Created by: wangkuiyi

It is reasonable to serve documents, especially API documents, of each version of Paddle, like what's been doing with Tensorflow:

My rough idea about the auto packaging and auto deployment of documents is as follows:

1. When we build documents, we generate a Docker image containing the document and an HTTP server, e.g., Nginx.
2. We publish this image to Dockerhub.com as other images.
3. On paddlepaddle.org, we run a Nginx instance and Docker containers of document images of all Paddle versions, and make this Nginx instance http_proxy to all these Docker containers.