Deploying machine learning models to production involves a distinct vocabulary — from running A/B tests to monitoring for drift and rolling back underperforming models. This exercise covers the collocations used by ML engineers, data scientists, and platform teams during deployment planning and review.
0 / 5 completed
1 / 5
The MLOps team plans to ___ the model to production using a blue-green deployment strategy.
Deploy the model is the standard ML operations collocation — models are 'deployed' to serving infrastructure. 'Release' is used for software versions; 'push' is informal (push to production); 'ship' is also used but more common in product discussions than MLOps.
2 / 5
We need to ___ model performance continuously after deployment to detect drift.
Monitor model performance is the precise MLOps collocation — monitoring implies automated, ongoing measurement with alerting. 'Track' is also correct but slightly less formal; 'watch' is informal; 'observe' is used in research contexts, not production monitoring.
3 / 5
Before promoting a new model version, we should ___ A/B tests to compare performance.
Run A/B tests is the natural collocation in ML deployment discussions — experiments and tests are 'run' in engineering contexts. 'Conduct' is more formal (conduct a study); 'perform' works but is less idiomatic; 'execute' sounds overly procedural.
4 / 5
The team decided to ___ the previous model version immediately after detecting accuracy regression.
Roll back the model version is the standard MLOps collocation — 'rollback' is the canonical term for reverting to a previous deployment. 'Revert' is used in version control; 'restore' implies recovering something lost; 'reinstate' is formal but uncommon in ML deployment language.
5 / 5
We'll ___ the model endpoint behind a feature flag before the full release.
Gate the model endpoint is the precise ML deployment collocation — feature gates control access to new capabilities. 'Shadow' is a specific deployment technique where traffic is mirrored without serving responses; 'hide' and 'wrap' are informal and imprecise in this context.