Skip to content

Deploy Transformer with InferenceService

Transformer is an InferenceService component which does pre/post processing alongside with model inference. It usually takes raw input and transforms them to the input tensors model server expects. In this example we demonstrate an example of running inference with a custom Transformer communicating by REST and gRPC protocol.

Create Custom Image Transformer

Implement pre/post processing with KServe Model API

KServe.Model base class mainly defines three handlers preprocess, predict and postprocess, these handlers are executed in sequence where the output of the preprocess handler is passed to the predict handler as the input. When predictor_host is passed, the predict handler makes a call to the predictor and gets back a response which is then passed to the postprocess handler. KServe automatically fills in the predictor_host for Transformer and hands over the call to the Predictor. By default transformer makes a REST call to predictor, to make a gRPC call to predictor, you can pass the --protocol argument with value grpc-v2.

To implement a Transformer you can derive from the base Model class and then overwrite the preprocess and postprocess handler to have your own customized transformation logic. For Open(v2) Inference Protocol, KServe provides InferRequest and InferResponse API object for predict, preprocess, postprocess handlers to abstract away the implementation details of REST/gRPC decoding and encoding over the wire.

from kserve import Model, ModelServer, model_server, InferInput, InferRequest
from typing import Dict
from PIL import Image
import torchvision.transforms as transforms
import logging
import io
import base64

logging.basicConfig(level=kserve.constants.KSERVE_LOGLEVEL)

def image_transform(byte_array):
    """converts the input image of Bytes Array into Tensor
    Args:
        instance (dict): The request input for image bytes.
    Returns:
        list: Returns converted tensor as input for predict handler with v1/v2 inference protocol.
    """
    image_processing = transforms.Compose([
        transforms.ToTensor(),
        transforms.Normalize((0.1307,), (0.3081,))
    ])
    image = Image.open(io.BytesIO(byte_array))
    tensor = image_processing(image).numpy()
    return tensor

# for v1 REST predictor the preprocess handler converts to input image bytes to float tensor dict in v1 inference REST protocol format
class ImageTransformer(kserve.Model):
    def __init__(self, name: str, predictor_host: str, headers: Dict[str, str] = None):
        super().__init__(name)
        self.predictor_host = predictor_host
        self.ready = True

    def preprocess(self, inputs: Dict, headers: Dict[str, str] = None) -> Dict:
        return {'instances': [image_transform(instance) for instance in inputs['instances']]}

    def postprocess(self, inputs: Dict, headers: Dict[str, str] = None) -> Dict:
        return inputs

# for v2 gRPC predictor the preprocess handler converts the input image bytes tensor to float tensor in v2 inference protocol format
class ImageTransformer(kserve.Model):
    def __init__(self, name: str, predictor_host: str, protocol: str, headers: Dict[str, str] = None):
        super().__init__(name)
        self.predictor_host = predictor_host
        self.protocol = protocol
        self.ready = True

    def preprocess(self, request: InferRequest, headers: Dict[str, str] = None) -> InferRequest:
        input_tensors = [image_transform(instance) for instance in request.inputs[0].data]
        input_tensors = np.asarray(input_tensors)
        infer_inputs = [InferInput(name="INPUT__0", datatype='FP32', shape=list(input_tensors.shape),
                                   data=input_tensors)]
        infer_request = InferRequest(model_name=self.model_name, infer_inputs=infer_inputs)
        return infer_request

Please see the code example here.

Transformer Server Entrypoint

For single model you just create a transformer object and register that to the model server.

if __name__ == "__main__":
    model = ImageTransformer(args.model_name, predictor_host=args.predictor_host,
                             protocol=args.protocol)
    ModelServer().start(models=[model])

For multi-model case if all the models can share the same transformer you can register the same transformer for different models, or different transformers if each model requires its own transformation.

if __name__ == "__main__":
    for model_name in model_names:
        transformer = ImageTransformer(model_name, predictor_host=args.predictor_host)
        models.append(transformer)
    kserve.ModelServer().start(models=models)

