/twm

Transformer-based World Models

Primary LanguagePythonMIT LicenseMIT

Transformer-based World Models Are Happy With 100k Interactions (ICLR 2023)

Currently WIP.

Implementation of our Transformer-based World Model (paper) in PyTorch.

Install packages from requirements.txt. Also make sure you have installed the Atari environments correctly.
For more information, see: https://github.com/openai/gym/releases/tag/v0.21.0

Execute the following command to run an experiment:

python -O twm/main.py --game Breakout --seed 0 --device cuda:0 --cpu_p 1.0 --wandb disabled

Use --wandb online to log the metrics in weights and biases.
To use other hyperparameters, edit the file twm/config.py.

If you find the code in this repository useful, please cite our paper:

@inproceedings{
  robine2023transformerbased,
  title={Transformer-based World Models Are Happy With 100k Interactions},
  author={Jan Robine and Marc H{\"o}ftmann and Tobias Uelwer and Stefan Harmeling},
  booktitle={The Eleventh International Conference on Learning Representations },
  year={2023},
  url={https://openreview.net/forum?id=TdBaDGCpjly}
}