Upload TensorFlow model to S3

This component uploads a TensorFlow network (top input) as saved model into the defined S3 working directory/bucket (bottom input). If not already existing, a directory with the model name and a subdirectory with the model version will be created. The files of the saved model will then be uploaded into /model-name/model-version/.

Note: TensorFlow Serving expects the model to have the 'serving' tag. If the model is not already tagged, you can use the "Add 'serving' tag to TensorFlow model" component.

Note: A running TensorFlow Serving server does not recognize the change of models with the same name and version, hence overwriting an existing model version is not possible. If you want to overwrite a model version, delete it first and wait about 5 minutes to make sure TensorFlow Serving has unloaded the model and will recognize the new one.

Input ports

1. Type: TensorFlow Deep Learning Network
   Port 1
   A TensorFlow network.
2. Type: File System
   Port 2
   An S3 file system connection. It has the bucket in which the model should be uploaded as working directory set.