이 MNIST training을 위한 tensorflow python code는 원래 /opt/DL/tensorflow/lib/python3.6/site-packages/tensorflow/examples/tutorials/layers/cnn_mnist.py 에 LMS를 적용한 것입니다. 보시다시피 hook를 사용하기 때문에 Estimator-based tensorflow training이고, 그에 따라 LMS가 적용되어 있습니다.
LMS가 동작하는 message 확인이라든가, 다중 사용자를 위한 permission 등을 위한 부분도 있습니다만 그건 LMS와는 무관한 부분이고, 그 부분들은 빨간색으로 표시를 했습니다. 실제 LMS 구현을 위한 부분은 굵은 파란색으로 표시했습니다. 의외로 간단하다는 것을 보실 수 있습니다. 해당 부분들을 제거하면 그냥 LMS 없는 평범한 MNIST training code가 됩니다.
이 example code도 PowerAI 5.2를 설치하면 딸려오는 /opt/DL/tensorflow/lib/python3.6/site-packages/tensorflow/contrib/lms/examples/cnn_mnist_lms.py 을 그대로 가져다 놓은 것입니다.
실제 수행해보면 다음과 같이 동작하며, 12개의 tensor가 host 서버의 RAM으로 swap-out/in 되는 것을 보실 수 있습니다.
[bsyu@p57a22 ~]$ cd /opt/DL/tensorflow/lib/python3.6/site-packages/tensorflow/contrib/lms/examples
[bsyu@p57a22 examples]$ source /opt/DL/tensorflow/bin/tensorflow-activate
[bsyu@p57a22 examples]$ python cnn_mnist_lms.py
...
INFO:tensorflow:[LMS][1] Tensor sparse_softmax_cross_entropy_loss/Sum_1:0 will be placed on /cpu:0
INFO:tensorflow:[LMS][1] Operation: adam_optimizer/gradients/sparse_softmax_cross_entropy_loss/div_grad/RealDiv_1, order 23, type RealDiv
INFO:tensorflow:[LMS][1] Tensor adam_optimizer/gradients/sparse_softmax_cross_entropy_loss/div_grad/RealDiv_1:0 will be placed on /cpu:0
INFO:tensorflow:[LMS][1] Consuming op adam_optimizer/gradients/sparse_softmax_cross_entropy_loss/div_grad/RealDiv_2 (order 24) swaps in adam_optimizer/gradients/sparse_softmax_cross_entropy_loss/div_grad/RealDiv_1:0
INFO:tensorflow:[LMS][1] No control dependency op needed for swap in of op adam_optimizer/gradients/sparse_softmax_cross_entropy_loss/div_grad/RealDiv_1.
INFO:tensorflow:[LMS][0] Edited model is valid and logically equivalent to the original one
INFO:tensorflow:[LMS][0] Added 25 ops into the model
INFO:tensorflow:[LMS][0] Editing model for LMS, took: 88.58513832092285 ms
INFO:tensorflow:[LMS][0] 12 tensors will be swapped out(in) to(from) the host
INFO:tensorflow:Graph was finalized.
...
{'accuracy': 0.9702, 'loss': 0.098796368, 'global_step': 20000}
전체 code 내용은 아래를 보시기 바랍니다.
[bsyu@p57a22 examples]$ cat /opt/DL/tensorflow/lib/python3.6/site-packages/tensorflow/contrib/lms/examples/cnn_mnist_lms.py
# Copyright 2016 The TensorFlow Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
"""Convolutional Neural Network Estimator for MNIST, built with tf.layers."""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import numpy as np
import tempfile # Change not related to LMS
import tensorflow as tf
tf.logging.set_verbosity(tf.logging.INFO) #LMS 기능과는 무관. LMS 메시지를 보기 위한 설정.
def cnn_model_fn(features, labels, mode):
"""Model function for CNN."""
# Input Layer
# Reshape X to 4-D tensor: [batch_size, width, height, channels]
# MNIST images are 28x28 pixels, and have one color channel
input_layer = tf.reshape(features["x"], [-1, 28, 28, 1])
# Convolutional Layer #1
# Computes 32 features using a 5x5 filter with ReLU activation.
# Padding is added to preserve width and height.
# Input Tensor Shape: [batch_size, 28, 28, 1]
# Output Tensor Shape: [batch_size, 28, 28, 32]
conv1 = tf.layers.conv2d(
inputs=input_layer,
filters=32,
kernel_size=[5, 5],
padding="same",
activation=tf.nn.relu)
# Pooling Layer #1
# First max pooling layer with a 2x2 filter and stride of 2
# Input Tensor Shape: [batch_size, 28, 28, 32]
# Output Tensor Shape: [batch_size, 14, 14, 32]
pool1 = tf.layers.max_pooling2d(inputs=conv1, pool_size=[2, 2], strides=2)