Build Transformer docker image

Under kserve/python directory, build the transformer docker image using Dockerfile

cd python
docker build -t $DOCKER_USER/image-transformer:latest -f transformer.Dockerfile .

docker push {username}/image-transformer:latest

Deploy the InferenceService with REST Predictor

Create the InferenceService

By default InferenceService uses TorchServe to serve the PyTorch models and the models can be loaded from a model repository in cloud storage according to TorchServe model repository layout. In this example, the model repository contains a MNIST model, but you can store more than one model there.

apiVersion: serving.kserve.io/v1beta1
kind: InferenceService
metadata:
  name: torch-transformer
spec:
  predictor:
    model:
      modelFormat:
        name: pytorch
      storageUri: gs://kfserving-examples/models/torchserve/image_classifier/v1
  transformer:
    containers:
      - image: kserve/image-transformer:latest
        name: kserve-container
        command:
          - "python"
          - "-m"
          - "model"
        args:
          - --model_name
          - mnist
apiVersion: serving.kserve.io/v1beta1
kind: InferenceService
metadata:
  name: torch-transformer
spec:
  predictor:
    pytorch:
      storageUri: gs://kfserving-examples/models/torchserve/image_classifier/v1
  transformer:
    containers:
      - image: kserve/image-transformer:latest
        name: kserve-container
        command:
          - "python"
          - "-m"
          - "model"
        args:
          - --model_name
          - mnist

Note

STORAGE_URI is a build-in environment variable used to inject the storage initializer for custom container just like StorageURI field for prepackaged predictors.

The downloaded artifacts are stored under /mnt/models.

Apply the InferenceService transformer-new.yaml

kubectl apply -f transformer-new.yaml

Expected Output

$ inferenceservice.serving.kserve.io/torch-transformer created

Run a prediction

First, download the request input payload.

Then, determine the ingress IP and ports and set INGRESS_HOST and INGRESS_PORT.

SERVICE_NAME=torch-transformer
MODEL_NAME=mnist
INPUT_PATH=@./input.json
SERVICE_HOSTNAME=$(kubectl get inferenceservice $SERVICE_NAME -o jsonpath='{.status.url}' | cut -d "/" -f 3)

curl -v -H "Host: ${SERVICE_HOSTNAME}" -H "Content-Type: application/json" -d $INPUT_PATH http://${INGRESS_HOST}:${INGRESS_PORT}/v1/models/$MODEL_NAME:predict

Expected Output

> POST /v1/models/mnist:predict HTTP/1.1
> Host: torch-transformer.default.example.com
> User-Agent: curl/7.73.0
> Accept: */*
> Content-Length: 401
> Content-Type: application/x-www-form-urlencoded
>
* upload completely sent off: 401 out of 401 bytes
Handling connection for 8080
* Mark bundle as not supporting multiuse
< HTTP/1.1 200 OK
< content-length: 20
< content-type: application/json; charset=UTF-8
< date: Tue, 12 Jan 2021 09:52:30 GMT
< server: istio-envoy
< x-envoy-upstream-service-time: 83
<
* Connection #0 to host localhost left intact
{"predictions": [2]}

Deploy the InferenceService calling Predictor with gRPC protocol

Comparing with REST, gRPC is faster due to the tight packing of the Protocol Buffer and the use of HTTP/2 by gRPC. In many cases, gRPC can be more efficient communication protocol between Transformer and Predictor as you may need to transmit large tensors between them.

Create InferenceService

Create the InferenceService with following yaml which includes a Transformer and a Triton Predictor. As KServe by default uses TorchServe serving runtime for PyTorch model, here you need to override the serving runtime to kserve-tritonserver for using the gRPC protocol. The transformer calls out to predictor with V2 gRPC Protocol by specifying the --protocol argument.

