Floating point exception in benchmark_score.py with mxnet-mkl


I’m trying to run the image-classification benchmark_score.py script using mxnet-mkl and getting a Floating point exception. This appears to happen while benchmarking alexnet for batch size 2. I have successfully run the benchmark for “base” mxnet (pip install, version 1.0.0).

Here’s the output from benchmark_score.py
$ python benchmark_score.py
INFO:root:network: alexnet
INFO:root:device: cpu(0)
/usr/local/lib/python2.7/dist-packages/mxnet/module/base_module.py:65: UserWarning: Data provided by label_shapes don’t match names specified by label_names ([] vs. [‘softmax_label’])
MKL Build:20171227
INFO:root:batch size 1, image/sec: 7.711693
Floating point exception (core dumped)

I’ve installed mxnet-mkl using pip; the version it gets is “mxnet-mkl (1.0.0.post4)”.

I’ve downloaded and installed mkl-dnn per instructions at https://software.intel.com/en-us/articles/intel-mkl-dnn-part-1-library-overview-and-installation including “make test” which passed, suggesting the library is working.