docker - 有自定义操作时如何使用docker image tensorflow/serving服务tensorflow模型？

Question

我正在尝试在这里找到的模型中使用tf-sentencepiece操作https://github.com/google/sentencepiece/tree/master/tensorflow

建立模型并获取包含变量和 Assets 的saved_model.pb文件没有问题。但是，如果我尝试将docker镜像用于tensorflow / serving，它会说

Loading servable: {name: model version: 1} failed: 
Not found: Op type not registered 'SentencepieceEncodeSparse' in binary running on 0ccbcd3998d1. 
Make sure the Op and Kernel are registered in the binary running in this process. 
Note that if you are loading a saved graph which used ops from tf.contrib, accessing 
(e.g.) `tf.contrib.resampler` should be done before importing the graph, 
as contrib ops are lazily registered when the module is first accessed.

我不熟悉如何手动构建任何东西，并希望我可以进行很多更改。

Answer 1

一种方法是:

拉docker开发镜像

$ docker pull tensorflow / serving:最新开发的

在容器中，更改代码

$ docker run -it tensorflow / serving:latest-devel

修改代码以添加op依赖项 here。

在容器中，构建TensorFlow Serving

容器:$ tensorflow_serving / model_servers:tensorflow_model_server && cp bazel-bin / tensorflow_serving / model_servers / tensorflow_model_server / usr / local / bin /

使用exit命令退出容器

查找容器ID:

$ docker ps

使用该容器ID提交开发镜像:

$ docker commit $ USER / tf-serving-devel-custom-op

现在使用开发容器作为源来构建服务容器

$ mkdir / tmp / tfserving

$ cd / tmp / tfserving

$ git clone https://github.com/tensorflow/serving。

$ docker build -t $ USER / tensorflow-serving --build-arg TF_SERVING_BUILD_IMAGE = $ USER / tf-serving-devel-custom-op -f tensorflow_serving / tools / docker / Dockerfile。

现在，您可以使用$ USER / tensorflow-serving在Docker instructions后面提供图像