thomelane commented on a change in pull request #15427: [TUTORIAL] Gluon performance tips and tricks URL: https://github.com/apache/incubator-mxnet/pull/15427#discussion_r299712376
########## File path: docs/tutorials/gluon/performance.md ########## @@ -0,0 +1,485 @@ +<!--- Licensed to the Apache Software Foundation (ASF) under one --> +<!--- or more contributor license agreements. See the NOTICE file --> +<!--- distributed with this work for additional information --> +<!--- regarding copyright ownership. The ASF licenses this file --> +<!--- to you 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. --> + +# Gluon Performance Tips & Tricks + +Compared to traditional machine learning methods, the field of deep-learning has increased model accuracy across a wide range of tasks, but it has also increased the amount of computation required for model training and inference. Specialised hardware chips, such as GPUs and FPGAs, can speed up the execution of networks, but it can sometimes be hard to write code that uses the hardware to its full potential. We will be looking at a few simple tips and trick in this tutorial that you can use to speed up training and ultimately save on training costs. + +We'll start by writing some code to train an image classification network for the CIFAR-10 dataset, and then benchmark the throughput of the network in terms of samples processed per second. After some performance analysis, we'll identify the bottlenecks (i.e. the components limiting throughput) and improve the training speed step-by-step. We'll bring together all the tips and tricks at the end and evaluate our performance gains. + + +```python +from __future__ import print_function +import multiprocessing +import time +import mxnet as mx +import numpy as np +from PIL import Image Review comment: Great catch! Added that to perform a slow rotation augmentation (that isn't an MXNet transform), but changed to a `sleep` instead. Switched out for MXNet function, removed dependency, and re-ran the notebook. ---------------------------------------------------------------- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: [email protected] With regards, Apache Git Services
