This repository has been archived by the owner on Nov 17, 2023. It is now read-only.
-
I found that the training speed of MxNet v1.5.0 built by pip and from source is different. I tested on 2 servers and 2 workers, ran resnet50, and recorded the average throughput of 100 iterations. However, the results are different, as 232.37 samples/sec (built from source) and 325.97 samples/sec (built by pip). Any suggestions? |
Beta Was this translation helpful? Give feedback.
Answered by
Zha0q1
Dec 8, 2020
Replies: 1 comment 8 replies
-
Hi, it might be due to that the build configurations are different. For example you might be using the mkl/cuda optimized pip wheel |
Beta Was this translation helpful? Give feedback.
8 replies
Answer selected by
Rivendile
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Hi, it might be due to that the build configurations are different. For example you might be using the mkl/cuda optimized pip wheel