ws-jiang/MetaPrompter_public

MetaPrompter

ENV

• environments:

  • torch=1.12.1
  • datasets=2.7.0
  • transformers=4.24.0
  • yaml=6.0
  • mysql-connector-python

• Tesla V100S-32GB x 8, NVIDIA A100-80GB x 8

• note: when the OOM problem happens, can reduce the number of query samples (in configs/few_shot_settings.yaml) or increase the GPU memory.

Experiments on RepVerb

• refer to code/run_verb.sh

Experiments on MetaPrompter

• refer to code/run_meta.sh

References

• MetaPrompting
• ContrastNet
• MLADA

bibtex

@InProceedings{jiang2023effective,
  author    = {Jiang, Weisen and Zhang, Yu and Kwok, James},
  booktitle = {International Conference on Machine Learning},
  title     = {Effective Structured Prompting by Meta-Learning and Representative Verbalizer},
  year      = {2023},
}