ai-app/modelzoo

Live Machine Learning Model Zoo

ModelZoo.Live

The final layer in the prediction serving stack.

ModelZoo.Live consists of three layers - a frontend UI written in JavaScript, a Go Routing Middleware, which includes both a gRPC server, and a HTTP Reverse Proxy for gRPC, and a Python Client to aid in using the service programmatically. It is intended to be used with the prediction serving framework of your choice, such as RayServe or Clipper.

It can be used to share and keep track of ML models across teams within your organization, and to organize metrics about your models.

For more info, about ModelZoo.Live, as well as what the team hopes it can do for you, check out this talk. The corresponding slides can be found here.

Go

To seed the model database, cd into modelzoo/go, run go build and finally ./go seed --data filename, where filename is a JSON file containing your seed data. Please check this example to get an idea for how this file should look.

To run the Go Server, cd into modelzoo/go, run go build and finally ./go serve. To query this, you can use the ModelZooConnection client API, written in Python.

To run the HTTP Reverse Proxy, cd into modelzoo/go, run go build and finally, ./go proxy.

The reverse proxy can be queried with cURL. Some examples follow. Please note that http://modelzoo.url should be replaced with the URL for your proxy. For example, if you ran it locally on port 9090, you would replace http://modelzoo.url with http://localhost:9090.

List Models

curl -X POST -d '{"body": ""}' http://modelzoo.url/get/models

Get Token

curl -X POST -d '{"body": ""}' http://modelzoo.url/get/token

Create User

curl -X POST -d '{"email":"myemail", "password":"mypassword"}' http://modelzoo.url/create/user

Get User (To Check if User exists. Also used to authenticate with the python client)

curl -X POST -d '{"email":"myemail", "password":"mypassword"}' http://modelzoo.url/get/user

Text Inference

curl -X POST -d '{"text": {"access_token": "0ad62eb5-c10f-4f09-acb9-509ebf654489", \
    "metadata": {},"model_name": "text_generation_mock","texts": ["123456","654321"]}, \
    "type": "TEXT"}' http://modelzoo.url/inference