Clipper: A prediction serving system for TensorFlow, PyTorch, PySpark and others

github.com

13 pointskot-behemoth5 comments

How are you different from other products that provides OCR as a service, object detection as a service, etc?

I should say that this is not my project. Nor am I a contributor — I only found it recently in my search for deploying TensorFlow models.

In terms of how this is different to other products that provides OCR as a service, object detection as a service: this is not an "as a service" product, but a framework that allows you to deploy ML models in a variety of configurations, such as Docker/K8s. I'd even say it's even further than a framework, as it's opinionated about what goes into the serving cluster: not just model workers, but query manager/load balancer, metrics aggregator and a config DB.

It's a bit amusing how every good name has been taken, sometimes multiple times. Clipper was the name of an xBase compiler, an ill-fated RISC CPU by Fairchild that powered Integraph computers. Besides that, it's also the name of that ill-conceived crypto backdoor chip.

I was hoping this was about the Clipper Card SF Bay area transit card + big data.

And clipper the polygon offsetting library!

Reply on news.ycombinator.com