torch.cuda.device_count# torch.cuda.device_count()[source]# Return the number of GPUs available. Note This API will NOT poison fork if NVML discovery succeeds. See Poison fork in multiprocessing for more details. Return type int