This repository contains the code and resources for the NAACL 2025 main paper titled "Specializing Large Language Models to Simulate Survey Response Distributions for Global Populations". The project focuses on adapting large language models (LLMs) to simulate survey response distributions across diverse global populations.

This project explores the use of large language models (LLMs) to simulate survey responses for global populations. By fine-tuning LLMs on survey data, we aim to generate realistic response distributions that reflect the diversity of global opinions.

To set up the environment, clone this repository and install the required dependencies:

pip install -r requirements.txt

Before running the experiments, you need to prepare the dataset. Run the following script to download and preprocess the data:

sh prepare_data.sh

or you can also directly use our processed data: Download Dataset.

To evaluate the model in a zero-shot setting, use the following script:

sh infer_slurm.sh

To fine-tune the model on the survey dataset, run the following script:

sh train_slurm.sh

After fine-tuning, you can evaluate the model's performance using the following script (config INFER_MODE as sft, and CKPT_PATH):

sh infer_slurm.sh

For any questions or issues, please open an issue on this repository or contact yongcao2018@gmail.com.