Benchmark PyTorch and Flux.jl

Platform and Version Info

Ubuntu 14.04 with Tesla K40 GPU.

Tool	CUDA	Python	PyTorch	Julia	Flux.jl	CuArrays.jl
Version	8.0	3.5	0.4.1	1.0.1	0.6.7	0.8.1

Feed Forward Neural Net

mlp

	GPU Usage	Time	Accuracy
PyTorch	287 MB	89 s	0.978
Flux	303 MB	27 s	0.979

Convolutional Neural Net

convnet

	GPU Usage	Time	Accuracy
PyTorch	414 MB	92 s	0.992
Flux	1302 MB	76 s	0.992

Residual Neural Net

resnet

	GPU Usage	Time	Accuracy
PyTorch	769 MB	256 s	0.782
Flux