Skip to content

Conversation

@danielsimig
Copy link

Previously EmbeddingBag was not able to deal with empty bags (that is, two consecutive elements in the offsets tensor). With this change, EmbeddingBag will return an all-zero vector as the embedding if it comes across an empty bag, regardless of the mode.

@colesbury
Copy link
Member

cc @cpuhrsch

@EthanSteinberg
Copy link
Contributor

This is a good idea. We really would need that GPU implementation though. Are you down for the GPU changes (I can make the GPU implementation changes if necessarily) ?

Copy link
Member

@colesbury colesbury left a comment

Choose a reason for hiding this comment

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

@adamlerer, you already reviewed this, right?

@danielsimig
Copy link
Author

danielsimig commented May 8, 2018

@colesbury the changes for max pooling have not been reviewed as max pooling was not synced at the time.

@danielsimig
Copy link
Author

@lalaland I'm not familiar with how the GPU implementation works and I don't think I'll have time for it, so if you could help out that'd be fantastic!

@colesbury
Copy link
Member

@pytorchbot retest this please

@soumith soumith merged commit b6adf68 into pytorch:master May 10, 2018
onnxbot added a commit to onnxbot/onnx-fb-universe that referenced this pull request May 10, 2018
@pjh5 pjh5 mentioned this pull request May 11, 2018
weiyangfb pushed a commit to weiyangfb/pytorch that referenced this pull request Jun 11, 2018
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