Skip to content

Conversation

@ngimel
Copy link
Collaborator

@ngimel ngimel commented Mar 14, 2018

uses pointer arithmetic to operate on 5 tensors with pointwiseApply3. Still pretty general because tensors are guaranteed to come from splitting an original tensor, thus can be addressed by pointer arithmetic.

@goldsborough
Copy link
Contributor

Do we still need gatedLinearCSigMul_functor?
Otherwise LGTM. Did you test that the result is identical?

@ngimel
Copy link
Collaborator Author

ngimel commented Mar 17, 2018

Yes, it is needed for forward. Self tests pass, as you see, and faiseq is training.

@soumith soumith merged commit f390a25 into pytorch:master Mar 17, 2018
@ngimel ngimel deleted the fused_glu branch April 25, 2018 20:10
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.

4 participants