KServe (Model serving)

KServe provides a Kubernetes Custom Resource Definition for serving predictive and generative machine learning (ML) models. It aims to solve production model serving use cases by providing high abstraction interfaces for Tensorflow, XGBoost, ScikitLearn, PyTorch, Huggingface Transformer/LLM models using standardized data plane protocols.

It encapsulates the complexity of autoscaling, networking, health checking, and server configuration to bring cutting edge serving features like GPU Autoscaling, Scale to Zero, and Canary Rollouts to your ML deployments. It enables a simple, pluggable, and complete story for Production ML Serving including prediction, pre-processing, post-processing and explainability. KServe is being used across various organizations.

Installation

Dependencies

KServe does not need any other component to run per se since it's a standalone controller that leverages Kubernetes CRDs.

However if you're going to serve models that are stored in S3 you will need to naturally deploy S3 and configure its access.

Deploy using helmfile

Helmfile is prepared for kserve.

cd helmfile
helmfile sync -f helmfile.yaml

Configuration

You must configure access to S3 if that is where you're storing your models.

You can change the tags and repositories for serving runtime images, as well as security parameters.

You can enable optional components:

ModelMesh: platform for high-scale, high-density serving of machine learning models. It's designed to optimize model serving for scenarios with frequently changing models and high request volumes.
LocalModel is a feature in KServe that allows you to serve machine learning models directly from local storage within your Kubernetes cluster. This capability is particularly useful for quick deployment and testing of models without relying on external storage solutions.

Usage

KServe main function is making deployment of machine learning models easy. Provided everything is configured correctly, creating a simple Custom Resource InferenceService should be enough, this will instantiate an inference server serving your chosen model in a pod with the appropriate parameters.

We will outline several examples on how to deploy an inference service. All of them will have to be stored in S3.

Scikit-learn Model

Download assets

https://console.cloud.google.com/storage/browser/kfserving-examples/models/sklearn

With:

gsutil -m cp -r "gs://kfserving-examples/models/sklearn" models

And upload in your s3 bucket kfserving-examples/models/sklearn

Run

Create this manifest:

sklearn-sample.yaml
apiVersion: "serving.kserve.io/v1beta1"
kind: "InferenceService"
metadata:
  annotations:
    serving.kserve.io/deploymentMode: RawDeployment
  name: "sklearn-iris"
spec:
  predictor:
    serviceAccountName: serving-s3
    model:
      modelFormat:
        name: sklearn
      storageUri: "s3://kfserving-examples/models/sklearn/1.0/model"

kubectl apply -f sklearn-sample.yaml -n kosmos-data

Test your inference service

Alias sklearn-iris-kosmos-data.kserve.kosmos.athea to your ingress hostname or VIP or LB.

Prepare your input:

iris-input.json
{
    "instances": [
        [6.8,  2.8,  4.8,  1.4],
        [6.0,  3.4,  4.5,  1.6]
    ]
}

curl -v -H "Host: sklearn-iris-kosmos-data.kserve.kosmos.athea" -H "Content-Type: application/json" "http://sklearn-iris-kosmos-data.kserve.kosmos.athea:80/v1/models/sklearn-iris:predict" -d @./iris-input.json

You should see two predictions returned (i.e. {"predictions": [1, 1]})

TensorFlow

Download assets

https://console.cloud.google.com/storage/browser/kfserving-examples/models/tensorflow

With:

gsutil -m cp -r "gs://kfserving-examples/models/tensorflow" models

And upload in your s3 bucket kfserving-examples/models/tensorflow

Run

Create this manifest:

tensorflow-sample.yaml
apiVersion: "serving.kserve.io/v1beta1"
kind: "InferenceService"
metadata:
  annotations:
    serving.kserve.io/deploymentMode: RawDeployment
  name: "flower-sample"
spec:
  predictor:
    serviceAccountName: serving-s3
    model:
      modelFormat:
        name: tensorflow
      storageUri: "s3://kfserving-examples/models/tensorflow/flowers"

kubectl apply -f tensorflow-sample.yaml -n kosmos-data

Test your inference service

Alias flower-sample-kosmos-data.kserve.kosmos.athea to your ingress hostname or VIP or LB.

Prepare your input

flower-input.json
{  
      "instances":[  
        {  
            "image_bytes":{  
              "b64":"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"
            },
            "key":"   1"
        }
      ]
  }

curl -v -H "Host: flower-sample-kosmos-data.kserve.kosmos.athea" -H "Content-Type: application/json" "http://flower-sample-kosmos-data.kserve.kosmos.athea:80/v1/models/flower-sample:predict" -d @./flower-input.json

You should see:

"predictions": [
        {
            "scores": [0.999114931, 9.20987877e-05, 0.000136786344, 0.00033725836, 0.000300534, 1.84813962e-05],
            "prediction": 0,
            "key": "   1"
        }
    ]

