Custom Neural Network Environment

I’m new in the field of reinforcement learning and currently searching for a suitable library to implement my deep reinforcement learning research project. It should be a Python library and use PyTorch for the deep learning integration. I would like to run multiple environments in parallel on (multiple) GPU. As my environment will contain a Neural Network it should be able to directly run on the GPU and work with tensors. I would like to hear your opinion and about your experience if TorchRL is the right (a suitable) way to go.