PRESTO

[NeurIPS 25] Official Implementation (Pytorch) of "PRESTO: Preimage-Informed Instruction Optimization for Prompting Black-Box LLMs"

Our code are based on the code from APE, InstructZero and INSTINCT.

Prepare the data

You can download the data for intrinsic induction from the github repo of InstructZero. You can download the dataset for GSM8K, AQUARAT, and SVAMP from the repo for APE.

Run our code

To run our code, you need to install the environment using conda: conda env create -f environment.yml

We provide bash scripts for running our experiments for instruction induction at Induction/experiments/run_neural_bandits.sh. To run it properly, you need to run the following in the terminal:

cd Induction
bash experiments/run_neural_bandits.sh