This repository provides out-of-the-box implementations for Chinese text classification using BERT and ERNIE pre-trained language models. The project fine-tunes these transformer models on a curated 10-class Chinese news headline dataset (THUCNews subset with 200K samples) and reports classification accuracy across multiple model variants including BERT, BERT+CNN, BERT+RNN, and BERT+RCNN. Pretrained model weights and configuration files are included for immediate use.