通用句子编码器中的 Tensorflow 会话错误

Question

我有以下通用句子编码器代码，一旦我将模型 加载到烧瓶中 api 并尝试，它就会出现以下错误（检查下方）击中它：

'''

import tensorflow.compat.v1 as tf
tf.disable_v2_behavior()

module_url = "https://tfhub.dev/google/universal-sentence-encoder-large/5"
model_2 = hub.load(module_url)
print ("module %s loaded" % module_url)

def embed(input):
    return model_2(input)


def universalModel(messages):
    accuracy = []
    similarity_input_placeholder = tf.placeholder(tf.string, shape=(None))
    similarity_message_encodings = embed(similarity_input_placeholder)
    with tf.Session() as session:
        session.run(tf.global_variables_initializer())
        session.run(tf.tables_initializer())
        message_embeddings_ = session.run(similarity_message_encodings, feed_dict={similarity_input_placeholder: messages})

        corr = np.inner(message_embeddings_, message_embeddings_)
        accuracy.append(corr[0,1])
    # print(corr[0,1])
    return "%.2f" % accuracy[0]

'''

将模型放入烧瓶时出现以下错误api： tensorflow.python.framework.errors_impl.InvalidArgumentError: Graph无效，包含1个节点的循环，包括：StatefulPartitionedCall 尽管此代码在 colab notebook 中运行时没有任何错误。

我正在使用 tensorflow 版本 2.2.0.

Answer 1

import tensorflow.compat.v1 as tf
tf.disable_v2_behavior()

这两行是为了让tensorflow2.x变成tensorflow1.x。

对于 Tensorflow 1.x，这是使用 flask、django 等服务时的常见问题。您必须定义用于推理的图形和会话，

将 tensorflow 导入为 tf 导入 tensorflow_hub 作为中心

# Create graph and finalize (finalizing optional but recommended).
g = tf.Graph()
with g.as_default():
  # We will be feeding 1D tensors of text into the graph.
  text_input = tf.placeholder(dtype=tf.string, shape=[None])
  embed = hub.Module("https://tfhub.dev/google/universal-sentence-encoder/2")
  embedded_text = embed(text_input)
  init_op = tf.group([tf.global_variables_initializer(), tf.tables_initializer()])
g.finalize()

# Create session and initialize.
session = tf.Session(graph=g)
session.run(init_op)

可以通过

处理输入请求

result = session.run(embedded_text, feed_dict={text_input: ["Hello world"]})

详情 https://www.tensorflow.org/hub/common_issues

对于 tensorflow 2.x 不需要会话和图表。

import tensorflow as tf
module_url = "https://tfhub.dev/google/universal-sentence-encoder-large/5"
model_2 = hub.load(module_url)
print ("module %s loaded" % module_url)

def embed(input):
    return model_2(input)
#pass messages as list
def universalModel(messages):
    accuracy = []
    message_embeddings_= embed(messages)
    corr = np.inner(message_embeddings_, message_embeddings_)
    accuracy.append(corr[0,1])
    # print(corr[0,1])
    return "%.2f" % accuracy[0]

通用句子编码器中的 Tensorflow 会话错误

Tensorflow session error in universal sentence encoder

python

tensorflow

tensorflow2.0

tensorflow-serving

sentence-similarity