Pedro Tabacof- Deploying Real-Time Machine Learning Models Using Serverless AWS | PyData London 2023

Deploying machine learning models using serverless AWS: a pragmatic approach to latency and scalability, with a focus on Docker and batch models.

Key takeaways

Docker is a good alternative to Flask for deploying machine learning models
Lambda is not suitable for applications with high latency requirements
Use boring technology, such as Docker, and avoid fancy solutions
Batch models are simpler to maintain and scale than real-time models
You should always go with the Docker path for deploying machine learning models
Batch models are more cost-effective than real-time models
Use CloudPico to store and deploy models, not Pico
Serverless is not always suitable for machine learning applications
Docker provides more control over the latency and scalability of your application
You should test whether you need real-time models or if batch models suffice

More talks