A slightly easier to use tensorflow-serving -- auto-construct models.config

!tensorflow-serving.md

README

docker.io/tensorflow/serving requires that you build a config file when working with more than one model. I find this a bit silly, this image builds it for you assuming you just have all the models in their separate directories.

• /models/model-a/1/
• /models/model-b/1/
• ...

Dockerfile

FROM tensorflow/serving
ADD entrypoint.sh /entrypoint.sh
RUN chmod +x /entrypoint.sh
ENTRYPOINT ["bash", "/entrypoint.sh"]
CMD []

entrypoint.sh

#/bin/bash

(
  echo "model_config_list: {"
  FIRST=true
  while IFS='\n' read MODEL; do
    if [ $FIRST = true ]; then
      FIRST=false;
    else
      echo ",";
    fi
    echo "  config: {"
    echo "    name: \"${MODEL}\","
    echo "    base_path: \"/models/${MODEL}\","
    echo "    model_platform: \"tensorflow\""
    echo "  }"
  done < <(find /models -mindepth 1 -maxdepth 1 -type d -printf "%f\n")
  echo "}"
) > /tmp/models.config

/usr/bin/tf_serving_entrypoint.sh --model_config_file=/tmp/models.config $@