xiangking/ark-nlp

Provides modular implementations of task-specific NLP architectures (text classification, NER, relation extraction, information extraction) built on PyTorch and Hugging Face Transformers, with pre-built components for tokenization, dataset handling, and training pipelines. Supports Chinese language processing through integration with jieba tokenization and models like ERNIE, NEZHA, and Roformer optimized for Chinese NLP tasks. Encapsulates loss functions, optimizers, and prediction utilities across six core modules (dataset, nn, processor, factory, model) to streamline rapid prototyping of production NLP systems.

301 stars. No commits in the last 6 months.