Running Pytorch master on travis-CI

I have to use Pytorch master for my project because I need to differentiate the torch.lgamma function, but it’s been hard getting travis-CI to use Pytorch master. So far I’ve tried compiling Pytorch (master) to a .egg file, then having travis wget that egg file and install it. Unfortunately, it seems like I’m always missing some dependencies on the travis system, even though when I compile pytorch locally, I turn off all the features I can (e.g., NO_CUDA=1, NO_MKL=1, …). Is there a way to build all the dependencies into the .egg file? Or is this the wrong way go about it? If so, what’s the easy way to get travis-CI to install pytorch master?


Can you add pytorch as your submodule?

Would pytorch master have to build on Travis then? I think it would time out. Travis cuts jobs off after 50 minutes and it runs much slower than a desktop computer.