Docker Containers for ML Models Quiz

Answers and explanations appear as you take the quiz.

1. Why copy a dependency lockfile before frequently changing application source?
   • Docker can reuse the dependency-install layer when only source code changes
   • It guarantees packages never need security updates
   • It makes model data part of the base image
   • It runs dependency installation on the GPU
2. Which problem does a multi-stage Docker build address?
   • Run every stage at the same time in production
   • Use build tools in one stage and copy needed runtime artifacts to another
   • Avoid declaring any base image
   • Store secrets in an intermediate layer
3. Why mount a large private dataset instead of copying it into the image?
   • It removes the need for data access controls
   • It ensures the model is deterministic
   • It avoids embedding data in an image that is distributed
   • It converts the dataset into a model layer automatically
4. What does digest pinning provide for a base image?
   • A smaller image size regardless of its layers
   • Automatic security updates without rebuilding
   • A record of the Python package versions only
   • A more precise identity for the referenced image content
5. Where should runtime secrets generally be supplied?
   • In source files copied into the image
   • Through a secure runtime mechanism rather than embedded image layers
   • In a public base-image tag
   • Inside the model's class labels
6. What does a running container add beyond an image?
   • The physical GPU driver inside the container
   • A new compiled base image automatically
   • Runtime configuration such as mounts, environment and network
   • A guarantee of bitwise-deterministic training
7. Why run an application as a non-root user when practical?
   • It makes the image digest shorter
   • It prevents every dependency vulnerability
   • It guarantees the model meets accuracy targets
   • It limits privileges if the process is compromised
8. What does a container image alone not guarantee?
   • Identical numerical results across different hardware
   • A packaged filesystem and metadata
   • A specified application command
   • Layered build inputs