pytorch, tensorflow and CUDA environment conflict

[jmwang0117]

Thank you for your excellent work!

To evaluate Moto-GPT on the SIMPLER benchmark, you need to install tensorflow[and-cuda]==2.15.1, which requires CUDA 12.2. However, Moto-GPT requires torch==2.4.1, which depends on CUDA 12.1. How can you resolve these conflicting CUDA version dependencies and utilize both torch==2.4.1 and tensorflow[and-cuda]==2.15.1 in moto_for_simpler?