lightseq cross entropy kernel is slower than original torch kernel.

[hyunwoongko]

CrossEntropyLoss:    0.0004372596740722656
LSCrossEntropyLayer: 0.0010995864868164062

Are there any benchmark results for this kernel?
As a result of my experiments, it seems to be slower than the original torch kernel.

[Taka152]

Have you tried our code for the speed test?

[hyunwoongko]

nope. I just used time.time() module.
I was wondering if my experiment was wrong, so I wanted to check any benchmark result.

[Taka152]

time.time() is testing on the CPU side, because GPU executes asynchronously, maybe you should ensure that the GPU task is finished. For example, retrieve the result at the end.

…

On Wed, Oct 13, 2021 at 7:53 PM Kevin Ko ***@***.***> wrote: nope. I just used time.time() module. I was wondering if my experiment was wrong, so I wanted to check any benchmark result. — You are receiving this because you commented. Reply to this email directly, view it on GitHub <#204 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AELIZAPVTTLQZLNDUSLEOQDUGVXNNANCNFSM5F4JX2UQ> .

[hyunwoongko]

Thank you for reply.
What does "retrieve the result at the end." means?

Thanks.

[Taka152]

For example, printing the result to the console will ensure that the GPU task is finished.

…

On Wed, Oct 13, 2021 at 10:43 PM Kevin Ko ***@***.***> wrote: Thank you for reply. What does "retrieve the result at the end." means? Thanks. — You are receiving this because you commented. Reply to this email directly, view it on GitHub <#204 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AELIZAODXKYT5ERUVXL73JTUGWLIXANCNFSM5F4JX2UQ> .

[hyunwoongko]

So is there a simple way to benchmark the time spent on the GPU?