ml_model_rest_server

This project implements a common rest server which can serve tensorflow-serving & xgboost models.

Architecture

Installation

Dependencies

• proxygen
• tensorflow
• tensorflow-serving
• xgboost

Steps

• ./configure
• cd debug
• make

Example

• To serve tf-serving model, start tf-serving first.

POST http://localhost:11000

{"input_type": "ints", "inputs": [1, 2, 4, 5], "model_name": "tensorflow-feature-sum-model", "model_signature_name": "tensorflow-feature-sum-model"}

• Output

{"y_output":"12"}

• To serve xgboost models, put the *.cpb files in the debug/src/frontends directory.

POST http://localhost:11000

{"input_type": "floats", "inputs": [1.0, 2.0, 4.0, 5.0], "model_name": "test.cpb", "model_signature_name": "xgboost"}

• Output

{"y_output":"0.0001"}

Sample models

A sample xgboost test.cpb model is present in resources directory.

Next steps

• docker integration