This repository implements named entity recognition for Chinese text using BERT and ALBERT pre-trained models. It provides three decoding approaches: Softmax classification, conditional random field (CRF) layers, and span-based extraction. The project supports adversarial training, focal loss, and label smoothing for improved performance. It evaluates on CLUENER and CNER datasets with benchmark F1 scores around 0.81 for BERT+Span.