python-tensorflowHow can I use graph_matcher from tensorflow.contrib.quantize.python to optimize my model?

graph_matcher from tensorflow.contrib.quantize.python provides a convenient way to optimize models. It allows you to apply a series of optimizations to the model's graph, such as quantizing weights and replacing certain operations with more efficient ones.

Example code

import tensorflow as tf
from tensorflow.contrib.quantize.python import graph_matcher

g = tf.Graph()
with g.as_default():
  # Create the model
  x = tf.placeholder(tf.float32, shape=[None, 784])
  W = tf.Variable(tf.random_normal([784, 10]))
  b = tf.Variable(tf.zeros([10]))
  y = tf.matmul(x, W) + b

# Match all nodes that have "Variable" as an op
matcher = graph_matcher.GraphMatcher(g, "Variable")

# Apply quantization to all the matched nodes
for match in matcher.match():
  tf.contrib.quantize.create_training_graph(input_graph=g, weight_bits=8)

The code above applies quantization to all the variables in the model's graph. This is just one of the optimizations that graph_matcher can perform. Other optimizations include replacing operations with more efficient ones, such as replacing tf.matmul with tf.linalg.matmul.

Code explanation

1. import tensorflow as tf: imports the TensorFlow library.
2. from tensorflow.contrib.quantize.python import graph_matcher: imports the graph_matcher module from TensorFlow's contrib library.
3. g = tf.Graph(): creates a new graph object.
4. with g.as_default():: sets the new graph as the default graph.
5. x = tf.placeholder(tf.float32, shape=[None, 784]): creates a placeholder for the input data.
6. W = tf.Variable(tf.random_normal([784, 10])): creates a variable for the weights of the model.
7. b = tf.Variable(tf.zeros([10])): creates a variable for the bias of the model.
8. y = tf.matmul(x, W) + b: creates an operation to calculate the output of the model.
9. matcher = graph_matcher.GraphMatcher(g, "Variable"): creates a GraphMatcher object to match nodes in the graph.
10. for match in matcher.match():: iterates through the matched nodes.
11. tf.contrib.quantize.create_training_graph(input_graph=g, weight_bits=8): applies quantization to the matched nodes.

Helpful links

• TensorFlow GraphMatcher API
• TensorFlow Quantization API

Edit this code on GitHub