RedisAI MLflow Demo

A Demo showing RedisAI-MLflow integration

This is a self contained demo built on Flask server that servers both frontend and backend. The server needs RedisAI instance running on the default port and on the same host. Follow the below steps for setting up and running the demo. But a detailed walk through is available at medium.com

Steps

• Run a RedisAI instance. You can follow the documentation for this but for quickly spinning up a RedisAI server, use the official docker image

docker run -p 6379:6379 redisai/redisai:latest

• Clone this repository and make a conda environment from the given yml file and activate it

conda env create -f env.yml
conda activate redisai_mlflow_demo

• Train the model and save it as torchscript model in MLflow. MLflow runs the train.py file which mock the training for the sake of this example and save the trained torchscript model.

mlflow run . --no-conda

• Deploy the trained model into RedisAI. For this, you need the Run ID from the previous step.

mlflow deployments create -t redisai -m runs:/<RUN ID PLACEHOLDER>/model --name gptmodel

• Now that MLflow has deployed the model in RedisAI, we can bring up our Flask application. It is a self contained application that serves both frontend and backend

cd server
flask run

• Go to the URL http://127.0.0.1:5000/ and you should see a UI similar to the one given below. Click on the Start button on top right and see the AI in action