Accelerate Model Training with an Easy to Use High-Performance...
summary
A reproducible stack is demonstrated that lets data scientists declare resources in Python and push distributed jobs without DevOps: CodeFlare SDK + operator manages MCAD and Instascale, creates Ray clusters via cluster.up, and submits batch jobs and interactive Ray sessions. Demo runs a PyTorch MNIST batch and a Hugging Face DistilBERT fine‑tune on separate Ray clusters, shows logs, cluster details and a completed long job; no numeric speedups are given, but orchestration and autoscaling behavior work end to end.
AI-written summary · about
chapters (13)
- 0:00Introduction
- 1:04Distributed ML challenges
- 6:06Project CodeFlare
- 10:03Ray in distributed computing
- 18:28Open Data Hub platform
- 20:37CodeFlare SDK workloads
- 25:16Demo prerequisites
- 27:46Jupyter notebook setup
- 30:39Ray cluster deployment
- 33:52Batch training job
- 35:37Interactive Ray workloads
- 39:51Monitoring training jobs
- 40:12Q&A
related talks
21mHow Ray and Anyscale Make it Easy to do Massive-scale ML on Aerial Imagery
33mOvercoming Distributed ML Challenges with Ray Train
20mThe Quick Journey to Using Ray: How We Implement Ray and Anyscale to Speed up our ML Processes
31mNext-Generation Data Science Workflows Using Ray
22mLarge Scale Data Loading and Data Preprocessing with Ray
30mScaling Machine Learning at Discord: Our Journey with Ray and Anyscale
01h 30mDistributed Python with Ray-Hands on with the Ray 2.0 APIs for scaling Python Workloads
23mAn Open Source Post-Training Stack: Kubernetes + Ray + PyTorch + vLLM