Documentation

create_iteration_per_loop_var

Description

This API is used in conjunction with load_iteration_per_loop_var to set the number of iterations per training loop every sess.run() call on the device side. This API is used to modify a graph and set the number of iterations per loop using load_iteration_per_loop_var.

Prototype

def create_iteration_per_loop_var(self, train_op)

Options

Option

Input/Output

Description

train_op

Input

Operation that updates the gradient

Returns

An operator for the user to call by using sess.run(op)

Example

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
from npu_bridge.npu_init import *

config = tf.ConfigProto(allow_soft_placement=True)
custom_op =  config.graph_options.rewrite_options.custom_optimizers.add()
custom_op.name =  "NpuOptimizer"
custom_op.parameter_map["enable_data_pre_proc"].b = True # If the GetNext operator exists on the network, offload it. GetNext operator offload is a prerequisite for iteration offload.
custom_op.parameter_map["iterations_per_loop"].i = 10  # Used for functional validation. Must be equal to iterations_per_loop set in load_iteration_per_loop_var.
config = npu_config_proto(config_proto=config)

# Train a model.
with tf.Session(config=config) as sess:
    sess.run(init)
    # Set the number of iterations per loop to 10 in sess.run mode.
    iteration = util.IterationPerLoop()   # Define the IterationPerLoop object.
    train_op = iteration.create_iteration_per_loop_var(optimizer) # Modify the graph.
    tf.train.Supervisor(logdir="/home/xxxx",init_op=init)  # Freeze the graph.
    iteration.load_iteration_per_loop_var(sess, 10)  # Set the number of iterations per loop.

    for epoch in range(training_epochs):
        avg_cost = 0
        total_batch = int(mnist.train.num_examples / batch_size)
 
        for i in range(total_batch):
            batch_xs, batch_ys = mnist.train.next_batch(batch_size)
            _, c = sess.run([train_op, cost], feed_dict={x: batch_xs, y: batch_ys})
 
            avg_cost += c / total_batch