Skip to content

Conversation

@zou3519
Copy link
Contributor

@zou3519 zou3519 commented Mar 15, 2018

Fixes #5801

Two fixes:

  1. Fix incorrect kldiv backward on CUDA
  2. Change Criterion testing to test that the criterion uses gradOutput. The problem with KLDiv is that it was not using gradOutput but we didn't have a test for this.

test/test_nn.py Outdated

This comment was marked as off-topic.

This comment was marked as off-topic.

@zou3519
Copy link
Contributor Author

zou3519 commented Mar 15, 2018

@pytorchbot reteset this please

@zou3519
Copy link
Contributor Author

zou3519 commented Mar 15, 2018

@pytorchbot retest this please

@ssnl
Copy link
Collaborator

ssnl commented Mar 16, 2018

you might want to rebase to make onnx pass..

@zou3519
Copy link
Contributor Author

zou3519 commented Mar 16, 2018

😢

@soumith soumith merged commit 04edb89 into pytorch:master Mar 17, 2018
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

KLDivLoss behaves differently on CPU/GPU

4 participants