Understanding When Tree of Thoughts Succeeds: Larger Models Excel in Generation, Not Discrimination

Official implementation for paper Understanding When Tree of Thoughts Succeeds: Larger Models Excel in Generation, Not Discrimination with code, prompts and datasets.

Setup

The enviroment.yaml file contains the required conda environment.

Code

• The dataset, prompts and task implementstations can be found under tot/data/, tot/prompts/ and tot/tasks/ respectively.
  
• tot/models.py provides LLMs as generator/discriminator modules for ToT.
  
• The code to execute ToT and the oracle implementation is under tot/methods/ directory. Since the oracle implementation relies on standard answers for different tasks, we tailor an implementation for different tasks.
  
• To experiment, run the run.py file with the required arguments.

Citation

If you find this work interesting/useful, you can cite this paper as:

@misc{chen2024understandingtreethoughtssucceeds,
      title={Understanding When Tree of Thoughts Succeeds: Larger Models Excel in Generation, Not Discrimination}, 
      author={Qiqi Chen and Xinpeng Wang and Philipp Mondorf and Michael A. Hedderich and Barbara Plank},
      year={2024},
      eprint={2410.17820},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2410.17820}, 
}