BERT-NER-Pytorch

Why we included this project

For anyone building Chinese named entity recognition with BERT, this repo is a rare complete starting point. It bundles softmax, CRF, and span-based decoding in one codebase, so you can compare all three approaches on the same data and see where CRF or span tagging beats a plain classifier. The included training recipes and per-epoch F1 tables for CLUENER and CNER give you concrete baselines to sanity-check your own pipeline before you swap in your own corpus. Practitioners with annotated Chinese text who want a reference implementation to extend will find this far easier to adapt than starting from a raw transformer library. Treat it as a research example rather than a maintained service: you'll integrate it into your own training flow, not drop it in as a turnkey tool.