How to use Apex to make subnetwork using FP16 and another subnetwork using FP32??

[youngfly11]

No description provided.

[ptrblck]

apex.amp is deprecated and you should use the native implementation via torch.cuda.amp as described here

[youngfly11]

apex.amp is deprecated and you should use the native implementation via torch.cuda.amp as described here

Hi, ptrblck;

Thanks for your response. Using native torch.cuda.amp is a good choice. But there is another question that:

Does the current native amp in PyTorch support gradient checkpointing and mixed precision training at the same time?