Deploying LLMs on Azure: OpenAI vs Machine Learning Endpoints

You developed a RAG application based on LLaMA-3-8B — now you need to serve it to hundreds of users. A local RTX 4090 handles debugging, but production requires a scalable endpoint with latency p99 <500 ms and autoscaling. Azure Machine Learning Managed Online Endpoints provide this capability — but proper configuration includes VNet integration, monitoring, and asynchronous deployment. We have deployed LLMs for 20+ companies, including a large fintech with strict data privacy requirements. A typical project: choosing between Azure OpenAI and Azure ML, configuring vLLM with PagedAttention, setting up RBAC and Private Endpoints. Infrastructure cost savings with irregular loads reach 50% compared to PAYG schemes.

Azure ML endpoint documentation

Problems we solve

Cold start and autoscaling. Without scale_settings configuration, the endpoint does not scale under sudden spikes. We set TargetUtilization, polling interval, and cooldown so that scaling from 1 to 8 instances takes <2 minutes without losing requests.

GPU memory management. OOM errors are a common issue when deploying LLaMA-3-70B. We use vLLM with PagedAttention and gpu_memory_utilization=0.90, as well as Tensor Parallelism across multiple GPUs.

Monitoring and alerting. Without collecting metrics (RequestsPerMinute, Latency P50/P99, GPU Utilization), you learn about problems only from users. We configure Azure Monitor + Application Insights with alert thresholds.

How to reduce latency p99?

For latency p99 <200 ms, we use vLLM with optimizations: max_num_batched_tokens=8192, --tensor-parallel-size 4 on A100. This yields throughput of 1500 tokens/sec for LLaMA-3-8B. In Azure OpenAI with PTU, latency p99 stays around 150 ms at a fixed TPM.

Why is autoscaling important?

Without autoscaling, you overpay for idle resources or lose users during spikes. We configure scale_settings: min_instances=1, max_instances=10 with target_utilization_percentage=70. The cost of these settings is zero — savings with irregular load reach up to 50%.

What is included in the work

• Requirements audit: load, latency SLA, compliance.
• Architecture design: service selection, region, GPU type (A100, V100), network isolation.
• Implementation: writing scoring script (vLLM or custom), configuring deployment configurations, CI/CD scripts.
• Load testing: measuring latency, throughput, identifying bottlenecks.
• Documentation: architecture description, operational instructions.
• Team training: workshop on monitoring and scaling.
• Support: one month after deployment.

How to choose: Azure OpenAI vs Azure ML Endpoints?

For production with GPT-4, we choose Azure OpenAI (SLA, PTU). For customization and open-source — Azure ML with vLLM.

Example GPU configurations for popular models

Process of work

1. Requirements analysis — load, latency SLA, budget, privacy requirements.
2. Infrastructure design — region selection, GPU type (A100, V100), network isolation.
3. Implementation — writing scoring script (vLLM or custom), configuring deployment configurations.
4. Load testing — measuring latency, throughput, identifying bottlenecks.
5. Deployment and monitoring — endpoint deployment, dashboard and alert configuration.

Timeline: from 2 to 4 weeks depending on complexity. Cost is calculated individually.

Example vLLM configuration for LLaMA-3-8B

model: meta-llama/Meta-Llama-3-8B-Instruct
tensor-parallel-size: 4
gpu-memory-utilization: 0.90
max-num-batched-tokens: 8192

Results and guarantees

• latency p99 <300 ms at batch size 1 for LLaMA-3-8B on A100.
• Autoscaling from 1 to 8 instances with custom rules.
• Savings up to 35% compared to Azure OpenAI PTU for high-load scenarios.
• 99.9% endpoint availability guarantee with proper configuration.

We guarantee delivery of all configurations, documentation, and training for your team. Support — one month after deployment.

How to order deployment?

Get a consultation: our engineers analyze your task and propose an architecture within one day. Contact us — we will deploy your LLM on Azure from scratch to production in 2–4 weeks. We hold Azure Solutions Architect certification and have 5+ years of MLOps experience.

