Skip to content

Conversation

@ytaous
Copy link
Contributor

@ytaous ytaous commented Oct 26, 2020

Eliminate Dropout op when ratio is 0 for ORT training.

Motivation and Context

  • In the experiment for Mainz model, when ratio = 0, it introduce some overhead with Cuda DtoD memcpy.
  • Allow Dropout elimination when:
  1. ratio input is an initializer of 0
  2. ratio input is not a graph input, so it cannot be overridden

@ytaous ytaous added the training issues related to ONNX Runtime training; typically submitted using template label Oct 26, 2020
@ytaous ytaous requested a review from SherlockNoMad October 26, 2020 01:28
@ytaous ytaous requested a review from a team as a code owner October 26, 2020 01:28
SherlockNoMad
SherlockNoMad previously approved these changes Oct 26, 2020
@ytaous ytaous merged commit 6f824c2 into master Oct 27, 2020
@ytaous ytaous deleted the ettao/dropout branch October 27, 2020 18:51
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

training issues related to ONNX Runtime training; typically submitted using template

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants