dongguanting / GZPL

GZPL

Code for our ACL2023 findings paper: Generative Zero-Shot Prompt Learning for Cross-Domain Slot Filling with Inverse Prompting

Requirement

# python==3.8
# torch==1.9.0
# openprompt==1.0
transformers==4.10.0
tqdm==4.62.2
yacs==0.1.8
protobuf==3.19.1
datasets==1.16.0
tensorboardX==2.4.1
sentencepiece==0.1.96
scikit-learn==0.24.2
nltk==3.6.5

直接cd根目录下，执行

pip install -r requirements.txt

这里对于原始的openprompt进行了一定的改动，安装只需要cd根目录下，执行

pip install .

Datasets

data文件夹需要从data下载，复制到prompt_sf目录下即可

Train&&Test

cd prompt_sf目录下，执行test.sh脚本即可

Citation

If you use any source codes or ideas included in this repository for your work, please cite the following paper.

@article{li2023generative,
  title={Generative zero-shot prompt learning for cross-domain slot filling with inverse prompting},
  author={Li, Xuefeng and Wang, Liwen and Dong, Guanting and He, Keqing and Zhao, Jinzheng and Lei, Hao and Liu, Jiachi and Xu, Weiran},
  journal={arXiv preprint arXiv:2307.02830},
  year={2023}
}