Coursera Flash Sale
40% Off Coursera Plus for 3 Months!
Grab it
Explore how Alibaba Cloud's AnalyticDB Ray enables high-performance, multi-modal AI pipelines directly within data warehouse environments in this 34-minute conference talk from Ray Summit 2025. Learn from Liang Lin and Fei Xue as they demonstrate how Ray's distributed compute capabilities integrate with cloud-native data warehouses to accelerate everything from ETL processes to large-scale inference and distributed fine-tuning. Discover three compelling real-world applications: optimizing advertising recommendation inference with heterogeneous compute that boosts GPU usage from 5% to 40% while achieving 2-3X faster data processing through dynamic storage scaling; accelerating LLM offline batch inference and data distillation using Ray Data and vLLM/SGLang to distill datasets from models like Qwen and DeepSeek, achieving 2-3X faster data loading and 50% performance gains; and implementing efficient distributed fine-tuning of multi-modal models through integration with Lance and Ray Data for processing large-scale image-text datasets, resulting in 3-5X faster distributed fine-tuning for Qwen-VL models. Understand how these use cases demonstrate the power of in-warehouse AI pipelines where multi-modal ETL and machine learning operations run side-by-side to create shorter paths from raw data to intelligent decision-making, offering complete end-to-end workflows from data labeling to training.
