Multi GPU training request

[sevaraMS]

In the request of Nvidia Driver >= 455.23 And I installed 460 versions of Nvidia Driver and automatically installed Cuda version 11.2 In the running time I get the Error message as this depository trained based on Nvidia Driver 465.19, then I reinstalled 470 versions of Nvidia Driver and automatically installed Cuda version 11.4.
When I run code:
sudo docker run --ipc=host -it ultralytics/yolov5:latest
I get the error that the PyTorch version is not satisfied, but for Cuda version 11.4 there is no PyTorch version available. Even for Nvidia Driver >= 455.23 Cuda driver version no available in PyTorch. What Can I do in this situation

[glenn-jocher]

@sevaraMS for most situations 460 is fine. See release v6.0 speed discussion #5141 (comment) with driver details. We used 460 with CUDA 11.2 running the Docker image to produce the README values.