We’re thrilled to announce the discharge of our latest collaborative features: pull requests and discussions on the Hugging Face Hub!
Pull requests and discussions can be found today under the community tab for all repository types: models, datasets, and Spaces. Any member of the community can create and take part in discussions and pull requests, facilitating collaborations not only inside teams, but in addition with everyone else in the neighborhood!
It’s the most important update ever done to the Hub, and we will not wait to see the community members start collaborating with it 🤩.
The brand new “Community” tab also aligns with proposals in ethical ML throughout the years. Feedback and iterations have a central place in the event of ethical machine learning software. We actually consider having it in the neighborhood’s toolset will unlock recent sorts of positive patterns in ML, collaborations, and progress.
Some example use cases for discussions and pull requests:
- Propose suggestions in model cards to enhance disclosures of ethical biases.
- Let users flag concerning generations of a given Space demo.
- Provide a venue through which model and dataset authors can have a direct discussion with community members.
- Allow others to enhance your repositories! For instance, users might want to supply TensorFlow weights!
Discussions
Discussions allow community members ask and answer questions in addition to share their ideas and suggestions directly with the repository owners and the community. Anyone can create and take part in discussions in the neighborhood tab of a repository.
Pull requests
Pull requests allow community members open, comment, merge, or close pull requests directly from the web site. The simplest method to open a pull request is to make use of the “Collaborate” button within the “Files and versions” tab. It would allow you to do single file contributions very easily.
Under the hood, our Pull requests don’t use forks and branches, but as a substitute, custom “branches” called refs which are stored directly on the source repo. This approach to avoids the necessity to create a forks for every new edition of the model/dataset.
How is that this different from other git hosts
At a high level, we aim to construct a less complicated version of other git hosts’ (like GitHub’s) PRs and Issues:
- no forks are involved: contributors push to a special
refbranch directly on the source repo - no hard distinction between issues and PRs: they’re essentially the identical so we display them in the identical lists
- streamlined for ML (i.e. models/datasets/Spaces repos), not arbitrary repos
What’s next
In fact, it’s only the start. We are going to take heed to the community feedback so as to add recent features and improve the community tab in the longer term. If you could have any feedback, you may join the discussion here. Today is the perfect time to hitch your first discussion and open a PR! 🤗
