Local AI Secure Server Knowledge Check
Table of Contents
Answer all ten questions without the key. A score of eight or higher prepares you for the capstone. Correct every missed security item before continuing.
Quiz Steps
- Create
local-ai-knowledge-check.md. - Record one answer and one sentence of reasoning per question.
- Use the key to score your work.
- Link each missed item to a repaired lesson record.
Questions
Which planning formula best represents local inference memory?
- A. Model file size only
- B. Weights plus context state, runtime buffers, concurrency margin, and operating margin
- C. Parameter count divided by CPU cores
- D. Storage space plus network speed
What does a model file digest establish?
- A. Accuracy on your task
- B. Safety from prompt injection
- C. Identity against a trusted reference digest
- D. License permission
Which measurement pair should stay separate?
- A. Prompt processing and token generation rates
- B. CPU name and operating system
- C. Model publisher and license
- D. Port and protocol
Why should the first API test use loopback?
- A. Loopback increases parameter count
- B. Loopback limits network reach during setup
- C. Loopback adds user authentication
- D. Loopback verifies model quality
Which result proves GPU offload occurred?
- A. A configuration request alone
- B. Runtime diagnostics and device memory activity
- C. A model card claim
- D. Faster typing in the terminal
Which server path has the strongest boundary?
- A. Public native runtime port
- B. LAN native runtime port with no identity
- C. Private route through an authenticated TLS gateway to a loopback runtime
- D. Port forwarding from a home router
Why limit context and concurrency together?
- A. Both affect memory allocation and service capacity
- B. Both change the model license
- C. Both rotate credentials
- D. Both select a GPU driver
What should a server access log record by default?
- A. Every full prompt and response
- B. Identity, route, model, sizes, decision, duration, and status
- C. User passwords
- D. Model weights
Which update package supports rollback?
- A. The newest model tag only
- B. A screenshot of the dashboard
- C. Prior runtime, model digest, configuration, and passing test record
- D. A larger context setting
What should happen when the native runtime listens on
0.0.0.0unexpectedly?
- A. Continue testing from the internet
- B. Stop the service, restore loopback binding, and recheck listeners
- C. Add another model
- D. Delete benchmark results
Answer Key
| # | Answer | Reason |
|---|---|---|
| 1 | B | Weights form one part of runtime memory use. |
| 2 | C | A digest supports artifact identity against a known reference. |
| 3 | A | Prompt evaluation and generation measure different phases. |
| 4 | B | Loopback keeps early service access on one host. |
| 5 | B | Observed diagnostics and device activity establish placement. |
| 6 | C | The gateway supplies identity and policy while the runtime stays off the network. |
| 7 | A | Parallel requests multiply active context and raise memory demand. |
| 8 | B | Metadata supports operations while reducing raw-content collection. |
| 9 | C | Rollback needs the prior artifacts, settings, and evidence. |
| 10 | B | Unexpected broad binding requires containment and verification. |
Expected Result
A passing score is 8 out of 10. Questions 4, 6, 8, 9, and 10 form the security gate. Answer all five correctly before the capstone.
Troubleshooting
- Memory answers stay unclear: Revisit Lesson 1 and your fit worksheet.
- Performance terms blend together: Revisit Lesson 3 and label both phases in your table.
- Network controls seem redundant: Map the native runtime and gateway as separate processes with separate duties.
- Rollback lacks a model artifact: Add the prior digest and storage location to your update record.
Verify Completion
Record your score, security-gate score, and corrections. Pass when your total reaches eight, all five security items are correct, and each missed question has a repaired artifact.
Continue with the Local AI Server Capstone .
