Context-Tuning

This is the repository for COLING 2022 paper "Context-Tuning: Learning Contextualized Prompts for Natural Language Generation". The implementation is completely based on our text generation library TextBox 2.0.

Installation

You should clone the TextBox repository and follow its instructions.

git clone https://github.com/RUCAIBox/TextBox.git && cd TextBox
bash install.sh

Dataset

The datasets ROCStories (roc), WritingPrompts (wp), WikiPlots (wikip), and ChangeMyView (cmv) can be downloaded at the link https://huggingface.co/datasets/RUCAIBox/Story-Generation.

Running Context-Tuning based on TextBox

For example, you can conduct Context-Tuning on roc dataset using this command:

python run_textbox.py --model=Context_Tuning --dataset=roc

You can use --dataset=xxx to specify the dataset name, such as roc, wp, wikip, and cmv.

Other hyperparameters can be changed in the yaml. The prompt_generator can be set to bert or roberta. The semantic_mapping can be set to True or False. The prompt_length of efficient_kwargs can also be changed at your will.

Reference

@inproceedings{tang-etal-2022-context,
    title = "Context-Tuning: Learning Contextualized Prompts for Natural Language Generation",
    author = "Tang, Tianyi  and
      Li, Junyi  and
      Zhao, Wayne Xin  and
      Wen, Ji-Rong",
    booktitle = "Proceedings of the 29th International Conference on Computational Linguistics",
    month = oct,
    year = "2022",
    address = "Gyeongju, Republic of Korea",
    publisher = "International Committee on Computational Linguistics",
    url = "https://aclanthology.org/2022.coling-1.552",
    pages = "6340--6354",
}