基于T5、BART、CPT的文本摘要生成
├── README.md
├── requirements.txt
├── scripts
│ ├── predict.sh
│ └── train.sh
├── src
│ ├── main.py
│ ├── modeling_cpt.py
│ ├── predict.py
│ └── SFT_utils.py
pip install -r requirements.txt
cd scripts
bash train.sh
cd scripts
bash predict.sh
SFT means squeeze and fine tuning, just to copy some layers from pretrained models, and fine tuning on the layers copied