We’re incredibly excited to share essentially the most impactful feature since Argilla joined Hugging Face: you possibly can prepare your AI datasets with none code, getting began from any Hub dataset! Using Argilla’s UI, you possibly can easily import a dataset from the Hugging Face Hub, define questions, and begin collecting human feedback.
Not accustomed to Argilla? Argilla is a free, open-source data-centric tool. Using Argilla, AI developers and domain experts can collaborate and construct high-quality datasets. Argilla is a component of the Hugging Face family and fully integrated with the Hub. Need to know more? Here’s an intro blog post.
Why is that this latest feature necessary to you and the community?
- The Hugging Face hub incorporates 230k datasets you should use as a foundation in your AI project.
- It simplifies collecting human feedback from the Hugging Face community or specialized teams.
- It democratizes dataset creation for users with extensive knowledge about a selected domain who’re unsure about writing code.
Use cases
This latest feature democratizes constructing high-quality datasets on the Hub:
- If you may have published an open dataset and need the community to contribute, import it right into a public Argilla Space and share the URL with the world!
- If you must start annotating a brand new dataset from scratch, upload a CSV to the Hub, import it into your Argilla Space, and begin labeling!
- If you must curate an existing Hub dataset for fine-tuning or evaluating your model, import the dataset into an Argilla Space and begin curating!
- If you must improve an existing Hub dataset to learn the community, import it into an Argilla Space and begin giving feedback!
How it really works
First, it’s essential to deploy Argilla. The really useful way is to deploy on Spaces following this guide. The default deployment comes with Hugging Face OAuth enabled, meaning your Space will probably be open for annotation contributions from any Hub user. OAuth is ideal to be used cases while you want the community to contribute to your dataset. If you must restrict annotation to you and other collaborators, check this guide for extra configuration options.
Once Argilla is running, check in and click on the “Import dataset from Hugging Face” button on the Home page. You possibly can start with certainly one of our example datasets or input the repo id of the dataset you must use.
In this primary version, the Hub dataset should be public. For those who are fascinated by support for personal datasets, we’d love to listen to from you on GitHub.
Argilla routinely suggests an initial configuration based on the dataset’s features, so that you don’t need to begin from scratch, but you possibly can add questions or remove unnecessary fields. Fields should include the information you would like feedback on, like text, chats, or images. Questions are the feedback you would like to gather, like labels, rankings, rankings, or text. All changes are shown in real time, so you possibly can get a transparent idea of the Argilla dataset you’re configuring.
When you’re joyful with the result, click “Create dataset” to import the dataset along with your configuration. Now you’re ready to present feedback!
You possibly can do that for yourself by following the quickstart guide. It takes under 5 minutes!
This latest workflow streamlines the import of datasets from the Hub, but you possibly can still import datasets using Argilla’s Python SDK when you need further customization.
We’d love to listen to your thoughts and first experiences. Tell us on GitHub or the HF Discord!
