End-to-End Multimodal AI Pipeline Overview
Foundations
Learn the end-to-end multi-modal AI pipeline, including how each component fits together and what the project provides. You’ll also gain hands-on experience running and exploring the implementation using Anyscale or the GitHub repository.
About this course
Course Modules
Overview
Get an end-to-end overview of the multi-modal AI pipeline and how its components fit together. You’ll learn what the project provides and how to run and explore the implementation via Anyscale or the GitHub repository.
Multi-modal AI pipeline
Batch Inference
Learn how to use Ray Data to ingest a large image dataset from cloud storage, enrich each record with labels, and run scalable batch inference by generating CLIP embeddings with `map_batches`. You’ll see how to structure preprocessing and model execution for efficient, streaming, distributed batch pipelines.
- Batch inference
- Data ingestion
- Batch embeddings
Distributed Training
Learn how to scale image model training across multiple workers using Ray Train, including setting up the runtime, ingesting and preprocessing datasets with Ray Data, and converting classes to numeric labels. By the end, you’ll have a distributed-ready training pipeline with reusable preprocessing (including optional embedding computation) for efficient large-scale training.
- Distributed training
- Preprocess
- Model
Online Serving
Learn how to deploy a trained image classification model as a scalable online API using Ray Serve and FastAPI, including configuring GPU resources and replica scaling. You’ll also integrate MLflow to load the best model artifacts and send real-time prediction requests via an HTTP `/predict` endpoint.
- Online serving
- Deployments
- Application