我们基于mT5-large，训练了它的中文版。为了加速训练，我们仅使用T5分词器(sentence piece)中的中英文对应的词表，并且使用了语料库自适应预训练(Corpus-Adaptive Pre-Training, CAPT)技术在悟道语料库(180G版本)继续预训练。预训练目标为破坏span。具体地，我们在预训练阶段中使用了封神框架大概花费了16张A100约96小时。

Based on mT5-large, we implement its Chinese version. In order to accelerate training, we only retrain the vocabulary and embedding corresponding to Chinese and English in T5tokenizer (sentence piece), and Corpus-Adaptive Pre-Training (CAPT) on the WuDao Corpora (180 GB version). The pretraining objective is span corruption. Specifically, we use the fengshen framework in the pre-training phase which cost about 96 hours with 16 A100 GPUs.

使用 Usage

python

from transformers import T5ForConditionalGeneration, AutoTokenizer
import torch
tokenizer=AutoTokenizer.from_pretrained('IDEA-CCNL/Randeng-T5-784M', use_fast=false)
model=T5ForConditionalGeneration.from_pretrained('IDEA-CCNL/Randeng-T5-784M')

引用 Citation

如果您在您的工作中使用了我们的模型，可以引用我们的论文：

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

text

@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}
}

也可以引用我们的网站:

You can also cite our website:

text

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