4 ms·
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 w
by vishvananda 8y ago
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/ https://oracle.github.io/graphpipe/
[1]: https://github.com/oracle/graphpipe-tf-py/blob/master/examples/model_server.py https://github.com/oracle/graphpipe-tf-py/blob/master/exampl...
[2]: https://oracle.github.io/graphpipe/#/guide/user-guide/quickstart https://oracle.github.io/graphpipe/#/guide/user-guide/quicks...
- iitzco 8y agoThanks 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.
- vishvananda 8y agoYes 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.