# Convolutional Layer #2
# Computes 64 features using a 5x5 filter.
# Padding is added to preserve width and height.
# Input Tensor Shape: [batch_size, 14, 14, 32]
# Output Tensor Shape: [batch_size, 14, 14, 64]
conv2 = tf.layers.conv2d(
inputs=pool1,
filters=64,
kernel_size=[5, 5],
padding="same",
activation=tf.nn.relu)
# Pooling Layer #2
# Second max pooling layer with a 2x2 filter and stride of 2
# Input Tensor Shape: [batch_size, 14, 14, 64]
# Output Tensor Shape: [batch_size, 7, 7, 64]
pool2 = tf.layers.max_pooling2d(inputs=conv2, pool_size=[2, 2], strides=2)
# Flatten tensor into a batch of vectors
# Input Tensor Shape: [batch_size, 7, 7, 64]
# Output Tensor Shape: [batch_size, 7 * 7 * 64]
pool2_flat = tf.reshape(pool2, [-1, 7 * 7 * 64])
# Dense Layer
# Densely connected layer with 1024 neurons
# Input Tensor Shape: [batch_size, 7 * 7 * 64]
# Output Tensor Shape: [batch_size, 1024]
dense = tf.layers.dense(inputs=pool2_flat, units=1024, activation=tf.nn.relu)
# Add dropout operation; 0.6 probability that element will be kept
dropout = tf.layers.dropout(
inputs=dense, rate=0.4, training=mode == tf.estimator.ModeKeys.TRAIN)
# Logits layer
# Input Tensor Shape: [batch_size, 1024]
# Output Tensor Shape: [batch_size, 10]
logits = tf.layers.dense(inputs=dropout, units=10)
predictions = {
# Generate predictions (for PREDICT and EVAL mode)
"classes": tf.argmax(input=logits, axis=1),
# Add `softmax_tensor` to the graph. It is used for PREDICT and by the
# `logging_hook`.
"probabilities": tf.nn.softmax(logits, name="softmax_tensor")
}
if mode == tf.estimator.ModeKeys.PREDICT:
return tf.estimator.EstimatorSpec(mode=mode, predictions=predictions)
# Calculate Loss (for both TRAIN and EVAL modes)
loss = tf.losses.sparse_softmax_cross_entropy(labels=labels, logits=logits)
# Configure the Training Op (for TRAIN mode)
if mode == tf.estimator.ModeKeys.TRAIN:
with tf.name_scope('adam_optimizer'):
optimizer = tf.train.GradientDescentOptimizer(learning_rate=0.001)
train_op = optimizer.minimize(
loss=loss,
global_step=tf.train.get_global_step())
return tf.estimator.EstimatorSpec(mode=mode, loss=loss, train_op=train_op)
# Add evaluation metrics (for EVAL mode)
eval_metric_ops = {
"accuracy": tf.metrics.accuracy(
labels=labels, predictions=predictions["classes"])}
return tf.estimator.EstimatorSpec(
mode=mode, loss=loss, eval_metric_ops=eval_metric_ops)
def main(unused_argv):
# Load training and eval data
mnist = tf.contrib.learn.datasets.load_dataset("mnist")
train_data = mnist.train.images # Returns np.array
train_labels = np.asarray(mnist.train.labels, dtype=np.int32)
eval_data = mnist.test.images # Returns np.array
eval_labels = np.asarray(mnist.test.labels, dtype=np.int32)
# The graph_location changes are not related to LMS enablement
# but rather allow multiple users to run the example without
# having permission issues on temp directories.
graph_location = tempfile.mkdtemp()
print('Saving graph to: %s' % graph_location)
# Create the Estimator
mnist_classifier = tf.estimator.Estimator(
model_fn=cnn_model_fn, model_dir=graph_location)
# Set up logging for predictions
# Log the values in the "Softmax" tensor with label "probabilities"
tensors_to_log = {"probabilities": "softmax_tensor"}
logging_hook = tf.train.LoggingTensorHook(
tensors=tensors_to_log, every_n_iter=50)
# Train the model
train_input_fn = tf.estimator.inputs.numpy_input_fn(
x={"x": train_data},
y=train_labels,
batch_size=100,
num_epochs=None,
shuffle=True)
# Hook for Large Model Support
from tensorflow.contrib.lms import LMSHook
lms_hook = LMSHook({'adam_optimizer'}, lb=3, debug=True)
mnist_classifier.train(
input_fn=train_input_fn,
steps=20000,
hooks=[logging_hook, lms_hook])
# Evaluate the model and print results
eval_input_fn = tf.estimator.inputs.numpy_input_fn(
x={"x": eval_data},
y=eval_labels,
num_epochs=1,
shuffle=False)
eval_results = mnist_classifier.evaluate(input_fn=eval_input_fn)
print(eval_results)
if __name__ == "__main__":
tf.app.run()
