Skip to content

Conversation

@ssnl
Copy link
Collaborator

@ssnl ssnl commented Mar 24, 2018

After discussing with @soumith , we decide to remove track_running_stats option from LayerNorm as it doesn't make much sense.

cc @soumith

@ssnl
Copy link
Collaborator Author

ssnl commented Mar 24, 2018

@pytorchbot retest this please

This comment was marked as off-topic.

This comment was marked as off-topic.

@ezyang ezyang requested a review from soumith March 26, 2018 18:10

This comment was marked as off-topic.

@ssnl
Copy link
Collaborator Author

ssnl commented Mar 29, 2018

@zou3519 does the new commit look good to you? :)

Copy link
Contributor

@zou3519 zou3519 left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Slightly concerned that we're not documenting torch.layer_norm and co's last argument (the cudnn_enabled) but it should be fine because users should interact with layer norm through nn.LayerNorm

image

@ssnl
Copy link
Collaborator Author

ssnl commented Mar 30, 2018

@zou3519 That's right. We have so many torch.* that are hidden from user.

@apaszke
Copy link
Contributor

apaszke commented Mar 30, 2018

If we’re not planning on making a function part of our interface then can we please prefix it with an underscore?

@ssnl
Copy link
Collaborator Author

ssnl commented Mar 30, 2018

@apaszke Having the non-underscored version helps our ATen users. I think this is the reason why BN isn't called at::_batch_norm

@soumith soumith merged commit 48ad454 into pytorch:master Mar 30, 2018
@ssnl ssnl deleted the ln branch March 30, 2018 16:55
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Projects

None yet

Development

Successfully merging this pull request may close these issues.

5 participants