Skip to main content

FastAI Sentiment

A custom predictor allows you to deploy your own prediction implementation, useful when the existing serving implementations don't fit your needs. If migrating from Cortex, the custom predictor work exactly the same way as PythonPredictor does in Cortex. Most PythonPredictors can be converted to custom predictor by copy pasting the code and renaming some variables.

The custom predictor is packaged as a Docker container. It is recommended, but not required, to keep large model files outside of the container image itself and to load them from a storage volume. This example follows that pattern. You will need somewhere to publish your Docker image once built. This example leverages Docker Hub, where storing public images are free and private images are cheap. Google Container Registry and other registries can also be used.

We're deploying a sentiment analyzer built with FastAI. It uses PyTorch under the hood, but working with the FastAI API is easier, which is why we wrap it instead of using a PyTorch predictor. The same principles can be applied to any framework.

Make sure you use a GPU enabled Docker image as a base, and that you enable GPU support when loading the model.

Tutorial source code​

To follow along with this tutorial, first clone the manifests from CoreWeave's kubernetes-cloud repository.

Get started​

After installing kubectl and adding your CoreWeave Cloud access credentials, clone the repository and folder for the tutorial source code.

This guide assumes use of a private container registry requiring authentication. Recommended solutions include using the Docker Registry application directly on CoreWeave Cloud, or else creating and using a Docker Hub account. Otherwise, another private container registry may work.

Once you have configured Docker Registry or created a Docker Hub account, create a new repository. For the rest of this guide, we'll assume that the name of the new repository is fastai-sentiment. No modifications are needed to any of these files to follow along.

Build the Docker image​

From the cloned source code directory, enter the custom-predictor directory. From here, build and push the Docker image.

 export DOCKER_USER=coreweave
 docker build -t $DOCKER_USER/fastai-sentiment:1
 docker push $DOCKER_USER/fastai-sentiment:1
important

The default Docker tag is latest. It is strongly discouraged to use this, as containers are cached on the nodes and in other parts of the CoreWeave stack. Once you have pushed to a tag, do not push to that tag again. Below, we use simple versioning by using tag 1 for the first iteration of the image.

Set up repository access​

Using kubectl create secret, create a new Secret using your login credentials, and name it docker-repository, or something that makes sense to you. This will be used by the nodes to pull your private image.

kubectl create secret docker-registry docker-hub --docker-server=https://index.docker.io/v1/ --docker-username=<your-name> --docker-password=<your-pword> --docker-email=<your-email>

Tell Kubernetes to use the newly created Secret by patching the ServiceAccount for your namespace to reference this Secret.

kubectl patch serviceaccounts default --patch "$(cat image-secrets-serviceaccount.patch.yaml)"

Download the model​

For performance reasons, the model is not bundled within the Docker image, so a Storage Volume is required, to which the pre-trained model will be downloaded. Storage Volumes are allocated using a Kubernetes PersistentVolumeClaim, or PVC. A simple container will also be deployed, which can be used to copy files to the newly created Volume.

From the custom-sentiment directory, use kubectl apply to deploy the PersistentVolumeClaim (model-storage-pvc.yaml).

$ kubectl apply -f model-storage-pvc.yaml
persistentvolumeclaim/model-storage created

Then, deploy the manifest for the sleep container (sleep-deployment.yaml).

$ kubectl apply -f sleep-deployment.yaml
deployment.apps/sleep created

The volume is mounted to /models inside the sleep container. Download the pre-trained model locally, then create a directory for it in the shared volume to upload it into. The name of the sleep Pod is assigned to a variable using kubectl. You can also get the name with kubectl get pods, as is done here:

curl -O https://bstore.coreweave.com/blobstore/models/fastai-sentiment/export.pkl

export SLEEP_POD=$(kubectl get pod -l "app.kubernetes.io/name=sleep" -o jsonpath='{.items[0].metadata.name}') # Export the value of the randomly-assigned variable to the $SLEEP_POD environment variable

kubectl exec -it $SLEEP_POD -- sh -c 'mkdir /models/sentiment' # Create a new directory inside the sleep pod at /models/senitment

kubectl cp ./export.pkl $SLEEP_POD:/models/sentiment/ # Copy the locally-downloaded model into that new directory

Optional – The model may also be downloaded from CoreWeave Object Storage or from Amazon S3. The Amazon S3 CLI utilities already exist in the sleep container to allow for this.

$ export SLEEP_POD=$(kubectl get pod -l "app.kubernetes.io/name=sleep" -o jsonpath='{.items[0].metadata.name}')
$ kubectl exec -it $SLEEP_POD -- sh
$# aws configure
$# mkdir /models/sentiment
$# aws s3 sync --recursive s3://mybucket /models/sentiment/

Deploy the model​

Modify the sentiment-inferenceservice.yaml to reference your own Docker image under .spec.predictor.containers.image.

spec:
  predictor:
    maxReplicas: 10
    minReplicas: 0
    containerConcurrency: 1
    containers:
    - name: kfserving-container
      image: coreweave/my-fastai-sentiment-container:1
      env:

Apply the resources using kubectl apply.

 $ kubectl apply -f sentiment-inferenceservice.yaml
 inferenceservice.serving.kubeflow.org/sentiment configured

List the Pods to ensure the Predictor has launched successfully. This may take a minute - wait for the READY column to indicate that all Pods are Running.

$ kubectl get pods

NAME                                                           READY   STATUS    RESTARTS   AGE
sentiment-predictor-default-px8xk-deployment-85bb6787d7-h42xk  2/2     Running   0          34s

If the predictor fails to init, check the logs for insights as to why. For example:

$ kubectl logs sentiment-predictor-default-px8xk-deployment-85bb6787d7-h42xk kfserving-container

Test the model​

Once all the Pods are running, the API endpoint for the model will be accessible as a URL provided in the InferenceService object. The API endpoints follow the Tensorflow V1 HTTP API.

$ kubectl get inferenceservices

NAME        URL                                                                             READY   DEFAULT TRAFFIC   CANARY TRAFFIC   AGE
sentiment   http://sentiment.tenant-test.knative.ord1.coreweave.cloud/v1/models/sentiment   True    100                                23h

The URL given in the output of kubectl get is the public API URL for your newly deployed model.

note

A HTTPS endpoint is also available, however the HTTPS endpoint bypasses any canary deployments. Retrieve it using kubectl get ksvc.

Run a test prediction by querying the acquired URL. Remember to add the :predict postfix.

 $ curl -XPOST -H @sample.json http://sentiment.tenant-test.knative.ord1.coreweave.cloud/v1/models/sentiment:predict
{"predictions": ["positive"]}

Clean up​

Delete the InferenceService. This will delete all the associated resources, except for your model storage and sleep Deployment.

$ kubectl delete inferenceservices sentiment
inferenceservice.serving.kubeflow.org "sentiment" deleted