About the batch size is 10 for each GPU.

[wtybest]

In you readme text, you write "Default training requires 4 GPUs with 11GB memory, and the batch size is 10 for each GPU. "
But I think it‘s 10 for all GPU. May not right, so I consult you.

[Yaoyi-Li]

I think it is 10 for each GPU. Could you give me any hint that makes you think it's 10 for all GPU?