What problem does it solve?
Reduces the complexity of publishing, operating, and debugging production inference APIs on Domino Data Lab by consolidating deployment steps, monitoring guidance, and troubleshooting practices into a single, actionable reference.
Core Features & Use Cases
- Publish Prediction Endpoints: Guidance to create model functions, register models, and publish REST prediction endpoints via Domino UI or API.
- Monitoring & Alerting: Instructions to access Domino's Grafana dashboards, key Prometheus queries for latency and error rates, and example alert rules for error rate and high latency.
- Scaling & GPU Inference: Recommendations for replica autoscaling, resource sizing, and NVIDIA Triton integration for high-performance GPU inference.
- Use Case: MLOps engineers deploying a classification model can publish a model API, validate performance with Grafana, configure alerts for P99 latency, and perform blue/green version rollouts.
Quick Start
Deploy a Domino model API named my-classifier from model.py using the predict function and ensure the default environment includes uwsgi.