Create TF serving configuration in python

model_config.proto

syntax = "proto3";

message ModelServerConfig {
  ModelConfigList model_config_list = 1;
}

message ModelConfigList {
  repeated ModelConfig config = 1;
}

message ModelConfig {
  string name = 1;
  string base_path = 2;
  ServableVersionPolicy model_version_policy = 7;
}

message ServableVersionPolicy {
  message Specific {
    repeated int64 versions = 1;
  }

  Specific specific = 102;
}

tf-config.py

from pathlib import Path

from google.protobuf import text_format, json_format

import model_config_pb2

def create_model_server_config(config):
    model_server_config = model_config_pb2.ModelServerConfig()
    for model_variant in config:
        model_config = model_server_config.model_config_list.config.add()
        model_config.name = model_variant["name"]
        model_config.base_path = model_variant["base_path"]
        specific = model_config.model_version_policy.Specific()
        versions = [int(v) for v in model_variant["model_version_policy"]["specific"]["versions"]]
        specific.versions.extend(versions)
        model_config.model_version_policy.specific.CopyFrom(specific)

    return model_server_config


def parse_proto_text(proto_text):
    model_config_list = model_config_pb2.ModelConfigList()
    model_config_list.ParseFromString(proto_text)

    return model_config_list

if __name__ == '__main__':
    with Path("model.fill.config").open() as f:
        message = text_format.Parse(f.read(), model_config_pb2.ModelServerConfig())
        config = json_format.MessageToDict(message, preserving_proto_field_name=True)
        config = config["model_config_list"]["config"]

    model_server_config = create_model_server_config(config)

    print(model_server_config)
    # print(text_format.Parse(text_format.MessageToString(model_server_config), model_config_pb2.ModelServerConfig()))