How to add a new model in tensorflow serving

问题:

I know there could be more than one model running at a time, which you can specify in a config file, as explained here.

In my case, I want to start the server with model_A, model_B and model_C, and in a future, add a new arbitrary model_D without restarting the server (since I don't want to interrupt the service for models A, B and C).

Is there a way to achieve this requirement?

回答1:

The following commit adds this functionality (and should be in TensorFlow Serving 1.7.0 and higher:

https://github.com/tensorflow/serving/commit/79d3354cd8a408fb981670512d661c2b7e89f5bf