Azure Functions for AI Inference

Run machine learning inference on Azure Functions with custom containers, Durable Functions for complex pipelines, and Blob Storage model caching.

Azure Functions Hosting Plans

Python Function with ML Model

import azure.functions as func
import json
import onnxruntime as ort
from azure.storage.blob import BlobServiceClient

app = func.FunctionApp()

# Load model at module level for reuse
session = ort.InferenceSession("model/sentiment.onnx")

@app.route(route="predict", methods=["POST"])
def predict(req: func.HttpRequest) -> func.HttpResponse:
    body = req.get_json()
    inputs = {session.get_inputs()[0].name: body["tokens"]}

    result = session.run(None, inputs)

    return func.HttpResponse(
        json.dumps({"sentiment": result[0].tolist()}),
        mimetype="application/json"
    )

Durable Functions for ML Pipelines

Durable Functions enable orchestrating multi-step inference pipelines where you need to chain preprocessing, inference, and postprocessing as separate functions with built-in retry logic and state management.

Model Management with Blob Storage

Store model versions in Azure Blob Storage and download them during function initialization. Use the function's local temporary storage as a model cache to avoid re-downloading on warm invocations.