ML Solutions in Production — Batches, NRT, RT: Which to Choose and What Pitfalls?
summary
Shows how to choose between batch, near-real-time (NRT), and real-time (RT) deployment architectures for ML models under high-load conditions. Covers the trade-offs and pitfalls of each approach for productionizing models, with a focus on system-level considerations for HighLoad environments.
AI-written summary · about
chapters (10)
- 0:00Introduction
- 1:29MLE role in ML systems
- 5:26MLE problems and mistakes
- 7:53Batch processing basics
- 11:48When to use batch
- 13:11Near Real Time trade-offs
- 16:20Real Time architecture challenges
- 19:55Metrics and A/B testing
- 22:01Beginner tips
- 23:51Q&A
related talks
31mMLOps In Practice – How To Run Your Machine Learning Models In Production At Enterprise Scale
26mPatterns of ML Models in Production
25mML in Production–What does “Production” even mean?
14mPerformance Optimization in High-Volume Transaction Systems
47mMachine Learning in Production: An Intro to MLOps.
44mML in Production
52mThe ML Lifecycle in Combat Conditions
29mDemystifying Machine Learning in Production: Reasoning about a Large-Scale ML Platform