Prodigy-HF: Direct Hugging Face integration for training and publishing models from Prodigy

AI Impact Summary

Prodigy now includes a plugin (Prodigy-HF) that directly bridges Prodigy with Hugging Face, enabling end-to-end model training on user-annotated data via hf.train.ner and hf.train.textcat, and enabling model reuse or publishing to the Hugging Face Hub with hf.upload. The workflow leverages AutoTokenizer and AutoModel, allowing fine-tuning on domain data with simple commands, which can substantially shorten the path from annotation to production-ready transformer models. Expect dependencies on the Hugging Face Transformers stack, local compute requirements, and Hub authentication when training or uploading models.