预训练张量流模型的烧瓶

Question

is there a way to use pre-trained char-lstm tensorflow with GPU (1.2) model within flask (0.12.2) app?

Model functions seamlessly when started through shell. I'm loading it with the following code below (without initializing tf variables, placeholders or similar):

....

saver = tf.train.Saver()

with tf.Session() as sess:

    saver.restore(sess, SAVE_PATH)
....

When trying to load the same pre-trained model (with 3 files: .meta , .index & .data ) through Flask it throws

ValueError: No variables to save

Is there a way to make it work? Many thanks !

Answer 1

I think this is because you need to initialize and use your variables properly.

Here's the simplest example I could find:

import tensorflow as tf

# Create TensorFlow object called hello_constant
hello_constant = tf.constant('Hello World!')

with tf.Session() as sess:
    # Run the tf.constant operation in the session
    output = sess.run(hello_constant)
    print(output)

A more complex example that could be more useful for you is:

def model_inputs():
    """
    Create TF Placeholders for input, targets, learning rate, and lengths of source and target sequences.
    :return: Tuple (input, targets, learning rate, keep probability, target sequence length,
    max target sequence length, source sequence length)
    """
    input = tf.placeholder(tf.int32,[None,None],name='input')
    target = tf.placeholder(tf.int32,[None,None])
    target_sequence_len = tf.placeholder(tf.int32, [None], name='target_sequence_length')
    max_target_len = tf.reduce_max(target_sequence_len,name='max_target_length' )
    source_sequence_len = tf.placeholder(tf.int32, [None], name='source_sequence_length')
    learning_rate = tf.placeholder(tf.float32)
    keep_prob = tf.placeholder(tf.float32, name='keep_prob')
    return input, target, learning_rate, keep_prob, target_sequence_len, max_target_len, source_sequence_len

And here's how to make the variables available in the session

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

The topic is important and I find it hard to summarize it efficiently in a single StackOverflow reply.
If you want a deeper look in how the system works I'd suggest to read the TensorFlow documentation about variables.