README.md

Auto-Instruct

This is the repository for Auto-Instruct, an automatic solution of generating and selecting instructions for prompting large language models (LLMs). Our method leverages the inherent generative ability of LLMs to produce diverse candidate instructions for a given task, and then ranks them using a scoring model trained on a variety of 575 existing NLP tasks. In experiments on 118 out-of-domain tasks, Auto-Instruct surpasses both human-written instructions and existing baselines of LLM-generated instructions. For more details, please refer to our paper "Auto-Instruct: Automatic Instruction Generation and Ranking for Black-Box Language Models" in EMNLP 2023 Findings.

The repository includes the following contents:

• data: the training / testing data files, meta-prompts, downstream prompts, and generated instructions.
• GPT-3/optimization: the source code for data, model training and model evaluation.
  • instruction_generation_templates: the templates of creating meta-prompts for each task (used for instruction generation)
  • instruction_generation: scripts for instruction generation
  • instruction_labeling: scripts for label the instructions for training / testing, as well as dataset pre-processing
  • run.py: entrance for model training / testing, see GPT-3/optimization/README.md
  • evaluation: evaluation scripts for the ranking model

Environment

pip install -r requirements.txt

Checkpoints

Checkpoints of instruction ranking models:

• Trained on instructions generated by text-davinci-003 under the few-shot setting: checkpoint
• Trained on instructions generated by text-davinci-003 under the zero-shot setting: checkpoint

Citation

If you find our work useful, please kindly cite our paper:

@inproceedings{Auto-Instruct,
  author = {Zhihan Zhang and
                  Shuohang Wang and
                  Wenhao Yu and
                  Yichong Xu and
                  Dan Iter and
                  Qingkai Zeng and
                  Yang Liu and
                  Chenguang Zhu and
                  Meng Jiang},
  title = {Auto-Instruct: Automatic Instruction Generation and 
                  Ranking for Black-Box Language Models},
  booktitle = {Findings of the 2023 Conference on Empirical 
               Methods in Natural Language Processing, {EMNLP} 2023, 
               Singapore, December 6-10, 2023},
  publisher = {Association for Computational Linguistics},
  year = {2023},
  url = {https://doi.org/10.48550/arXiv.2310.13127}
}