PyTorch Windows version

Windows specific changes:

Bug fixes

1. Errors in backward leads to deadlock
2. Memory leak in multiprocessing using DataLoader
3. An indention bug in torch.cuda

Features

1. Add support for new CUDA and cuDNN versions
2. Add support for Ninja and clcache when compiling

Known issues

1. Some tests will fail
2. There is no support for torch.distributed, nccl and magma
3. No support for python versions other than 3.5 and 3.6
4. Don't set num_worker to a number more than 1. ( 0 is safer ) Wrap your main code entry with an if statement like this.

if __name__ == '__main__':

The packages are accessible here.