TensorFlow 2.2 in SageMaker not saving output model for deployment

Question

I am experiencing an issue when using TensorFlow version 2.2 in Amazon SageMaker to train my model. The output model is not being saved and cannot be deployed (no model artifact in /output folder). However, when using TensorFlow version 1.15, the model is trained and saved successfully. Can anyone provide insight or a solution for this issue?
I am using : 
                     framework_version='2.2.0'
                      py_version="py37"

Thank you in advance

Divyam Chauhan · Answer

There could be multiple reasons for the issue you're facing:

* Incorrect code: Make sure you've correctly saved the model using tf.saved_model.save or tf.keras.models.save_model functions.
* Incompatible versions: TensorFlow 2.2 may have compatibility issues with the Amazon SageMaker environment you're using. Try using a different version of TensorFlow or a different environment.
* Missing dependencies: Make sure you've installed all required dependencies for TensorFlow 2.2.
* Disk space: Check if the disk space is sufficient for saving the model.
* Permission issues: Ensure that the user running the training script has the necessary permissions to write to the /output folder.

Try these potential solutions and see if any of them resolve your issue.

TensorFlow 2.2 in SageMaker not saving output model for deployment

Relevant content