-
Notifications
You must be signed in to change notification settings - Fork 1.1k
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
TypeError: An op outside of the function building code is being passed a Graph tensor #620
Comments
I can still replicate this. @jburnim to triage. |
@srvasude Yes, this bug still exists in TensorFlow 2.1 and TFP 0.9. This is a big issue that should be solved as soon as possible because apparently the only workaround to it (described in this comment tensorflow/tensorflow#33729 (comment)) will not be available in TensorFlow 2.2 (see this issue tensorflow/tensorflow#35138). |
We're looking into this. As a potential workaround, at least some cases it appears that calling |
@davmre That doesn't seem to work with my example above using TF 2.1 and TFP 0.9.0. Can you provide an example where that trick avoids the problem? |
…_tf_function=True` - Make `_DenseVariational` layers eager compatible. PiperOrigin-RevId: 264872770
This error doesn't occur anymore in the nightly versions of TF and TFP. |
…work. Need to fix Convolutional1D not using 'causal' padding. Also need to fix disabling eager_execution required because of known issue with Conv1DFlipout in current tf 2.2 and tfp 0.9 versions. Problem identified here: tensorflow/probability#620
@nbro I'm still getting this with tensorflow 2.2.0, which should have included the nightly fixes as of Apr 28 2020. |
@mcmar Do you have a small, self-contained example of this? Which layers were you using? |
This issue of using the eager execution still persists with TFlow 2.4.0 and TFlow-Prob: 0.12.1. Update
|
System information
Describe the current behavior
I am getting the error
After having gotten the exception
See the detailed traceback below.
Describe the expected behavior
No error.
Code to reproduce the issue
Other info / logs
The problem is apparently related to the layer
tfp.layers.Convolution2DFlipout
. I know that if I usetf.compat.v1.disable_eager_execution()
after having imported TensorFlow, I do not get the mentioned error anymore, but I would like to use TensorFlow's eager execution, avoid sessions or placeholders.I opened the same issue here: tensorflow/tensorflow#33729.
The text was updated successfully, but these errors were encountered: