-
Notifications
You must be signed in to change notification settings - Fork 655
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
[pytorch] Add system property to config GraphExecutorOptimize #2156
Conversation
Codecov ReportBase: 72.08% // Head: 71.52% // Decreases project coverage by
Additional details and impacted files@@ Coverage Diff @@
## master #2156 +/- ##
============================================
- Coverage 72.08% 71.52% -0.57%
- Complexity 5126 6339 +1213
============================================
Files 473 626 +153
Lines 21970 27972 +6002
Branches 2351 3017 +666
============================================
+ Hits 15838 20006 +4168
- Misses 4925 6492 +1567
- Partials 1207 1474 +267
Help us with your feedback. Take ten seconds to tell us how you rate us. Have a feature suggestion? Share it here. ☔ View full report at Codecov. |
0373713
to
cee50cf
Compare
This reverts commit cee50cf.
Add system property to config GraphExecutorOptimize. If true, during inference, an initial warm up in the first inference will be needed, which cost ~3.6s latency. With GraphExecutorOptimize=True, the rest of the latencies are reduced.
This seems to have apparent effect on resnet18 on gpu, while other models are not much affected.
This utilizes
setGraphExecutorOptimize
relavant PR #904Solves the issue #2151
#2153
Update:
Tested in gpu docker image.