saved_model

使用 saved_model_cli 检查及执行 saved_model

CLI to inspect and execute SavedModel

查看 saved_model 详情

# 查看有哪些  tag-sets
saved_model_cli show --dir /tmp/saved_model_dir

# 查看 tag-set 里有哪些 SignatureDef
saved_model_cli show --dir /tmp/saved_model_dir --tag_set serve

# 查看一个具体的 signature_def 详情
saved_model_cli show \
    --dir /tmp/saved_model_dir --tag_set serve \
    --signature_def serving_default

# 查看所有信息
saved_model_cli show --dir /tmp/saved_model_dir --all

执行 saved_model

saved_model#run_command

### 把 checkpoint 转成 saved_model [Serving Inception Model with TensorFlow Serving and Kubernetes] 可以用链接中写的脚本， 或者直接 docker run ```bash docker run --rm -it --network=host -v D:\models\inception-v3\:/models/inception registry.docker-cn.com/tensorflow/serving:nightly-devel bash # in docker container bazel build -c opt \ tensorflow_serving/example:inception_saved_model bazel-bin/tensorflow_serving/example/inception_saved_model \ --checkpoint_dir=inception-v3 --output_dir=models/inception ```

[Serving Inception Model with TensorFlow Serving and Kubernetes]: https://www.tensorflow.org/serving/serving_inception