Hugging Face

Hugging Face is a platform for publishing and consuming machine learning artifacts. A model repository holds weights and a configuration; a dataset repository holds data and a loader script; a Space holds a running application. All three are versioned with git, so an artifact has a commit history and a revision that can be pinned. The library ecosystem is the other half. Transformers, Datasets, and the surrounding packages provide the loading code that the repositories assume, so a model card and its weights are usable with a few lines. That pairing - a hosting convention plus the library that reads it - is why the hub became the default place to publish an open model. Spaces host small applications, often a demo of a model in the same repository. Because a Space is a container with a web interface, a model can be tried before it is downloaded, which shortens the path from reading about a model to evaluating it.