Ask HN: How do you serve Torch models in production?

I have been using Tensorflow for my latest project. I have been very happy with it but there is a piece of the project in which it would be easiest to use Torch, for the pretrained models. Only thing I am stuck on is how people are serving Torch models in production. Tensorflow has a nice serving system, but I can't seem to find any recommendations on how to best serve Torch models in a low-latency application. Thanks!