For the SSD resnet, there is having some limitation and causes the training failed.
Regards this ,we are still debugging the issue. In the meanwhile, could you try with other model first?
Will keep you update once we have solution for it.
Thanks and sorry for the inconvenience