Hugging Face Hub Community Update: Pull Requests and Discussions

Hugging Face has launched pull requests and discussions on the Hugging Face Hub, providing a centralized collaborative environment for models, datasets, and Spaces. These features allow any community member to propose changes, ask questions, and interact directly with repository owners to accelerate ML development and ethical software iterations.

Collaborative Features: Discussions and Pull Requests

Hugging Face has integrated a new "Community" tab across all repository types, including models, datasets, and Spaces. This tab serves as the primary interface for two new collaboration tools:

Discussions

Discussions enable community members to ask questions, provide answers, and share ideas or suggestions directly with repository owners and the broader community. These are open to any user, facilitating direct communication within the repository's context.

Pull Requests

Pull requests allow users to open, comment on, merge, or close contributions directly through the website. To simplify the process, Hugging Face provides a "Collaborate" button in the "Files and versions" tab, which is specifically designed to facilitate single-file contributions.

Technical Implementation and Git Workflow

Hugging Face's implementation of pull requests differs from traditional git hosts like GitHub in three primary ways to better suit machine learning workflows:

  • No Forks Required: Instead of requiring contributors to fork a repository, the Hub uses custom "branches" called refs stored directly on the source repository. This removes the overhead of creating forks for every new version of a model or dataset.
  • Unified Interface: There is no hard distinction between issues and pull requests; they are treated as essentially the same and are displayed in the same lists.
  • ML-Specific Optimization: The system is streamlined specifically for ML repositories (models, datasets, and Spaces) rather than arbitrary general-purpose repositories.

Implications for Ethical ML and Community Contribution

The introduction of the Community tab aligns with ethical ML proposals by providing a transparent venue for feedback and iteration. Specific use cases for these tools include:

  • Improving Disclosures: Proposing suggestions in model cards to better disclose ethical biases.
  • Reporting Issues: Flagging concerning generations within a Space demo.
  • Direct Communication: Establishing a direct line of communication between dataset/model authors and the community.
  • Repository Enhancement: Allowing the community to contribute improvements, such as providing TensorFlow weights for a model.

Sources