ONNX

Download assets

https://console.cloud.google.com/storage/browser/kfserving-examples/models/onnx

With:

gsutil -m cp -r "gs://kfserving-examples/models/onnx" models

Remove triton folder and upload in your s3 bucket kfserving-examples/models/onnx

Run

Create your manifest:

onnx-sample.yaml
apiVersion: "serving.kserve.io/v1beta1"
kind: "InferenceService"
metadata:
  annotations:
    serving.kserve.io/deploymentMode: RawDeployment
  name: "style-sample"
spec:
  predictor:
    serviceAccountName: serving-s3
    model:
      protocolVersion: v2
      modelFormat:
        name: onnx
      storageUri: "s3://kfserving-examples/models/onnx"

kubectl apply -f onnx-sample.yaml -n kosmos-data

vLLM

vLLM has more restrictive hardware requirements, if your GPU cannot run vLLM, you can always switch your backend to huggingface using spec.model.args

kubectl apply -f vllm-sample.yaml -n kosmos-data

facebook/opt-125m
qwen/qwen2.5-1.5B

vllm-sample.yaml
apiVersion: "serving.kserve.io/v1beta1"
kind: InferenceService
metadata:
  annotations:
    serving.kserve.io/deploymentMode: RawDeployment
  name: facebook-opt-125
spec:
  predictor:
    serviceAccountName: serving-s3
    runtimeClassName: nvidia
    model:
      modelFormat:
        name: huggingface
      storageUri: "s3://kfserving-examples/models/huggingface/facebook"
      args:
        - --model_name=facebook/opt-125m
        - --model_dir=/mnt/models/opt-125m
        #- --backend=huggingface # for hf instead of vllm
        # - --enforce-eager
        # - --dtype=half
      resources:
        requests:
          cpu: '1'
          memory: 4Gi
          nvidia.com/gpu: '1'
        limits:
          cpu: '1'
          memory: 32Gi
          nvidia.com/gpu: '1'

Test:

curl http://facebook-opt-125-kosmos-data.kserve.kosmosml.wip/openai/v1/completions -H "Content-Type: application/json" -d '{
        "model": "facebook/opt-125m",
        "prompt": "San Francisco is a",
        "max_tokens": 7,
        "temperature": 0
    }'

vllm-sample-qwen.yaml
apiVersion: "serving.kserve.io/v1beta1"
kind: InferenceService
metadata:
  annotations:
    serving.kserve.io/deploymentMode: RawDeployment
  name: qwen
spec:
  predictor:
    serviceAccountName: serving-s3
    runtimeClassName: nvidia
    model:
      modelFormat:
        name: huggingface
      storageUri: "s3://kfserving-examples/models/huggingface/qwen"
      args:
        - --model_name=qwen/qwen2.5-1.5B
        - --model_dir=/mnt/models/qwen2.5-1.5B
        #- --backend=huggingface
        # - --enforce-eager
        # - --dtype=half
      resources:
        requests:
          cpu: '1'
          memory: 4Gi
          nvidia.com/gpu: '1'
        limits:
          cpu: '1'
          memory: 32Gi
          nvidia.com/gpu: '1'

Test:

curl http://qwen-kserve-kosmos-data.kserve.kosmos.athea/openai/v1/completions -H "Content-Type: application/json" -d '{
        "model": "qwen/qwen2.5-1.5B",
        "prompt": "San Francisco is a",
        "max_tokens": 7,
        "temperature": 0
    }'

Mlflow

In order to retrieve models from Mlflow model registry and serve them for inference. You just have to change the storageUri to support Mlflow schema:

"mlflow:/<model_name_in_registry>/<model_version>"

We first need to register models to our Mlflow model registry from inside the Kubernetes cluster:

First define some environment variables:

export AWS_ACCESS_KEY_ID="minio-access-key"
export AWS_SECRET_ACCESS_KEY="minio-secret-key"
export AWS_ENDPOINT_URL="http://minio.kosmos-s3"
export AWS_DEFAULT_REGION="us-east-1"
export MLFLOW_TRACKING_URI="http://mlflow.kosmos-data"
export MLFLOW_S3_ENDPOINT_URL="$AWS_ENDPOINT_URL"
export MLFLOW_TRACKING_USERNAME="admin"
export MLFLOW_TRACKING_PASSWORD="<mlflow_admin_password>"
export MLFLOW_TRACKING_INSECURE_TLS="True"
export MLFLOW_S3_IGNORE_TLS="True"

You can follow these two examples:

Scikit-Learn
Qwen

Let's retrieve sklearn model and push it to Mlflow using its python library

exp_name = "sklearn"
registered_name = "sklearn"

mlflow.set_experiment(exp_name)
mlflow.start_run()

# log the model to the run, this pushes it to s3
mlflowd.log_model("sklearn", artifact_path="model")
# push model to registry, this links the registry model and version to an s3 path
run_id = mlflow.active_run().info.run_id
mlflow.register_model(model_uri=f"runs:/{run_id}/model", name=registered_name)
mlflow.end_run()

