Mlflow-TorchServe

A plugin that integrates TorchServe with MLflow pipeline. mlflow_torchserve enables mlflow users to deploy the mlflow pipeline models into TorchServe . Command line APIs of the plugin (also accessible through mlflow's python package) makes the deployment process seamless.

Prerequisites

Following are the list of packages which needs to be installed before running the TorchServe deployment plugin

1. torch-model-archiver
2. torchserve
3. mlflow

Installation

Plugin package which is available in pypi and can be installed using

pip install mlflow-torchserve

Installation from Source

Plugin package could also be installed from source using the following commands

python setup.py build
python setup.py install

What does it do

Installing this package uses python's entrypoint mechanism to register the plugin into MLflow's plugin registry. This registry will be invoked each time you launch MLflow script or command line argument.

Create deployment

The create command line argument and create_deployment python APIs does the deployment of a model built with MLflow to TorchServe.

CLI

mlflow deployments create -t torchserve -m <model uri> --name DEPLOYMENT_NAME -C 'MODEL_FILE=<model file path>' -C 'HANDLER=<handler file path>'

Python API

from mlflow.deployments import get_deploy_client
target_uri = 'torchserve'
plugin = get_deploy_client(target_uri)
plugin.create_deployment(name=<deployment name>, model_uri=<model uri>, config={"MODEL_FILE": <model file path>, "HANDLER": <handler file path>})

Update deployment

Update API can used to modify the configuration parameters such as number of workers, version etc., of an already deployed model. TorchServe will make sure the user experience is seamless while changing the model in a live environment.

CLI

mlflow deployments update -t torchserve --name <deployment name> -C "min-worker=<number of workers>"

Python API

plugin.update_deployment(name=<deployment name>, config={'min-worker': <number of workers>})

Delete deployment

Delete an existing deployment. Excepton will be raised if the model is not already deployed.

CLI

mlflow deployments delete -t torchserve --name <deployment name / version number>

Python API

plugin.delete_deployment(name=<deployment name / version number>)

List all deployments

Lists the names of all the models deployed on the configured TorchServe.

CLI

mlflow deployments list -t torchserve

Python API

plugin.list_deployments()

Get deployment details

Get API fetches the details of the deployed model. By default, Get API fetches all the versions of the deployed model.

CLI

mlflow deployments get -t torchserve --name <deployment name>

Python API

plugin.get_deployment(name=<deployment name>)

Run Prediction on deployed model

Predict API enables to run prediction on the deployed model.

For the prediction inputs, DataFrame, Tensor and Json formats are supported. The python API supports all of these three formats. When invoked via command line, one needs to pass the json file path that contains the inputs.

CLI

mlflow deployments predict -t torchserve --name <deployment name> --input-path <input file path> --output-path <output file path>

output-path is an optional parameter. Without output path parameter result will be printed in console.

Python API

plugin.predict(name=<deployment name>, df=<prediction input>)

Plugin help

Run the following command to get the plugin help string.

CLI

mlflow deployments help -t torchserve