Hi

Thank you for your impressive work on. It has been very helpful for my research.

I am writing to inquire about two things:

CUDA Memory Occupation by Sub-processes When running the provided code with multiple environments (e.g., num_envs=10), I noticed that the sub-processes seem to initialize the CUDA context, resulting in significant VRAM consumption for each child process. Have you encountered this behavior during your experiments? I am wondering if this is expected or if there is a recommended configuration to avoid it.

Thank you very much for your time and help.

Best regards