Integrating Feast Online Feature Store with KFServing
summary
Demo shows an end-to-end integration: Feast v0.12.1 runs with an S3 offline store and registry, a Redis online store, and a materialize cron job that moves features to Redis; features are defined in a GitHub repo and populated from a Jupyter notebook using pip install feast and the Feast SDK, then retrieved via curl. A KFServing transformer container calls the Feast feature server, then forwards augmented inputs to the predictor; the deployed service returns a concrete ranking (driver 3 chosen), so the pipeline works end-to-end.
AI-written summary · about
chapters (17)
- 0:00Intro
- 0:35Agenda
- 1:04Why Feature Store
- 2:07Feature Stores Solutions
- 3:58Feast History
- 9:05Feast Infrastructure
- 10:50Adopters and Contributors to Feast
- 11:36Feast Deployment in Kubernetes Cluster
- 18:16KFServing Introduction
- 20:40KFServing Stack
- 21:57Supported Frameworks and Storage
- 22:52KFServing Inference Service
- 24:33Transformers
- 25:54KFServing with Feast Transformer
- 28:37Demo
- 33:26Summary
- 36:05Open Source Summit North America Talks
related talks
29mIntegrating High Performance Feature Stores with KServe Model Serving
32mBuilding a Cloud Native Feature Store with Feast on Kubeflow
31mFeast as a Cloud Native Data System: Bridging Batch Processing and Online Serving...
27mFeast: feature store for Machine Learning
34mSolving the ML Pain You Forgot to Mention
39mIntroducing KFServing: Serverless Model Serving on Kubernetes
46mScaling ML Pipelines with Feast, Ray and Kubeflow
23mFeathr: An Enterprise-Grade, High Performance Feature Store