Not at the moment. The TRL library is designed for CUDA and many things do not work on other accelerators, like Mac Metal. I am also on Mac, so I face the same problem as you. This is the reason why I use Modal here, so the code is runnable by anyone.
Thanks for the good 😊
Is it possible to train it on a Mac m3?
Not at the moment. The TRL library is designed for CUDA and many things do not work on other accelerators, like Mac Metal. I am also on Mac, so I face the same problem as you. This is the reason why I use Modal here, so the code is runnable by anyone.