模型:
uer/bart-large-chinese-cluecorpussmall
这个模型是由 UER-py 进行预训练的,该模型在 this paper 中进行了介绍。
你可以从 UER-py Modelzoo page 中下载一套中文BART模型,也可以通过以下链接从HuggingFace上下载:
| Link | |
|---|---|
| BART-Base | 1238321 |
| BART-Large | 1239321 |
您可以使用此模型直接进行文本生成的管道化处理(以BART-Base为例):
>>> from transformers import BertTokenizer, BartForConditionalGeneration, Text2TextGenerationPipeline
>>> tokenizer = BertTokenizer.from_pretrained("uer/bart-base-chinese-cluecorpussmall")
>>> model = BartForConditionalGeneration.from_pretrained("uer/bart-base-chinese-cluecorpussmall")
>>> text2text_generator = Text2TextGenerationPipeline(model, tokenizer)
>>> text2text_generator("中国的首都是[MASK]京", max_length=50, do_sample=False)
[{'generated_text': '中 国 的 首 都 是 北 京'}]
训练数据来自 CLUECorpusSmall 。
模型在 Tencent Cloud 上由 UER-py 进行预训练。我们使用512的序列长度进行了1000000步的预训练。以BART-Base为例
python3 preprocess.py --corpus_path corpora/cluecorpussmall.txt \
--vocab_path models/google_zh_vocab.txt \
--dataset_path cluecorpussmall_bart_seq512_dataset.pt \
--processes_num 32 --seq_length 512 \
--data_processor bart
python3 pretrain.py --dataset_path cluecorpussmall_bart_seq512_dataset.pt \
--vocab_path models/google_zh_vocab.txt \
--config_path models/bart/base_config.json \
--output_model_path models/cluecorpussmall_bart_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 5e-5 --batch_size 8 \
--span_masking --span_max_length 3
最后,我们将预训练的模型转换为Huggingface的格式:
python3 scripts/convert_bart_from_uer_to_huggingface.py --input_model_path cluecorpussmall_bart_base_seq512_model.bin-1000000 \
--output_model_path pytorch_model.bin \
--layers_num 6
@article{lewis2019bart,
title={Bart: Denoising sequence-to-sequence pre-training for natural language generation, translation, and comprehension},
author={Lewis, Mike and Liu, Yinhan and Goyal, Naman and Ghazvininejad, Marjan and Mohamed, Abdelrahman and Levy, Omer and Stoyanov, Ves and Zettlemoyer, Luke},
journal={arXiv preprint arXiv:1910.13461},
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}
}