kserve/python/custom_model/model.py

# Copyright 2021 The KServe Authors.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#    http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.

import argparse

from fastapi.middleware.cors import CORSMiddleware
from torchvision import models, transforms
from typing import Dict, Union
import torch
import time
from PIL import Image
import base64
import io
import numpy as np

from kserve import (
    Model,
    ModelServer,
    model_server,
    InferRequest,
    InferOutput,
    InferResponse,
    logging,
)
from kserve.errors import InvalidInput
from kserve.model_server import app
from kserve.utils.utils import generate_uuid


# This custom predictor example implements the custom model following KServe REST v1/v2 protocol,
# the input can be raw image base64 encoded bytes or image tensor which is pre-processed by transformer
# and then passed to the custom predictor, the output is the prediction response.
class AlexNetModel(Model):
    def __init__(self, name: str):
        super().__init__(name, return_response_headers=True)
        self.model = None
        self.ready = False
        self.load()

    def load(self):
        self.model = models.alexnet(pretrained=True, progress=False)
        self.model.eval()
        # The ready flag is used by model ready endpoint for readiness probes,
        # set to True when model is loaded successfully without exceptions.
        self.ready = True

    def preprocess(
        self, payload: Union[Dict, InferRequest], headers: Dict[str, str] = None
    ) -> torch.Tensor:
        raw_img_data = None
        if isinstance(payload, Dict) and "instances" in payload:
            headers["request-type"] = "v1"
            if "data" in payload["instances"][0]:
                # assume the data is already preprocessed in transformer
                np_array = np.asarray(payload["instances"][0]["data"])
                input_tensor = torch.Tensor(np_array)
                return input_tensor.unsqueeze(0)
            elif "image" in payload["instances"][0]:
                # Input follows the Tensorflow V1 HTTP API for binary values
                # https://www.tensorflow.org/tfx/serving/api_rest#encoding_binary_values
                img_data = payload["instances"][0]["image"]["b64"]
                raw_img_data = base64.b64decode(img_data)
        elif isinstance(payload, InferRequest):
            infer_input = payload.inputs[0]
            if infer_input.datatype == "BYTES":
                if payload.from_grpc:
                    raw_img_data = infer_input.data[0]
                else:
                    raw_img_data = base64.b64decode(infer_input.data[0])
            elif infer_input.datatype == "FP32":
                # assume the data is already preprocessed in transformer
                input_np = infer_input.as_numpy()
                return torch.Tensor(input_np)
        else:
            raise InvalidInput("invalid payload")

        input_image = Image.open(io.BytesIO(raw_img_data))
        preprocess = transforms.Compose(
            [
                transforms.Resize(256),
                transforms.CenterCrop(224),
                transforms.ToTensor(),
                transforms.Normalize(
                    mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225]
                ),
            ]
        )
        input_tensor = preprocess(input_image)
        return input_tensor.unsqueeze(0)

    def predict(
        self,
        input_tensor: torch.Tensor,
        headers: Dict[str, str] = None,
        response_headers: Dict[str, str] = None,
    ) -> Union[Dict, InferResponse]:
        start = time.time()

        output = self.model(input_tensor)
        torch.nn.functional.softmax(output, dim=1)
        values, top_5 = torch.topk(output, 5)
        result = values.flatten().tolist()
        response_id = generate_uuid()
        infer_output = InferOutput(
            name="output-0", shape=list(values.shape), datatype="FP32", data=result
        )
        infer_response = InferResponse(
            model_name=self.name, infer_outputs=[infer_output], response_id=response_id
        )
        end = time.time()

        # Example for custom response headers
        if response_headers is not None:
            response_headers.update(
                {"prediction-time-latency": f"{round((end - start) * 1000, 9)}"}
            )

        if "request-type" in headers and headers["request-type"] == "v1":
            return {"predictions": result}
        else:
            return infer_response


parser = argparse.ArgumentParser(parents=[model_server.parser])
args, _ = parser.parse_known_args()

if __name__ == "__main__":
    if args.configure_logging:
        logging.configure_logging(args.log_config_file)
    model = AlexNetModel(args.model_name)
    model.load()
    app.add_middleware(
        CORSMiddleware,
        allow_origins=["*"],
        allow_credentials=True,
        allow_methods=["*"],
        allow_headers=["*"],
    )
    ModelServer().start([model])