README.md

LM CLRS

GitHub for the CLRS tasks in language form.

Reference

How to use

1. Install Requirements:

   pip install -r requirements.txt

2. Generate data:

   The options for <TASK> are defined in the tasks array in data_schema.py

   python data_generator.py --task <TASK> --part <train or test> --num_samples <number of generated samples>

3. (Optional) Create txt files of inputs:

   This only works for 2d array inputs.

   python inputs_to_text_files.py --task <TASK> --dp <number of decimal places to round data to>

4. Generate prompts:

   We give a framework to generate prompts automatically. You can then add your prompt for each task, to the respective function in prompt_generator.py.

   python prompt_generator.py --task <TASK> --part <train or test> --num_samples <number of generated samples>`

   You can then call prompt_gen(task, partition, max_samples) to generate prompts at test time.

Pointers

Some CLRS outputs are pointers, for example the output of: [13, 13, 7, 0, 4, 12, 11, 13, 14, 13, 2, 7, 0, 13, 0, 7] depicts the tree in the diagram below. I.e. A[i] = parent of node i

Citation

@article{mcleish2024benchmarking,
  title={Benchmarking ChatGPT on Algorithmic Reasoning},
  author={McLeish, Sean and Schwarzschild, Avi and Goldstein, Tom},
  journal={arXiv preprint arXiv:2404.03441},
  year={2024}
}

Origional GNN CLRS

https://github.com/google-deepmind/clrs