This is the repository containing code for "Zero-shot Learning by Generating Task-specific Adapters" (arXiv). This is a beta version and we will add more details in the future.

We modified the code in shmsw25/bart-closed-book-qa (Thanks to the authors!).

Following their instructions, please install the environment with these commands:

pip install torch==1.1.0
pip install git+https://github.com/huggingface/transformers.git@7b75aa9fa55bee577e2c7403301ed31103125a35

Download ZEST dataset from here and place (zest_{train|dev|test_unanswered}.jsonl) in ./data.

See ./scripts/zest_bart_large.sh and ./scripts/zest_grouped_bart_large_from_trained.sh

@article{Ye2021ZeroshotLB,
  title={Zero-shot Learning by Generating Task-specific Adapters},
  author={Qinyuan Ye and Xiang Ren},
  journal={ArXiv},
  year={2021},
  volume={abs/2101.00420}
}