改变训练参数

清理模型

$>rm -rf /tmp/mnist_model

训练（100迭代）并导出mnist模型

$>bazel build -c opt //tensorflow_serving/example:mnist_saved_model
$>bazel-bin/tensorflow_serving/example/mnist_saved_model --training_iteration=100 --model_version=1 /tmp/mnist_model

训练(2000迭代)并导出mnist模型

$>bazel-bin/tensorflow_serving/example/mnist_saved_model --training_iteration=2000 --model_version=2 /tmp/mnist_model

查看模型导出目录

$>ls /tmp/mnist_model
1  2

改变服务发布参数

$>mkdir /tmp/monitored
$>cp -r /tmp/mnist_model/1 /tmp/monitored
$>bazel build -c opt //tensorflow_serving/model_servers:tensorflow_model_server
$>bazel-bin/tensorflow_serving/model_servers/tensorflow_model_server --enable_batching --port=9000 --model_name=mnist --model_base_path=/tmp/monitored

增加并行请求

$>bazel build -c opt //tensorflow_serving/example:mnist_client
$>bazel-bin/tensorflow_serving/example/mnist_client --num_tests=1000 --server=localhost:9000 --concurrency=10

将model/2移入模型检测目录

$>cp -r /tmp/mnist_model/2 /tmp/monitored
$>bazel-bin/tensorflow_serving/example/mnist_client --num_tests=1000 --server=localhost:9000 --concurrency=10

ModelServer会自动发现更新的版本，并将其发布为服务。