pytorch-bert-crf-ner

Why we included this project

Extracting named entities from Korean text is the use case this repository targets. It builds on KoBERT and adds a CRF layer so the model outputs a consistent sequence of entity tags instead of independent guesses per token. The included notebooks run the tagger on real sentences, which makes the training and inference flow easy to follow if you want to adapt it to your own labels. It reads as a well-organized reference implementation rather than a supported production service, so expect to treat it as a starting point for your own pipeline.