Randeng T5 Char 700M Chinese

🚀 Randeng-T5-Char-700M-Chinese

A Chinese version of T5-large, proficient in handling NLT tasks, leveraging BertTokenizer and a Chinese character-level vocabulary.

• Main Page: Fengshenbang
• Github: Fengshenbang-LM

🚀 Quick Start

Model Taxonomy

Model Information

Compared with T5-small, we trained its Chinese version. To better suit Chinese tasks, we only used BertTokenzier and a vocabulary supporting both Chinese and English. Additionally, we employed the Corpus-Adaptive Pre-Training (CAPT) technique for continued pre-training on the WuDao corpus (180G version). The pre-training objective was span corruption. Specifically, during the pre-training phase, we used the Fengshen framework, which took approximately 96 hours with 16 A100 GPUs.

✨ Features

Good at handling NLT tasks, this model is a Chinese T5-large that uses BertTokenizer and a Chinese vocabulary.

💻 Usage Examples

Basic Usage

from transformers import T5ForConditionalGeneration, BertTokenizer
import torch

tokenizer=BertTokenizer.from_pretrained('IDEA-CCNL/Randeng-T5-Char-700M-Chinese', add_special_tokens=False)
model=T5ForConditionalGeneration.from_pretrained('IDEA-CCNL/Randeng-T5-Char-700M-Chinese')

📄 License

This project is licensed under the Apache-2.0 license.

📚 Documentation

Citation

If you are using the resource for your work, please cite our paper:

@article{fengshenbang,
  author    = {Jiaxing Zhang and Ruyi Gan and Junjie Wang and Yuxiang Zhang and Lin Zhang and Ping Yang and Xinyu Gao and Ziwei Wu and Xiaoqun Dong and Junqing He and Jianheng Zhuo and Qi Yang and Yongfeng Huang and Xiayu Li and Yanghan Wu and Junyu Lu and Xinyu Zhu and Weifeng Chen and Ting Han and Kunhao Pan and Rui Wang and Hao Wang and Xiaojun Wu and Zhongshen Zeng and Chongpei Chen},
  title     = {Fengshenbang 1.0: Being the Foundation of Chinese Cognitive Intelligence},
  journal   = {CoRR},
  volume    = {abs/2209.02970},
  year      = {2022}
}

You can also cite our website:

@misc{Fengshenbang-LM,
  title={Fengshenbang-LM},
  author={IDEA-CCNL},
  year={2021},
  howpublished={\url{https://github.com/IDEA-CCNL/Fengshenbang-LM}},
}