Benchmarking on CIFAR-10:

The following mentioned model definition files are under the folder: models/cifar10/ .


DLBENCH_ROOT="path to the root directory of this benchmark"


Run TensorFlow with its default MNIST setting:

cd $DLBENCH_ROOT/models/cifar10/tensorflow/
python > train_log.txt 2>&1

After the completion of training, run the following command to test the tranined model:

python > test_log.txt 2>&1

The Accuracy will appear after completion of And the Training Time and Testing Time can be extracted from the train_log.txt and test_log.txt.


Similarly, the NN network structure of Caffe is shown as follows:

Run Caffe with its default setting:

cd $DLBENCH_ROOT/models/cifar10/caffe
./ > log.txt 2>&1

The Training Time, Testing Time and Accuracy can be extracted from the log.txt file.


cd $DLBENCH_ROOT/models/cifar10/torch

Run on CPU:

th train-on-cifar-10.lua


Note: The implementation for Theano on CIFAR-10 derived from Reslab Theano tutorial (10 February 2015)

cd $DLBENCH_ROOT/models/cifar10/theano
THEANO_FLAGS=device=cpu python