Warning: Grad strides do not match bucket view strides. 可能影响DDP性能
1.transpose或permute造成内存不连续。#beforeoutput_tensor=in_tensor.transpose(1,3)#afteroutput_tensor=in_tensor.transpose(1,3).contiguous()2.切片操作造成内存不连续。#beforeinput_tensor=input_tensor[:,:H,:W,:]#afterinput_te