We are excited to announce the release of Prodigy-HF, a new integration that streamlines the connection between Prodigy, our annotation tool, and the Hugging Face model hub. This direct link enables users to load models from Hugging Face directly into Prodigy for active learning, evaluation, and fine-tuning tasks.
Prodigy-HF simplifies the workflow by allowing you to pull in pre-trained models, use them for real-time predictions during annotation, and export your annotated data back to formats compatible with Hugging Face. This integration is designed to speed up the iterative process of improving NLP models through human-in-the-loop annotation.
The integration supports a wide range of tasks including text classification, named entity recognition, and question answering. Users can leverage the extensive catalog of models available on Hugging Face to bootstrap their annotation projects, reducing the time needed to achieve high-quality labeled datasets.
For existing Prodigy users, the new integration is available as an extension that can be easily installed via pip. Detailed documentation and example workflows are provided to help teams get started quickly. We believe this will be a valuable tool for researchers and engineers working on custom NLP solutions.