This repository contains the code for SCULPT: Systematic Tuning of Long Prompts.
To set up the project and run the different methods included in this repository, follow these steps:
-
Install the required dependencies using the
requirements.txtfile:pip install -r requirements.txt
-
Run the desired method using the provided shell scripts and specify the task name as an argument:
- To run the SCULPT method:
./run_sculpt.sh <task_name>
- To run the Protegi method:
./run_protegi.sh <task_name>
- To run the OPRO method:
./run_opro.sh <task_name>
- To run the APEX method:
./run_apex.sh <task_name>
- To run the APE method:
./run_ape.sh <task_name>
- To run the LongAPE method:
./run_longape.sh <task_name>
Replace
<task_name>with the desired task (e.g.,formal_fallacies,causal_judgment,disambiguation_qa,salient_translation,go_emotions,beaver_tails). - To run the SCULPT method:
If you use this code or the concepts from our work, please cite our paper published in the main conference of ACL 2025:
SCULPT: Systematic Tuning of Long Prompts arXiv: https://arxiv.org/pdf/2410.20788
@article{kumar2024sculpt,
title={SCULPT: Systematic Tuning of Long Prompts},
author={Kumar, Shanu and Venkata, Akhila Yesantarao and Khandelwal, Shubhanshu and Santra, Bishal and Agrawal, Parag and Gupta, Manish},
journal={arXiv preprint arXiv:2410.20788},
year={2024}
}