Get tensorflow serving response directly in output proto format

Question

I have an output vector coming from tensorflow model which is a 50x1 vector of probabilities [0.4 0.3 0.4 ... 0.5] where first 20 dimensions are user vectors and last 30 are