply.
Watch out Sharing the weights does not make a model open source. Weights you can download still come with their own terms, and some do not allow commercial use. Read the license before using them for work.
Source: Google for Developers, "Machine Learning Glossary"; Open Source Initiative, "The Open Source AI Definition – 1.0"