GT-740M，3.0 plus nvidia GPU should be work, not tested
file are uploaded to baidu pan, because of file size limit of github and pypi
https://pan.baidu.com/s/1OEpY8zWqQhIGmZwHne4BAg?pwd=4qik
as 2G vga ram is too small for an normal project, just for fun, and learning deep learning toy data, enjoy
just find a pytorch-LMS can use ram for gpu，but not actively developed already, however, if used for study，maybe a good solution to let 'out of memory' go
see below:
Maybe try to compile it someday! https://github.com/mtbrandy/pytorch#binaries
koila https://github.com/rentruewang/koila
https://github.com/facebookresearch/fairscale/
It seems solution 2 is better, regardless of time using.