We’re on a journey to make our libraries more responsible, one commit at a time!
As a part of the Diffusers library documentation, we’re proud to announce the publication of an ethical framework.
Given diffusion models’ real case applications on the earth and potential negative impacts on society, this initiative goals to guide the technical decisions of the Diffusers library maintainers about community contributions. We want to be transparent in how we make decisions, and above all, we aim to make clear what values guide those decisions.
We see ethics as a process that leverages guiding values, concrete actions, and continuous adaptation. For that reason, we’re committed to adjusting our guidelines over time, following the evolution of the Diffusers project and the helpful feedback from the community that keeps it alive.
Ethical guidelines
- Transparency: we’re committed to being transparent in managing PRs, explaining our selections to users, and making technical decisions.
- Consistency: we’re committed to guaranteeing our users the identical level of attention in project management, keeping it technically stable and consistent.
- Simplicity: with a desire to make it easy to make use of and exploit the Diffusers library, we’re committed to keeping the project’s goals lean and coherent.
- Accessibility: the Diffusers project helps lower the entry bar for contributors who may also help run it even without technical expertise. Doing so makes research artifacts more accessible to the community.
- Reproducibility: we aim to be transparent in regards to the reproducibility of upstream code, models, and datasets when made available through the Diffusers library.
- Responsibility: as a community and thru teamwork, we hold a collective responsibility to our users by anticipating and mitigating this technology’s potential risks and dangers.
Safety features and mechanisms
As well as, we offer a non-exhaustive – and hopefully constantly expanding! – list of safety features and mechanisms implemented by the Hugging Face team and the broader community.
-
Community tab: it enables the community to debate and higher collaborate on a project.
-
Tag feature: authors of a repository can tag their content as being “Not For All Eyes”
-
Bias exploration and evaluation: the Hugging Face team provides a Space to reveal the biases in Stable Diffusion and DALL-E interactively. On this sense, we support and encourage bias explorers and evaluations.
-
Encouraging safety in deployment
-
Secure Stable Diffusion: It mitigates the well-known issue that models, like Stable Diffusion, which can be trained on unfiltered, web-crawled datasets are inclined to suffer from inappropriate degeneration. Related paper: Secure Latent Diffusion: Mitigating Inappropriate Degeneration in Diffusion Models.
-
Staged released on the Hub: in particularly sensitive situations, access to some repositories needs to be restricted. This staged release is an intermediary step that enables the repository’s authors to have more control over its use.
-
-
Licensing: OpenRAILs, a brand new form of licensing, allow us to make sure free access while having a set of restrictions that ensure more responsible use.
