Hugging Face is an Artificial Intelligence (AI) company which develops social AI-run chatbot applications. For this, Hugging Face developed its own Natural Language Processing (NLP) model called Hierarchical Multi-Task Learning (HMTL). Furthermore, can be more recognized for the Transformers which is a library of state-of-the-art pre-trained models for PyTorch, TensorFlow and JAX.

Moreover, Hugging Face offers a Git-based hosting platform to store Models, Datasets and Spaces, called the Hugging Face Hub.

Hugging Face Hub

The platform has the potential (or it already has) to build an active, healthy and strong Open Source community around the AI field. The dynamic this platform follows is similar to other code hosting platforms such as GitHub. However, Hugging Face Hub repositories can interact with each other, that is, a model uploaded by any user can use a dataset uploaded by any other user. Moreover, there are repositories that host demo apps of models so that users can understand how the model works. These repositories are called Spaces.

Repository Example

Hugging Face repositories use Git repositories and some users may find a similar interface with other code hosting platforms. The repositories are organized in three pages:

• Model card, where is given the information of the repository (readME, some stats, and repositories related, such as datasets or spaces for a model repository).
• Files and versions, similar to other code hosting platforms, it is where the files, commit history and branches of the repository are shown.
• Community, newly added, it is built as a simpler version of other git hosts' pull requests and issues. There are two types of threads: pull requests and discussions, but there is no hard distinction between them.

To visualize how a repository is organized, in the following we show the model repository bert-base-uncased.