Anyscale Courses

Agentic Workload Inference at Scale - ByteDance's AIBrix and DeerFlow

Discover ByteDance's AIBrix & DeerFlow: open-source infrastructure for scalable LLM inference with smart autoscaling, KVCache management, and agentic workload support.

• YouTube
• 31 minutes
• Self-Paced
• Free Video

Pinterest's Approach to Real-Time ML Experimentation Using Ray

Discover how Pinterest transformed ML experimentation from weeks to days using Ray's real-time streaming pipeline, achieving 10x faster model updates and saving hundreds of thousands in costs.

• YouTube
• 30 minutes
• Self-Paced
• Free Video

How Prime Intellect Builds Scalable Infrastructure for Agentic RL

Discover how Prime Intellect architects distributed reinforcement learning infrastructure at scale, featuring async-first trainers, fault-tolerant execution, and multi-cloud compute platforms.

• YouTube
• 30 minutes
• Self-Paced
• Free Video

How Workday Achieved 50x Cheaper Model Serving with Ray Serve

Discover how Workday rebuilt their ML model-serving architecture with Ray Serve, achieving 50x cost savings while scaling to tens of thousands of models across multiple environments.

• YouTube
• 29 minutes
• Self-Paced
• Free Video

Scaling LLM Post-Training at Character.AI

Discover how Character.AI scales LLM post-training for millions using Ray ecosystem, Rayman platform, and open-source RL libraries for AI entertainment at global scale.

• YouTube
• 32 minutes
• Self-Paced
• Free Video

MLOps with Ray on Anyscale

Discover how to streamline ML workflows using Ray on Anyscale for scalable data processing, distributed training, hyperparameter tuning, and production deployment.

• YouTube
• 21 minutes
• Self-Paced
• Free Video

How to Get Started with Distributed Training at Scale

Master distributed training strategies for scaling deep learning models using data, model, and pipeline parallelism with PyTorch and Ray to overcome bottlenecks and system failures.

• YouTube
• 31 minutes
• Self-Paced
• Free Video

Maximizing Compute Efficiency on Anyscale

Discover advanced observability tools for Ray distributed AI workloads, featuring scalable dashboards, debugging techniques, and the new open-source Ray Export API for better optimization.

• YouTube
• 26 minutes
• Self-Paced
• Free Video

RLlib - Lessons from the V2 Stack and Road Ahead

Discover RLlib v2's architecture redesign for massive distributed reinforcement learning, scaling to 10,000+ environment runners with enhanced reliability and performance insights.

• YouTube
• 32 minutes
• Self-Paced
• Free Video

Ray Observability Upgrades - Debug, Optimize, and Scale Faster

Discover Ray's latest observability upgrades for debugging distributed AI workloads, featuring scalable dashboards, persistent monitoring, and the new open-source Export API.

• YouTube
• 32 minutes
• Self-Paced
• Free Video

Building Fault-Tolerant Massive Ray Clusters on Anyscale

Discover how Ray's distributed runtime handles failures gracefully in 10,000+ node clusters, covering fault tolerance, state management, recovery, and workload-aware scheduling strategies.

• YouTube
• 31 minutes
• Self-Paced
• Free Video

Ray Data for Structured Workloads - Deep Dive

Explore RLlib v2's redesigned architecture for large-scale reinforcement learning, featuring 10,000+ environment runners and enhanced scalability for modern AI systems.

• YouTube
• 32 minutes
• Self-Paced
• Free Video

Ray - Last Year's Progress and the Road Ahead

Discover Ray's latest performance, resiliency, and observability enhancements plus get a preview of the 2026 roadmap for next-generation AI systems.

• YouTube
• 29 minutes
• Self-Paced
• Free Video

How Ray Data Powers Scalable AI Workloads

Discover how Ray Data enables large-scale AI workloads with GPU-accelerated batch inference, distributed training preparation, and high-performance multimodal data processing.

• YouTube
• 30 minutes
• Self-Paced
• Free Video

Motional's Blueprint for High-Performance ML Systems in Autonomous Driving

Discover how Motional rebuilt their ML system using Ray to process terabyte-scale autonomous vehicle data in hours instead of weeks, featuring their novel "1-actor-per-node" pattern.

• YouTube
• 30 minutes
• Self-Paced
• Free Video