[Opening Keynote] Anyscale Demo: Machine Learning Application from Dev to Prod
summary
Shows a runnable pipeline that trains an intent classifier with Ray Tune and XGBoost on Ray, runs 100 hyperparameter trials across ~155 CPUs on 10 worker machines, and saves the best model to S3. Builds a Ray Serve + FastAPI inference flow that queries two models in parallel, deploys to Anyscale with automated dependency syncing and a remote debugger, and pushes a CI-driven zero‑downtime update while load testing with stable latency. Highlights infinite laptop experience and zero downtime deployments.
AI-written summary · about
chapters (14)
- 0:00Introduction to demo
- 2:26Building ML support bot
- 5:57Training model pipeline
- 9:51Debugging with remote debugger
- 13:10Monitoring training job
- 15:57Recap of training phase
- 16:59Deploying inference pipeline
- 19:13Testing support bot
- 20:59Fixing bot response logic
- 22:04Setting up CI workflow
- 23:28Zero-downtime deployment
- 24:10Monitoring staging deployment
- 25:34Verifying production fix
- 26:16Summary and next steps
related talks
31mKeynote: Anyscale Product Demo
30mScaling Machine Learning at Discord: Our Journey with Ray and Anyscale
16mKeynote: An Open-Source Stack for AI Compute: Kubernetes + Ray + PyTorch + vLLM
21mStreamlining AI Workflows with Apache Airflow and Ray
05mInterview
28mScaling AI & Machine Learning Workloads With Ray on AWS, Kubernetes, & BERT
29mScaling ML Pipelines for Autonomous Driving with Ray
18mBeyond Inference: Bringing ML into Production