The Enterprise HiggingFace Hub plan recently introduced Storage Regions support as part of its offerings.
Storage Regions provide organizations with the capability to select the location for storing their models and datasets. This enhancement presents two primary advantages:
- Ensuring adherence to regulatory and legal standards, thereby reinforcing digital sovereignty.
- Boosting performance through enhanced download and upload speeds and reduced latency.
At present, the supported regions include:
- The United States 🇺🇸
- The European Union 🇪🇺
Additionally, there are plans to extend support to the Asia-Pacific region 🌏 in the near future.
The following section of the blog post will outline the steps for configuring this feature within the settings of an organization.
Organization Settings
If your organization is not an HiggingFace Enterprise Hub org yet, you will see the following screen:
Upon subscription, the Regions settings page will become accessible.
The page displays:
- A comprehensive audit detailing the current locations of the organization’s repositories.
- Dropdown menus to specify the preferred creation sites for future repositories.
Additionally, a ‘Repository Tag’ feature is implemented:
Each repository (be it a model or a dataset) situated in a non-standard location will have its Region exhibited prominently as a tag. This ensures that members of the organization can quickly identify the locations of repositories.
Regulatory and Legal Compliance
Certain regulated industries are obligated to store data within specific geographic boundaries.
For entities operating within the European Union, this feature enables the construction of machine learning processes in adherence to GDPR regulations, ensuring that datasets, models, and inference endpoints are all housed within EU data centers.
Enterprise Hub clients seeking more information on compliance are encouraged to make contact for assistance.
Performance
Locating models and datasets in proximity to an organization’s team and infrastructure can lead to marked enhancements in performance. This is particularly significant during uploads and downloads, given the substantial size of model weights and dataset files.
For instance, if your operations are based in Europe and you choose to store your repositories within the EU region, you could anticipate experiencing upload and download speeds that are approximately 4 to 5 times faster than you would if the repositories were located in the US.
Read related articles: