Error when trying distributed training

Hello, 

I am trying to perform distributed training when finetuning with some of my own images. It runs fine on a single GPU but fails for multi GPU. Could you please help? Thank you for the wonderful code!

- Anirudh