apiVersion: serving.kserve.io/v1beta1
kind: InferenceService
metadata:
  name: torch-grpc-transformer
spec:
  predictor:
    model:
      modelFormat: 
        name: pytorch
      storageUri: gs://kfserving-examples/models/torchscript
      runtime: kserve-tritonserver
      runtimeVersion: 20.10-py3
      ports:
      - name: h2c
        protocol: TCP
        containerPort: 9000
  transformer:
    containers:
    - image: kserve/image-transformer:latest
      name: kserve-container
      command:
      - "python"
      - "-m"
      - "model"
      args:
      - --model_name
      - cifar10
      - --protocol
      - grpc-v2
apiVersion: serving.kserve.io/v1beta1
kind: InferenceService
metadata:
name: torch-grpc-transformer
spec:
  predictor:
    triton:
      storageUri: gs://kfserving-examples/models/torchscript
      runtimeVersion: 20.10-py3
      ports:
      - name: h2c
        protocol: TCP
        containerPort: 9000
  transformer:
    containers:
    - image: kserve/image-transformer:latest
      name: kserve-container
      command:
      - "python"
      - "-m"
      - "model"
      args:
      - --model_name
      - cifar10
      - --protocol
      - grpc-v2

Apply the InferenceService grpc_transformer.yaml

kubectl apply -f grpc_transformer.yaml

Expected Output

$ inferenceservice.serving.kserve.io/torch-grpc-transformer created

Run a prediction

First, download the request input payload.

Then, determine the ingress IP and ports and set INGRESS_HOST and INGRESS_PORT

SERVICE_NAME=torch-grpc-transformer
MODEL_NAME=cifar10
INPUT_PATH=@./image.json
SERVICE_HOSTNAME=$(kubectl get inferenceservice $SERVICE_NAME -o jsonpath='{.status.url}' | cut -d "/" -f 3)

curl -v -H "Host: ${SERVICE_HOSTNAME}" -H "Content-Type: application/json" -d $INPUT_PATH http://${INGRESS_HOST}:${INGRESS_PORT}/v1/models/$MODEL_NAME:predict

Expected Output

*   Trying ::1...
* TCP_NODELAY set
* Connected to localhost (::1) port 8080 (#0)
> POST /v1/models/cifar10:predict HTTP/1.1
> Host: torch-transformer.default.example.com
> User-Agent: curl/7.64.1
> Accept: */*
> Content-Length: 3394
> Content-Type: application/x-www-form-urlencoded
> Expect: 100-continue
>
Handling connection for 8080
< HTTP/1.1 100 Continue
* We are completely uploaded and fine
< HTTP/1.1 200 OK
< content-length: 222
< content-type: application/json; charset=UTF-8
< date: Thu, 03 Feb 2022 01:50:07 GMT
< server: istio-envoy
< x-envoy-upstream-service-time: 73
<
* Connection #0 to host localhost left intact
{"predictions": [[-1.192867636680603, -0.35750141739845276, -2.3665435314178467, 3.9186441898345947, -2.0592284202575684, 4.091977119445801, 0.1266237050294876, -1.8284690380096436, 2.628898859024048, -4.255198001861572]]}* Closing connection 0

Performance Comparison between gRPC and REST

From the following latency stats of both transformer and predictor you can see that the transformer to predictor call takes longer time(92ms vs 55ms) for REST than gRPC, REST takes more time serializing and deserializing 3*32*32 shape tensor and with gRPC it is transmitted as tightly packed numpy array serialized bytes.

# from REST v1 transformer log
2023-01-09 07:15:55.263 79476 root INFO [__call__():128] requestId: N.A., preprocess_ms: 6.083965302, explain_ms: 0, predict_ms: 92.653036118, postprocess_ms: 0.007867813
# from REST v1 predictor log
2023-01-09 07:16:02.581 79402 root INFO [__call__():128] requestId: N.A., preprocess_ms: 13.532876968, explain_ms: 0, predict_ms: 48.450231552, postprocess_ms: 0.006914139
# from REST v1 transformer log
2023-01-09 07:27:52.172 79715 root INFO [__call__():128] requestId: N.A., preprocess_ms: 2.567052841, explain_ms: 0, predict_ms: 55.0532341, postprocess_ms: 0.101804733
# from gPPC v2 predictor log
2023-01-09 07:27:52.171 79711 root INFO [__call__():128] requestId: , preprocess_ms: 0.067949295, explain_ms: 0, predict_ms: 51.237106323, postprocess_ms: 0.049114227
Back to top