Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Fix BERT EmbeddingPostprocessor #7706

Merged
merged 1 commit into from Oct 23, 2019

Conversation

Vooblin
Copy link
Contributor

@Vooblin Vooblin commented Oct 22, 2019

Change one_hot+matmul on gather. It's important for conversion to tflite because one_hot+matmul seriously change model predictions after conversion in tflite format. Gather does the same operations and doesn't have the same effect.

It is related to problem in tflite. Issue: tensorflow/tensorflow#33613.
My fix allow to convert model to tflite before this problem will be solved

@saberkun
Copy link
Member

  • Renjie from TFLite team to understand the issue.
    Thanks for the PR. I think the PR itself is good and I think it is trivial to GPU/TPU.
    But I heard tflite is not happy with tf.gather?

@renjie-liu
Copy link
Member

Gather is supported in tflite

@saberkun
Copy link
Member

I see. That's great.
Let me test against internal deps.

@saberkun saberkun added the ready to pull ready to pull (create internal pr review and merge automatically) label Oct 23, 2019
tf-models-copybara-bot pushed a commit that referenced this pull request Oct 23, 2019
@tf-models-copybara-bot tf-models-copybara-bot merged commit c34c439 into tensorflow:master Oct 23, 2019
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
cla: yes ready to pull ready to pull (create internal pr review and merge automatically)
Projects
None yet
Development

Successfully merging this pull request may close these issues.

None yet

6 participants