I did this because of the request from Pytorch's team or the people from them(I am not sure), for which they wanted me to do a pull request example for them. It took me 2 days to rush it.

Generally, there are two ways(or more) to deploy it in the web. First is put everything in the browser, so you will download the model in the browser and you can run it offline. I like this way but you need to solve issue like running c++, python, or other language for javascript because bowser can only run javascript. I usually use this way because I like Reactjs. The downside is people can download the model that you train and build.

However, this repo uses the second way. That is running the model in the backend and running the Reactjs as the front end. 

https://github.com/JonathanSum/TorchServeReact