Is Mamba Compatible with Trajectory Optimization in Offline Reinforcement Learning?

Instructions

1. Python 3.8.19
   conda create -n your_env_name python==3.8
2. Activate Env
   conda activate your_env_name
3. install the requirements of dt or pip install -r requirements.txt
4. upgrade torch(torch==1.2.0+cu121) and install mamba from our source
   cd mamba and pip install -e .
5. move to subdir and run script respectively.

Notice

1. Every time you switch to or switch out of the cu_seqlens branch, you need to reinstall mamba pip install -e ., this is because the cu_seqlens branch has modified the C files, so recompilation is necessary.

Acknowledgements

This repository is based on decision-transformer, Mamba, Mamba-minimal and HiddenMambaAttn. Thanks for their wonderful works.

To-Do

1. add support for cu_seqlens.
2. how to install the env of atari and the version of atari.
3. run command of attention, atari, rnn-like dema.
4. code of maze navigation.