AI Inference Latency Monitoring: Metrics & Alerts Setup

Monitoring AI Inference Latency: Metrics & Alerts Setup

A user sends a request to an LLM and waits for the first token. If TTFT exceeds two seconds, UX drops sharply — customers leave. We, as MLOps engineers with experience in vLLM and TGI, set up latency monitoring turnkey: from metric collection to Grafana alerts. Contact us — we'll evaluate your project.

What Are P50, P95, P99 Latency and How to Interpret Them?

LLM inference latency is measured in percentiles. P50 is median latency, P95 means 95% of requests are faster than this value, P99 means 99%. For real-time services, P99 is critical; for batch, P95 is sufficient. Example: if P99 total latency > 30s, 1% of users experience unacceptable delay. Monitoring percentiles helps identify outliers and long-tail latency.

Problems We Solve

LLM inference latency consists of three components: Queuing time (time in the runtime queue), Prefill time (processing input context), and Decode time (token generation). Each requires its own metric and alert threshold. For example, long system prompts increase prefill time, requiring KV-cache caching. Comparison: vLLM with PagedAttention reduces decode latency up to 2x compared to naive implementation.

Why TTFT Monitoring Is Critical for LLM Services?

TTFT is the first sign of inference issues. If p50 TTFT > 1s, users massively leave the service. We configure alerts on p95 > 3s. Empirically: when TTFT > 5s, conversion drops by up to 40%.

How to Set Up Alerts on P99 Latency in Grafana?

We use Prometheus Alertmanager with rules based on histogram_quantile. Example alert:

- alert: LLMHighTTFT
  expr: histogram_quantile(0.95, rate(llm_time_to_first_token_seconds_bucket[5m])) > 3
  for: 5m
  annotations:
    summary: "TTFT p95 > 3 seconds"

- alert: LLMHighTotalLatency
  expr: histogram_quantile(0.99, rate(llm_total_latency_seconds_bucket[5m])) > 30
  for: 5m
  annotations:
    summary: "Total latency p99 > 30 seconds"

How We Do It: Tech Stack and Configs

The key tool is Prometheus histograms. Buckets are selected based on typical latency:

from prometheus_client import Histogram, Summary
import time

# Latency histograms
TTFT_HISTOGRAM = Histogram(
    "llm_time_to_first_token_seconds",
    "Time to first token",
    buckets=[0.1, 0.3, 0.5, 1.0, 2.0, 5.0, 10.0]
)

TOTAL_LATENCY = Histogram(
    "llm_total_latency_seconds",
    "Total request latency",
    labelnames=["model", "endpoint"],
    buckets=[0.5, 1.0, 2.0, 5.0, 10.0, 30.0, 60.0]
)

TPOT_HISTOGRAM = Histogram(
    "llm_time_per_output_token_ms",
    "Time per output token in milliseconds",
    buckets=[5, 10, 20, 50, 100, 200]
)

class LatencyTracker:
    def track_streaming_request(self, request_id: str, model: str):
        start = time.time()
        first_token_time = None

        def on_first_token():
            nonlocal first_token_time
            first_token_time = time.time()
            TTFT_HISTOGRAM.observe(first_token_time - start)

        def on_complete(total_tokens: int):
            end = time.time()
            total_latency = end - start
            TOTAL_LATENCY.labels(model=model, endpoint="/v1/chat").observe(total_latency)

            if first_token_time and total_tokens > 1:
                decode_time = end - first_token_time
                tpot_ms = (decode_time / (total_tokens - 1)) * 1000
                TPOT_HISTOGRAM.observe(tpot_ms)

        return on_first_token, on_complete

Additionally, we collect vLLM metrics: vllm:time_to_first_token_seconds, vllm:time_per_output_token_seconds, vllm:e2e_request_latency_seconds — they are already broken down by percentiles.

Metric Type Comparison

Runtime Comparison: vLLM vs TGI

What Is Included in the Work

• Documentation for all configured metrics and dashboards.
• Grafana dashboard code (JSON export).
• Deployment and alerting instructions.
• Team training: how to read dashboards and respond to alerts.
• 2 weeks of post-release support.

Estimated Timeline

From 5 to 15 working days, depending on infrastructure complexity and the number of models.

Common Mistakes

Inference Degradation Detection: Sliding Windows and Anomalies

A single latency spike is not a reason to panic. A sustained trend is a reason to act. We configure degradation detection using sliding windows:

• 7-day vs 30-day window: if the average p95 TTFT over a week grows by 30% relative to the month — automatic warning alert.
• Hourly anomalies: isolation forest on metrics per hour identifies abnormal periods (growth after a new model deployment, degradation when changing batch size).
• Correlation with GPU metrics: when TTFT increases, we check GPU utilization and memory. Increased latency + low GPU utilization = queue problem. Increased latency + high GPU memory = model doesn't fit.

Automated correlation of these metrics finds the root cause of an incident 5x faster than manual analysis.

We guarantee SLA: alert response time no more than 30 seconds. Our team's experience: 5+ years in MLOps, over 20 inference monitoring projects. Certified AWS and GCP specialists. Get a consultation — we'll evaluate your project. Order a monitoring audit — we'll prepare a plan.

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.