Show HN: TFServe – Simple and easy HTTP Server for tensorflow models

github.com

4 pointsiitzco3 comments

I like how you made this ultra-simple. It makes it very easy to understand and use. There are times when a python+json server isn't performant enough. To help with this, my team recently built a protocol for efficient model serving called GraphPipe[1] that allows you to do serving like this. There is an example for serving from python[1]. The easiest way to get started is to use the go server in a docker container[2].

[0]: https://oracle.github.io/graphpipe/ [1]: https://github.com/oracle/graphpipe-tf-py/blob/master/exampl... [2]: https://oracle.github.io/graphpipe/#/guide/user-guide/quicks...

Thanks for your comment! I've heard about graphpipe and I will definitely check it out.

My idea was to provide a simple and easy way to deploy a simple (yet limited) server to consume the models through HTTP. Mainly considering that tensorflow Serving is not that accesible for people not familiarized with tensorflow.

Yes I totally understand. In general model serving doesn't seem to be something people are thinking about at the moment. I expect it gets more attention in the next couple of years.

Reply on news.ycombinator.com