Incontro DevOps Italia 2024
When a Cloud Engineer has to do a review the code of its colleague Data Scientist for production environment, it is always important to understand where it is best to put the focus. Often, the best approach is to promote the resources awareness to be used and to find a framework to split the work together.
https://2024.incontrodevops.it/talks_speakers/
33. DevOps & MLOps - IDI 2024
Data Scientist & Cloud Engineer
1
How to survive with colleagues with different skills, without hindering the table football
game during the lunch break.
2 Versioning of configuration,
training and inference data
statistics, models, ..
3
Define tools and services:
libraries, images, storages, ..
4 Define guidelines for who does
what: review, support,
deploying the model, ..
Promoting resources
awareness to be used
35. DevOps & MLOps - IDI 2024
Step Functions vs SageMaker Pipeline
36. DevOps & MLOps - IDI 2024
AWS services
Which AWS services for which role, when we want to automate a solution from local to
production: depends on the life cycle of the model but above all by the actors involved.
Data Scientists (DS) Cloud Engineer (CE) DS AWS Services CE AWS Services
many one AWS Step Functions AWS CloudFormation
one many Amazon SageMaker
Pipeline
AWS CloudFormation
& Step Functions
37. DevOps & MLOps - IDI 2024
AWS services
Which AWS services for which model, when we have a small model, we can also use
something like AWS Lambda, but when we need GPU, we must to use an Endpoint.
AWS Service Provisioning type Resources Cost per hour
Endpoint Provisioned 2 CPU + 4GB RAM $0.056
Endpoint Serverless 4096MB $0.288
AWS Lambda Serverless 4096MB $0.00024012 +GBsec