Easy MLOps Engineer questions

34 questions · 5 points each for a correct first answer.

1. 1.What a model registry is formodel-registryeasy
2. 2.Why version datareproducibility, data-versioningeasy
3. 3.Why put a model in a containerdockereasy
4. 4.What experiment tracking recordsexperiment-trackingeasy
5. 5.What data drift ismonitoring, drifteasy
6. 6.When batch inference fitsmodel-servingeasy
7. 7.Same code, different resultsreproducibilityeasy
8. 8.Canary deploymentdeploymenteasy
9. 9.Data used to tune hyperparametersmodel-evaluationeasy
10. 10.Recall from prediction countsclassification-metricseasy
11. 11.What concept drift ismodel-monitoring, drifteasy
12. 12.Blue-green deploymentdeployment-strategieseasy
13. 13.A/B test versus canarydeployment-strategies, experimentationeasy
14. 14.What CI should check for a modelci-cdeasy
15. 15.Image versus containerdockereasy
16. 16.Traffic before the model has loadedkubernetes, model-servingeasy
17. 17.What a pipeline DAG isorchestrationeasy
18. 18.Reading a cron scheduleschedulingeasy
19. 19.Where API keys belongsecurity, secrets-managementeasy
20. 20.What p99 latency meanslatency, monitoringeasy
21. 21.Throughput of four replicascapacity-planningeasy
22. 22.Memory for 7B parameters in fp16gpu-memory, llm-servingeasy
23. 23.Why quantise a modelquantisation, inference-optimisationeasy
24. 24.Why export a model to ONNXmodel-serving, inference-optimisationeasy
25. 25.Logging a metric every epochexperiment-trackingeasy
26. 26.Making a training run repeatablereproducibilityeasy
27. 27.What training-serving skew istraining-serving-skew, feature-engineeringeasy
28. 28.What a model card is forgovernanceeasy
29. 29.Scaling a stateless model APIscalingeasy
30. 30.Bad data reached trainingdata-validation, pipelineseasy
31. 31.Protocol for internal inference callsmodel-serving, apiseasy
32. 32.Pod stuck in CrashLoopBackOffkubernetes, debuggingeasy
33. 33.What a vector database doesvector-databases, rageasy
34. 34.Cost of a GPU training jobcost-managementeasy