ValueError: Output tensors to a Model must be the output of a TensorFlow Layer

@bstriner
Hello Ben,
Apologies for just popping up in your github now. I'm just desperate and I saw that you have already solved a similar problem.

My problem seems to be more serious. I have successfully trained and tested a model outside of aws sagemaker. But soblab I try to train it with a TensorFlow gpu instance the following error pops up.

**ValueError: Output tensors to a Model must be the output of a TensorFlow Layer (thus holding past layer metadata). Found: Tensor("dense/truediv:0", shape=(?, 2, 209), dtype=float32)** by calling

**model = tf.keras.Model(inputs=[numerical_input, os_input, browser_input, action_input], outputs = next_n_actions)**

https://stackoverflow.com/questions/66542710/valueerror-output-tensors-to-a-model-must-be-the-output-of-a-tensorflow-layer

https://gitlab.com/patricksardin08/data-science/-/tree/master/

I would be very happy if you find time to look at it
Thank You!

No repository file or test is named. Start by checking whether aws-sagemaker-remote contains the TensorFlow or SageMaker entry point involved in the reported model construction, then compare it with the linked external project and reproduce the error. Done would require a confirmed repository-scoped fix and a regression test, neither of which the report currently defines.

Written by the indexing model from the issue text.