一、安装docker
- 由于apt官方库里的docker版本可能比较旧,所以先卸载可能存在的旧版本:
$ sudo apt-get remove docker docker-engine docker-ce docker.io
- 更新apt包索引:
$ sudo apt-get update
- 安装以下包以使apt可以通过HTTPS使用存储库(repository):
$ sudo apt-get install -y apt-transport-https ca-certificates curl software-properties-common
- 添加Docker官方的GPG密钥:
$ curl -fsSL https://download.docker.com/linux/ubuntu/gpg | sudo apt-key add -
- 使用下面的命令来设置stable存储库:
$ sudo add-apt-repository "deb [arch=amd64] https://download.docker.com/linux/ubuntu $(lsb_release -cs) stable"
- 再更新一下apt包索引:
$ sudo apt-get update
- 安装最新版本的Docker CE:
$ sudo apt-get install -y docker-ce
二、模型转为pb文件
def save_model_for_production(model, version='1', path='prod_models'):
tf.keras.backend.set_learning_phase(1)
if not os.path.exists(path):
os.mkdir(path)
export_path = os.path.join(
tf.compat.as_bytes(path),
tf.compat.as_bytes(version))
builder = tf.compat.v1.saved_model.builder.SavedModelBuilder(export_path)
model_input = tf.compat.v1.saved_model.build_tensor_info(model.input)
model_output = tf.compat.v1.saved_model.build_tensor_info(model.output)
prediction_signature = (
tf.compat.v1.saved_model.signature_def_utils.build_signature_def(
inputs={'inputs': model_input},
outputs={'outputs': model_output},
method_name=tf.compat.v1.saved_model.signature_constants.PREDICT_METHOD_NAME))
with tf.compat.v1.keras.backend.get_session() as sess:
builder.add_meta_graph_and_variables(
sess=sess, tags=[tf.compat.v1.saved_model.tag_constants.SERVING],
signature_def_map={
'predict':prediction_signature,
tf.saved_model.signature_constants.DEFAULT_SERVING_SIGNATURE_DEF_KEY:prediction_signature
})
builder.save()
三、开启服务
1、准备TF Serving的Docker环境
docker pull tensorflow/serving
2、开启服务
docker run -dt -p 8501:8501 -v /home/wangdong/TextClassification/export_model:/models/export_model -e MODEL_NAME=export_model tensorflow/serving &
3、验证是否开启成功
curl -d '{"instances":[[0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,1,38,508,980,35,140]]}' -X POST http://localhost:8501/v1/models/export_model:predict
注意:其中export_model为存储.pb文件的文件夹名称
四、tensorflow serving发布多个模型
详文见此博客:https://blog.csdn.net/JerryZhang__/article/details/86516428
1、创建一个多模型配置文件
#从tensorflow-serving git文档上抄下来的
sudo vim {你的tf-serving配置文件根目录}/models.config
贴入以下内容
model_config_list: {
config: {
name: "half_plus_two",
base_path: "/models/half_plus_two",
model_platform: "tensorflow"
},
config: {
name: "half_plus_three",
base_path: "/models/half_plus_three",
model_platform: "tensorflow"
}
}
2、启动一个docker container
sudo docker run -p 8500:8500 -p 8501:8501 \
--mount type=bind,source={你的tf-serving根目录}/serving/tensorflow_serving/servables/tensorflow/testdata/saved_model_half_plus_two_cpu,target=/models/half_plus_two\
--mount type=bind,source={你的tf-serving根目录}/serving/tensorflow_serving/servables/tensorflow/testdata/saved_model_half_plus_three,target=/models/half_plus_three\
--mount type=bind,source={你的tf-serving配置文件根目录}/models.config,target=/models/models.config \
-t tensorflow/serving --model_config_file=/models/models.config&
3、测试多个模型调用
curl -d '{"instances": [1.0, 2.0, 5.0]}' -X POST http://127.0.0.1:8501/v1/models/half_plus_two:predict
curl -d '{"instances": [1.0, 2.0, 5.0]}' -X POST http://127.0.0.1:8501/v1/models/half_plus_three:predict
注意,注意,注意:版本很重要,我使用的版本是tensorflow1.13.1,其他版本很可能不成功。