中国飞马

模型描述

该模型由 UER-py 进行预训练，其介绍见 this paper 。

你可以从 UER-py Modelzoo page 下载一套中文PEGASUS模型，或者通过以下链接从HuggingFace获取：

使用方法

你可以直接使用此模型进行文本生成的流程：

>>> from transformers import BertTokenizer, PegasusForConditionalGeneration, Text2TextGenerationPipeline
>>> tokenizer = BertTokenizer.from_pretrained("uer/pegasus-base-chinese-cluecorpussmall")
>>> model = PegasusForConditionalGeneration.from_pretrained("uer/pegasus-base-chinese-cluecorpussmall")
>>> text2text_generator = Text2TextGenerationPipeline(model, tokenizer)  
>>> text2text_generator("内容丰富、版式设计考究、图片华丽、印制精美。[MASK]纸箱内还放了充气袋用于保护。", max_length=50, do_sample=False)
    [{'generated_text': '书 的 质 量 很 好 。'}]

训练数据

CLUECorpusSmall 用作训练数据。

训练过程

该模型由 UER-py 在 Tencent Cloud 上进行预训练。我们以512的序列长度进行1000000个步骤的预训练。以PEGASUS-Base为例：

python3 preprocess.py --corpus_path corpora/cluecorpussmall.txt \
                      --vocab_path models/google_zh_vocab.txt \
                      --dataset_path cluecorpussmall_pegasus_seq512_dataset.pt \
                      --processes_num 32 --seq_length 512 \
                      --data_processor gsg --sentence_selection_strategy random

python3 pretrain.py --dataset_path cluecorpussmall_pegasus_seq512_dataset.pt \
                    --vocab_path models/google_zh_vocab.txt \
                    --config_path models/pegasus/base_config.json \
                    --output_model_path models/cluecorpussmall_pegasus_base_seq512_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 8

最后，我们将预训练模型转换为Huggingface的格式：

python3 scripts/convert_pegasus_from_uer_to_huggingface.py --input_model_path cluecorpussmall_pegasus_base_seq512_model.bin-250000 \                                                                
                                                           --output_model_path pytorch_model.bin \                                                                                            
                                                           --layers_num 12

BibTeX条目和引用信息

@inproceedings{zhang2020pegasus,
  title={Pegasus: Pre-training with extracted gap-sentences for abstractive summarization},
  author={Zhang, Jingqing and Zhao, Yao and Saleh, Mohammad and Liu, Peter},
  booktitle={International Conference on Machine Learning},
  pages={11328--11339},
  year={2020},
  organization={PMLR}
}

@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}
}