Continuous deployment for ML: The new software development lifecycle
summary
Machine learning teams spend 75% of their time on infrastructure not modeling, because dozens of languages, framework versions, hardware drivers and per-framework deployment tools create brittle pipelines. Training and inference are distinct workloads with different scale and latency needs, and models are often pipelined together so composability matters. Demo shows a fully automated flow using GitHub Actions + Algorithmia to push a pickled model from repo to live API, proving continuous deployment for ML can work. The recommended approach treats models as modular services and automates versioning and governance.
AI-written summary · about
chapters (10)
- 0:00Intro
- 1:02Survey results
- 3:18ML composability
- 4:48ML training vs ML production
- 6:16Build test deploy monitor
- 7:26Compliance and governance
- 8:19Continuous deployment for ML
- 10:11git push
- 10:42GitHub action
- 11:56Demo
related talks
18mContinuous deployment of Machine Learning models
24mFrom Model to Production: Automating the ML Lifecycle with CI/CD
30mChallenges of machine learning: The new ML lifecycle
56mDevOps for Machine Learning
39mContinuous Delivery for Machine Learning — The Future of MLOps
57mDevOps for Machine Learning
42mCI/CD for Machine Learning
33mContinuous Delivery Principles for Machine Learning