How to Solve the Model Serving Component of the MLOps Stack
Publish date: 2022-12-04 Sunday
Last updated: 2022-12-04 Sunday
Last updated: 2022-12-04 Sunday
How to Solve the Model Serving Component of the MLOps Stack
Is a paper about serving models and can be found here.
When people talk about productionizing ML models, they use the term serving rather than simply deployment. So what does this mean? To serve a model is to expose it to the real world and ensure it meets all your production requirements, aka your latency, accuracy, fault-tolerance, and throughput are all at the “business is happy” level.
Shows a basic setup, intermediate and advanced setup.
what do I think about it
Great article about model serving! Shows many of the tradeoffs and gives practical advice.