MLOps: Infrastructure for Training, Deploying, and Monitoring ML Models

The model is trained, metrics — F1 0.94 on validation. Three months later in production, quality drops by 12%. No one knows when — there is no monitoring. It's impossible to retrain quickly — the training script is in a Jupyter notebook of a data scientist who has already left. Data for retraining is collected manually from three disparate systems. About half of the projects come to us with this pain. We build a turnkey MLOps platform: from experiment tracking to automatic deployment and data drift monitoring. We will assess your infrastructure in 1–2 weeks, and in 4–6 weeks you will get a basic MLOps core running in production. Our team has 10+ years of experience in ML infrastructure, over 50 implementations.

How does MLOps infrastructure benefit your ML projects?

Experiment Tracking and Reproducibility

Without tracking, an ML project turns into chaos: it's unclear which checkpoint is better, which hyperparameters were used, which dataset. Reproducing a result a month later is a quest.

Why is experiment tracking the foundation of reproducibility?

MLflow is an open source standard for tracking. It logs parameters, metrics, artifacts (models, graphs), and code. MLflow Model Registry is a centralized model storage with versioning and lifecycle stages (Staging → Production → Archived). Deployment via MLflow Serving or integration with external systems.

Typical initialization in code:

import mlflow

mlflow.set_experiment("fraud-detection-v2")
with mlflow.start_run():
    mlflow.log_params({"learning_rate": 3e-4, "batch_size": 64, "epochs": 10})
    mlflow.log_metric("val_f1", val_f1, step=epoch)
    mlflow.pytorch.log_model(model, "model")

This is the minimum. In production, we add logging of system metrics (GPU utilization, memory), dataset (hash, version), code (git commit hash). Weights & Biases — richer UI, collaboration features, sweep for hyperparameter optimization. MLflow — for on-premise deployment without external dependencies.

DVC (Data Version Control) — versioning of data and models on top of git. Data is stored in S3/GCS/Azure Blob, only metadata (hashes) in git. dvc repro reproduces the entire pipeline from raw data to metrics.

To ensure reproducibility of training, fix random seeds (torch.manual_seed, numpy.random.seed, random.seed) and record them in experiment metadata. Without this, debugging irregular results is painful. Log the dataset version (DVC hash) and git commit — then any experiment can be reproduced down to the byte.

Pipeline Orchestration: Kubeflow, Airflow, Prefect

A pipeline orchestrator becomes necessary when: A 100-line training script in cron is fine for simple tasks. But as soon as you have a multi-step pipeline (data loading → preprocessing → feature engineering → training → validation → deployment if quality above threshold), you need an orchestrator with retry logic, visualization, and alerts.

Kubeflow — Kubernetes-native orchestrator for ML (see Kubeflow). Each step is a Docker container. Supports parallel steps, conditional branches, artifacts between steps. Integrates with Katib (AutoML), KServe (serving), Feast (feature store).

Apache Airflow — more general DAG orchestrator. Wide ecosystem of operators (S3, Spark, DBT, Kubernetes). Easier to deploy if Airflow already exists in the company.

Prefect / Metaflow — less boilerplate. Prefect 2.x with @flow and @task decorators — quick start for small teams.

Typical training pipeline architecture on Kubeflow:

1. Data ingestion component — fetches data from S3/DB, validates schema via Great Expectations
2. Preprocessing component — transformations, normalization, train/val/test split
3. Training component — training on GPU, logging to MLflow
4. Evaluation component — metric calculation, comparison with baseline in Model Registry
5. Conditional deployment — deploy only if new model is better than current by >2% F1

Each component is a separate Docker image. Pipeline is versioned in git. Scheduled run (retraining once a week on new data) or manual.

Model Registry and Lifecycle Management

Model Registry is not just a checkpoint store. It is a centralized system that knows:

• Which model is currently in production (and with what metrics)
• History of all versions with training parameters
• Metadata: dataset, git commit, validation results
• Lifecycle stage: None → Staging → Production → Archived

MLflow Model Registry — standard. For enterprise — Vertex AI Model Registry (GCP), SageMaker Model Registry (AWS), Azure ML Model Registry.

Model promotion through stages: automatically move model to Staging after successful eval, then manual or automatic (during A/B test) promotion to Production. Rollback — switch to previous Production version in seconds.

