Pytorch versions during deployment

[fzoric8]

Hi,

I'm trying to deploy mask-rcnn on the jetson by using model trained on the custom data.
Model exhibits satisfactory behavior and it is trained with the pytorch 1.12 version, and when I check predictions on PC I've used to train it.

I'm trying to deploy it on a jetson that has pytorch 1.13. After the conversion, model doesn't predict at all.
Conversion of the exemplary model (retinanet18) works

Can I use different pytorch versions for training/deployment?