ChineseGLUE

Why we included this project

ChineseGLUE adapts the GLUE benchmark formula to Chinese, bringing together the evaluation datasets, baseline scores, and pretrained checkpoints that mark a reasonable standard on Chinese language-understanding tasks. The task spread is wider than a single model eval: news classification, sentence-pair similarity, natural language inference, sentiment, named entity recognition, and extractive reading comprehension, in both simplified and traditional Chinese. The leaderboard tables let you compare BERT, RoBERTa, ERNIE, XLNet, and ALBERT variants across those tasks before committing to training your own model, which helps when choosing a starting checkpoint. For teams building Chinese NLP systems, the dataset links and formatting notes save the trouble of hunting down each corpus and standing up comparisons from scratch. The project has largely moved on to the CLUE benchmark, but this classic version stays a coherent snapshot of the original task set.