I have built a web application that calculates text semantic similarity between two user-given sentences. I am using 'Universal Sentence Encoder' by Google to calculate semactic vectors of each sentence and calculate cosine similarity scores. In the application, I have to use
tensorflow, tensorflow-hub,pandas in the application for the text encoder. While the web app runs perfectly in local environment, when I publish it through Google App engine on Google Cloud, it returns a 500 Server Error when I run it. I believe this is due to the limitation of computational resources in Google App Engine.
I tried the highest instance class of the standard environment and flex environment for the Google App Engine as well, but they all did not work as expected. I cannot use the ML Engine, because though I am using TensorFlow on my app, I am not create a ML model.
I have been exploring other options such as Cloud Dataflow, but using Cloud Dataflow is rather much of an overshot for a simple web app, because I can only send data over through DataStore instead of just directly sending data over.
Can you give any suggestions? I am sorry if my question got too verbose. Let me know if clarification should be made.