Final Exam: MLOps and LLMOps Complete

This comprehensive final exam covers all 14 modules. You have 50 questions to test your mastery of MLOps and LLMOps concepts.

Time Recommended: 90 minutes
Passing Score: 80% (40/50)

---

Part 1: MLOps Foundations (Questions 1-25)

Question 1

What is the primary goal of MLOps?

1. To train models faster
2. To deploy and maintain ML models in production reliably
3. To reduce model size
4. To collect more data

---

Question 2

Which tool is used for data version control in ML projects?

1. Git only
2. DVC (Data Version Control)
3. Docker
4. Kubernetes

---

Question 3

What does experiment tracking typically include?

1. Only model weights
2. Only training data
3. Parameters, metrics, artifacts, and code versions
4. Only deployment configs

---

Question 4

What is a feature store?

1. A retail store for ML products
2. A centralized repository for ML features
3. A model registry
4. A database for raw data

---

Question 5

What is data drift?

1. Data moving between databases
2. Changes in the statistical properties of input data over time
3. Data being deleted
4. Data format changes

---

Question 6

Which is NOT a common model serving framework?

1. TorchServe
2. TensorFlow Serving
3. Jupyter Notebooks
4. Triton Inference Server

---

Question 7

What is Kubeflow primarily used for?

1. Web development
2. ML workflows on Kubernetes
3. Database management
4. Network monitoring

---

Question 8

What triggers model retraining?

1. Only manual requests
2. Only scheduled intervals
3. Performance degradation, drift, schedule, or new data
4. Every code commit

---

Question 9

What is P99 latency?

1. Average response time
2. Response time that 99% of requests are faster than
3. Maximum response time
4. Minimum response time

---

Question 10

What is the purpose of a model card?

1. Physical identification for models
2. Documentation of model purpose, performance, and limitations
3. Credit card for cloud billing
4. Access control

---

Questions 11-25

(Additional MLOps questions covering versioning, pipelines, CI/CD, monitoring, distributed training, containerization, and best practices)

---

Part 2: LLMOps Specialization (Questions 26-50)

Question 26

What makes LLMOps different from traditional MLOps?

1. LLMOps is simpler
2. Token-based costs, prompt management, and text output evaluation
3. LLMOps doesn't need monitoring
4. LLMOps only works on premises

---

Question 27

What is hallucination in LLMs?

1. Model crashes
2. Generating plausible but incorrect information
3. Slow response times
4. Memory errors

---

Question 28

What is vLLM's PagedAttention?

1. A documentation format
2. Memory management technique for efficient KV cache
3. A pagination API
4. A new attention mechanism

---

Question 29

What is quantization used for?

1. Increasing model accuracy
2. Reducing model memory footprint and increasing speed
3. Adding more parameters
4. Data preprocessing

---

Question 30

What is red teaming in LLM safety?

1. A deployment strategy
2. Testing to find vulnerabilities and harmful outputs
3. Color coding outputs
4. Team organization

---

Question 31

What does RAGAS evaluate?

1. Training speed
2. RAG pipeline quality (faithfulness, relevancy)
3. Hardware performance
4. Cost efficiency

---

Question 32

Why version prompts?

1. To make them longer
2. For reproducibility, testing, and rollback capability
3. To reduce tokens
4. To speed up inference

---

Question 33

What is the purpose of semantic caching for LLMs?

1. Store model weights
2. Cache similar queries to avoid redundant API calls
3. Improve model quality
4. Encrypt responses

---

Question 34

What is a model router in LLMOps?

1. Network hardware
2. System that directs requests to appropriate models
3. Training optimizer
4. Data pipeline

---

Question 35

Why is PII detection important in LLM applications?

1. To improve accuracy
2. To protect sensitive data and ensure compliance
3. To reduce latency
4. To save costs

---

Questions 36-50

(Additional LLMOps questions covering deployment, evaluation, safety, prompt optimization, enterprise features, and best practices)

---

🎓 Exam Complete!

Congratulations on completing the MLOps and LLMOps Complete course final exam!

Score Interpretation:

• 45-50 correct (90-100%): Excellent! You've mastered MLOps and LLMOps.
• 40-44 correct (80-89%): Great job! Strong understanding of the material.
• 35-39 correct (70-79%): Good, but review weak areas before certification.
• Below 35 (< 70%): Recommend reviewing modules and retaking the exam.

---

📜 Certificate

Upon achieving a passing score of 80% or higher, you are eligible for the MLOps and LLMOps Complete certification.

Contact your instructor to receive your certificate.

---

Thank you for completing this course!

We hope you've gained valuable knowledge for building production-grade ML systems. Good luck in your MLOps and LLMOps journey! 🚀