Glossaire

Modèle Servir

Apprends l'essentiel du service de modèles - déploie des modèles d'IA pour des prédictions en temps réel, l'évolutivité et l'intégration transparente dans les applications.

Entraîne les modèles YOLO simplement
avec Ultralytics HUB

En savoir plus

Once a Machine Learning (ML) model is trained and validated, the next critical step is making it available to generate predictions on new data. This process is known as Model Serving. It involves deploying a trained model into a production environment, typically behind an API (Application Programming Interface) endpoint, allowing applications or other systems to request predictions in real-time. Model serving acts as the bridge between the developed model and its practical application, transforming it from a static file into an active, value-generating service within the broader Machine Learning Lifecycle.

Importance du service de modèles

Model serving is fundamental for operationalizing ML models. Without it, even the most accurate models, like state-of-the-art Ultralytics YOLO object detectors, remain isolated in development environments, unable to impact real-world processes. Effective model serving ensures that the insights and automation capabilities developed during training are accessible and usable. It enables real-time inference, allowing applications to respond dynamically to new data, which is crucial for tasks ranging from object detection in videos to natural language processing (NLP) in chatbots. Ultimately, model serving is essential for realizing the return on investment (ROI) of AI initiatives.

Model Serving vs. Model Deployment

While often used interchangeably, Model Serving is technically a specific component within the broader process of Model Deployment. Model deployment encompasses all steps needed to take a trained model and make it operational in a live production environment, including packaging, infrastructure setup, integration, and monitoring. Model Serving focuses specifically on the infrastructure and software layer that hosts the model and handles incoming prediction requests, making the model's functionality available as a service, often via network protocols like REST or gRPC. View our guide on Model Deployment Options for more details.

Applications dans le monde réel

Le service de modèle permet d'innombrables fonctionnalités pilotées par l'IA avec lesquelles nous interagissons quotidiennement. En voici deux exemples :

Composantes clés du service modèle

Implementing a robust model serving system involves several components working together:

Platforms like Ultralytics HUB aim to simplify this entire workflow, offering integrated solutions for training, versioning, deploying, and serving computer vision models, aligning with MLOps (Machine Learning Operations) best practices. Key considerations include scalability to handle load changes, security (Data Security), and maintainability.

Tout lire