A plugin that integrates RedisAI with MLflow pipeline. mlflow_redisai enables you to
use mlflow to deploy the models built and trained in mlflow pipeline into RedisAI without any
extra effort from the user. This plugin provides few command line APIs, which is also accessible
through mlflow's python package, to make the deployment process seamless.
For installing and activating the plugin, you only need to install this package which is available in pypi and can be installed with
pip install mlflow_redisaiInstalling this package uses python's amazing 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.
This plugin allows you to interact with RedisAI deployment through MLflow using below given options. All of these options are accessible through command line and python API, although the predict command line option is not reliable in the current release. if you are connecting to a non-local RedisAI instance or if your RedisAI instance needs non-default connection parameters such as username or password, take a look at the connection parameters section
Deploy the model to RedisAI. The create command line argument and create_deployment python
APIs does the deployment of a model built with MLflow to RedisAI. It fetches the information, such as
which framework the model is built on, from the model configuration file implicitly.
mlflow deployments create -t redisai --name <rediskey> -m <model uri> -C <config option>from mlflow.deployments import get_deploy_client
target_uri = 'redisai' # host = localhost, port = 6379
redisai = get_deploy_client(target_uri)
redisai.create_deployment(rediskey, model_uri, config={'device': 'GPU'})Update deployment API has a very similar signature to the create API. In face, update deployment does exactly the same operation as create API except that the model should already be deployed. If the model is not present already, it raises an exception. Update API can be used to update a new model after retraining. This type of setup is useful if you want to change the device on which the inference is running or if you want to change the autobatching size, or even if you are doing live training and updating the model on the fly . RedisAI will make sure the user experience is seamless while changing the model in a live environment.
mlflow deployments update -t redisai --name <rediskey> -m <model uri> -C <config option>redisai.update_deployment(rediskey, model_uri, config={'device': 'GPU'})Delete an existing deployment. Error will be thrown if the model is not already deployed
mlflow deployments delete -t redisai --name <rediskey>redisai.delete_deployment(rediskey)List the names of all the deployments. This name can then be used in other APIs or can be used in the get deployment API to get more details about a particular deployment. Currently, it displays every deployment, not just the deployment made through this plugin
mlflow deployments list -t redisairedisai.list_deployments()Get API fetches the meta data about a deployment from RedisAI. This metadata includes
- BACKEND : the backend used by the model as a String
- DEVICE : the device used to execute the model as a String
- TAG : the model's tag as a String
- BATCHSIZE : The maximum size of any batch of incoming requests. If BATCHSIZE is equal to 0 each incoming request is served immediately. When BATCHSIZE is greater than 0, the engine will batch incoming requests from multiple clients that use the model with input tensors of the same shape.
- MINBATCHSIZE : The minimum size of any batch of incoming requests.
- INPUTS : array reply with one or more names of the model's input nodes (applicable only for TensorFlow models)
- OUTPUTS : array reply with one or more names of the model's output nodes (applicable only for TensorFlow models)
mlflow deployments get -t redisai --name <rediskey>redisai.get_deployment(rediskey)MLflow integration also made a handy help API which is specific to any plugins you install rather than the help string of the mlflow command itself. For quickly checking something out, it'd be easy to use the help API rather than looking out for this document.
PS: Since help is specific to the plugin and not really an attribute of the client object itself,
it's not available under client object (redisai variable in the above examples)
mlflow deployments help -t redisaiIf you are new to RedisAI and trying it out for the first time, you might not know the setup already
(although it's quite easy to setup a local RedisAI instance). The local-run API is there to help
you, if that's the case. It pulls the latest docker image of RedisAI (yes, you need docker installed in your
machine for this to work), run it on the default port, deploy the model you specified.
PS: It's IMPORTANT to note that this API leaves the docker container running. You would need to manually stop the container once you are done with experimentation. Also, remember that if you trying to run this API twice without killing the first container, it throws an error saying the port is already in use.
mlflow deployments run-local -t redisai --name <rediskey> -m <model uri> -C <config option>For connecting to a RedisAI instance with non-default connection parameters, you'd either need to
supply the necessary (and allowed) parameters through environmental variables or modify the target
URI that you'd pass through the command line using -t or -target option.
Currently this plugin allows five options through environmental variables which are given below (without description because they are self explanatory)
- REDIS_HOST
- REDIS_PORT
- REDIS_DB
- REDIS_USERNAME
- REDIS_PASSWORD
A template for quick verification is given below
redisai:/[[username]:[password]]@[host]:[port]/[db]
where the default value each would be
- username = None
- password = None
- host = localhost
- port = 6379
- db = 0