You should use model format sklearn

You can find the manifest to deploy mlflow-sklearn.yaml

mlflow-sklearn.yaml
apiVersion: "serving.kserve.io/v1beta1"
kind: "InferenceService"
metadata:
  name: "mlflow-sklearn"
  namespace: kosmos-data
spec:
  predictor:
    model:
      modelFormat:
        name: sklearn
      storageUri: "mlflow:/sklearn/1"
      resources:
        requests:
          cpu: '1'
          memory: 100Mi
        limits:
          cpu: '1'
          memory: 1Gi

Then once it is deployed you can query it via python's requests like so:

cURL
Python

Prepare the input file input.json

{
  "instances": [
    [6.8,  2.8,  4.8,  1.4],
    [6.0,  3.4,  4.5,  1.6]           
  ]
}

curl -H "Content-Type: application/json" "http://mlflow-sklearn-kosmos-data.kserve.kosmos.athea/v1/models/model:predict" -d @./input.json 

import requests
import json

headers= {
  "Content-Type": "application/json"
}
# example of data to submit for prediction
sample_data = [
    [6.8,  2.8,  4.8,  1.4],
    [6.0,  3.4,  4.5,  1.6]
]

# convert it to string
data = json.dumps({ "instances": sample_data})

# submit a prediction
response = requests.post("http://mlflow-sklearn-kosmos-data.kserve.kosmos.athea/v1/models/model:predict", headers=headers, data=data)

predictions = response.json()["predictions"]
print(predictions)

Retrieve the model from S3 bucket kfserving-examples/models/huggingface/qwen/qwen2.5-1.5B/

Then use Mlflow to push it to registry as is:

exp_name = "qwen"
registered_name = "qwen"

mlflow.create_experiment(exp_name)
mlflow.set_experiment(exp_name)
mlflow.start_run()

# log the model to the run, this pushes it to s3, artifact will be pushed under the name "model"
mlflow.log_artifact("qwen2.5-1.5B", "model")
# push model to registry, this links the registry model and version to an s3 path
run_id = mlflow.active_run().info.run_id
mlflow.register_model(model_uri=f"runs:/{run_id}/model", name=registered_name)
mlflow.end_run()

You should use model format huggingface

You can find the manifest to deploy mlflow-qwen.yaml

mlflow-qwen.yaml
apiVersion: "serving.kserve.io/v1beta1"
kind: "InferenceService"
metadata:
  name: "mlflow-qwen"
  namespace: kosmos-data
spec:
  predictor:
    runtimeClassName: nvidia
    model:
      modelFormat:
        name: huggingface
      storageUri: "mlflow:/qwen/1"
      args:
      - "--backend=huggingface"
      resources:
        requests:
          cpu: '1'
          memory: 4Gi
        limits:
          cpu: '1'
          memory: 32Gi

Then once it is deployed you can query it via python's requests or curl like so:

cURL
Python

Prepare input in input.json

{
  "model": "mlflow-qwen",
  "prompt": "Paris est une ville",
  "stream": false,
  "max_tokens": 90
}

curl -k "http://mlflow-qwen-kosmos-data.kserve.kosmos.athea/openai/v1/completions" -H "Content-Type: application/json" -d @./input.json

It'll return something like this where the actual inference result is in .choices[0].text

{
  "id": "9d9197c3-1aba-4310-8cb5-555f5ad1ec18",
  "choices": [
    {
      "finish_reason": "length",
      "index": 0,
      "logprobs": null,
      "text": " de 1 000 000 d'habitants, la plus grande ville de France. Elle est située sur la rive gauche de la Seine, dans le département de la Seine-Saint-Denis, à 10 km au sud de Paris. Elle est la capitale de la région Île-de-France, la plus peuplée de France et la plus grande ville d'Europe. Elle"
    }
  ],
  "created": 1737968725,
  "model": "mlflow-qwen",
  "system_fingerprint": null,
  "object": "text_completion",
  "usage": {
    "completion_tokens": 90,
    "prompt_tokens": 4,
    "total_tokens": 94
  }
}

import requests

# Request data
data = {
    "model": "mlflow-qwen",
    "prompt": "Paris est une ville",
    "stream": False,
    "max_tokens": 90
}

# Headers
headers = {
    "Content-Type": "application/json"
}

# Make the request
url = "http://mlflow-qwen-kosmos-data.kserve.kosmos.athea/openai/v1/completions"
response = requests.post(
    url=url,
    headers=headers,
    json=data,  # requests will automatically handle JSON serialization
    verify=False  # equivalent to curl's -k flag
)

completion = response.json()['choices'][0]['text']
print(completion)

Installation​

Dependencies​

Deploy using helmfile​

Configuration​

Usage​

Scikit-learn Model​

Download assets​

Run​

Test your inference service​

TensorFlow​

Download assets​

Run​

Test your inference service​

ONNX​

Download assets​

Run​

vLLM​

Mlflow​

Installation

Dependencies

Deploy using helmfile

Configuration

Usage

Scikit-learn Model

Download assets

Run

Test your inference service

TensorFlow

Download assets

Run

Test your inference service

ONNX

Download assets

Run

vLLM

Mlflow