Multiprocessing with docker and GPU

Hi

I am having an issue with multiprocessing and inferencing multiple streams inside a docker container, I note another topic here but it hasnt been answered as yet Multi-Process freeze with docker.

My set up is Orin Nano, Jetpack 5.1.2 I am using base image of dustynv/l4t-pytorch:r35.4.1. I have been using this set up for over a year and it works perfectly, there is no issue with running single streams. In addition if i use the same code outside of the docker container I can run multiple streams and inference multiple streams with no issue. Inside the docker container I can run multiple streams if i force them to use CPU i.e. the issue is not a multiprocessing one or memory etc. there appears to be an issue with how docker and gpu act with multiprocesssing, any guidance here?