Run the experiment with distributed mode

[hwang595]

Hi @JiahuiYu, thanks a lot for releasing the code for your awesome work! I'm inspired a lot by reading through your implementation.

I have one quick question about running the slimmable ResNet-50 experiment using distributed mode. In the current configuration file, it triggers this line by default. However, I have a server with 16 GPUs where using DataParallel is not efficient. How can I enable the training with distributed or distributed_all_reduce e.g. as implemented here?

Thanks a lot in advance.

Best,
Hongyi

[JiahuiYu]

Hi Hongyi,

Just adding distributed: True in config file and you should be ready to run.