domino-model-endpoints

Domino Model Endpoints Skill

This skill provides comprehensive knowledge for deploying and monitoring model API endpoints in Domino Data Lab.

Key Concepts

Model Endpoints Overview

Domino Model Endpoints provide:

• REST API for model predictions
• Automatic scaling and load balancing
• Version management
• Built-in monitoring with Grafana
• Authentication via API tokens

Endpoint Lifecycle

Train Model → Register → Deploy Endpoint → Monitor → Update Version

Related Documentation

• DEPLOY-ENDPOINT.md - Creating model APIs
• MONITORING.md - Grafana, metrics, alerts
• SCALING.md - GPU inference, Triton, scaling

Environment Requirements

Important: Model APIs use the default environment set for your project. The environment must have the uwsgi Python package installed for model endpoints to work.

Required Package

# Add to your environment's Dockerfile instructions
RUN pip install uwsgi

Or in requirements.txt:

uwsgi

Setting Default Environment

1. Go to Project Settings → Execution Preferences
2. Set the Default Environment that includes uwsgi
3. This environment will be used for all Model API deployments

Quick Start

1. Create Endpoint Function

# model.py
def predict(features):
    """
    Domino calls this function for predictions.

    Args:
        features: Input data (dict, list, or primitive)

    Returns:
        JSON-serializable prediction result
    """
    import pickle

    # Load model (cached after first call)
    with open('model.pkl', 'rb') as f:
        model = pickle.load(f)

    prediction = model.predict([features])
    return {"prediction": prediction.tolist()}

2. Deploy via Domino UI

1. Go to Publish → Model APIs
2. Click New Model
3. Configure:
   • Name: my-classifier
   • File: model.py
   • Function: predict
   • Environment: Select compute environment
4. Click Publish