Introduction

The minimal TensorFlow application for benchmarking.

Start predict server

./tensorflow_model_server --port=9000 --model_name=minimal --model_base_path=./model

Start predict client

./predict_client.py --host 127.0.0.1 --port 9000 --model_name minial --model_version 1

cloudml models predict -n minial -s 127.0.0.1:9000 -f ./data.json

Benchmark

Test local predict

./benchmark_predict.py --benchmark_test_number 10 --benchmark_batch_size 1

Test python gRPC client

./benchmark_predict_client.py --host 127.0.0.1 --port 9000 --model_name minial --model_version 1 --benchmark_batch_size 100  --benchmark_test_number 10000