search

keras-team / keras-io

Keras documentation, hosted live at keras.io
Apache License 2.0
2.76k stars 2.04k forks source link

Training time doubles after replacing Conv2D with SeparableConv2D #1225

Closed coelho-k closed 1 year ago

coelho-k commented 1 year ago

Everything about the architecture remains constant. Using Tensorflow 2.9.0 within a Docker container, with cudnn 8

grasskin commented 1 year ago

Hi @coelho-k, could you provide a minimal colab example for reproduction?

coelho-k commented 1 year ago

Hi @grasskin, It is the exact code from: https://keras.io/examples/vision/mobilevit/ However, I just replaced all the Conv2D layers with SeparableConv2D

SuryanarayanaY commented 1 year ago

Hi @coelho-k ,

Apologies for the delay. I have tested the attached example with Conv2D and SeparableConv2D layers respectively and got training completed in 1219.6 seconds and 1236.8 seconds respectively which is almost same time for both cases. Please refer to attached gist.

github-actions[bot] commented 1 year ago

This issue is stale because it has been open for 14 days with no activity. It will be closed if no further activity occurs. Thank you.

github-actions[bot] commented 1 year ago

This issue was closed because it has been inactive for 28 days. Please reopen if you'd like to work on this further.