Serving: From FastAPI to Triton Inference Server

Simple case. FastAPI + PyTorch/ONNX on one server — 80% of production ML deployments are exactly that. Sufficient for most tasks with load up to 100 req/s.

from fastapi import FastAPI
import onnxruntime as ort

app = FastAPI()
session = ort.InferenceSession("model.onnx", providers=["CUDAExecutionProvider"])

@app.post("/predict")
async def predict(request: PredictRequest):
    inputs = preprocess(request.text)
    outputs = session.run(None, {"input_ids": inputs})
    return {"label": postprocess(outputs)}

Triton Inference Server — production standard for high loads (500+ req/s). Dynamic batching, concurrent model execution, model ensemble. Supports TensorRT, ONNX, PyTorch TorchScript, TensorFlow SavedModel.

KServe — Kubernetes-native ML serving with autoscaling, canary deployments, A/B testing out of the box. Scale-to-zero for inactive models — savings on infrastructure up to 40% annually for a project with 10 models.

Monitoring: Data Drift, Model Drift, Infrastructure Metrics

Monitoring — what is usually done last and regretted first. Three levels.

Infrastructure monitoring. Latency (P50/P95/P99), throughput (req/s), error rate (4xx, 5xx), GPU/CPU utilization. Prometheus + Grafana — standard. Alert when P99 latency > threshold or error rate > 1%.

Data drift monitoring. Distribution of input data changes over time. Detect via PSI (Population Stability Index) for numerical features: PSI > 0.2 — strong drift. Chi-squared test for categorical, Kolmogorov-Smirnov test for continuous. Evidently AI — open source library with ready-made drift tests.

Model drift monitoring. If ground truth is delayed (e.g., we know conversion after a week) — monitor real metrics. If not — surrogate metrics: distribution of prediction scores, proportion of confident predictions.

Alerting. Three levels: INFO (minor drift, log it), WARNING (significant, notify team), CRITICAL (quality dropped below threshold — automatic switch to fallback model).

Why is data drift monitoring important?

Without it, you learn about model degradation only from user complaints or ringing SLA. A drift alert allows you to retrain the model in advance, before errors start causing losses. In one of our projects, PSI monitoring detected drift 2 days after a data source change — this saved the campaign.

Feature Store

Feature Store solves the training-serving skew problem. If preprocessing during training and inference is implemented in two different places — divergence is inevitable.

A Feature Store is needed when:

• Several models use the same features
• Features are computed from streaming data (real-time)
• Large team with different people on feature engineering and model training

Feast — open source Feature Store. Offline store (S3 + Parquet) for training, online store (Redis, DynamoDB) for low-latency inference. Feature definitions as code, materialization job syncs offline → online.

Tecton (commercial), Vertex AI Feature Store (GCP), SageMaker Feature Store (AWS) — managed options with less ops overhead.

CI/CD for ML

ML CI/CD is regular CI/CD plus specific ML steps.

ML-specific checks in CI:

• Reproducibility check: run training with a fixed seed, result must match
• Data validation: Great Expectations or Pandera on schema/distribution checks
• Model performance check: automatic eval on holdout, block merge if degradation > threshold
• Latency regression test: inference must meet SLA

GitOps for deployment. Merge to main → CI triggers training → eval → if passes → automatic deployment to Staging → smoke tests → manual promotion to Production or automatic upon successful canary.

Tools: GitHub Actions / GitLab CI for CI, ArgoCD for GitOps deployment on Kubernetes.

What's Included in MLOps Platform Development

We provide a full cycle of work, documentation, and team training.

Total time from audit to full MLOps platform: 3–5 months. Also possible phased launch: basic level (tracking + serving) in 4–6 weeks.

Cost is calculated individually based on data volume, number of models, and infrastructure requirements. Order an MLOps infrastructure audit — get a roadmap in 1–2 weeks. Contact us for a project assessment — we will send a preliminary estimate within 2 business days.

Note: warranty on architectural solutions — 12 months. We provide integration certificates with major cloud providers (AWS, GCP, Azure). During our work, we have not lost a single client after the first implementation — the experience of 50+ successful MLOps projects speaks for itself. Get a consultation on building an MLOps platform today.