uie_pytorch

Why we included this project

This is a PyTorch port of UIE, the universal information extraction model that PaddleNLP originally shipped, aimed at anyone who works with Chinese text and needs structured facts out of messy documents. You describe what you want in plain language, naming entity types, relations, events, or opinion targets, and the model runs zero-shot extraction with no training data, which makes it useful for structuring medical reports, legal judgments, financial statements, or accident records. The repo also includes scripts for fine-tuning on your own labeled data, evaluating results, converting doccano annotations, and exporting to ONNX for deployment. Teams already on the PyTorch and transformers stack get UIE without having to adopt PaddlePaddle, and a range of pretrained checkpoints covers base, medical, and English variants.