Training can only be run on Linux systems with CUDA drivers and a compatible CUDA GPU. If required, this can be bypassed by adding the device='cpu' flag in the botv2.py This is tested and confirmed ...