Chinese word-based RoBERTa Miniatures

Model description

This is the set of 5 Chinese word-based RoBERTa models pre-trained by UER-py, which is introduced in this paper. Besides, the models could also be pre-trained 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.

Most Chinese pre-trained weights are based on Chinese character. Compared with character-based models, word-based models are faster (because of shorter sequence length) and have better performance according to our experimental results. To this end, we released the 5 Chinese word-based RoBERTa models of different sizes. In order to facilitate users in reproducing the results, we used a publicly available corpus and word segmentation tool, and provided all training details.

You can download the 5 Chinese RoBERTa miniatures either from the UER-py Modelzoo page, or via HuggingFace from the links below:

Compared with char-based models, word-based models achieve better results in most cases. Here are scores on the devlopment set of six Chinese tasks:

For each task, we selected the best fine-tuning hyperparameters from the lists below, and trained with the sequence length of 128:

• epochs: 3, 5, 8
• batch sizes: 32, 64
• learning rates: 3e-5, 1e-4, 3e-4

How to use

You can use this model directly with a pipeline for masked language modeling (take the case of word-based RoBERTa-Medium):

>>> from transformers import pipeline
>>> unmasker = pipeline('fill-mask', model='uer/roberta-medium-word-chinese-cluecorpussmall')
>>> unmasker("[MASK]的首都是北京。")
[
    {'sequence': '中国 的首都是北京。',
     'score': 0.21525809168815613, 
     'token': 2873, 
     'token_str': '中国'}, 
    {'sequence': '北京 的首都是北京。', 
     'score': 0.15194718539714813, 
     'token': 9502, 
     'token_str': '北京'}, 
    {'sequence': '我们 的首都是北京。', 
     'score': 0.08854265511035919, 
     'token': 4215, 
     'token_str': '我们'},
    {'sequence': '美国 的首都是北京。', 
     'score': 0.06808705627918243, 
     'token': 7810, 
     'token_str': '美国'}, 
    {'sequence': '日本 的首都是北京。', 
     'score': 0.06071401759982109, 
     'token': 7788, 
     'token_str': '日本'}
]

Here is how to use this model to get the features of a given text in PyTorch:

from transformers import AlbertTokenizer, BertModel
tokenizer = AlbertTokenizer.from_pretrained('uer/roberta-medium-word-chinese-cluecorpussmall')
model = BertModel.from_pretrained("uer/roberta-medium-word-chinese-cluecorpussmall")
text = "用你喜欢的任何文本替换我。"
encoded_input = tokenizer(text, return_tensors='pt')
output = model(**encoded_input)

and in TensorFlow:

from transformers import AlbertTokenizer, TFBertModel
tokenizer = AlbertTokenizer.from_pretrained('uer/roberta-medium-word-chinese-cluecorpussmall')
model = TFBertModel.from_pretrained("uer/roberta-medium-word-chinese-cluecorpussmall")
text = "用你喜欢的任何文本替换我。"
encoded_input = tokenizer(text, return_tensors='tf')
output = model(encoded_input)

Since BertTokenizer does not support sentencepiece, AlbertTokenizer is used here.

Training data

CLUECorpusSmall is used as training data. Google's sentencepiece is used for word segmentation. The sentencepiece model is trained on CLUECorpusSmall corpus:

>>> import sentencepiece as spm
>>> spm.SentencePieceTrainer.train(input='cluecorpussmall.txt',
             model_prefix='cluecorpussmall_spm',
             vocab_size=100000,
             max_sentence_length=1024,
             max_sentencepiece_length=6,
             user_defined_symbols=['[MASK]','[unused1]','[unused2]',
                '[unused3]','[unused4]','[unused5]','[unused6]',
                '[unused7]','[unused8]','[unused9]','[unused10]'],
             pad_id=0,
             pad_piece='[PAD]',
             unk_id=1,
             unk_piece='[UNK]',
             bos_id=2,
             bos_piece='[CLS]',
             eos_id=3,
             eos_piece='[SEP]',
             train_extremely_large_corpus=True
            )

Training procedure

Models are pre-trained by UER-py on Tencent Cloud. We pre-train 1,000,000 steps with a sequence length of 128 and then pre-train 250,000 additional steps with a sequence length of 512. We use the same hyper-parameters on different model sizes.

Taking the case of word-based RoBERTa-Medium

Stage1:

python3 preprocess.py --corpus_path corpora/cluecorpussmall.txt \
                      --spm_model_path models/cluecorpussmall_spm.model \
                      --dataset_path cluecorpussmall_word_seq128_dataset.pt \
                      --processes_num 32 --seq_length 128 \
                      --dynamic_masking --data_processor mlm

python3 pretrain.py --dataset_path cluecorpussmall_word_seq128_dataset.pt \
                    --spm_model_path models/cluecorpussmall_spm.model \
                    --config_path models/bert/medium_config.json \
                    --output_model_path models/cluecorpussmall_word_roberta_medium_seq128_model.bin \
                    --world_size 8 --gpu_ranks 0 1 2 3 4 5 6 7 \
                    --total_steps 1000000 --save_checkpoint_steps 100000 --report_steps 50000 \
                    --learning_rate 1e-4 --batch_size 64 \
                    --data_processor mlm --target mlm

Stage2:

python3 preprocess.py --corpus_path corpora/cluecorpussmall.txt \
                      --spm_model_path models/cluecorpussmall_spm.model \
                      --dataset_path cluecorpussmall_word_seq512_dataset.pt \
                      --processes_num 32 --seq_length 512 \
                      --dynamic_masking --data_processor mlm

python3 pretrain.py --dataset_path cluecorpussmall_word_seq512_dataset.pt \
                    --spm_model_path models/cluecorpussmall_spm.model \
                    --pretrained_model_path models/cluecorpussmall_word_roberta_medium_seq128_model.bin-1000000 \
                    --config_path models/bert/medium_config.json \
                    --output_model_path models/cluecorpussmall_word_roberta_medium_seq512_model.bin \
                    --world_size 8 --gpu_ranks 0 1 2 3 4 5 6 7 \
                    --total_steps 250000 --save_checkpoint_steps 50000 --report_steps 10000 \
                    --learning_rate 5e-5 --batch_size 16 \
                    --data_processor mlm --target mlm

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

python3 scripts/convert_bert_from_uer_to_huggingface.py --input_model_path models/cluecorpussmall_word_roberta_medium_seq512_model.bin-250000 \
                                                        --output_model_path pytorch_model.bin \
                                                        --layers_num 8 --type mlm

BibTeX entry and citation info

@article{devlin2018bert,
  title={BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding},
  author={Devlin, Jacob and Chang, Ming-Wei and Lee, Kenton and Toutanova, Kristina},
  journal={arXiv preprint arXiv:1810.04805},
  year={2018}
}

@article{turc2019,
  title={Well-Read Students Learn Better: On the Importance of Pre-training Compact Models},
  author={Turc, Iulia and Chang, Ming-Wei and Lee, Kenton and Toutanova, Kristina},
  journal={arXiv preprint arXiv:1908.08962v2 },
  year={2019}
}

@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}
}

@article{zhao2023tencentpretrain,
  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},
  pages={217},
  year={2023}