Which Serverless Prediction Platforms on Google Cloud Platform?
summary
Compares Google Cloud options for serving TensorFlow models: recommends Cloud Run for online prediction because it provides container portability, scale-to-zero and millisecond billing; warns GPU VMs are beta, lack SLA and don’t scale-to-zero so incur baseline costs. Describes BigQuery ML for batch scoring via SQL with billing by bytes scanned, and explains training limits (use Kubernetes jobs or AI Platform for long-running training since Cloud Run has short request limits); provides cost tradeoffs and guidance when each setup is appropriate.
AI-written summary · about
chapters (9)
- 0:00Introduction
- 1:26Key concepts
- 6:26Serverless platforms on GCP
- 9:58Cloud Run for containers
- 15:18GKE container manager
- 19:56Serverless prediction comparison
- 23:26Batch with BigQuery ML & Vertex AI
- 30:51Batch pros and cons
- 32:13Resources & FAQ
related talks
23mBuilding Scalable, Observable MLOps Systems on Google Cloud
25mAn Introduction To Google Cloud Platform
32mMachine Learning with Google Cloud Platform
38mServerless tools for machine learning, big data and app backends
31mProductionizing ML with ML Ops and Cloud AI
45mServerless Machine Learning: a beginner's guide
43mMachine Learning at Scale
29mML Ops on Google Cloud Platform