Do you have a dockerfile that works with DDP using torch==1.12.1+cu113 and torchvision==0.13.1+cu113? Or maybe another version of torch with CUDA <= 11.4?
Want to make sure that mine is not missing anything
Do you have a specific restriction on building your own container or would using a prebuilt container work for you? If so you could check out the 21.10 NGC container which should have CUDA 11.4: PyTorch Release 21.10 - NVIDIA Docs
Prebuilt container works for me. Thank you 