Hi there, I’ve done a benchmark for MXNet and Pytorch with cifar10 dataset. And it looks like MXNet is over 1.5 times faster than Pytorch. Congrats to all our community and thanks for all contributions so far.
Thank you @mouryarishik four you insights.
I’m interested how the inference speed using different batch-sizes (e.g. 1, 64) compares between the frameworks on your setup, too.
I’ve found similar benchmarks which confirm that training speed can be faster on MXNET:
I’d like to point out that the following benchmark is actually untrustworthy:
The user has taken implementations found from different sources, without checking that the code is comparable or not.