def forward( self, x ):
start = time.time()
out = self.Rconv2d( x )
cost = time.time()-start
out = self.maxpool1( out )
out = self.block1( out )
out = self.block2( out )
out = self.block3( out )
out = F.avg_pool2d( out, 7 )
out = out.view( out.size(0), -1 )
out = self.out_layer( out )
total_cost = time.time()-start
Finally, use the time module to check the execution efficiency of this layer and find that this transformation layer module accounts for 99.9% of the model’s time
Could you please use an online translate service, so that we could support you?
Custom layers can run on the GPU. However, since CUDA operations are asynchronous, you would have to synchronize the code via torch.cuda.synchronize() before starting and stopping the timer.
Also, please post code snippets by wrapping them into three backticks ```, which makes debugging easier.