mxnet-deepmark
Deepmark on MXNet.
Feb 16, 2017, Note: Upstream stopped maintaining this repo.
This forked repo adds CPU specific changes, and new models for testing.
Quickstart
No dataset download is needed, all tests are done using dummy/random generated data. Please refer to benchmark.py for supported models.
# inference benchmark for mnist, resnet-20(cifar10), and resnet-50
./bmark.sh --network mnist,resnet-20,resnet-50
# training benchmark for mnist, resnet-20(cifar10), and resnet-50
./bmark.sh --network mnist,resnet-20,resnet-50 --is-train true
# inference benchmark for resnet-50 with different batch sizes
./bmark.sh --network resnet-50 --batch-size 1,8,64,128