metadata

language: Chinese
widget:
  - text: 北京上个月召开了两会

Chinese RoBERTa-Base Models for Text Classification

Model description

This is the set of 5 Chinese RoBERTa base models fine-tuned by UER-py.

You can download the 5 Chinese RoBERTa base models either from the links below:

corpus	Link
JD full	roberta-base-finetuned-jd-full-chinese
JD binary	roberta-base-finetuned-jd-binary-chinese
Dianping	roberta-base-finetuned-dianping-chinese
Ifeng	roberta-base-finetuned-ifeng-chinese
Chinanews	roberta-base-finetuned-chinanews-chinese

How to use

You can use this model directly with a pipeline for text classification (take the case of roberta-base-finetuned-chinanews-chinese):

>>> from transformers import AutoModelForSequenceClassification,AutoTokenizer,pipeline
>>> model = AutoModelForSequenceClassification.from_pretrained('uer/roberta-base-finetuned-chinanews-chinese')
>>> tokenizer = AutoTokenizer.from_pretrained('uer/roberta-base-finetuned-chinanews-chinese')
>>> text_classification = pipeline('sentiment-analysis', model=model, tokenizer=tokenizer)
>>> text_classification("北京上个月召开了两会")
    [{'label': 'mainland China politics', 'score': 0.7211663722991943}]

Training data

We use 5 Chinese text classification datasets which are collected by Glyph project.

Training procedure

Models are fine-tuned by UER-py on Tencent Cloud. We fine-tune three epochs with a sequence length of 512 on the basis of the pre-trained model chinese_roberta_L-12_H-768. At the end of each epoch, the model is saved when the best performance on development set is achieved.

Taking the case of roberta-base-finetuned-chinanews-chinese

python3 run_classifier.py --pretrained_model_path models/cluecorpussmall_roberta_base_seq512_model.bin-250000 \ 
                          --vocab_path models/google_zh_vocab.txt \
                          --train_path Glyph/Chinanews_train.txt \
                          --dev_path Glypg/Chinanews_test.txt \
                          --output_model_path models/Chinanews_model.bin \
                          --learning_rate 3e-5 --batch_size 32 --epochs_num 3 \
                          --seq_length 512 --embedding word_pos_seg --encoder transformer --mask fully_visible

Finally, we convert the pre-trained model into Huggingface's format:

python3 scripts/convert_bert_text_classification_from_uer_to_huggingface.py --input_model_path models/Chinanews_model.bin \
                                                                            --output_model_path pytorch_model.bin \
                                                                            --layers_num 12

BibTeX entry and citation info

@article{zhao2019uer,
  title={UER: An Open-Source Toolkit for Pre-training Models},
  author={Zhao, Zhe and Chen, Hui and Zhang, Jinbin and Zhao, Xin and Liu, Tao and Lu, Wei and Chen, Xi and Deng, Haotang and Ju, Qi and Du, Xiaoyong},
  journal={EMNLP-IJCNLP 2019},
  pages={241},
  year={2019}
}