GenerationPrograms: Fine-grained Attribution with Executable Programs

by David Wan, Eran Hirsch, Elias Stengel-Eskin, Ido Dagan, Mohit Bansal

This repo contains the code for our paper GenerationPrograms: Fine-grained Attribution with Executable Programs.

Data

Please refer to the respective repository for instructions on downloading the data. We expect a json file with the following keys:

• context: List[str], where each string is a document
• question: str

Code

The code is located in the src directory.

The files are:

1. program.py: Contains the Program object, which constructs the tree and keeps a record of module inputs and outputs.
2. run.py: Script to run the baseline ALCE method.
3. run_program.py: Script to run GenerationPrograms.
4. util.py: Utilities for parsing and splitting sentences.
5. model.py: Model wrapper. It creates a Model() object, and running the run(prompt) method will execute the model.

Both inference scripts take the same arguments. Here is an example of how to run the code:

dataset="lfqa"
python src/run.py --data data/${dataset}.json --model gpt4o \
    --prompt prompts/${dataset}_fewshot.txt --fewshot_examples prompts/${dataset}_fewshots_doc.json \
    --output output/gpt4o_${dataset}_alce.json

python src/run_program.py --model gpt4o --add_citation \
    --data data/${dataset}.json  \
    --prompt prompts/program_fewshot.txt  --fewshot_examples prompts/${dataset}_fewshots_program.json \
    --output output/gpt4o_${dataset}_program.json

Citation

If you find our project useful in your research, please cite the following paper:

@misc{wan2025generationprogramsfinegrainedattributionexecutable,
      title={GenerationPrograms: Fine-grained Attribution with Executable Programs}, 
      author={David Wan and Eran Hirsch and Elias Stengel-Eskin and Ido Dagan and Mohit Bansal},
      year={2025},
      eprint={2506.14580},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2506.14580}, 
}