Question

我复制并使用了这段代码。网址：http://fdahms.com/2017/03/05/tensorflow-serving-jvm-client/

但部署版本时发生错误。

Model validation failed: Serving metagraph must contain exactly one SignatureDef with key: serving_default

我试图通过引用https://github.com/tensorflow/serving/blob/master/tensorflow_serving/example/mnist_saved_model.py

来修复代码

import tensorflow as tf


x = tf.placeholder(tf.float32, shape=(None))
y = tf.placeholder(tf.float32, shape=(None))

three = tf.Variable(3, dtype= tf.float32)
z = tf.scalar_mul(three, x) + y

import os
from tensorflow.python.util import compat

model_version = 1
path = os.path.join("three_x_plus_y", str(model_version))
builder = tf.saved_model.builder.SavedModelBuilder(path)

legacy_init_op = tf.group(tf.tables_initializer(), name='legacy_init_op')

tensor_info_x = tf.saved_model.utils.build_tensor_info(x)
tensor_info_y = tf.saved_model.utils.build_tensor_info(y)
tensor_info_z = tf.saved_model.utils.build_tensor_info(z)

prediction_signature = (
      tf.saved_model.signature_def_utils.build_signature_def(
          inputs= {'egg': tensor_info_x, 'bacon': tensor_info_y},
          outputs= {'spam': tensor_info_z},
          method_name=tf.saved_model.signature_constants.PREDICT_METHOD_NAME))


with tf.Session() as sess:
    sess.run(tf.global_variables_initializer())

    builder.add_meta_graph_and_variables(
        sess,[tf.saved_model.tag_constants.SERVING],
        signature_def_map= {
            "magic_model": prediction_signature},
        legacy_init_op=legacy_init_op
        )
    builder.save()

但是我得到了同样的错误......

我正在使用Google Cloud Machine Running Engine＆＃39; 我需要帮助..谢谢你的阅读。

Answer 1

将signature_def_map中的密钥从magic_model更改为serving_default。

with tf.Session() as sess:
    sess.run(tf.global_variables_initializer())

    builder.add_meta_graph_and_variables(
        sess,[tf.saved_model.tag_constants.SERVING],
        signature_def_map= {
            "serving_default": prediction_signature},
        legacy_init_op=legacy_init_op
        )

云ml引擎版本创建错误3

1 个答案: