This repository is based on the Long Range Arena (LRA) datasets from Google Research. The original repository from Google is available here. The paper is available on ArXiv.
In this repo we provide researchers with code for some of the simpler tasks. It also has the advantage of using PyTorch and HuggingFace Transformers instead of Jax/Flax, with which less are familiar. In theory, this (arguably) makes our code easier to understand and extend for some researchers.
- Run
sh ./get_data.sh. It will create a newdatasetsdirectory in the project and will populate it with the required datasets (make sure to run this in the project's root directory as the script uses relative paths) - To test the code with simple models, just run:
python run_model.py