Open weights (a model whose trained weights are published; how it differs from open source)
Japanese オープンウェイト
An open-weights model is one whose trained weights, the numbers an AI learned, are published for anyone to download. It is not the same as open source, which also needs training information and code, so read the terms before use.
Open weights means the weights of an AI model have been published. The weights are the numbers an AI ends up with after learning from a huge amount of text, a bit like the feel a cook gains after years of practice. With them, you can run the model on your own computer or company server. It does not mean the recipe book, meaning the training data and the way it was made, is shared too.
Watch out Open weights and open source are not the same thing. The Open Source AI Definition asks for freedom to use, study, change and share, plus information about the training data and the code. Weights can be public while the license still has limits, so read the license before using a model for work.
Source: Open Source Initiative, The Open Source AI Definition 1.0; Meta’s LLaMa license is not Open Source