is distributed data parallel possible?

[sompark05]

I want to train the model, but training on 1 GPU is too slow.
So I want to use 4 GPUs.

Is distributed data parallel possible? If so, could you please explain how to set for ddp?

I really appreciate any help you can provide.

[junseok520]

I am sorry but there is no option for DDP training. Sorry for the inconvenience.

[ynlgcn]

I want to run with 2 cards, so I type in the device 0,1 command and get the error: "ValueError: AMP equipped with DataParallel has to manually write autocast() for each forward function, you can choose to do this by yourself", may I ask the author, if I want to run with 2 cards, how should I do it?