Saving model for Tensorflow Serving while using Tensorflow distributed training

Question

We are planning to implement distributed training in Tensorflow. For that, we are using Tensorflow Distributed (https://www.tensorflow.org/deploy/distributed). We are able to achieve distributed training using "asynchronous between graph replication training". Below is the code snippet.

.....
ps_hosts = FLAGS.ps_hosts.split(",")
worker_hosts = FLAGS.worker_hosts.split(",")
# Create a cluster from the parameter server and worker hosts.
cluster = tf.train.ClusterSpec({"ps": ps_hosts, "worker": worker_hosts})

# Create and start a server for the local task.
server = tf.train.Server(cluster,
                       job_name=FLAGS.job_name,
                       task_index=FLAGS.task_index)
img_width, img_height = 124, 124

if FLAGS.job_name == "ps":
server.join()
elif FLAGS.job_name == "worker":

####### Assigns ops to the local worker by default.#######
with tf.device(tf.train.replica_device_setter(
    worker_device="/job:worker/task:%d" % FLAGS.task_index,
    cluster=cluster)):

######## set Keras learning phase to train #######
  K.set_learning_phase(1)
  # do not initialize variables on the fly
  K.manual_variable_initialization(True)

  if K.image_data_format() == 'channels_first':
      input_shape = (3, img_width, img_height)
  else:
      input_shape = (img_width, img_height, 3)

  X= tf.placeholder(tf.float32, shape= [None, img_width, img_height, 3], name= "X")
  Y= tf.placeholder(tf.float32, shape= [None, n_classes], name="Y")

  print("Building keras model")

  ....
  ....

  ####### Defining our total loss ####### 

  ###### Defining our TF Optimizer and passing hyperparameters ######      
  .......
  ...........
  ...............

We have defined our training supervisor like as per below.

sv = tf.train.Supervisor(is_chief=(FLAGS.task_index == 0),
                           logdir=logdir,
                           init_op=init_op,
                           saver=saver,
                           summary_op=summary_op,
                           global_step=global_step)

We have also initialized the supervisor using below code snippet.

with sv.prepare_or_wait_for_session(server.target) as sess:

Then we have passed on different batches during the training. Till this part everything works fine. But when we are trying save/export the model for tensorflow serving, it's not generating right set of check point files so that we can serve it in production. While hosting the checkpoint files through tensorflow_model_server, we are getting below error.

 Loading servable: {name: default version: 2} failed: Invalid argument: 
 Cannot assign a device for operation 'init_all_tables': Operation was 
 explicitly assigned to /job:worker/task:0 but available devices are [ 
 /job:localhost/replica:0/task:0/cpu:0 ]. Make sure the device specification 
 refers to a valid device.[[Node: init_all_tables = 
 NoOp[_device="/job:worker/task:0"]()]]

Just to note, we have also tried following methods to save the trained graph.

i) SavedModelBuilder

builder = saved_model_builder.SavedModelBuilder(export_path)

ii) Model exporter

export_path = "/saved_graph/"
model_exporter.export(export_path, sess)

iii) tf.train.Saver - function

tf.train.Saver

but with above none of the scenarios we have seen success.

We couldn't find any direct article which shows a complete example or explain things in details.We have gone through below reference links.

https://github.com/tensorflow/tensorflow/issues/5439 https://github.com/tensorflow/tensorflow/issues/5110 Running distributed Tensorflow with InvalidArgumentError: You must feed a value for placeholder tensor 'Placeholder' with dtype float

Any suggestions or reference would be of great help.

Thank you.

---------------------------------------------------------------------

As per the suggestions, We have tried to use "clear_devices=True" while exporting the model but that didn't help. Below is the code snippet which we have used.

    for epoch in range(training_epochs):

        epoch_num=0

        batch_count = int(num_img/batch_size)
        count = 0

        for i in range(batch_count):

            epoch_num=0

          # This will create batches out of out Training dataset and it will 
           pe passed to the feed_dict    
            batch_x, batch_y = 
            next_batch(batch_size,epoch_num,train_data,train_labels,num_img)

          # perform the operations we defined earlier on batch
            _, cost, step = sess.run([train_op, cross_entropy, global_step],

            feed_dict={X: batch_x, Y: batch_y})
            sess.run(tf.global_variables_initializer())
            builder = tf.saved_model.builder.SavedModelBuilder(path)
            builder.add_meta_graph_and_variables(
            sess,
            [tf.saved_model.tag_constants.SERVING],
            signature_def_map= {
            "magic_model": 
             tf.saved_model.signature_def_utils.predict_signature_def(
             inputs= {"image": X},
             outputs= {"prediction": preds})
             }, clear_devices=True)
            builder.save()

            sv.stop()
            print("Done!!")

We are getting below error when we are using clear_devices=True..

Error: Traceback (most recent call last):
File "insulator_classifier.py", line 370, in 
tf.app.run()
File "/root/anaconda3/lib/python3.6/site-
packages/tensorflow/python/platform/app.py", line 48, in run
_sys.exit(main(_sys.argv[:1] + flags_passthrough))
File "insulator_classifier.py", line 283, in main
}, clear_devices=False)
File "/root/anaconda3/lib/python3.6/site-
packages/tensorflow/python/saved_model/builder_impl.py", line 364, in 
add_meta_graph_and_variables
allow_empty=True)
File "/root/anaconda3/lib/python3.6/site-
packages/tensorflow/python/training/saver.py", line 1140, in __init__
self.build()
File "/root/anaconda3/lib/python3.6/site-
packages/tensorflow/python/training/saver.py", line 1172, in build
filename=self._filename)
File "/root/anaconda3/lib/python3.6/site-
packages/tensorflow/python/training/saver.py", line 677, in build
filename_tensor = constant_op.constant(filename or "model")
File "/root/anaconda3/lib/python3.6/site-
packages/tensorflow/python/framework/constant_op.py", line 106, in constant
attrs={"value": tensor_value, "dtype": dtype_value}, name=name).outputs[0]
File "/root/anaconda3/lib/python3.6/site-
packages/tensorflow/python/framework/ops.py", line 2582, in create_op
self._check_not_finalized()
File "/root/anaconda3/lib/python3.6/site-
packages/tensorflow/python/framework/ops.py", line 2290, in 
_check_not_finalized
raise RuntimeError("Graph is finalized and cannot be modified.")
RuntimeError: Graph is finalized and cannot be modified.

Anything we are missing in here ??

---------------------------------------------------------------------

Further update :

We can see that it works with suggestion 2) from @Tianjin Gu Please see below code snippet

    X= tf.placeholder(tf.float32, shape= [None, img_width, img_height, 3], name= "X")
    Y= tf.placeholder(tf.float32, shape= [None, n_classes], name="Y")

    ....
    ....

    model_exporter = exporter.Exporter(saver)
          model_exporter.init(
     tf.get_default_graph().as_graph_def(),
    named_graph_signatures={
             'inputs': exporter.generic_signature({'input': X}),
             'outputs': exporter.generic_signature({'output': Y})}, clear_devices=True)
          export_path = "/export_path"

As we exporter we see this warning -

    WARNING:tensorflow:From test_classifier.py:283: Exporter.export (from tensorflow.contrib.session_bundle.exporter) is deprecated and will be removed after 2017-06-30.

So we should ideally use "tf.saved_model.builder.SavedModelBuilder" - but that's not working for some reason.

Any further suggestion ?

Thank you.

Saving model for Tensorflow Serving while using Tensorflow distributed training

---------------------------------------------------------------------

---------------------------------------------------------------------

Further update :

Answers (1)

Update:

Related Questions