Edit model card

Chinese RoBERTa-Base Model for NER

Model description

The model is used for named entity recognition. It is fine-tuned by UER-py, which is introduced in this paper. Besides, the model could also be fine-tuned by TencentPretrain introduced in this paper, which inherits UER-py to support models with parameters above one billion, and extends it to a multimodal pre-training framework.

You can download the model either from the UER-py Modelzoo page, or via HuggingFace from the link roberta-base-finetuned-cluener2020-chinese.

How to use

You can use this model directly with a pipeline for token classification :

>>> from transformers import AutoModelForTokenClassification,AutoTokenizer,pipeline
>>> model = AutoModelForTokenClassification.from_pretrained('uer/roberta-base-finetuned-cluener2020-chinese')
>>> tokenizer = AutoTokenizer.from_pretrained('uer/roberta-base-finetuned-cluener2020-chinese')
>>> ner = pipeline('ner', model=model, tokenizer=tokenizer)
>>> ner("江苏警方通报特斯拉冲进店铺")
       {'word': '江', 'score': 0.49153077602386475, 'entity': 'B-address', 'index': 1, 'start': 0, 'end': 1}, 
       {'word': '苏', 'score': 0.6319217681884766, 'entity': 'I-address', 'index': 2, 'start': 1, 'end': 2}, 
       {'word': '特', 'score': 0.5912262797355652, 'entity': 'B-company', 'index': 7, 'start': 6, 'end': 7},
       {'word': '斯', 'score': 0.69145667552948, 'entity': 'I-company', 'index': 8, 'start': 7, 'end': 8}, 
       {'word': '拉', 'score': 0.7054660320281982, 'entity': 'I-company', 'index': 9, 'start': 8, 'end': 9}

Training data

CLUENER2020 is used as training data. We only use the train set of the dataset.

Training procedure

The model is fine-tuned by UER-py on Tencent Cloud. We fine-tune five 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.

python3 finetune/run_ner.py --pretrained_model_path models/cluecorpussmall_roberta_base_seq512_model.bin-250000 \
                            --vocab_path models/google_zh_vocab.txt \
                            --train_path datasets/cluener2020/train.tsv \
                            --dev_path datasets/cluener2020/dev.tsv \
                            --label2id_path datasets/cluener2020/label2id.json \
                            --output_model_path models/cluener2020_ner_model.bin \
                            --learning_rate 3e-5 --epochs_num 5 --batch_size 32 --seq_length 512

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

python3 scripts/convert_bert_token_classification_from_uer_to_huggingface.py --input_model_path models/cluener2020_ner_model.bin \
                                                                             --output_model_path pytorch_model.bin \
                                                                             --layers_num 12

BibTeX entry and citation info

  title={Roberta: A robustly optimized bert pretraining approach},
  author={Liu, Yinhan and Ott, Myle and Goyal, Naman and Du, Jingfei and Joshi, Mandar and Chen, Danqi and Levy, Omer and Lewis, Mike and Zettlemoyer, Luke and Stoyanov, Veselin},
  journal={arXiv preprint arXiv:1907.11692},

  title={CLUENER2020: Fine-grained Name Entity Recognition for Chinese},
  author={Xu, Liang and Dong, Qianqian and Yu, Cong and Tian, Yin and Liu, Weitang and Li, Lu and Zhang, Xuanwei},
  journal={arXiv preprint arXiv:2001.04351},
  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},

  title={TencentPretrain: A Scalable and Flexible Toolkit for Pre-training Models of Different Modalities},
  author={Zhao, Zhe and Li, Yudong and Hou, Cheng and Zhao, Jing and others},
  journal={ACL 2023},
Downloads last month

Space using uer/roberta-base-finetuned-cluener2020-chinese 1