fix start index of statistic arrays #2926
Conversation
Codecov ReportAll modified and coverable lines are covered by tests ✅
Additional details and impacted files@@ Coverage Diff @@
## develop #2926 +/- ##
========================================
Coverage 91.83% 91.83%
========================================
Files 479 479
Lines 18340 18340
========================================
Hits 16842 16842
Misses 1498 1498 ☔ View full report in Codecov by Sentry. |
| @@ -263,12 +263,12 @@ def main(): | |||
|
|
|||
| if flags.QPS: | |||
| print("resnet50, Rate = {} QPS ".format( | |||
| format((((flags.batch) / (sum(latency[1:]) / len(latency[1:])))), | |||
| format((((flags.batch) / (sum(latency[0:]) / len(latency[0:])))), | |||
There was a problem hiding this comment.
Good find.
I assume the reason why it started with [1:] was that it considered first run as the warmup run.
If you doing --run 1 it wouldn't give you the most accurate performance numbers.
I think it is better to do warm up runs as a separate calls.
There was a problem hiding this comment.
Yes, we needed the warmup run since there was overhead during setup.
This is required as this step also includes the additional run. A better approach here would be to set the amount of runs to run 1 + 1 so that you still do a warmup run, but run two runs under the hood.
|
@jasberc why was this closed? |
This fixes corruption of "python resnet50.py" with "--run 1" and correct statistic results.