What is the inference speed?

[qijiezhao]

Hi, I ran a script to test the inference speed:

net=shufflenetv2().cuda()
data=torch.autograd.Variable(torch.ones(8,3,224,224))
t1=time.time()
for i in range(100):
    _=net(data)
t2=time.time()
print((t2-t1)/float(100))

It gets ~4.4ms, slower than ~3ms, which is proposed in the original paper.
So, how is the difference?

[lxtGH]

I got the same problem, have you solved this problem

[HITerStudy]

I got the same problem, have you solved this problem

Hello, why my test result was 20ms? Please give me some helps!Thanks.