-
Notifications
You must be signed in to change notification settings - Fork 74k
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
tf.contrib.rnn.GLSTMCell is hilariously broken #16703
Comments
/CC @panyx0718, can you take a look? |
Right. this is a bug introduced during a refactoring. Perhaps we should cache one Linear for each group_id. |
/CC @asimshankar, can you fix this? |
Doh! Thanks for pointing this out @akhti - will send out a fix. |
No worries - the bug was here since October and nobody noticed :) |
* Fix issue described here: tensorflow#16703 * Make it work correctly for input sizes!=num_units
Please have a look at this PR #16788 |
In 3f57902 an "if" was added that caches linear transformation weights:
https://github.com/tensorflow/tensorflow/blob/r1.5/tensorflow/contrib/rnn/python/ops/rnn_cell.py#L2316
The problem is that this _linear is inside a loop. And so the change tied weights of all these linear transformations.
CC @okuchaiev
The text was updated successfully, but these errors were encountered: