SBM-Transformer

1. Our experiment environment is listed in requirements.txt.

2. Download code from the LRA repo and place long-range-arena/ in datasets/

3. Download and unzip lra_release.gz into datasets/. The resulting directory structure should be

datasets/long-range-arena
datasets/lra_release

4. Run datasets/create_datasets.sh to create the train-, dev-, and test-split pickle files for each LRA task.
5. To run baseline methods on LRA, run code/run_baselines.sh
6. To run SBM-transformer on LRA, run code/run_tasks.sh