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
model.run_eagerly=False is much slower than model.run_eagerly=True #35052
Comments
Issue is replicating with Tf 2.0. |
@doldre I was able to reproduce the issue with 1.default
Please close this issue if it was resolved by |
@jvishnuvardhan thanks for your help. I reproduced your result with tf-nightly(2.1.0-dev20191125). It looks like the problem was fixed. |
I found this problem is still existed when using RMSprop optimizer. |
This issue has been fixed, @doldre you can verify it through tf-nightly. If this is not do-able, you can also wait for 2.2. |
Please make sure that this is a bug. As per our GitHub Policy, we only address code/doc bugs, performance issues, feature requests and build/installation issues on GitHub. tag:bug_template
System information
Describe the current behavior
I try to implemente a simple fm algorithm in tensorflow 2.0. I found use keras.fit is very slow in default params.
If I change the model.run_eagerly to True the performance will be better.
Then I tried turn off eager_execution by tf.compat.v1.disable_eager_execution(), the performance is the same as tf1.14 with estimator.
Describe the expected behavior
I think keras.fit with model.run_eagerly=False will use tf.function to wrap the training loop, and it's performance should be close to the disable eager execution. But it's perform awfully, it even slower than model.run_eagerly=True.
Code to reproduce the issue
Other info / logs
Include any logs or source code that would be helpful to diagnose the problem. If including tracebacks, please include the full traceback. Large logs and files should be attached.
The text was updated successfully, but these errors were encountered: