Knowledge check

Knowledge check

Complete

Which ML platform architecture is best suited for deploying multiple models with differing operational requirements across teams?

A) Monolithic on-premise deploymentB) Serverless architecture onlyC) Microservices deployed via container orchestration (e.g., Kubernetes)D) Notebook-based local training

Why might a hybrid cloud architecture be preferable for a healthcare ML solution?

A) It increases vendor lock-in.B) It ensures all data remains in public cloud storage.C) It allows sensitive data to remain on-premises while leveraging cloud scalability for training.D) It is always more cost-effective than full cloud.

What is the primary reason model parallelism is used in ML workloads?

A) To improve inference speed for small models.B) To split training jobs across cloud regions.C) To handle models too large to fit into the memory of a single device.D) To reduce latency in batch predictions.

A fraud detection system must classify 10,000+ transactions per second with sub-100 ms latency. What operational requirement is most critical?

A) Model sizeB) UptimeC) Cost efficiencyD) Throughput and latency

A recommendation model with 500 million parameters is being retrained nightly. Which resource constraint is most likely to arise?

A) Limited bandwidthB) GPU memory availabilityC) Logging overheadD) User session length

Which auto-scaling strategy is best for adjusting the number of running containers based on real-time CPU usage?

A) Cluster autoscalerB) Spot instance scalingC) Horizontal Pod Autoscaler (HPA)D) Manual scaling via kubectl

Why might a team use serverless functions for ML inference?

A) To deploy large models with heavy GPU needs.B) To batch process millions of records.C) To handle sporadic, low-latency inference tasks without maintaining infrastructure.D) To train large models cost-effectively.

Which of the following is a benefit of using Infrastructure as Code (IaC) tools like Terraform for ML platforms?

A) Reduced training timeB) Improved model accuracyC) Reproducible infrastructure deploymentD) Interactive data labelling

A team schedules non-critical retraining jobs on preemptible instances at night. What is the main benefit?

A) Improved model accuracyB) Reduced retraining timeC) Lower compute cost for fault-tolerant tasksD) Guaranteed high availability

What's the primary advantage of defining resource requests and limits in Kubernetes for ML workloads?

A) Faster model convergenceB) Preventing low-priority workloads from starving critical jobsC) Automatically converting models to microservicesD) Encrypting model weights

Evaluate your understanding of this unit by completing the Knowledge Check.

Knowledge check

Action item: Knowledge check

In progress