macanv/BERT-BiLSTM-CRF-NER

Implements a production-ready NER pipeline combining BERT embeddings with BiLSTM-CRF sequence tagging, optimized for Chinese text processing but adaptable to other languages through minimal code changes. Features a deployable server architecture with Flask/gRPC support for both NER and text classification tasks, enabling real-time inference via client-server interaction. Includes pre-trained model checkpoints and handles complete workflows from dataset preparation through model freezing and serving, with configurable label schemas and sequence length parameters.

4,900 stars. No commits in the last 6 months.